Ventilation system energy-saving optimization method and system based on neural network

Abstract

The invention provides a ventilation system energy-saving optimization method and system based on a neural network, and the method comprises the steps: collecting the historical operation data of a fan, and carrying out the mean interpolation and wavelet transform filtering preprocessing of the historical operation data; optimizing the human evolutionary optimization algorithm HEOA by adopting a Gaussian mutation strategy to obtain an improved human evolutionary optimization algorithm IHEOA; parameters of the SOFTS model are optimized through an IHEOA algorithm; an IHEOA-SOFTS ventilation system model is constructed, and the relation between the air volume opening degree and the indoor temperature and the relation between the air volume opening degree and the energy consumption are established; establishing a comprehensive objective function considering indoor temperature control precision and total power consumption, taking minimization of the comprehensive objective function as an objective, utilizing an IHEOA algorithm to continuously optimize and adjust the opening degree of each fan in the IHEOA-SOFTS ventilation system model, determining the optimal relation between the opening degree of the fan and the indoor temperature and the total power consumption, and calculating the indoor temperature control precision and the total power consumption of the fan according to the optimal relation between the opening degree of the fan and the indoor temperature and the total power consumption. Therefore, accurate control of fan parameters and energy-saving optimization of the system are realized.

Description

Technical Field

[0001] The present application relates to the technical field of ventilation system energy-saving optimization, and in particular to a ventilation system energy-saving optimization method and system based on neural network. Background Art

[0002] With the continuous increase in global energy demand and the increasingly serious environmental problems, how to effectively save energy and reduce energy consumption has become the focus of attention of various industries. As an important part of building environmental control, the energy consumption of building ventilation system accounts for a considerable proportion of the total energy consumption of buildings.

[0003] Traditional ventilation systems mainly rely on preset operating strategies and simple feedback control, which makes it difficult to respond to environmental changes and dynamically adjust in real time, resulting in energy waste and inefficient operation. Summary of the invention

[0004] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0005] To this end, the first purpose of this application is to propose a ventilation system energy-saving optimization method based on neural network.

[0006] The second purpose of this application is to propose a ventilation system energy-saving optimization system based on neural network.

[0007] The third objective of the present application is to provide an electronic device.

[0008] A fourth objective of the present application is to provide a computer-readable storage medium.

[0009] A fifth object of the present application is to provide a computer program product.

[0010] To achieve the above-mentioned purpose, the first embodiment of the present application proposes a ventilation system energy-saving optimization method based on a neural network, comprising:

[0011] Collect historical operation data of the fan, and perform mean interpolation and wavelet transform filtering preprocessing on the historical operation data; the historical operation data includes the opening degree, external temperature, humidity, temperature at the fan and total energy consumption of each fan;

[0012] The Gaussian mutation strategy is used to optimize the human evolution optimization algorithm HEOA, and the improved human evolution optimization algorithm IHEOA is obtained;

[0013] The parameters of the SOFTS model are optimized by the IHEOA algorithm; the parameters include embedding dimension, core representation dimension, number of encoder layers, prediction window length, and learning rate;

[0014] An IHEOA-SOFTS ventilation system model is constructed to establish the relationship between the air volume opening, indoor temperature and energy consumption; the IHEOA-SOFTS ventilation system model takes the opening of each fan, the external temperature and humidity as inputs, and takes the temperature at each fan and the total energy consumption as outputs;

[0015] A comprehensive objective function considering indoor temperature control accuracy and total power consumption is established. With the goal of minimizing the comprehensive objective function, the IHEOA algorithm is used to continuously optimize and adjust the opening of each fan in the IHEOA-SOFTS ventilation system model to determine the optimal relationship between the fan opening and the indoor temperature and total power consumption.

[0016] Optionally, the performing mean interpolation and wavelet transform filtering preprocessing on the historical operation data includes:

[0017] Perform mean interpolation on the historical running data, identify missing values ​​in the data set, calculate the mean of all non-missing values ​​in the data set, and fill the calculated mean into the missing value position. The calculation formula is as follows:

[0018]

[0019] Among them, x i is the non-missing value, n is the total number of non-missing values, and mean is the mean of the positions filled with missing values;

[0020] The mean interpolated data set is subjected to wavelet transform filtering, and Daubechies wavelet is selected as the wavelet function;

[0021] Perform discrete wavelet transform on the filtered data to obtain the wavelet coefficients. The calculation formula is as follows:

[0022]

[0023] ψ j,k (m) = 2 -j / 2 ψ(2 -j ·mk)

[0024] Among them, W j,k is the wavelet coefficient at scale j and position k, j is the scale, k is the position parameter, m is the position index of the original data, ψ j,k (m) is the scaled and translated wavelet function. Scaling and translation are obtained by scaling the original wavelet function ψ by 2 -j / 2 Multiply and translate 2 -j mk, where f(m) is the value of the original data at position m;

[0025] The wavelet coefficients are subjected to hard threshold processing to remove noise and unimportant information. The formula is as follows:

[0026]

[0027] Among them, λ is the set threshold, if the wavelet coefficient W j,k If the absolute value of is greater than the threshold λ, the coefficient is retained, otherwise, it is set to 0;

[0028] The wavelet coefficients after threshold processing are used to reconstruct the data. The formula is as follows:

[0029]

[0030] Among them, f′(m) is the value of the reconstructed data at position m, which is obtained by dividing all the reconstructed wavelet coefficients W th,j,k The corresponding wavelet function ψ j,k (m) and sum it over all scales j and positions k;

[0031] The overall reconstructed data is represented as:

[0032] f′={f′(m)∣m=1,2,…,M}

[0033] Where M is the length of the signal and f′ is the overall reconstructed data.

[0034] Optionally, the Gaussian mutation strategy is used to optimize the human evolution optimization algorithm HEOA to obtain an improved human evolution optimization algorithm IHEOA, including:

[0035] To simulate the chaotic stage in the early stage of human evolution, the population is initialized using the logical chaos mapping method. The calculation formula is as follows:

[0036] x i =α·x i-1 ·(1-x i-1 ),0≤x0≤1,i=1,2,...,N,α=4

[0037] Among them, x i represents the i-th iteration value, x i-1 represents the previous iteration value, N is the population size, α is the control parameter of the chaotic mapping; the chaotic value x i Mapped to the search space, the position of the initial population is obtained, and the calculation formula is:

[0038]

[0039] in, represents the initial position of the i-th individual, lb and ub are the lower and upper limits of the search space respectively;

[0040] In each iteration of the human evolution optimization algorithm, the individual fitness of each individual in the population is calculated, and a Gaussian mutation strategy is applied to generate a normally distributed random number perturbation so that the new solution is fine-tuned near the original solution. The update formula is as follows:

[0041] X new =X old +Gaussian(0,σ)

[0042] Among them, X new is the new solution after mutation, X old is the original solution before mutation, Gaussian(0,σ) means the mean is 0 and the variance is σ 2 Gaussian distribution random number; in the human exploration stage, individuals search through Gaussian mutation strategy, and the formula for updating the position is as follows:

[0043]

[0044] Among them, β is the adaptive function, dim is the dimension of the problem, and Respectively represent the current position and the next position after update of the i-th individual, X best This is the best location I have explored so far. is the average position of the current population, Levy represents levy distribution, f jump is the jump coefficient, rand is a random number in the range [0,1];

[0045] In the human development stage, the population is divided into leaders, explorers, followers, and losers, and a Gaussian mutation strategy is applied to each role, and the position is updated using the corresponding strategy;

[0046] Check the termination condition. If the termination condition is met, stop the iteration and output the optimal solution.

[0047] Optionally, in the human development stage, the population is divided into leaders, explorers, followers, and losers, a Gaussian mutation strategy is applied to each role, and the position is updated using the corresponding strategy, including:

[0048] For each leader i in the population, the Gaussian mutation strategy is applied, and the formula is as follows:

[0049]

[0050] Among them, X i+1 is the new solution after mutation, X i is the original solution before mutation, The mean is 0 and the variance is A Gaussian random variable of

[0051] The top 40% of pre-adapted individuals in the population are designated as leaders, and the positions are updated using the leader strategy and the Gaussian mutation strategy is added. The formula is as follows:

[0052]

[0053] in, is the position of the i-th individual in the t+1th generation, is the position of the i-th individual in the t-th generation, Rn represents a random number following a normal distribution, the function ones(1, dim) represents the generation of a row vector containing dim elements, where each element is set to 1, R is a random number in the range [0, 1], representing the complexity of the situation related to the leader, A is the situation evaluation value, which is used to determine whether the leader changes its search strategy, ω is the knowledge acquisition difficulty coefficient, which gradually decreases over time, and the expression is:

[0054]

[0055] The top 40% to 80% of the pre-adapted individuals in the population are designated as explorers, and the explorer strategy is used to update the position and add the Gaussian mutation strategy. The formula is as follows:

[0056]

[0057] in, is the position of the i-th individual in the t+1th generation, is the position of the i-th individual in the t-th generation, is the least fit individual in the population in generation t, The mean is 0 and the variance is Gaussian random variation;

[0058] The top 80% to 90% of the pre-adapted individuals in the population are designated as followers and the Gaussian mutation strategy is applied. The formula is as follows:

[0059]

[0060] in, represents the position of the most adaptable individual in the human population at the tth iteration, Rd represents a random number in the range [1, dim], The mean is 0 and the variance is A Gaussian random variable of

[0061] The last 10% of pre-adapted individuals in the population are designated as losers, and the population replenishment equation is improved by the Gaussian mutation strategy:

[0062]

[0063] Among them, X i+1 is the position of the newly generated population individual, X best is the position of the individual with the highest fitness in the population, X i is the position of the loser in the population, Rn represents a random number that follows a normal distribution, σ is a parameter that controls the Gaussian variation, called the standard deviation, is a standard normal distribution.

[0064] Optionally, the optimizing the parameters of the SOFTS model by using the IHEOA algorithm includes:

[0065] The objective function f(θ) is defined according to the mean square error of the SOFTS model, where θ is a set of parameters to be optimized, including the embedding dimension d, the core representation dimension d′, the number of encoder layers N, the prediction window length H, and the learning rate L; the formula of the objective function f(θ) is as follows:

[0066]

[0067] Where N is the number of samples, Y i is the actual value, is the model prediction value;

[0068] During initialization, the initial parameter set θ = {d, d′, N, H, L} is generated by the population initialization method of the IHEOA algorithm;

[0069] Define the prediction model, the formula is:

[0070]

[0071] Among them, X i is the input feature, SOFTS(θ,X i ) is the prediction result of the SOFTS model based on the parameter set θ;

[0072] In each iteration, the exploration and development phases of the IHEOA algorithm are applied to update the parameter set θ;

[0073] By calculating the current fitness, the performance of the SOFTS model corresponding to each parameter set is calculated, and the optimal parameter set is updated;

[0074] When the termination condition is reached, the optimal parameter set is output and applied to the SOFTS model.

[0075] Optionally, the prediction process of the IHEOA-SOFTS ventilation system model includes:

[0076] Input the preprocessed data set X into the ventilation system model t ={A t ,Text,t ,H ext,t}, where X t is the total model input at time step t, A t is the opening degree of each fan at time step t, T ext,t is the external temperature at time step t, H ext,t is the external humidity at time step t;

[0077] In the ventilation system model, the input data is embedded into a high-dimensional space through an embedding layer, and the formula is:

[0078] S0=Embedding(X t )

[0079] Among them, S0 is the output feature representation of the embedding layer;

[0080] The output feature representation of the embedding layer is input into the STAR module, and the output feature representation of the i-th layer is obtained through time series processing. The formula is:

[0081] S i =STAR(S i-1 )

[0082] Among them, S i represents the feature representation of the output of the i-th layer in the STAR module, S i-1 Represents the output feature representation of the i-1th layer in the STAR module;

[0083] The output feature representation of the i-1th layer is processed by a multi-layer perceptron to extract the core representation. The formula is as follows:

[0084] o i =MLP(S i-1 )

[0085] Among them, i is the core representation of the i-th layer;

[0086] Repeat and concatenate the core representation and the feature representation to generate a combined feature representation. The formula is:

[0087] F i =Repeat_Concat(S i-1 ,o i )

[0088] Among them, F i It is the combined feature of the feature representation input to the i-th layer of multi-layer perceptron and the core representation of the output;

[0089] The combined feature representation is processed through the encoder layer, and the original feature and the processed feature representation are combined through the residual, and the formula is as follows:

[0090] S i =MLP(F i )+S i-1

[0091] Among them, S i Feature representation of the encoder layer output;

[0092] The feature representation output by the encoder layer is input into the linear layer and the prediction result is output, which is expressed as:

[0093]

[0094] in, represents the prediction result, S L It represents the final feature representation obtained after layer-by-layer processing by a multi-layer neural network. represents the temperature at each fan at the prediction time step t, E total,t represents the total power consumption at the prediction time step t.

[0095] Optionally, the comprehensive objective function considering indoor temperature control accuracy and total power consumption is established, and the opening of each fan in the IHEOA-SOFTS ventilation system model is continuously optimized and adjusted by using the IHEOA algorithm to determine the optimal relationship between the fan opening and the indoor temperature and total power consumption, including:

[0096] A comprehensive objective function f(θ) is established. The comprehensive objective function f(θ) takes into account both indoor temperature control accuracy and total power consumption and is defined as follows:

[0097]

[0098] Among them, T target is the target temperature, is the mean square error of temperature control, is the total power consumption, w1, w2 are weight coefficients used to balance the impact of different objectives;

[0099] Use the IHEOA intelligent algorithm to initialize the population and generate the initial fan opening solution set A t , each solution is represented by a combination of fan opening degrees, and the population initialization formula is as follows:

[0100]

[0101] For each solution in the population Calculate its fitness value, which is determined by the comprehensive objective function Determine, the formula is:

[0102]

[0103] The population is screened according to the fitness value, and the better solution is selected to form a new candidate population. The screened population is expressed as:

[0104] P′=Select(P)

[0105] Based on the screened population, a crossover operation is performed to generate new candidate solutions. The crossover formula is:

[0106] P″=Crossover(P′)

[0107] Perform mutation operations on the new solutions generated by crossover, make small adjustments to each candidate solution, and the update formula is:

[0108] P″′=Mutate(P″)

[0109] The mutated solution is merged with the current population to generate an updated population. The update formula is:

[0110] P = Update(P, P″′)

[0111] When the predetermined number of iterations is reached or the objective function value converges, the algorithm is terminated and the final optimized fan opening parameter set is output, which is expressed as:

[0112]

[0113] Use optimized fan opening Control the ventilation system, adjust the relationship between the fan opening and the indoor temperature and total power consumption, and achieve optimal control of fan parameters and energy-saving optimization of the system.

[0114] To achieve the above-mentioned purpose, the second embodiment of the present application proposes a ventilation system energy-saving optimization system based on a neural network, comprising:

[0115] A preprocessing module is used to collect historical operation data of the fan and perform mean interpolation and wavelet transform filtering preprocessing on the historical operation data; the historical operation data includes the opening degree, external temperature, humidity, temperature at the fan and total energy consumption of each fan;

[0116] The first optimization module is used to optimize the human evolution optimization algorithm HEOA by using a Gaussian mutation strategy to obtain an improved human evolution optimization algorithm IHEOA;

[0117] The second optimization module is used to optimize the parameters of the SOFTS model through the IHEOA algorithm; the parameters include embedding dimension, core representation dimension, number of encoder layers, prediction window length, and learning rate;

[0118] A model building module is used to build an IHEOA-SOFTS ventilation system model and establish the relationship between the air volume opening and the indoor temperature and energy consumption; the IHEOA-SOFTS ventilation system model takes the opening of each fan, the external temperature and the humidity as inputs, and takes the temperature at each fan and the total energy consumption as outputs;

[0119] The optimal relationship output module is used to establish a comprehensive objective function that takes into account the indoor temperature control accuracy and total power consumption. With the goal of minimizing the comprehensive objective function, the IHEOA algorithm is used to continuously optimize and adjust the opening of each fan in the IHEOA-SOFTS ventilation system model to determine the optimal relationship between the fan opening and the indoor temperature and total power consumption.

[0120] To achieve the above-mentioned purpose, the third aspect of the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0121] The memory stores computer-executable instructions;

[0122] The processor executes the computer-executable instructions stored in the memory to implement the method as described in any one of the first aspects.

[0123] To achieve the above-mentioned purpose, the fourth aspect embodiment of the present application proposes a computer-readable storage medium, in which computer-readable storage medium is stored computer execution instructions, and when the computer execution instructions are executed by a processor, they are used to implement the method as described in any one of the first aspects.

[0124] To achieve the above-mentioned purpose, the fifth aspect of the present application proposes a computer program product, which implements any method in the first aspect when executed by a processor.

[0125] The technical solution provided by the embodiments of the present application brings at least the following beneficial effects:

[0126] Through the energy-saving optimization method of ventilation system based on neural network and using IHEOA-SOFTS model, the total energy consumption of ventilation system can be significantly reduced while taking into account the indoor temperature control accuracy; by optimizing the fan opening parameters and establishing the optimal relationship between fan opening, indoor temperature and total power consumption, the fan parameters can be precisely controlled, and the energy saving and operation efficiency of the system can be improved. This application can effectively balance the comfort and energy consumption goals, and provide an efficient solution for the intelligent optimization of ventilation system.

[0127] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0128] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0129] Figure 1 A simplified flowchart of a ventilation system energy-saving optimization method based on a neural network provided in an embodiment of the present application;

[0130] Figure 2 A detailed flowchart of a ventilation system energy-saving optimization method based on a neural network provided in an embodiment of the present application;

[0131] Figure 3 A schematic diagram of the structure of a ventilation system energy-saving optimization system based on a neural network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0132] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0133] In view of the problems of high energy consumption, low control accuracy and inability to accurately respond to environmental changes in existing ventilation system energy-saving optimization methods, the embodiment of the present application provides a ventilation system energy-saving optimization method based on a neural network. Figure 1 and Figure 2 A schematic diagram of a flow chart of a ventilation system energy-saving optimization method based on a neural network provided in an embodiment of the present application. Figure 1 and Figure 2 , the method comprises the following steps:

[0134] Step S1, collecting historical operation data of the wind turbine, and performing mean interpolation and wavelet transform filtering preprocessing on the historical operation data.

[0135] In the embodiment of the present application, the historical operation data includes the opening degree of each fan, the external temperature, the humidity, the temperature at the fan and the total energy consumption. By preprocessing the collected data, noise can be removed and missing values ​​can be filled, providing high-quality data input for the construction of subsequent models.

[0136] It includes the following sub-steps:

[0137] Step S11, perform mean interpolation on the historical operation data, identify missing values ​​in the data set, calculate the mean of all non-missing values ​​in the data set, and fill the calculated mean into the missing value position. The calculation formula is as follows:

[0138]

[0139] Among them, x i is the non-missing value, n is the total number of non-missing values, and mean is the mean of the positions filled with missing values. Mean interpolation can effectively reduce the impact of missing data on model accuracy and provide complete data input for subsequent steps.

[0140] Step S12, select Daubechies wavelet as wavelet function, and perform wavelet transform filtering on the interpolated data. Wavelet transform can decompose the signal into different scales, thereby extracting characteristic information in the data and removing high-frequency noise, providing a smooth data representation for subsequent analysis.

[0141] Step S13, performing discrete wavelet transform on the filtered data to obtain wavelet coefficients, and the calculation formula is as follows:

[0142]

[0143] ψ j,k (m) = 2 -j / 2 ψ(2 -j ·mk)

[0144] Among them, W j,k is the wavelet coefficient at scale j and position k, j is the scale, k is the position parameter, m is the position index of the original data, ψ j,k (m) is the scaled and translated wavelet function. Scaling and translation are obtained by scaling the original wavelet function ψ by 2 -j / 2 Multiply and translate 2 -j ·mk, where f(m) is the value of the original data at position m. Through discrete wavelet transform, multi-scale features in the data can be extracted to provide support for subsequent processing.

[0145] Step S14, in order to remove noise and unimportant information, hard threshold processing is performed on the wavelet coefficients, and the formula is as follows:

[0146]

[0147] Among them, λ is the set threshold, if the wavelet coefficient W j,k If the absolute value of is greater than the threshold λ, the coefficient is retained, otherwise it is set to 0 and those smaller coefficients are removed, which usually correspond to noise or unimportant signal features.

[0148] This process can effectively remove the noise components in the data.

[0149] Step S15, reconstructing a smooth data signal based on the wavelet coefficients after threshold processing, the formula is as follows:

[0150]

[0151] Among them, f′(m) is the value of the reconstructed data at position m, which is obtained by dividing all the reconstructed wavelet coefficients W th,j,k The corresponding wavelet function ψ j,k (m) and sum it over all scales j and positions k.

[0152] The overall reconstructed data is represented as:

[0153] f′={f′(m)∣m=1,2,…,M}

[0154] Where M is the length of the signal and f′ is the overall reconstructed data.

[0155] The reconstructed data is smoothed and noise-free, providing high-quality input data for subsequent steps.

[0156] Step S2, using Gaussian mutation strategy to optimize the human evolution optimization algorithm HEOA to obtain an improved human evolution optimization algorithm IHEOA.

[0157] In the embodiment of the present application, in order to improve the global optimization ability of the algorithm, the human evolution optimization algorithm HEOA is improved by introducing a Gaussian mutation strategy to obtain an improved human evolution optimization algorithm IHEOA. The IHEOA algorithm generates a population in the initial stage and adjusts the positions of the individuals in the population through multiple iterations to optimize the objective function.

[0158] It includes the following sub-steps:

[0159] Step S21, initializing the population by logical chaos mapping method to generate an initial population with randomness and uniform distribution. The calculation formula of logical chaos mapping is as follows:

[0160] x i =α·x i-1 ·(1-x i-1 ),0≤x0≤1,i=1,2,...,N,α=4

[0161] Among them, x i represents the i-th iteration value, x i-1 represents the previous iteration value, N is the population size, and α is the control parameter of the chaotic mapping. i Mapped to the search space, the position of the initial population is obtained, and the calculation formula is:

[0162]

[0163] in, represents the initial position of the i-th individual, lb and ub are the lower and upper limits of the search space respectively.

[0164] By initializing the population, the uniformity of the distribution of the initial solution is ensured, providing a good foundation for subsequent optimization.

[0165] Step S22, in each iteration of the human evolution optimization algorithm, the individual fitness of each individual in the population is calculated, and a Gaussian mutation strategy is applied to generate a normally distributed random number perturbation so that the new solution is fine-tuned around the original solution. The update formula is as follows:

[0166] X new =X old +Gaussian(0,σ)

[0167] Among them, X new is the new solution after mutation, X old is the original solution before mutation, Gaussian(0,σ) means the mean is 0 and the variance is σ 2 Gaussian distributed random numbers.

[0168] At the same time, in the human exploration stage, facing undeveloped areas and limited knowledge, individuals tend to adopt a unified search strategy. In the embodiment of the present application, individuals update their positions through a Gaussian mutation strategy combined with an adaptive search method, and the update formula is as follows:

[0169]

[0170] Among them, β is the adaptive function, dim is the dimension of the problem, and Respectively represent the current position and the next position after update of the i-th individual, X best This is the best location I have explored so far. is the average position of the current population, Levy represents levy distribution, f jump is the jump coefficient, and rand is a random number in the range [0,1].

[0171] Average position The calculation formula is as follows:

[0172]

[0173] The adaptive function β function is used to adjust parameters according to the number of iterations and the current position, reflecting the gradual difficulty and group characteristics of human exploration of knowledge. The calculation formula is as follows:

[0174]

[0175] In order to simulate the knowledge acquisition and spiral development in the exploration phase, Levy distribution is used. The expression of Levy distribution is as follows:

[0176]

[0177] Among them, μ~N(0,D), v~N(0,D), The value of γ is 1.5.

[0178] In order to improve the dispersion of the search position, a jumping strategy is introduced in the exploration phase, and the jumping coefficient f jump The expression is as follows:

[0179]

[0180] Step S22, in the human development stage, the population is divided into leaders, explorers, followers, and losers, a Gaussian mutation strategy is applied to each role, and the position is updated using the corresponding strategy.

[0181] (1) Leader: For each leader i in the population, the Gaussian mutation strategy is applied, and the formula is as follows:

[0182]

[0183] Among them, X i+1 is the new solution after mutation, X i is the original solution before mutation, The mean is 0 and the variance is is a Gaussian random variable.

[0184] The leader has rich knowledge and is usually located in the optimal area. In the embodiment of the present application, the top 40% of pre-adapted individuals in the population are designated as leaders, and the leader strategy is used to update the position and add the Gaussian mutation strategy. The formula is as follows:

[0185]

[0186] in, is the position of the i-th individual in the t+1th generation, is the position of the i-th individual of the t-th generation, Rn represents a random number following a normal distribution, the function ones(1,dim) represents the generation of a row vector containing dim elements, where each element is set to 1, R is a random number in the range [0,1], representing the complexity of the situation related to the leader, A is the situation evaluation value, which is used to determine whether the leader changes its search strategy. In this embodiment of the present application, the value of A can be 0.6. According to the complexity of the situation at a specific position, the leader will select an appropriate search strategy.

[0187] ω is the coefficient of difficulty in acquiring knowledge, which decreases gradually with the passage of time. The expression is:

[0188]

[0189] When R < A, the position update of an individual is affected by the current generation t, a random number rand, and the maximum number of iterations Max iter The exponential decay term ensures that as time goes by, the amplitude of the position update gradually decreases, which helps to converge to the local optimal solution.

[0190] When R ≥ A, the position update of an individual is determined by the weight system ω and an offset of a random number Rn based on the normal distribution. This offset is a vector of all 1s, indicating that the explorer makes a certain amplitude of movement in each dimension. This strategy increases the diversity of the search space and helps to jump out of the local optimum.

[0191] (2) Explorer: The explorer plays a crucial role in venturing into unknown territory to discover the global optimal solution. In the embodiments of this application, the individuals in the population with pre-adaptation ranking in the top 40% to 80% are designated as explorers. The explorer strategy is used to update the position and the Gaussian mutation strategy is added. The explorer is brave enough to enter the unknown territory. The formula is as follows:

[0192]

[0193] Where, is the position of the i-th individual in the (t + 1)-th generation, is the position of the i-th individual in the t-th generation, is the least adaptable individual in the population in the t-th generation, represents a Gaussian random variable with a mean of 0 and a variance of .

[0194] When the position X i of an individual is quite different from the position X worst of the worst individual, the update amplitude is large, encouraging the individual to move away from the current poor solution and towards a better solution.

[0195] (3) Follower: In the embodiments of this application, the individuals in the population with pre-adaptation ranking in the top 80% to 90% are designated as followers. The Gaussian mutation strategy is applied. The formula is as follows:

[0196]

[0197] Where, represents the position of the most adaptable individual in the human population at the t-th iteration. Rd represents a random number within the range of [1, dim]. The introduction of the random number Rd ensures that each update is not completely deterministic, thus avoiding all individuals concentrating at one position and maintaining the diversity of the population. represents a Gaussian random variable with a mean of 0 and a variance of .

[0198] (4) Losers: In the present embodiment, the last 10% of individuals in the population that have been pre-adapted are designated as losers. For each loser i in the population, those individuals that have poor adaptability and remain in the population are called losers. These losers with poor adaptability and unable to integrate into society will be eliminated, and the population will be supplemented by breeding in areas suitable for human development. The population supplementation equation after the Gaussian mutation strategy is improved is:

[0199]

[0200] Among them, X i+1 is the position of the newly generated population individual, X best is the position of the individual with the highest fitness in the population, X i is the position of the loser in the population, Rn represents a random number that follows a normal distribution, σ is a parameter that controls the Gaussian variation, called the standard deviation, is a standard normal distribution.

[0201] Step S24, after each iteration, check whether the termination condition is met, such as reaching the maximum number of iterations or the objective function value converges. If the condition is met, stop the algorithm and output the optimal solution.

[0202] Step S3, optimizing the parameters of the SOFTS model by using the IHEOA algorithm.

[0203] In the embodiment of the present application, the key parameters of the SOFTS model are optimized by the IHEOA algorithm to improve the prediction performance and generalization ability of the model. The optimized parameters include the embedding dimension d, the core representation dimension d′, the number of encoder layers N, the prediction window length H, and the learning rate L.

[0204] It includes the following sub-steps:

[0205] Step S31, define the objective function f(θ) according to the mean square error of the SOFTS model, where θ is a set of parameters to be optimized, including the embedding dimension d, the core representation dimension d′, the number of encoder layers N, the prediction window length H, and the learning rate L. The formula of the objective function f(θ) is as follows:

[0206]

[0207] Where N is the number of samples, Y i is the actual value, is the model prediction value.

[0208] It can be understood that minimizing the objective function f(θ) can ensure that the prediction error of the SOFTS model on the training data is minimized, thereby improving the model performance.

[0209] During initialization, the population initialization method of the IHEOA algorithm is used to generate an initial parameter set θ = {d, d′, N, H, L}. The initial parameter set provides an initial solution for subsequent optimization iterations.

[0210] Step S32: predict the input features using the SOFTS model. The prediction formula is as follows:

[0211]

[0212] Among them, X i is the input feature, SOFTS(θ,X i ) is the prediction result of the SOFTS model based on the parameter set θ. This prediction model provides a basis for subsequent fitness calculation and parameter optimization.

[0213] Step S33, in each iteration, the exploration phase and development phase of the IHEOA algorithm are applied to update the parameter set θ. The exploration phase is used to search for the global optimal solution, and the development phase is used for local optimization. In each round of iteration, the algorithm fine-tunes the parameter set through the Gaussian mutation strategy and adjusts the position of the solution in combination with the fitness calculation, thereby improving the optimization effect.

[0214] Step S34, evaluate the performance of the SOFTS model corresponding to each parameter set by calculating the current fitness. The fitness is calculated based on the objective function f(θ), that is, the prediction error of the current parameter set on the training data. According to the fitness value, the optimal parameter set is updated to ensure that the parameter solution with the best performance is always retained during the optimization process.

[0215] Step S35, when the preset termination condition is reached (for example, the maximum number of iterations is reached or the objective function value converges), the algorithm terminates and outputs the optimal parameter set, and applies the optimized parameter set to the SOFTS model for the final prediction task of the model to improve the prediction accuracy and generalization ability.

[0216] Step S4, constructing an IHEOA-SOFTS ventilation system model, and establishing the relationship between air volume opening, indoor temperature and energy consumption.

[0217] In the embodiment of the present application, the IHEOA-SOFTS ventilation system model is constructed to establish a mapping relationship between the opening degree of each fan and the indoor temperature and total energy consumption. The model takes the opening degree of each fan, the external temperature and humidity as input, and the temperature and total energy consumption at each fan as output, providing an accurate prediction basis for fan parameter optimization and system energy saving.

[0218] It includes the following sub-steps:

[0219] Step S41: taking the preprocessed historical operation data as input, the input data set is represented by Xt ={A t ,T ext,t ,H ext,t}, where X t is the total model input at time step t, A t is the opening degree of each fan at time step t, T ext,t is the external temperature at time step t, H ext,t is the external humidity at time step t.

[0220] Step S42: In the ventilation system model, the input data X is embedded in the ventilation system model through the embedding layer. t ={A t ,T ext,t ,H ext,t} is embedded into a high-dimensional space for subsequent processing. The output feature of the embedding layer is expressed as:

[0221] S0=Embedding(X t )

[0222] Among them, S0 is the output feature representation of the embedding layer.

[0223] The purpose of the embedding layer is to convert the raw input data into a high-dimensional vector representation so that the model can better capture the complex relationships between the input features.

[0224] Step S43, the output feature representation of the embedding layer is input into the STAR module to capture the long-term and short-term dependencies of the time series data layer by layer. The output representation of the STAR module is:

[0225] S i =STAR(S i-1 )

[0226] Among them, S i represents the feature representation of the output of the i-th layer in the STAR module, S i-1 Represents the output feature representation of the i-1th layer in the STAR module.

[0227] Step S44, the output feature representation of the i-1th layer of the STAR module is processed by a multi-layer perceptron to extract the core representation, and the formula is as follows:

[0228] o i =MLP(S i-1 )

[0229] Among them, i is the core representation of the i-th layer, which is used to further capture the core information of the features.

[0230] Step S45, repeat and concatenate the core representation and the corresponding feature representation to generate a combined feature representation. The specific formula is:

[0231] F i =Repeat_Concat(S i-1 ,o i )

[0232] Among them, F i It is the combined feature of the feature representation input to the i-th layer of the multi-layer perceptron and the core representation of the output.

[0233] Step S46, the combined feature representation is processed through the encoder layer, and the original features are combined with the processed features through residual connections to enhance the expression ability of the model. The encoder layer output is represented as:

[0234] S i =MLP(F i )+S i-1

[0235] Among them, S i It is the feature representation of the encoder layer output.

[0236] Step S47, the final output feature representation S output by the encoder layer L Input to the linear layer and output the prediction result, expressed as:

[0237]

[0238] in, represents the prediction result, S L It represents the final feature representation obtained after layer-by-layer processing by a multi-layer neural network. represents the temperature at each fan at the prediction time step t, E total,t represents the total power consumption at the prediction time step t.

[0239] Through the above steps, the IHEOA-SOFTS ventilation system model can accurately predict the relationship between each fan opening, indoor temperature and total power consumption, providing efficient support for energy-saving optimization of the ventilation system.

[0240] When applied, the trained model is used to input the air volume opening, external temperature and humidity data to predict the temperature and total power consumption of each fan. The formula is:

[0241]

[0242] in, represents the predicted temperature of each fan at the t+kth step, represents the predicted total power consumption at the t+kth step, θ * Represents the set of parameters of the optimized SOFTS model.

[0243] Step S5, establish a comprehensive objective function that takes into account the indoor temperature control accuracy and total power consumption, and use the IHEOA algorithm to continuously optimize and adjust the opening of each fan in the IHEOA-SOFTS ventilation system model with the goal of minimizing the comprehensive objective function, and determine the optimal relationship between the fan opening and the indoor temperature and total power consumption.

[0244] In an embodiment of the present application, the key parameters (fan opening) of the IHEOA-SOFTS ventilation system model are optimized through the IHEOA intelligent algorithm to achieve optimal energy saving and control effects.

[0245] The process mainly includes the following sub-steps:

[0246] Step S51, first, combining the two optimization goals of indoor temperature control accuracy and total power consumption, define a comprehensive objective function f(θ). The expression of the objective function is as follows:

[0247]

[0248] Among them, T target is the target temperature, is the mean square error of temperature control, is the total power consumption, w1 and w2 are weight coefficients, which respectively represent the weights of temperature control and energy consumption. The specific values ​​can be adjusted according to actual needs.

[0249] Step S52, using the IHEOA intelligent algorithm to initialize the population and generate an initial set of fan opening solutions. Each solution is represented by a combination of fan openings, and the population initialization formula is as follows:

[0250]

[0251] Among them, N is the population size, which serves as the initial input of the optimization process and provides the basis for subsequent fitness calculations.

[0252] Step S53, after initializing the population, Calculate the fitness value, which is determined by the comprehensive objective function The specific steps are to use the ventilation system model to predict the temperature and total power consumption of each fan according to the input parameters. The prediction formula is:

[0253]

[0254] After the fitness value calculation is completed, it provides input for the next screening process.

[0255] Step S54, after calculating the fitness value, the population is screened according to the fitness value, and the better solution is selected to form a new candidate population. The screened population is represented as:

[0256] P′=Select(P)

[0257] The selected solutions have higher fitness values ​​and are more likely to generate new solutions that meet the optimization objectives. Based on the selected population, a crossover operation is performed to generate new candidate solutions. The crossover formula is:

[0258] P″=Crossover(P′)

[0259] This operation generates new solutions by combining the features of better solutions, providing diversity for subsequent optimization processes.

[0260] Step S55: In order to further increase the diversity of the population, a mutation operation is performed on the new solution generated by crossover. The mutation operation makes a small adjustment to each candidate solution to prevent the population from falling into a local optimal solution. The mutated population is represented as:

[0261] P″′=Mutate(P″)

[0262] This step ensures that the optimization process can explore a larger search space, thereby improving the global search ability of the algorithm.

[0263] Step S56, merge the mutated solution with the current population to generate an updated population. The update formula is:

[0264] P = Update(P, P″′)

[0265] Step S57: When the predetermined number of iterations is reached or the objective function value converges, the algorithm is terminated and the final optimized fan opening parameter set is output, which is expressed as:

[0266]

[0267] Step S58, finally, use the optimized fan opening Control the ventilation system, and optimize the fan parameters and the energy saving of the system by adjusting the relationship between the fan opening and the indoor temperature and total power consumption, so as to achieve the goal of system energy saving and precise control.

[0268] In order to implement the above embodiments, the present application also proposes a ventilation system energy-saving optimization system based on a neural network. Figure 3 The schematic diagram of the structure of a ventilation system energy-saving optimization system 10 based on a neural network provided in an embodiment of the present application. Figure 3 As shown, the system includes:

[0269] The preprocessing module 100 is used to collect historical operation data of the fan and perform mean interpolation and wavelet transform filtering preprocessing on the historical operation data; the historical operation data includes the opening degree, external temperature, humidity, temperature at the fan and total energy consumption of each fan;

[0270] The first optimization module 200 is used to optimize the human evolution optimization algorithm HEOA by using a Gaussian mutation strategy to obtain an improved human evolution optimization algorithm IHEOA;

[0271] The second optimization module 300 is used to optimize the parameters of the SOFTS model through the IHEOA algorithm; the parameters include embedding dimension, core representation dimension, number of encoder layers, prediction window length, and learning rate;

[0272] The model building module 400 is used to build an IHEOA-SOFTS ventilation system model and establish the relationship between the air volume opening and the indoor temperature and energy consumption; the IHEOA-SOFTS ventilation system model takes the opening of each fan, the external temperature and the humidity as inputs, and takes the temperature at each fan and the total energy consumption as outputs;

[0273] The optimal relationship output module 500 is used to establish a comprehensive objective function that takes into account the indoor temperature control accuracy and the total power consumption, and to minimize the comprehensive objective function. The IHEOA algorithm is used to continuously optimize and adjust the opening of each fan in the IHEOA-SOFTS ventilation system model to determine the optimal relationship between the fan opening and the indoor temperature and the total power consumption.

[0274] Regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0275] In order to implement the above embodiments, the present application also proposes an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method provided by the above embodiments.

[0276] In order to implement the above embodiments, the present application also proposes a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the methods provided by the above embodiments.

[0277] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which implements the methods provided by the above embodiments when executed by a processor.

[0278] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution of this application can be achieved, and this document is not limited here.

[0279] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A ventilation system energy-saving optimization method based on neural network, characterized in that: The following steps are involved: Collecting historical operation data of the fan, and performing mean interpolation and wavelet transform filtering preprocessing on the historical operation data; The historical operation data includes the opening degree of each fan, external temperature, humidity, temperature at the fan and total energy consumption; The Gaussian mutation strategy is used to optimize the human evolution optimization algorithm HEOA, and the improved human evolution optimization algorithm IHEOA is obtained; The parameters of the SOFTS model are optimized by the IHEOA algorithm; the parameters include embedding dimension, core representation dimension, number of encoder layers, prediction window length, and learning rate; An IHEOA-SOFTS ventilation system model is constructed to establish the relationship between the air volume opening, indoor temperature and energy consumption; the IHEOA-SOFTS ventilation system model takes the opening of each fan, the external temperature and humidity as inputs, and takes the temperature at each fan and the total energy consumption as outputs; A comprehensive objective function considering indoor temperature control accuracy and total power consumption is established. With the goal of minimizing the comprehensive objective function, the IHEOA algorithm is used to continuously optimize and adjust the opening of each fan in the IHEOA-SOFTS ventilation system model to determine the optimal relationship between the fan opening and the indoor temperature and total power consumption.

2. The method according to claim 1, characterized in that The performing mean interpolation and wavelet transform filtering preprocessing on the historical operation data includes: Perform mean interpolation on the historical running data, identify missing values ​​in the data set, calculate the mean of all non-missing values ​​in the data set, and fill the calculated mean into the missing value position. The calculation formula is as follows: Among them, x i is the non-missing value, n is the total number of non-missing values, and mean is the mean of the positions filled with missing values; The mean interpolated data set is subjected to wavelet transform filtering, and Daubechies wavelet is selected as the wavelet function; Perform discrete wavelet transform on the filtered data to obtain the wavelet coefficients. The calculation formula is as follows: ψ j,k (m)=2 -j / 2 ψ(2 -j ·mk) Among them, W j,k is the wavelet coefficient at scale j and position k, j is the scale, k is the position parameter, m is the position index of the original data, ψ j,k (m) is the scaled and translated wavelet function. Scaling and translation are obtained by scaling the original wavelet function ψ by 2 -j / 2 Multiply and shift by 2 -j mk, where f(m) is the value of the original data at position m; The wavelet coefficients are subjected to hard threshold processing to remove noise and unimportant information. The formula is as follows: Among them, λ is the set threshold, if the wavelet coefficient W j,k If the absolute value of is greater than the threshold λ, the coefficient is retained, otherwise, it is set to 0; The wavelet coefficients after threshold processing are used to reconstruct the data. The formula is as follows: Among them, f′(m) is the value of the reconstructed data at position m, which is obtained by dividing all the reconstructed wavelet coefficients W th,j,k The corresponding wavelet function ψ j,k (m) and sum it over all scales j and positions k; The overall reconstructed data is represented as: f′={f′(m)∣m=1,2,…,M} Where M is the length of the signal and f′ is the overall reconstructed data.

3. The method according to claim 2, characterized in that The Gaussian mutation strategy is used to optimize the human evolution optimization algorithm HEOA to obtain an improved human evolution optimization algorithm IHEOA, including: To simulate the chaotic stage in the early stage of human evolution, the population is initialized using the logical chaos mapping method. The calculation formula is as follows: x i =α·x i-1 ·(1-x i-1 ),0≤x0≤1,i=1,2,...,N,α=4 Among them, x i represents the i-th iteration value, x i-1 represents the previous iteration value, N is the population size, α is the control parameter of the chaotic mapping; the chaotic value x i Mapped to the search space, the position of the initial population is obtained, and the calculation formula is: in, represents the initial position of the i-th individual, lb and ub are the lower and upper limits of the search space respectively; In each iteration of the human evolution optimization algorithm, the individual fitness of each individual in the population is calculated, and a Gaussian mutation strategy is applied to generate a normally distributed random number perturbation so that the new solution is fine-tuned near the original solution. The update formula is as follows: X new =X old +Gaussian(0,σ) Among them, X new is the new solution after mutation, X old is the original solution before mutation, Gaussian(0,σ) means the mean is 0 and the variance is σ 2 Gaussian distribution random number; in the human exploration stage, individuals search through Gaussian mutation strategy, and the formula for updating the position is as follows: Among them, β is the adaptive function, dim is the dimension of the problem, Respectively represent the current position and the next position after update of the i-th individual, X best This is the best location I have explored so far. is the average position of the current population, Levy represents levy distribution, f jump is the jump coefficient, rand is a random number in the range [0,1]; In the human development stage, the population is divided into leaders, explorers, followers, and losers, and a Gaussian mutation strategy is applied to each role, and the position is updated using the corresponding strategy; Check the termination condition. If the termination condition is met, stop the iteration and output the optimal solution.

4. The method according to claim 3, characterized in that In the human development stage, the population is divided into leaders, explorers, followers, and losers. Gaussian mutation strategies are applied to each role, and the positions are updated using corresponding strategies, including: For each leader i in the population, the Gaussian mutation strategy is applied, and the formula is as follows: Among them, X i+1 is the new solution after mutation, X i is the original solution before mutation, The mean is 0 and the variance is A Gaussian random variable of The top 40% of pre-adapted individuals in the population are designated as leaders, and the positions are updated using the leader strategy and the Gaussian mutation strategy is added. The formula is as follows: in, is the position of the i-th individual in the t+1th generation, is the position of the i-th individual in the t-th generation, Rn represents a random number following a normal distribution, the function ones(1, dim) represents the generation of a row vector containing dim elements, where each element is set to 1, R is a random number in the range [0, 1], representing the complexity of the situation related to the leader, A is the situation evaluation value, which is used to determine whether the leader changes its search strategy, ω is the knowledge acquisition difficulty coefficient, which gradually decreases over time, and the expression is: The top 40% to 80% of the pre-adapted individuals in the population are designated as explorers, and the explorer strategy is used to update the position and add the Gaussian mutation strategy. The formula is as follows: in, is the position of the i-th individual in the t+1th generation, is the position of the i-th individual in the t-th generation, is the least fit individual in the population in generation t, The mean is 0 and the variance is Gaussian random variation; The top 80% to 90% of the pre-adapted individuals in the population are designated as followers and the Gaussian mutation strategy is applied. The formula is as follows: in, represents the position of the most adaptable individual in the human population at the tth iteration, Rd represents a random number in the range [1, dim], The mean is 0 and the variance is A Gaussian random variable of The last 10% of pre-adapted individuals in the population are designated as losers, and the population replenishment equation is improved by the Gaussian mutation strategy: Among them, X i+1 is the position of the newly generated population individual, X best is the position of the individual with the highest fitness in the population, X i is the position of the loser in the population, Rn represents a random number that follows a normal distribution, σ is a parameter that controls the Gaussian variation, called the standard deviation, is a standard normal distribution.

5. The method according to claim 4, characterized in that The optimization of the parameters of the SOFTS model by the IHEOA algorithm includes: The objective function f(θ) is defined according to the mean square error of the SOFTS model, where θ is a set of parameters to be optimized, including the embedding dimension d, the core representation dimension d′, the number of encoder layers N, the prediction window length H, and the learning rate L; the formula of the objective function f(θ) is as follows: Where N is the number of samples, Y i is the actual value, is the model prediction value; During initialization, the initial parameter set θ = {d, d′, N, H, L} is generated by the population initialization method of the IHEOA algorithm; Define the prediction model, the formula is: Among them, X i is the input feature, SOFTS(θ,X i ) is the prediction result of the SOFTS model based on the parameter set θ; In each iteration, the exploration and development phases of the IHEOA algorithm are applied to update the parameter set θ; By calculating the current fitness, the performance of the SOFTS model corresponding to each parameter set is calculated, and the optimal parameter set is updated; When the termination condition is reached, the optimal parameter set is output and applied to the SOFTS model.

6. The method according to claim 5, characterized in that The prediction process of the IHEOA-SOFTS ventilation system model includes: Input the preprocessed data set X into the ventilation system model t ={A t ,T ext,t ,H ext,t }, where X t is the total model input at time step t, A t is the opening degree of each fan at time step t, T ext,t is the external temperature at time step t, H ext,t is the external humidity at time step t; In the ventilation system model, the input data is embedded into a high-dimensional space through an embedding layer, and the formula is: S0=Embedding(X t ) Among them, S0 is the output feature representation of the embedding layer; The output feature representation of the embedding layer is input into the STAR module, and the output feature representation of the i-th layer is obtained through time series processing. The formula is: S i =STAR(S i-1 ) Among them, S i represents the feature representation of the output of the i-th layer in the STAR module, S i-1 Represents the output feature representation of the i-1th layer in the STAR module; The output feature representation of the i-1th layer is processed by a multi-layer perceptron to extract the core representation. The formula is as follows: O i =MLP(S i-1 ) Among them, i is the core representation of the i-th layer; Repeat and concatenate the core representation and the feature representation to generate a combined feature representation. The formula is: F i =Repeat_Concat(S i-1 ,o i ) Among them, F i It is the combined feature of the feature representation input to the i-th layer of multi-layer perceptron and the core representation of the output; The combined feature representation is processed through the encoder layer, and the original feature and the processed feature representation are combined through the residual, and the formula is as follows: S i =MLP(F i )+S i-1 Among them, S i Feature representation of the encoder layer output; The feature representation output by the encoder layer is input into the linear layer and the prediction result is output, which is expressed as: in, represents the prediction result, S L It represents the final feature representation obtained after layer-by-layer processing by a multi-layer neural network. represents the temperature at each fan at the prediction time step t, E total,t represents the total power consumption at the prediction time step t.

7. The method according to claim 6, characterized in that The method establishes a comprehensive objective function that takes into account indoor temperature control accuracy and total power consumption, takes minimizing the comprehensive objective function as the goal, uses the IHEOA algorithm to continuously optimize and adjust the opening of each fan in the IHEOA-SOFTS ventilation system model, and determines the optimal relationship between the fan opening and the indoor temperature and total power consumption, including: A comprehensive objective function f(θ) is established. The comprehensive objective function f(θ) takes into account both indoor temperature control accuracy and total power consumption and is defined as follows: Among them, T target is the target temperature, is the mean square error of temperature control, is the total power consumption, w1, w2 are weight coefficients used to balance the impact of different objectives; Use the IHEOA intelligent algorithm to initialize the population and generate the initial fan opening solution set A t , each solution is represented by a combination of fan opening degrees, and the population initialization formula is as follows: For each solution in the population Calculate its fitness value, which is determined by the comprehensive objective function Determine, the formula is: The population is screened according to the fitness value, and the better solution is selected to form a new candidate population. The screened population is expressed as: P′=Select(P) Based on the screened population, a crossover operation is performed to generate new candidate solutions. The crossover formula is: P″=Crossover(P′) Perform mutation operation on the new solution generated by crossover, make small adjustments to each candidate solution, and the update formula is: P″′=Mutate(P″) The mutated solution is merged with the current population to generate an updated population. The update formula is: P = Update(P, P″′) When the predetermined number of iterations is reached or the objective function value converges, the algorithm is terminated and the final optimized fan opening parameter set is output, which is expressed as: Use optimized fan opening Control the ventilation system, adjust the relationship between the fan opening and the indoor temperature and total power consumption, and achieve optimal control of fan parameters and energy-saving optimization of the system.

8. A ventilation system energy-saving optimization system based on neural network, characterized in that: include: A preprocessing module, used for collecting historical operation data of the fan, and performing mean interpolation and wavelet transform filtering preprocessing on the historical operation data; The historical operation data includes the opening degree of each fan, external temperature, humidity, temperature at the fan and total energy consumption; The first optimization module is used to optimize the human evolution optimization algorithm HEOA by using a Gaussian mutation strategy to obtain an improved human evolution optimization algorithm IHEOA; The second optimization module is used to optimize the parameters of the SOFTS model through the IHEOA algorithm; the parameters include embedding dimension, core representation dimension, number of encoder layers, prediction window length, and learning rate; A model building module is used to build an IHEOA-SOFTS ventilation system model and establish the relationship between the air volume opening and the indoor temperature and energy consumption; the IHEOA-SOFTS ventilation system model takes the opening of each fan, the external temperature and the humidity as inputs, and takes the temperature at each fan and the total energy consumption as outputs; The optimal relationship output module is used to establish a comprehensive objective function that takes into account the indoor temperature control accuracy and total power consumption. With the goal of minimizing the comprehensive objective function, the IHEOA algorithm is used to continuously optimize and adjust the opening of each fan in the IHEOA-SOFTS ventilation system model to determine the optimal relationship between the fan opening and the indoor temperature and total power consumption.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

Images