Design of commercial environment space based on digital media technology

The world has entered the Information Era along with the fast social and technological advancement. Also known as the "Digitized Era," it means that the production and transfer of digital information are the major operating patterns of our society. Under the impetus of this tendency, electronic data has taken the place of the necessary material elements, the mode of production has been converted from human work to computer, the culture has realized the transition from the tangible to the informational, and the global economy has become the transition from the physical to the digital.

Based on the macro-view of DMM, many researchers have probed into The Times's impact and social impact. Marshall McLuhan's comprehension of the spread of media, "Media information," and "Mass Extension" shows that the cultural and living techniques of mankind are very important. Because of its precise analysis and forecast, it is a theory foundation for the enlightenment of digital media[1]. In the era of digital existence, "Information" is replacing fundamental communication with social media, demonstrating that the debate on ICT and ICT has a significant impact on people's study, job, and leisure lives [2]. In his study, Russell Norman studied the influence of two ways of communicating in an industry age on the influence of two ways of communicating with each other [3]. Steffen's Digital Media: The Link Between Man and Technique analyses nine examples of NTD from a phenomenology point of view and concludes that ICT and its creation are not neutral but intended to transform the world and alter our ability in different ways [4]. Zhang Wenjun's "Introduction to Digital New Media" is a collection of achievements in the Digital Technology and Media field[5]. Based on the Developing Situation of Information Visualization Design under New Media Background, Anjun illustrates the impact of the Digital Media Technique on Information Visualization and the Features of New Media Visual Vision.[6]. Wang Jing's dissertation "Philosophical Reflections on New Media Technology" studies digital media technology from three aspects: ontology, epistemology, and axiology, respectively, and puts forward her views on how to correctly understand the technology and deal with the relationship between people and technology [7].

For the study of digital media technology and commercial environment space design means and technology, this paper first analyzes the word embedding algorithm and differential algorithm and optimizes the differential algorithm; secondly, according to the algorithm, it proposes the environment space layout design system and tests the system to make references for related research.

2

Differential algorithm

2.1

Interactive differential algorithm

2.1.1

Differential evolution algorithm

As a population-based stochastic algorithm, the differential evolution algorithm usually starts its optimization process by generating the initial population, composed of NP individuals; each individual is randomly generated in the search space and is a multidimensional vector, also known as an individual vector or target vector[8]. In the n-dimensional search space, the i th individual in the initial population can be represented by the formula (1).

(1)

X_{i} (t) = [X_{i, 1} (t), X_{i, 2} (t), X_{i, 3} (t), \dots, X_{i, n} (t)] i = 1, 2, 3, \dots, N P

Where NP represents the population size, t represents the number of population evolutions, and t=0 during the initialization stage.

Different optimization problems may have different ranges of variables; the range of each variable needs to be preset before initializing the population, and each dimensional value of all vectors is within the set interval. Each individual of the initial population was generated using formula (2): (2) $X_{i, j} (t) = L_{\min} + rand (0, 1) (L_{\max} - L_{\min}), i = 1, 2, 3, \dots, N P, j = 1, 2, \dots, n$ After initialization, the three operators of variation, crossing, and selection are performed repeatedly to find the optimal solution. In the t-iteration, the individual vector X_i(t) generates the variant vector H_i(t) by variant manipulation. The variation vector mainly comprises the basis vector and the difference vector. Different mutation operators have different emphases, and the common mutation operators are listed below[9-11].

(3)

DE / rand / 1 : H_{i} (t) = X_{n} (t) + F (X_{r 2} (t) - X_{n} (t))

(4)

DE / rand / 2 : H_{i} (t) = X_{n} (t) + F (X_{n} (t) - X_{n} (t)) + F (X_{r} (t) - X_{r} (t))

(5)

DE / best / 1 : H (t) = X_{est} (t) + F (X_{n 2} (t) - X_{n n} (t))

(6)

DE / current-to-rand / 1 : H (t) = X (t) + F (X_{n} (t) - X (t)) + F (X_{2} (t) - X_{3} (t))

(7)

DE / current-to-best / 1 : H (t) = X (t) + F (X_{bear} (t) - X (t)) + F (X_{11} (t) - X_{r_{2}} (t))

Where the indices r₁, r₂, r₃, r₄, and r₅, are mutually exclusive integers randomly selected from the set {1,2,..., NP} and different from the subscript I of the current individual, the vector Xbest stands for the optimal individuals in the present group. T indicates the number of generations that have evolved. Parameter F is the scaling coefficient of a differential vector, which influences the capability of searching. F is a big, diverse group, with an emphasis on the whole world, while F is a small one, with an emphasis on the local search and the rate of convergence. The common initial set is either 0.5 or 0.6.

After the population passes through the mutation operation, the cross operator selects the dimension between the target vector and its offspring mutation vector to produce the trial vector according to the probability. The DE algorithm is mainly binomial crossover, generating a random number independently for each dimension of the binomial crossover vector. Each independent dimension of the selected target vector or variation vector is stored in the trial vector, and the binomial crossover can be described as: (8) $U_{i, j} (t) = {\begin{array}{l} H_{i, j} (t), & if rand (0, 1) < C R or j = j rand \\ X_{i, j} (t), & else \end{array}$ Where U_i,j(t) represents the j-dimension value of Ith trial vector in generation t, CR is the crossover probability, the value range is in the interval (0,1), and j rand is a random integer within [1, n].

2.1.2

Basic principles of the interactive differential evolution algorithm

Interactive differential evolution algorithm, a technique that optimizes and retrieves candidate solutions using users' perceptual information, integrates attributes such as human knowledge, experience, and preferences into the optimization process, thus searching for solutions using systems based on human evaluation. Its essence is to improve the differential evolution algorithm, take advantage of the global optimization ability of the DE algorithm, and introduce artificial subjective intervention. The most important difference between the two is that the IDE algorithm does not need to calculate the adaptive value of the evolving individual but changes the adaptive value function to the user subjectivity for pairing comparison; the user chooses the individual more in line with their favorite, that is, completes the comparison of individual adaptive value. It is not difficult to understand that pairing comparison is a two-choice and simpler than the traditional assignment mechanism. In addition, it can be seen from formula (8) that the selection operation of the DE algorithm itself implies the evaluation method of pairing comparison, so it is suitable as the evaluation method of human-computer interaction in the IDE algorithm.

The basic idea of the interactive difference algorithm is that the population is randomly initialized, the chromosome phenotype is presented through the interactive interface or other forms, and the user evaluates the individual according to their emotional preferences. If the user's requirements are not met, the population is crossed and varied, the test individuals are generated, and then the specific phenotypes are presented through the interaction interface or other methods; the user matches the target individual and the test individual, selects the individuals that are more consistent with their emotional needs and retains to the next generation population. After many iterations, find satisfactory solutions in the search space. Algorithm 2 is the pseudo-code for the interactive differential evolution algorithm. Interactive differential evolution algorithm inherits the characteristics of the DE algorithm; in addition, it has its characteristics, mainly including the following: a small number of populations and iterations; IDE as a man-machine collaboration method, introduced the interactive system, generally interactive interface; with user subjectivity, user evaluation differences, the best individual results are not unique.

2.2

Multiobjective optimization of the differential algorithm

2.2.1

Multiobjective optimization approach

A typical multiobjective problem consists of a set of decision vectors, a set of objective vectors, a set of inequality constraint vectors, and a set of equality constraint vectors, mathematically described as: (9) $\min (F (x)) = {[f_{1} (x), f_{2} (x), \dots, f_{k} (x)]}^{T}$ (10) $s .t . g_{1} (x) \leq 0, i = 1, 2, \dots, p$ (11) $h_{j} (x) = 0, j = 1, 2, \dots, q$ In equation (11), X is the decision vector; x_{j min} and x_{j max} are the lower and upper limits of the j ; f_ix are the target vector; g_i and h_k h are the inequality constraint vector and the equation constraint vector, respectively.

2.2.2

Comparative analysis of the multiobjective difference algorithm and other improved algorithms

To illustrate the performance of the multiobjective difference algorithm in this paper, it was tested in a MATLAB environment with five benchmark test functions and compared with DE, PSO, DKPSO, and MODE algorithms.

1

Sphere

(12)

f_{1} (x) = \sum_{i = 1}^{n} x_{i}^{2}_{x i} \in [- 100, 100]

obvious,f₁(x^*) = 0The global best advantage of f₁(x) is x^* = (0,0, ⋯ 0), and the optimal value is f₁(x^*) = 0

2

Rosenbrock function

(13)

f_{2} (x) = \sum_{i = 1}^{n} (100 \times {(x_{i + 1} - x_{i}^{2})}^{2} + {(x_{i} - 1)}^{2}) x_{i} \in [- 30, 30]

Thef₂(x) has a global minimum point x^* = (1,1, ⋯ 1)and an optimal value f₂(x^*) = 0

3

Rastrigin

(14)

f_{3} (x) = \sum_{i = 1}^{n} (x_{i}^{2} - 10 \cos (2 π x_{i}) + 10) x_{i} \in [- 5.12, 5.12]

The f₃(x) has a global minimum point x^* = (0,0, ⋯ 0) and the optimal valuef₃(x^*) = 0

4

Griewank

(15)

f_{4} (x) = \frac{1}{4000} \sum_{i = 1}^{n} x_{i}^{2} - \prod_{i = 1}^{n} \cos (| \frac{x_{i}}{\sqrt{i}} |) + 1, x_{i} \in [- 600, 600]

The f₄(x) consists of a large number of local extreme points, with a global minimum points^* = (0,0, ⋯ 0), f₄(x^*) = 0

5

Ackley

(16)

f_{5} (x) = - 20 \exp (- 0.2 \sqrt{\frac{1}{n} \sum_{i = 1}^{n} x_{i}^{2}}) - \exp (\frac{1}{n} \sum_{i = 1}^{n} \cos (2 π x_{i}) + 20 + e), x_{i} \in [- 32, 32]

The f₅(x) has a global minimum point x^* = (0,0, ⋯ 0) and the optimal value f 5 (x) =0f₅(x^*) = 0

6

Schwefel

(17)

f_{6} (x) = - \sum_{i = 1}^{n} (x_{i} \sin \sqrt{| x_{i} |}), x_{i} \in [- 500, 500]

Thef₅(x) has a global minimum point x^* = 420.9687, the optimal valuef₅(x^*) = –418.9829n Figures 1 and 2 show the evolutionary plots of the six test functions, as shown below:

As can be seen from Figure 1 and Figure 2, Compared with the others, the proposed method has a higher convergence rate and higher precision, which can prevent premature aging and enhance its performance. The MODE method achieves the optimum location rapidly and does not fluctuate near the optimum location. In the multi-peak Rastrigin test, CPSODE is more powerful in the Rastrigin function, while the MODE method is more likely to get the best solution. So, the MODE method performs better and more efficiently than the others.

2.2.3

Improvement of multiobjective differential evolution algorithms

Due to the frequent problem of unbalanced exploration and development, this paper first uses the mirror point screening method to optimize the initial population. Chaos variation is then introduced during the development process. The final data comparison and simulation results prove that the present algorithm is effective in improving the algorithm's performance.

1

Adaptive and chaotic variation

Chaos variation is a widespread nonlinear phenomenon found in nature and society, which has attracted wide attention due to its ergodicity and other characteristics. For multiobjective problems, the effectiveness of chaotic operators in improving population diversity and jumping out of the local optima has been demonstrated. Common chaotic maps are Logistic, Sinusoidal, and Tent. Logistic mapping and Tent mapping are applied to the adaptive chaotic mutation strategy in this chapter. These two mappings are shown below:

2

Logistic

(18)

r^{n + 1} = {ar}^{n} (1 - r^{n})

3

Tent

(19)

r^{n + 1} = {\begin{matrix} 2 r^{n} & 0 \leq r^{n} \leq 0.5 \\ 2 (1 - r^{n}) & 0.5 < r^{n} \leq 1 \end{matrix}

rⁿ⁺¹ is the random number between 0-1, and n represents the number of iterations.

4

Reference points and mirror points

In multiobjective optimization, it is often necessary to assist in creating a new person, and Das and Dennis have put forward a common approach for creating a reference point. The number of reference points W depends upon a destination spatial size M and a border dividing number p. This paper adopts a two-level reference point generating algorithm, which keeps the variety efficient and decreases the computation cost. The method of the number of reference points W is indicated as follows: (20) $W = (\begin{matrix} M + p_{1} - 1 \\ p_{1} \end{matrix}) + (\begin{matrix} M + p_{2} - 1 \\ p_{2} \end{matrix})$ Where p1 and p2 are the number of boundary and inner layers, respectively.

The reflection point is the symmetry of the reference point concerning the original, and the line that passes through the reference point RP to the specified mirror MP is defined as NBI. A mirror point can be represented as follows: (21) $m p_{i} = r p_{i} - 1 fori = 1, \dots, M$ Figure 3 shows the concept of the reference point mirror point and its corresponding direction with the generation method of the two layers of reference points and mirror points, where the blue solid point represents the reference point, the black solid point represents the mirror point, and the mirror point concerning the origin.

5

Mirror-point generation strategy

Based on the proposed model, the pose evolution algorithm has achieved outstanding results in multiobjective optimization. However, the majority of experiments have been concerned with the question of the regular Pareto border. However, their performance declines when confronted with a degenerate, a disjunction, an inversion, a strong convex, or a strong concave Pareto front. One example is shown in Fig. 3-3-a), in which a group of uniform distribution of reference points corresponds to a group of equally distributed optimum solutions. But when the Peña front has an irregular shape, the Peña front does not coincide with the PBI, so the PBI does not necessarily cross the Pareo front completely. As illustrated in Fig. 3 (1), the optimum solution is not uniformly distributed across Pele's front, and a few places do not intersect with PBI. Fig. 3 (2) indicates that no matter how geometric, a group of NBI directions can obtain uniform distribution Pareto optimum solutions. The Acute Angular Measurement measures the PBI Direction of Diversity, whereas the Euclidean Distance Measurement is an NBI Direction Measurement for Convergence and Homogeneity. Then, PBI and NBI are used in the selection policy so that the diversity and homogeneity of the population can be improved simultaneously.

2.2.4

Simulation analysis of a multiobjective differential evolution improvement algorithm

To make the superiority of the algorithm more convincing, this section also plots the Pareto front after the AMODE-MPS and contrast algorithm experiences 30000 on partial functions. Figure 4-6 shows the Pareto front obtained by the AMODE-MPS and the contrast algorithm after running on several typical test functions.

As can be seen from Figure 4-6, MOEA-PC has not fully converged on ZDT 3 and DTLZ7, but not distributed on UF 7, MSEA on DTLZ7, PREA on ZDT 3 and DTLZ7, and some fronts did not converge. Compared with other algorithms, AMODE-MPS is fully converged and well distributed. This demonstrates the great potential of AMODE-MPS in handling complex multiple targets.

Two strategies were used to improve the population diversity. One is mirror point screening, a strategy that screens well-distributed parents to produce better offspring. The other is an adaptive chaotic mutation strategy that adds adaptive chaotic operators to population exploration to increase diversity. Two control algorithms were added to verify the effectiveness of these two strategies. Among the two control algorithms, AMODE-MPS 1 only adds the mirror point screening strategy and DE/hand / 1 / bin in the populations in development and exploration. AMODE-MPS 2 adds only adaptive chaotic mutations without a mirror point screening strategy.

Figure 7 and Figure 8 show the IGD trends of AMODE-MPS 1, AMODE-MPS 2, and AMODE-MPS on some functions. The horizontal axis is the 100 uniform sampling points during the operation, and the vertical axis is the corresponding log (IGD) value. It can be seen that AMODE-MPS 1 and AMODE-MPS 2 did not converge worse than AMODE-MPS throughout evolution and even better than AMODE-MPS for many functions, indicating that the addition strategy does not slow down the rate of evolution. And that AMODE-MPS converges faster on most test functions, especially for the UF series functions, indicating the great potential of AMODE-MPS in processing complex frontiers.

3

Interior layout design based on the combination of word embedding algorithm and differential algorithm

3.1

Interior layout design based on an interactive differential algorithm

3.1.1

Design of code

In the process of building plan design, the overall frame outline of the house type is usually determined first. Then, the internal space area is divided, and finally, the drawings are generated. According to the family's different development stages and personnel composition, the space needs often become the stage characteristics [12-13].

3.1.2

Algorithm process

The whole indoor layout system can be divided into four parts: (1) the input of parameters and rules; (2) the interaction interface of manual scoring and target presentation; (3) the algorithm subject; (4) the generated population of each iteration and the output of the final target solution. Specifically, the parameter cross probability Pc and variant probability Pm users can input themselves on the interaction interface.

The algorithm is divided into the following steps:

Step 1: The user enters the parameters through the interactive interface and clicks the "Start" button;

Step 2: Generate the initial population according to the rules and test the legitimacy explicit representation of the interaction interface;

Step 3: The user scores the population of individuals through the interactive interface;

Step 4: Through the roulette method, select individuals to join the new population in turn;

Step 5: Apply the recombination operator to the new population;

Step 6: Small probability of individual genetic variation;

Step 7: k-means clustering of the new population and explicit representation on the interaction interface;

Step 8: The user selects the final solution or returns to Step 3.

3.1.3

Design system testing

This paper performs an experimental study under simulation to prove the effectiveness of DGA and GA in interior layout design. All characteristic values have been normalized to the range of 0 to 1.0. Parameters are preconditioned so that every iteration has NP = 12, Junior is 0.3, Scale Factor F is 0.7, and Cross Probability CR is 0.5. The empirical value was chosen by the set of variables F and CR [14-16].

In this paper, we assume that the user is not able to make manual adjustments but can get a general idea of the layout drawing produced by this method. The user's final satisfaction is expected, as illustrated in Figure 3.6. The distance from the optimal solution to the phylogenetic tree is considered the adaptation value, not the user assessment.

Compared with the conventional GA method, IDE adopts a pair comparison method to choose the social behavior of the next generation of people. The interactive model enables the user to judge more quickly and decreases the number of mistakes. Moreover, the conventional DGA method is adopted to compute the individual score of a group, and the IDE adopts human choice to substitute for it. Conventional self-adaptation algorithms can not be used anymore. To reduce the error to a minimum, select time indicates personal quality ranking and a base for selecting P and Best. In the case of the group, the more satisfied the customer is, the higher the personal score, and the easier it is to be chosen by the user.

Assuming a final satisfying solution, as illustrated in Fig. 3.6, Euclid's distance from an individual to a solution is taken as an adaptive value. A good variant operator can greatly affect the performance of this algorithm. Finally, we compare 4 classic variation operators and find out which is most suitable for the inner arrangement model. Then, the experimental subjects were separated into four groups, which were treated differently. Five tests were carried out for the sake of the accuracy of the test data, and the mean was chosen as the test data. The experimental results are shown in Figure 9.

DE/Currency to Pay/1 and DE/Currency to Bet/1 from the comparative results shown in Figure 9

They are among the top performers because they perform better in convergence than DE/RAND/1 and DE/BEST/1.

To decrease the mistake of selecting a time sequence, DE/Currency to Bet/1 is preferable.

Taking a hypothetical target vector as a benchmark, Fig. 10 illustrates the variation of the mean convergence profile of each of these algorithms in ten simulations. Experiments indicate no obvious difference between these two methods at the beginning stage of the iteration at t = 20. The IGA convergence is faster in the latter part of the iteration, and IDE and IDE-BO are more efficient.

Compared with the others, IDE-BO is much faster and more accurate when t = 60. To confirm the accuracy of the proposed method, we use t = 150 times in the simulated test. However, in actual HCI, t 50 times is more suitable to prevent user fatigue and influence the end outcome due to user fatigue.

The iterative times needed by IDE-BO and the other two conventional algorithms are presented in Table 1. As we can see, IDE-BO usually takes about t = 125, and IDE needs about 133 iterations. Although IGA is much more rapid in convergence than IDE-BO, it is found in Fig. 3.8 that this method is subject to local optimization. For statistical comparison of the performance of the IDE-BO versus that of the rivals, the Wilcoxon signature rank test was conducted at 0.05. Table 3.3 compares IDE-BO and two other algorithms. The IDE-BO p values are significantly lower than 0.05, and there is a small gap between IDE and IGA, whereas IGA is more convergent. Generally speaking, IDEBO's performance differs significantly from those of the others.

Table 1.

statistical comparisons of the Wilcoxon test

IDE-BOVS	IDE	IGA
P Values	0.008	0.000

3.2

Spatial layout design based on word embedding algorithm and multiobjective optimization algorithm

3.2.1

Word embedding algorithm

Word embedding algorithm refers to the general term of feature learning techniques and language models in natural language processing. The text corpus is transformed into a mathematical representation through a mathematical model. In the early word embedding, it is to code the words of the whole corpus, and each word is represented as a long vector. After removing all the words in the corpus, the number of words obtained is the dimension of the vector. In contrast, the vector of a word is expressed as 1 in the corresponding one dimension, and the value in other dimensions is expressed as. Assuming that the corpus only "I like to play football and play games" in this sentence, the corpus word segmentation is heavy after only seven words, so for each word in this sentence, the vector dimension is 7, and the value in the vector is only 0 and 1, namely the words for the dimension value is 1, another dimension value is 0.

After controlling the dimension of the word vector in two dimensions, the word vector representing "animal" will be closer in the coordinates, the two words indicating the adjective "good" are closer, and other words that are not close to each other are also far apart in the coordinate system.

Therefore, the distributed representation of words can better quantify the semantic and grammar approximate relationship of words in the corpus, which is more meaningful than the single-hot encoding form, while the generalized word embedding method refers to the distributed representation of words:

3.2.2

Design system testing

In this article, we plan to build a good VIP dining room environment site layout and turn it into a mirror. Then, we can modify the code beginning point to expand the code. Ten thousand will be divided into a large number, and 30,000 will be added to the database. The ratio between the training set, the verification group, and the experimental group will be 8: 1. The precision is computed below: N is the sum of the experimental data, and N is the exact number of series in which the forecast results are correct. It should be noted that the data set for the Layout Net Model includes the Layout Data of the Precise Segment Segment. If there is no accurate division of the length of the area in the existing layout, then it will be regarded as incorrect.

The layout net model has the same structure as the partition net model. This paper gives a case study on the parameter adjusting procedure of the distribution net model and analyses the way of determining the parameters in the experiment.

1

Word embedding vector dimensionOUTPUT_DIM

Different OUTPUT _ DIM is applied to the alignment test, and the precision and loss curves of FNN are illustrated in Fig. 11. Learning speed is 0.1, HIDDEN _ SIZE is 128, and basic _ size is 256. When OUTPUT _ DIM is 256 or 512, and the training cycle Epochs is 20, the training curve becomes saturated, and the optimum precision is 100 percent. To decrease the model's complexity and increase the training rate, we use OUTPUT _ DIM as 256 to carry on the next comparative test.

3000 building layout data verified the precision of the FNN model. Test After _ Embedding Net Model and Pre _ Embedding Net Model Precision as illustrated in Table 2, the Precision of a Single Family Segment is used to describe Single- Functional Precision Series. For the entire Family Segment Series, if one Family Segment Forecast Mistake, Consider the Family Segment Series not Accurate. Based on four experiments, the mean precision of one layout model with no character embedding layer is 89.75%, and the total block order is 73.48%. The mean precision of the individual placement model with no character embedding layer is 98.60%, and the mean precision for the entire building section is 92.16%.

Table 2.

Test set accuracy

Test set accuracy%	First, for the first time	Second time	Third time	The fourth time	Average accuracy
After adding the embedded layer (Single house section)	98.54	99.01	97.63	99.22	98.60
After adding the embedded layer (Whole house section)	91.40	90.79	93.31	93.12	92.16
Join the embedding layer before (Single house section)	90.24	88.54	91.06	89.14	89.75
Join the embedding layer before (Whole house section)	72.93	75.67	70.66	74.67	73.48

See Chart 4.6 for the Embedding Network Model and the Embedding Net Model. The Following _ Embedding Network Model has an average Test Rate of 17.88 s/3000 Copies, Pre _ Embedding.

The average test speed of the network model is 309.43s/3000 copies of the data.

Based on the arrangement arithmetic, it is suggested that the planar arrangement should be consistent with the internal layout. Fig. 12, there is no division between the building segments in (1) and (4), and in (2), the building segments in (3) are separated into several segments and then use the distribution net model to define their respective functions.

3

The space design strategy of the commercial environment under digital technology

4.1

Integrative strategy

Contemporary digital architecture technology for the building form integrated design strategy provides basic tools, building software reduces the designer of the multifaceted geometry, surface integrity greater the operation difficulty of geometric space, the combination of topological deformation technology and digital technology for the architectural form integration provides feasible smooth strategy, building bionics and other related disciplines established the correlation between biological prototype and architectural form, and provides the translation strategy. Therefore, the integrated architectural form design strategy is scientific and operable in the design and practice[17].

The benign development of the contemporary city can not be separated from the reasonable planning and layout. In contrast, the connection between the architecture and the city directly shows the spiritual outlook of the city. The design strategy of architectural environment integration requires public buildings not only to focus on the noumenon but also to systematically think about the integrated connection of architectural space, site, and urban space, to seek the divine and shape integration of architecture and environment, and to form a benign combination benefit with the whole space inside and outside[18].

In decreasing land resources, growing commercial value, intensive development strategy requires public buildings and urban cohesion not only stay on the horizontal dimension, underground, ground, the ground on the integration of the vertical integration and horizontal interweave processing to complete the rational development and utilization of space resources, but this also needs from the functional organization, space combination way and architectural and environment integration design level and type of comprehensive discussion.

4.2

Dissolution of spatial level and interface

4.2.1

Grade resolution of the planar organization

Traditional public architecture continues modernist architecture with clear functional partition and streamlined design, space on the primary and secondary relationship, and the internal and external relationship is very clear, the behavior and path have a clear guide, with the organization to classify, can be summarized into the tree structure group, linear structure of series and center radiation. This design thinking is dominated by reason and has a complete graphical logic chain. The design of the contemporary public architectural space is influenced by the nonlinear theory guided by complexity science, So for the process of design translation into architecture, More considering the complex and diverse relationships between the self-organization of space and population behavior, In nature, Whether it's biological structure, urban form, or population behavior patterns, Its organization mode is mostly a network structure. Without human intervention, no clear center or primary and secondary structure exists. Therefore, The factors in the architectural design process are not linear but a symbiotic system with complexity. This is not simply anti-tradition and unconventional slogans, but the actual demand of the rich contemporary cultural context[19-20].

The contemporary public building space is a network structure in the organizational relationship. The previous chapter discusses the structure of the system of multivalent characteristics and traffic space multidimensional organization characteristics and general function design features on the technical level and organization structure level for space mesh premise, mesh, three-dimensional transportation organization to ensure the overall space network structure of barrier-free operation, eliminate the primary and secondary relationship of space, and formed the space within the organization of ecosystem. Sister Island and the design of Kanazawa gallery in the functional partition and streamline design are adopted network structure organization, network traffic space connects the large and small square and cylinder space, to avoid a single dull decentralization and no directional, the monomer volume of different sizes in different shapes and similar volume together, to achieve the characteristics of building interior space uniformity, different functional partition interface with traffic space integration, and provide more choices for crowd behavior.

4.2.2

Resolution and spatial topology of the layers

As shown in Table 3,The spatial integrated design of the vertical dimension eliminates a "layer" and opens up the spatial vertical dimension. As a new design medium, digital technology plays an important role in promoting this process, among which the most representative is the exploration of spatial topology by combining digital technology and topology.

Table 3.

spatial strategy

Tactics	Merit	Shortcoming
Homogenization	Simple in structure and easy to implement	Significant loss of information and less detail
Like meta polymerization	Retain more information and reduce the noise	May cause boundary blur and is not suitable for complex terrain
Space averaging method	Reduce the local outliers, more smooth processing	The information is vague and can not retain the spatial characteristics
Target selection	Important information can be selected flexibly, according to the analysis requirements	Selection criteria need to be specified, and important data may be missed
Multiscale analysis	Be able to understand spatial phenomena from multiple perspectives	Computing complexity is high, and the integration is difficult
Weighted assessment	Be able to consider the importance of each data source comprehensively	Weight setting requires professional knowledge and has a high degree of uncertainty

5

Conclusion

This paper aims to study the means and technology of space design in the business environment under digital media technology. First, this paper thoroughly analyzes the word embedding and differential algorithm and optimizes the differential algorithm. Secondly, based on the above algorithm, an environment spatial layout design system is proposed and tested. The specific conclusions are as follows:

1) A new frame is built based on IGA, and a restriction and a space element model are proposed. The design procedure is described in detail.

2) In this paper, we can get the location information from a well-arranged guest dining room in an indoor space. Based on the data expansion, we can make ten thousand long-section data with tags and thirty thousand home-style scenes with tags as the data collection.

3) The precision of a single-building segment means that the precision of one segment is predicted by one segment of the order of building type. If there is an error in any part of a building, then the predicted result of the block order will be wrong, which will be 0. Based on four experiments, the mean precision of one layout model with no character embedding layer was 89.75%, and the mean precision was 73.48%. The mean precision was 98.60% for a single-layout model with no text embedding layer, and the mean precision for the entire building section was 92.16%.

4) From two aspects, the solution of integrated strategy and spatial grade resolution strategy is proposed. The strategy and specific methods of the spatial form design of contemporary public buildings are comprehensively analyzed. The internal causes of the transformation of spatial form design are sorted out, and the identification degree of the design of the spatial form of public buildings is established.

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

Design of commercial environment space based on digital media technology

Sai Yin

Online veröffentlicht: 27. Feb. 2025

Eingereicht: 25. Okt. 2024

Akzeptiert: 29. Jan. 2025

DOI: https://doi.org/10.2478/amns-2025-0111

Schlüsselwörterword embedding algorithm, differential algorithm, spatial design, digital media

© 2025 Sai Yin, published by Sciendo

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

Schlüsselwörter
word embedding algorithm, differential algorithm, spatial design, digital media