Optimization of power regulation strategy for distributed photovoltaic users based on improved genetic algorithm

With the large-scale grid connection of new energy power plants, they play an increasingly important role in the new power system, especially distributed photovoltaic is gradually increasing its proportion in the power system. Photovoltaic power generation is a renewable energy technology that utilizes solar cells to convert light energy into electricity, which can significantly change the traditional energy structure, reduce dependence on fossil fuels, and reduce resource consumption [1-3]. Photovoltaic power generation does not produce greenhouse gases and harmful substances such as carbon dioxide, reducing the risk of air pollution and environmental damage. By promoting photovoltaic power generation, the demand for fossil fuels can be reduced, the extraction and consumption of limited resources such as oil and coal can be reduced, and the transformation of energy structure and sustainable development can be realized [4-6]. Distributed photovoltaic (PV) decentralizes the deployment of power generation units near the users, reducing the energy loss in power transmission. Compared with traditional centralized power generation, distributed PV has higher flexibility and dispatchability, which can better cope with fluctuations in power demand and power supply during peak hours, and guarantee the stable operation of the power grid [7-9]. However, the large number of distributed PV access makes the distribution network transform from traditional passive network to active network, which will inevitably have an impact on the traditional grid, such as insufficient carrying capacity for in situ consumption, equipment voltage rise over the limit, and system peaking difficulties [10-12]. In order to cope with the impact of distributed PV access on the power grid, we can try to optimize the distributed PV user power regulation strategy by using improved genetic algorithms, in order to effectively reduce the voltage overrun caused by the PV grid connection, improve the shortcomings of centralized control of the traditional grid, and improve the optimization effect of the system reactive power, which is of important research significance and application promotion value [13-15].

This paper builds up a distributed PV user power regulation model, proposes the scheduling of loads on the basis of load classification and the scheduling based on the battery storage system BESS, and realizes the construction of the power regulation optimization model of distributed PV through the transfer of the usage time of the appliances and the optimal charging-discharging method of the BESS, respectively. Facing the solving of the model in this paper, the NSGA-II algorithm is improved, and the combined crossover operator and combined variation operator are introduced to address the shortcomings of insufficient global searching ability of the crossover variation operator and the lack of diversity of the population. A dynamic crowding strategy is introduced to address the fact that the classical crowding calculation method is prone to unevenly distributing the retained individuals. The performance of the algorithm before and after the improvement is tested by three standard test functions to analyze the uniformity of the Pareto front and the degree of approximation to the real Pareto front, which proves that the improved algorithm has better search accuracy and individual uniformity. Application practices are carried out from various aspects such as PV forecasting, PV scheduling strategy optimization, and scheduling optimization economic benefits to explore the effectiveness and practicality of the distributed PV user power regulation model in this paper.

2

Overview

Based on the current energy development trend, the continuous promotion of distributed renewable energy development has become a necessary path for energy conservation, emission reduction, and low-carbon transformation. Literature [16] proposed a grid-connected photovoltaic/wind hybrid distributed generation system intelligent control strategy based on the meta-heuristic firefly algorithm, and experimentally verified that the proposed strategy has good robustness and performance, which can solve some technical problems caused by the integration of distributed power generation into the electric power system, and maintains the stability of the power generation system. Literature [17] designed an independent solar distributed energy system based on genetic algorithms, and simulation experiments confirmed the superior performance of the designed system, which can maximize the benefits of the system. Literature [18] designed a PV power generation system performance optimization control strategy based on proportional-integral (PI) controller and Whale Optimization Algorithm (WOA) and the simulation experiments carried out in PSCAD/EMTDC environment, the results verified the validity of the proposed control strategy, which solves the high penetration problem of the grid and enables the PV power generation system to operate in an optimal state. Literature [19] proposed a hierarchical design scheme for distributed batteries in solar shared building communities using genetic algorithms and nonlinear planning methods, i.e., a distributed energy storage system based on solar-PV-electricity sharing, and verified the feasibility of the designed scheme through actual case studies, which can significantly reduce the battery capacity and power loss in the sharing process, and has a significant impact on optimizing the positive energy area in terms of energy efficiency, energy production and flexibility in three major functions. Literature [20] proposed a two-layer voltage control strategy based on distribution network partitioning in order to realize network partitioning and voltage coordinated control of distributed PV units with high penetration of distribution network, and verified the scientific and practicality of the proposed strategy through simulation experiments. Literature [21] proposed a distribution network cluster optimization and control method considering the heterogeneous characteristics of distributed PV resources, and simulation verification is carried out as an example of the improved IEEE33 node system distribution network, and the results confirm the validity and feasibility of the proposed method, which plays a certain effect in the reduction of the functional loss, the optimization of the power reversal and the management of the voltage overruns, and maintains the economic security of the distribution network operation.

3

Distributed photovoltaic user power regulation model

With the energy crisis and environmental pollution becoming more and more serious, traditional energy sources no longer satisfy people’s growing energy demand, and abundant and clean renewable energy sources are the future development direction, and distributed renewable power generation technology has gained rapid progress. However, after the introduction of distributed renewable power sources, the complexity of microgrids and the imbalance between the supply and demand side increase dramatically, which brings difficulties to the economic optimization of various types of microgrids for scheduling. In this paper, the optimization of power regulation strategy for distributed photovoltaic (PV) users is taken as the research objective, and a power regulation model for distributed PV users is constructed, and the corresponding solution method of the model is proposed based on the improved genetic algorithm NSGA-II.

3.1

Modeling

3.1.1

Multiple scheduling goals

1)

Minimization of electricity consumption cost

For most households, the first consideration in scheduling electricity consumption is the economy. Thus, it is meaningful to establish an electricity consumption model with the goal of minimizing electricity consumption costs.

The total cost of electricity consumption is calculated as shown in equations (1)-(3): (1) $P_{L} (t) = P_{u} (t) + \sum_{a = 1}^{n} P_{n} (t)$ (2) $P_{t o t a l} (t) = P_{L} (t) - P_{P V} (t)$ (3) $C (t) = {\begin{array}{l} P_{t o t a l} (t) \cdot P r i c e_{R T P} (t) & i f P_{t o t a l} (t) < 0 \\ P_{t o t a l} (t) \cdot P r i c e_{s a l e} (t) & i f P_{t o t a l} (t) \geq 0 \end{array}$

Among them, P_L(t) represents the power of all household loads at the moment of t, P_tomi(t) represents the power required by the grid at the moment of t, P_u(t) is the power of all uncontrollable loads at the moment of t, $\sum_{n = 1}^{n} P_{n} (t)$ is the total power of all controllable loads at the moment of t, and P_PV(t) is the power of photovoltaic power generation at the moment of t. C(t) is the cost of power consumption at the moment of t, Price_RTP(t) is the price of power purchased from the grid by the user at the moment of t, and Price_Sale(t) is the price of power sold from the grid by the user at the moment of t. When P_otal(t) < 0, it means that the PV power generation is not enough for the user to use, and needs to purchase power from the grid; P_total(t) ≥ 0, it means that the PV power generation is surplus, and the surplus power is sold to the grid.

The objective function to minimize the cost of electricity consumption in a day is shown in Eq. (4): (4) $\min_{\begin{matrix} t = 1 \end{matrix}}^{24} C (t)$

2)

PV maximization of consumption objective

Compared with the above electricity cost minimization model, although both are using PV as much as possible, the two are not equivalent, this is due to the use of real-time tariffs, and the objective function of maximizing PV power consumption is shown in Equation (5): (5) $\min | \sum_{t = 1}^{24} P_{t o t a l} (t) |$

3.1.2

Optimized scheduling model based on appliances

1)

Optimize the scheduling strategy

Although the introduced PV power generation can reduce the amount of electricity purchased by users from the grid to a certain extent, PV power generation is unstable and susceptible to weather, so there will be a mismatch between the high and low peak periods of PV power generation and the high and low peak periods of household load.

For this reason the above problem can be solved by shifting the usage time of controllable appliances to the peak period of PV power generation. Specifically, it is divided into the following cases:

(1) When P_pV(t) – P_L(t) > 0, that is, the PV power generation is greater than the user needs, priority will be given to the PV power supply load use, if there is still excess power, the excess power will be sold to the grid, so as to gain revenue.

(2) When P_pV(t) – P_L(t) < 0, that is, the PV power generation is not enough for the use of electrical appliances, in the allowable time frame, the moment of the controllable load transfer to the price of electricity in the valley time, in order to reduce the cost of electricity.

2)

Model objective function and constraints

For each device a, define the energy consumption scheduling vector $x_{a} = [x_{a}^{1}, \dots, x_{a}^{T}]$ , where T is the number of time slots in a day, i.e., the number of 24-hour divisions in a day, $x_{a}^{'}$ is the scheduling energy consumption of the appliance a at the t th moment, and E_a is the total energy consumption required for the operation of the appliance a. However, considering the user’s living habits, each appliance has an approximate scheduling interval, and the user sets the permissible operation interval of appliance a to [α_a,β_a], which means that appliance a can only be scheduled within this interval. In order to provide the required energy to the appliance within the interval [α_a,β_a], there is: (6) $\sum_{t = α_{a}}^{β_{a}} x_{a}^{t} = E_{a}$

In addition to the constraints in the above equation, for any t < α_a and t > β_a, the appliances a do not run and x_a = 0, and hence β_a – α_a, are greater than or equal to the number of hours required for the appliances to run normally.

For each of the appliances, there is also a maximum and minimum power constraint, so there is: (7) $γ_{a}^{\min} \leq x_{a}^{t} \leq γ_{a}^{\max}$

The real-time tariff (RTP) is provided by the grid company and the user sets the parameters $α_{a}, β_{a}, E_{a}, γ_{a}^{\min}, γ_{a}^{\max}$ for each appliance a. Given the feasible energy scheduling set and RTP, the key is how to schedule appliances based on time-varying tariffs so that the cost of electricity consumption is minimized and the total electricity payment cost to the user is: (8) $\sum_{t = 1}^{T} P r i c e_{R T P} (t) \cdot (\sum_{a \in A} x_{a}^{t})$

Then the objective function of the optimal scheduling model based on electrical appliances is: (9) $\min \sum_{t = 1}^{T} P r i c e_{R T P} (t) \cdot (\sum_{a \in A} x_{a}^{t})$

3.1.3

Optimized scheduling model based on energy storage system

1)

Optimized scheduling strategy

In this paper, a scheduling strategy based on energy storage system (BESS) is proposed, and an optimal scheduling model based on this strategy is also developed, which manages the household energy by determining the best charging and discharging scheme as well as the optimal capacity of the energy storage system, which combines the storage system with the PV generation system [22]. The optimal energy capacity and charging/discharging method are determined for the energy storage system to minimize the electricity bill. Then an improved genetic algorithm is applied to solve the problem and some simulations and analysis are implemented [23].

Based on the above scheduling ideas, the following specific scenarios are analyzed respectively.

(1)

High electricity price period

If the time period is in the high electricity price period, it will correspond to the following three scenarios:

If the PV power generation is greater than the load demand, i.e., when P_PV(t)–P_L(t) > 0, both the PV surplus power and the battery surplus power are sold to the grid for revenue;

If the PV power generation is equal to the load demand, i.e., P_PV(t)–P_L(t) = 0, then the PV power generation can realize self-sufficiency, and the power can be sold to the grid through the energy storage system to obtain revenue;

If the PV power generation is less than the load demand, i.e., P_PV(t)–P_L(t) < 0, there are two cases. When the excess power in the energy storage system is enough to meet the load use, priority is given to using the energy storage system to discharge; when the power in the energy storage system is not enough for the load use, the remaining power demand must be purchased from the grid.

(2)

Low tariff period

If the time period is in the low tariff period, it will correspond to the following three scenarios:

If the PV power generation is greater than the load demand, i.e., P_PV(t)–P_L(t) > 0, charging of the storage system is prioritized, which is divided into two cases. When the surplus power is sufficient for the storage system to be fully charged, the surplus PV power is sold to the grid for revenue; when the surplus power is insufficient for the storage system to be fully charged, additional power is purchased from the grid to charge the storage system;

If the PV power generation is equal to the load demand, i.e., P_PV(t)–P_L(t) = 0, then it can realize self-sufficiency, i.e., there is no need to purchase or sell electricity from the grid, and there is no need for the storage system to charge or discharge.

If the PV power generation is less than the load demand, i.e., P_PV(t)–P_L(t) < 0, there are two cases. When the excess power of the energy storage system is enough to satisfy the load use, priority is given to using the energy storage system for discharging; when the remaining power of the energy storage system is not enough for the load use, power must be purchased from the grid.

2)

Model objective and constraints

The optimization objective is to minimize the cost of electricity consumption by considering the charging and discharging cycles of the PV power generation and energy storage system, because the focus of this chapter is on the optimization of distributed energy scheduling, so the calculations do not take into account the purchase of photovoltaic and storage batteries, installation costs, such as Equation (10) for the objective function: (10) $\min \sum_{t = 1}^{T} ((P_{L} (t) - P_{P V} (t) - P_{B} (t)) \cdot \Pr i c e_{R T P} (t)) + P_{P V} (t) \cdot C_{P V} + P_{B} (t) \cdot C_{B})$

Where, P_L(t) is the total load power at the moment t, P_pV(t) is the PV power at the moment t, P_B(t) is the charging (discharging) power of the battery at the moment t, Price_RTP(t) is the real-time tariff at the moment t, C_PV is the operating cost of the PV panels, and C_B is the operating cost of the battery.

The charging and discharging power as well as the capacity of the energy storage system need to be less than or equal to the rated value, i.e: (11) $P_{B} (t) \leq P_{B}^{r} E_{B} (t) \leq E_{B}^{r} P_{L} (t) + P_{B} (t) - P_{P V} (t) = P_{g r i d} (t)$

Where $P_{B}^{r}$ is the rated power of the energy storage system, $E_{B}^{r}$ is the rated capacity of the energy storage system, which is related to the specifications of the selected energy storage equipment. P_grid(t) is the amount of electricity that the user interacts with the grid at the moment t, when P_grid(t) > 0 indicates the purchase of electricity from the grid; when P_grid(t) < 0 indicates the sale of electricity to the grid.

3.2

Model solving

3.2.1

Multi-objective optimization problems

A multi-objective optimization problem consists of n decision variable, k objective functions to be optimized, m inequality constraints, and p equality constraints: (12) $\begin{matrix} * \min / * \max y = f (x) = (f_{1} (x), f_{2} (x), \dots, f_{k} (x)), k \geq 2 \\ s . t . g_{i} (x) \leq 0, i = 1, 2, \dots, m \\ h_{j} (x) = 0, j = 1, 2, \dots, p \end{matrix}$

Here x = (x₁,x₂,…,x_n) is the n -dimensional decision variable of x ∈ Rⁿ, f(x) is the objective function, g_i(x) is the m inequality constraint function, h_j(x) is the p equality constraint function, and g_i(x) and h_j(x) form the feasible solution region.

3.2.2

Constraints processing

Intelligent optimization algorithms are stochastic optimization algorithms proposed for unconstrained optimization problems, and the existence of constraints must be considered in solving practical optimization problems; constraints lead to the existence of infeasible domains in the search space of the decision variables, and multi-objective optimization with constraints needs to consider both the constraints and the objective function at the same time.

In this paper, the hierarchical penalty function method is used to deal with some of the constraint problems, the expression is given in the following equation: (13) $Φ (x) = f (x) + \sum_{i = 1}^{N} {R_{k, i} \times \max {[0, g_{i} (x)]}^{2}}$

Where, R_k,i is the penalty factor of the segmentation, the degree of violation of the constraint is divided into several levels, different levels use different penalty factors, the higher the degree of violation of the constraint, the greater the value of the penalty factor.

3.2.3

Data normalization

Data normalization is one of the important steps in the data preprocessing stage, which can make the objective function converge faster. Linear function normalization is to do a linear transformation of the initial data to normalize the range of each value to [0, 1], the normalization formula is as follows: (14) $x_{0} = \frac{x - x_{\min}}{x_{\max} - x_{\min}}$

Back-normalize the data according to the following equation: (15) $x = (x_{\max} - x_{\min}) x^{'} + x_{\min}$

Where, x₀ is the normalized value, x is the original value, x_min is the minimum value of the sample data and x_max is the maximum value of the sample data.

3.2.4

NSGA-II Algorithm

Non-dominated Sorting Genetic Algorithm (NSGA-II) is a multi-objective evolutionary algorithm that is commonly used to solve multi-objective optimization problems [24]. NSGA-II is an improvement of Non-dominated Sorting Genetic Algorithm (NSGA). NSGA algorithm has limitations, such as the lack of elitism, the need to define the shared parameters, and the computational complexity is large. NSGA-II algorithm introduces the elitism strategy, the fast non dominated sorting method and the mechanism of congestion operator, the computational complexity is greatly reduced and the computational efficiency is improved.

The NSGA-II algorithm focuses on fast non-dominated sorting, elite strategy, congestion distance, and selection operation, which are briefly described below.

1)

Fast non-dominated sorting

Assuming that there are m objective functions f_i(x), i for any integer of 1,2,3,…,m, j for any integer of 1,2,3,…,m, i is not equal to j, here the minimum value of the dominant objective, for x₁ and x₂ there is f_i(x₁) < f_i(x₂) constant, that is, the individual x₁ is said to dominate the individual x₂; for x₁ and x₂ there is f_i(x₁) ≤ f_i(x₂) constant and at least one objective to satisfy the function f_i(x₁) < f_i(x₂), then it is said that the individual x₁ weakly dominate the individual x₂; if there is an objective function f_i(x₁) ≤ f_i(x₂) and at least one objective function f_i(x₁) > f_i(x₂) exists, then x₁ and x₂ are said to not dominate each other.

2)

Crowding distance operator

Individual crowding distance is used to calculate the density around an individual, which is calculated as the accumulation of the normalized difference between two neighboring bodies of an individual in different target directions. When comparing two individuals of the same Pareto class with different crowding levels, the individual with the larger crowding level is considered to be more “independent”. The equation below provides the formula for calculating crowding: (16) $I (i) = \sum_{j = 1}^{m} \frac{f_{j}^{i + 1} - f_{j}^{i - 1}}{f_{j}^{\max} - f_{j}^{\min}}$

where I(i) is the congestion distance at i, $f_{j}^{i + 1}$ is the value of individual i + 1 on objective function j, $f_{j}^{i - 1}$ is the value of individual i – 1 on objective function j, and $f_{j}^{\max}$ and $f_{j}^{\min}$ are the maximum and minimum values of objective function j.

3)

Elite strategy

Parent population P_t generates offspring population Q_t through championship selection, crossover and mutation, merges the new population Q_θ generated by t generations with parent population P_t to generate R_t populations with population size 2N, and generates Pareto levels 1, 2, 3… When the number of individuals in a certain level exceeds N, select the individuals with high crowding distance to be placed until the t + 1 th generation population is filled.

3.2.5

Improvement of NSGA-II algorithm

The classical NSGA-II algorithm suffers from the disadvantages of insufficient search accuracy and particle uniformity, in order to solve these shortcomings, this paper introduces a Y-NSGA-II algorithm to optimize the crossover and mutation operators in the genetic operation and the strategy of calculating the distance of crowding degree in the nondominated sorting.

1)

Combined crossover operator

The NSGA-II algorithm is encoded using the simulated binary crossover operator (SBX), which selects x₁ and x₂ parent individuals and generates offspring individuals c₁ and c₂ according to the following equation: (17) ${\begin{array}{l} c_{1, i} = \frac{(1 + β) x_{1, i}}{2} + \frac{(1 - β) x_{2, i}}{2} \\ c_{2, i} = \frac{(1 - β) x_{1, i}}{2} + \frac{(1 + β) x_{2, i}}{2} \end{array}$

where x_1,i and x_2,i are the i rd generation parent individuals x₁ and x₂, respectively, and c_1,i and c_2,i are the i th generation offspring individuals c₁ and c₂, respectively.

β is the uniform distribution factor calculated as follows: (18) $β = {\begin{matrix} {(2 γ)}^{\frac{1}{μ + 1}}, γ \leq 0.5 \\ {[\frac{1}{2 (1 - γ)}]}^{\frac{1}{μ + 1}}, γ > 0.5 \end{matrix}$

where γ is a uniform random number of (0, 1) and μ is the cross-distribution index.

The SBX operator has a small search space and poor global search capability. To address the shortcomings of the SBX operator, this paper introduces a combinatorial crossover operator NDX to improve the algorithm’s global search ability and convergence speed, and introduces the positive-taiwanese distribution into the simulated binary crossover operator, where the uniform distribution factor β is replaced by 1.481*N, and N is the positive-taiwanese distribution of the positive number of random variables, which is given in the following formulas: (19) ${\begin{array}{l} c_{1, i} = \frac{x_{1, i} + x_{2, i}}{2} + 1.481 N \frac{x_{1, i} - x_{2, i}}{2} \\ c_{2, i} = \frac{x_{1, i} + x_{2, i}}{2} - 1.481 N \frac{x_{1, i} - x_{2, i}}{2} \end{array}$

2)

Combined Mutation Operator

The variation operator in NSGA-II algorithm often adopts polynomial variation, in order to obtain better performance in the process of variation optimization, this paper uses a new type of combined variation operator, which is able to select the most effective variation in different iteration cycles. The specific method is as follows: (20) $u_{k + l} = {\begin{array}{l} u_{k} + U (- σ_{1}, σ_{2}) & t \leq T / 4 \\ u_{k} + C (0, σ_{2}) & T / 4 < t \leq 3 T / 4 \\ u_{k} + N (0, σ_{3}) & 3 T / 4 < t \leq T \end{array}$

3)

Dynamic congestion degree strategy

In order to solve the defects in practical applications, this paper introduces a dynamic crowding degree strategy. After recording the individual with the largest crowding degree, the individual is eliminated and the crowding degree of the remaining individuals in the layer is re-calculated and sorted, then the individual with the largest crowding degree is recorded and eliminated in the new sorting, the crowding degree of the remaining individuals is updated, and the above process of recording, elimination, and sorting is repeated until the number of recorded individuals meets the requirements.

3.2.6

Algorithm testing

When improving a multi-objective algorithm, it is necessary to test the performance of the algorithm by evaluating and comparing the algorithm with a test function. In this paper, the classical test functions ZDT1, ZDT2 and ZDT3 are used to test the effectiveness of the algorithm improvement.

1)

ZDT1 function: (21) $\begin{matrix} f_{1} (x) = x_{1} \\ f_{2} (x, g) = g (x) \cdot (1 - \sqrt{\frac{f_{1}}{g (x)}}) \\ g (x) = 1 + \frac{9}{n - 1} \cdot \sum_{i = 2}^{n} x_{i} \end{matrix}$

2)

ZDT2 function: (22) $\begin{matrix} f_{1} (x) = x_{1} \\ f_{2} (x, g) = g (x) \cdot (1 - {(\frac{f_{1}}{g (x)})}^{2}) \\ g (x) = 1 + \frac{9}{n - 1} \cdot \sum_{i = 2}^{n} x_{i} \end{matrix}$

3)

ZDT3 function: (23) $\begin{matrix} f_{1} (x) = x_{1} \\ f_{2} (x, g) = g (x) \cdot (1 - \sqrt{\frac{f_{1}}{g (x)}} - \frac{f_{1}}{g (x)} \cdot \sin (10 π f_{1})) \\ g (x) = 1 + \frac{9}{n - 1} \cdot \sum_{i = 2}^{n} x_{i} \end{matrix}$

The ZDT1-3 function expression is shown in the above equation, the search space of all three test functions is 0 ≤ x_i ≤ 1, n is 30, i takes values ranging from 1, 2, …, 30 integers.

Schott proposed the evaluation metric of spacing (SP) to test whether the individuals in the Pareto front of the algorithm are evenly distributed, and the smaller the value of SP, the more evenly distributed: (24) $S P = \sqrt{\frac{1}{n - 1} \sum_{i = 1}^{n} {(d_{i} - \bar{d})}^{2}}$ (25) $d_{i} = \min (| f_{1}^{j} (x) - f_{1}^{i} (x) | + | f_{2}^{j} (x) - f_{2}^{i} (x) |, i, j = 1, \dots\dots, n$

n is the number of solutions, and d_i is the Euclidean distance between the i rd and nearest individual in the Pareto front of the test function, where $\bar{d}$ is the mean of d_i: (26) $G D = \frac{\sqrt{\sum_{i = 1}^{n} d_{i}^{2}}}{n}$

GD (Generation Distance) is a measure of the proximity of the Pareto front of the test algorithm to the true Pareto front; the larger the value, the farther it deviates from the true Pareto front and the worse the convergence.

4

Distributed photovoltaic user power regulation optimization practice

In this chapter, the effectiveness of the distributed PV customer power regulation model constructed in this paper will be discussed and analyzed in depth from various aspects, such as PV forecasting and the application of scheduling strategies.

4.1

PV Forecasting Application Practices

In this section, the PV forecasting function of the distributed PV customer power regulation model constructed in this paper is tested for application. In this section, the PV prediction application practice is applied to the PV power data from a station area from November 5 to November 30, 2023, the 26 days of data as a historical data selection database, selected a typical sunny day (November 21), cloudy day (November 27) for prediction, to 20min as the prediction time step, respectively, at 0:00, 7:00, 12:00, 15:00 0 output the prediction results. The comparison results between the predicted and actual values are specifically shown in Fig. 1. Figures a and b show the PV prediction during sunny and cloudy days, respectively; the curves in the figures are the actual output, while the bars are the predicted values. At 0:00 every day, the predicted PV power based on the historical data has a large deviation due to the different weather on that day from the previous day, especially when the change of cloudy and sunny is obvious, while at 7:00 and 12:00 based on the historical data and the sunny index on that day, the predicted value tends to the actual value, and the prediction of tracking for the instantaneous weather change is realized.

4.2

PV Scheduling Strategy Application Practice

Distributed photovoltaic grid control to switch between the fixed strategy and the economic dispatch strategy, as far as the former is concerned, the most important thing is to make the photovoltaic power generation to get the maximum utilization rate at the same time, do not make the grid due to random fluctuations in the impact; the latter not only can make the AC load to be satisfied, but also make the increase of distributed photovoltaic power grid system to earn as much profit as possible. Random switching between the two in different situations can maximize the benefits. In this section, the utility of the distributed PV customer power regulation model constructed in this paper will be examined when a fixed strategy or economic dispatch strategy is used.

1)

Fixed strategy

The simulation results of the distributed PV customer power regulation model constructed in this paper under the fixed strategy are shown in Fig. 2. When the PV generation power is greater than the AC load, the energy storage absorbs the excess power; and when the PV generation power is less than the AC load, the power shortage is compensated by the energy storage, and the whole simulation result meets the goal of the fixed strategy, which indicates that the distributed PV user power regulation model constructed in this paper is able to carry out the corresponding power regulation and control in accordance with the fixed strategy.

2)

Economic scheduling strategy

The results of power regulation simulation for the distributed PV user model constructed in this paper under the economic dispatch strategy are specifically shown in Fig. 3. The difference with the simulation results under the fixed strategy is that the energy storage in the valley time tariff and the first usual tariff time period, the power is always -0.45kW in 43 hours, has been charging, the power is negative. While the grid supplies power to AC loads during the valley time tariff period, the usual tariff period compensates for the power deficit between the PV and AC loads. The grid supplemental power decreases to 0kW after 43 hours, while the storage output power shows an overall increase, with the highest power output reaching 13.1kW. The reason for keeping the storage energy always charging is to store the power during low tariff periods and to utilize the power output during peak tariff periods for the purpose of The reason why the energy storage is always charged is to store power during the low tariff period and utilize it to supply power to the AC load during the peak tariff period, especially when the PV generation power is smaller than the AC load, thus reducing the need to buy power from the grid, which correspondingly saves the cost of the distributed PV grid system operation and enhances the economic efficiency of the microgrid, which is consistent with the goal of adopting the economic dispatch strategy, indicating that the distributed PV user power regulation model constructed in this paper can be carried out according to the economic dispatch strategy accordingly. This is consistent with the goal of using economic dispatch strategy, which indicates that the distributed PV customer power regulation model constructed in this paper can also carry out the corresponding system regulation according to the economic dispatch strategy.

In summary, regardless of whether a fixed strategy or an economic dispatch strategy is used, the distributed PV user power regulation model built in this paper can be optimized and regulated accordingly to operate in accordance with the objectives to be achieved by the strategy.

4.3

Power dispatch optimization economic benefit simulation experiment

In order to further verify the effectiveness of the distributed photovoltaic user power regulation model in this paper, multi-day power consumption data of a user in real northern China are collected, and the obtained user load power consumption data are used to conduct simulation experiments of power dispatch optimization. The one-day power purchase cost of the actual user is compared with the simulated power consumption curve of the whole day after the transferable load scheduling using the optimization algorithm. The simulated power consumption curve after optimal scheduling of electricity consumption is specifically shown in Fig. 4. Figures (a) and (b) show the simulated power consumption curves before and after optimal scheduling of electricity consumption. It can be seen that the power consumption curve is mainly shifted from the time of high price of electricity to the time of low price, which reduces the user’s cost of purchasing electricity, proving the effectiveness of the power consumption optimization scheduling algorithm proposed in this paper in meeting the user’s improved economic efficiency.

At the same time, this paper compares the economic benefits of “self-generation and self-consumption of residual power on-grid” and “direct full feed-in” two PV grid-connected modes. The comparison of the economic benefits of the two grid-connected modes before and after the increase in PV capacity is shown in Table 1. Before the increase of PV capacity, there is a maximum output power limit of 1kW for PV, and after the increase of PV capacity, there is a maximum output power limit of 2kW for PV. Before the increase of PV capacity, the self-generated self-consumption of residual power on-grid mode its purchase and sale of electricity costs are lower, direct full Internet access mode purchase and sale of electricity costs are relatively high, but the overall benefit is higher than the self-generated self-consumption of residual power on-grid mode. In reality, PV household projects are often easier to mobilize users’ installation enthusiasm after generating revenue. At the same time, a higher overall return means that the user’s power consumption is less dependent on power generated from fossil fuel sources. Obviously, if the PV is present with a maximum output power of 1kW, the user’s comprehensive income can be negative. In this paper, after increasing the user’s simulated PV capacity to 2kW, the income from electricity sales significantly increases, and the comprehensive income is positive, which is more conducive to mobilizing the user’s motivation to install. And direct full Internet access to the final comprehensive income is higher.

Table 1.

Economic benefit

Photovoltaic capacity	Parallel pattern	Purchase cost	Sales revenue	Integrated income
1kW	Spontaneous use of spare electricity	9.52	8.71	-0.81
1kW	Direct online	12.72	12.38	-0.34
2kW	Spontaneous use of spare electricity	9.58	14.86	5.28
2kW	Direct online	12.85	20.05	7.2

5

Conclusion

This paper constructs a distributed PV user power regulation model with two distributed energy scheduling strategies of electrical appliances and energy storage system BSEE as the core content to provide optimization direction for distributed PV user power regulation strategy. The effectiveness and practicality of the model are discussed in terms of PV forecasting, PV scheduling strategies, and economic benefits of power scheduling optimization.

In the prediction of PV power data from November 5 to November 30, 2023 in a station area, the predicted values of this paper’s model converge to the actual values on both typical sunny days (November 21) and cloudy days (November 27), showing the predictive tracking of the immediate weather changes. In the face of two different PV scheduling strategies, the fixed strategy and the economic scheduling strategy, the model in this paper can be optimized according to the strategies, and run according to the goals to be achieved by the strategies. The multi-day power consumption data of a user in northern China is selected as a sample to carry out simulation experiments to analyze the economic benefits of power dispatch optimization, and under the consideration of distributed PV, the power consumption curve is shifted from the time of high electricity price to the time of low electricity price, and the user’s cost of purchasing electricity is reduced. Comparing the economic benefits of “self-generation and self-consumption of residual power on the Internet” and “direct full Internet access”, when the PV capacity is increased from 1kW to 2kW, the income from the sale of electricity of both PV grid-connected modes increases significantly, and the comprehensive income is positive, which is conducive to the mobilization of users’ motivation to install. Mobilizing users’ installation enthusiasm.

Sprache:: Englisch

Zeitrahmen der Veröffentlichung:: 1 Hefte pro Jahr
Fachgebiete der Zeitschrift:: Biologie, Biologie, andere, Mathematik, Angewandte Mathematik, Mathematik, Allgemeines, Physik, Physik, andere

Zeitschrift RSS Feed

Optimization of power regulation strategy for distributed photovoltaic users based on improved genetic algorithm

Yangrui Zhang

Kai Liu

Peng Ren

Chao Yan

Online veröffentlicht: 19. März 2025

Eingereicht: 06. Okt. 2024

Akzeptiert: 03. Feb. 2025

DOI: https://doi.org/10.2478/amns-2025-0548

SchlüsselwörterDistributed photovoltaic, Genetic algorithm, NSGA-II, Dynamic congestion strategy, Power regulation

© 2025 Yangrui Zhang et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Schlüsselwörter
Distributed photovoltaic, Genetic algorithm, NSGA-II, Dynamic congestion strategy, Power regulation