Subway station environment control wind-water linkage energy-saving optimization control method based on deep learning

Abstract

A subway station environment control wind and water linkage energy-saving optimization control method based on deep learning comprises the steps that the train operation state, the station environment data and the environment control equipment data are preprocessed, and a time sequence data matrix is obtained; and decomposing the time sequence data matrix through non-negative matrix decomposition, extracting disturbance characteristics in the time sequence data matrix, and outputting a blind source separation result matrix. And performing convolution operation on the blind source separation result matrix through a convolutional neural network, extracting a spatial pattern of disturbance characteristics, and generating a spatial feature vector through pooling dimension reduction. And capturing a time change rule of dynamic disturbance through a bidirectional GRU structure in the recurrent neural network, and generating an implicit state sequence. And an attention mechanism is introduced into an output layer of the recurrent neural network, an implicit state when the train enters the station is concerned, and optimized disturbance feature representation is generated through weight weighting. And taking the optimized disturbance characteristic representation as the input of a full connection layer, and generating and outputting a corresponding energy-saving strategy category.

Description

Technical Field

[0001] The present invention belongs to the technical field of subway station environmental control, and more specifically, relates to a subway station environmental control and feng shui linkage energy-saving optimization control method based on deep learning. Background Art

[0002] In the operation of modern subway stations, in order to ensure passenger comfort and the stability of equipment in the station, the environmental control and Feng Shui linkage system has become a key infrastructure. The system achieves dynamic balance and energy-saving operation of the micro-environment in the station by comprehensively regulating the air flow, temperature and humidity, and water cooling cycle in the station.

[0003] Current technologies mostly use strategies based on traditional control models and empirical parameters, which have achieved certain results in adjusting constant passenger flow and stable environmental parameters. However, when trains enter and exit the station, the platform area will be impacted by instantaneous pressure waves and thermal disturbances (typical pressure fluctuation range is about ±50 Pa, and the temperature of the train compartment is usually 2-3℃ higher than the platform temperature), resulting in drastic fluctuations in system energy consumption and delayed regulation. This traditional control method has significant deficiencies in dealing with fast, nonlinear, and strongly coupled dynamic changes, and it is difficult to achieve refined and real-time energy-saving optimization control. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the purpose of the present invention is to solve the above-mentioned defects and further propose a subway station environmental control and feng shui linkage energy-saving optimization control method based on deep learning.

[0005] The present invention adopts the following technical solution.

[0006] The first aspect of the present invention discloses a deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method, the method comprising: Acquire train operation status, station environment data and environmental control equipment data, and pre-process the train operation status, station environment data and environmental control equipment data to obtain a time series data matrix; Decomposing the time series data matrix by non-negative matrix decomposition to extract disturbance characteristics in the time series data matrix, and outputting a blind source separation result matrix; Performing a convolution operation on the blind source separation result matrix through a convolutional neural network to extract the spatial pattern of the disturbance characteristics, and generating a spatial feature vector through pooling dimensionality reduction; The spatial feature vector is used as the input of a bidirectional GRU structure in a recurrent neural network to capture the temporal variation of the dynamic disturbance and generate a hidden state sequence; Based on the implicit state sequence, an attention mechanism is introduced into the output layer of the recurrent neural network to focus on the implicit state when the train enters the station, and an optimized disturbance feature representation is generated through weighting; The optimized perturbation feature representation is used as the input of the fully connected layer to generate and output the corresponding energy-saving strategy category; Among them, the train operation status includes the train speed and the train entry time, the station environment data includes air pressure, temperature and humidity, the environmental control equipment data includes fan speed, cold water flow and air conditioning refrigeration load, the disturbance characteristics include pressure wave intensity and thermal disturbance pattern, and the spatial pattern includes pressure wave distribution and heat transfer path in different areas of the subway station.

[0007] Furthermore, the station environment data and environmental control equipment data are data collected by sensors or equipment arranged according to the environmental control requirements of the subway station, and are used to reflect the pressure waves and thermal disturbances caused by trains entering and leaving the station; The train running status, station environment data and environmental control equipment data are obtained, and the train running status, station environment data and environmental control equipment data are preprocessed to obtain a time series data matrix, including: The station environment data and the environmental control equipment data are aligned according to the timestamps at the same time interval to obtain a collection data vector. The expression of the collection data vector is: In the formula, Indicates at time The collected data vector, For time The air pressure data, For time Temperature data, For time Humidity data, For time The fan speed data, For time Cold water flow data, For time Refrigeration load data; The high-frequency noise in the collected data vector is filtered out by low-pass filtering, the signal within a preset frequency range is retained by a band-pass filter, and outliers are removed by a triple standard deviation rule, and the removed outliers are replaced by linear interpolation; The collected data vector after filtering, denoising and outlier removal is sliced ​​by time window, and the segmented time series is stored in matrix form to obtain the time series data matrix. The expression of the time series data matrix is: In the formula, is the time series data matrix after segmentation, is the amount of data at the sampling point. In the time series data matrix, each row represents the time series of a variable, and each column represents any time point. All variables.

[0008] Furthermore, the time series data matrix is ​​decomposed by non-negative matrix decomposition to extract disturbance characteristics in the time series data matrix, and output a blind source separation result matrix, including: Obtain the definition dimension of the time series data matrix, and decompose the time series data matrix based on the definition dimension by the non-negative matrix decomposition, the decomposition expression is: In the formula, is the time series data matrix, with a dimension of m×n; is the feature basis matrix with dimension m×r, which is used to represent the weights of r feature bases, and each row corresponds to m variables; is an activation matrix with a dimension of r×n, which is used to characterize the strength of each feature basis at n time points; r is the number of decomposed features and satisfies ; The matrix reconstruction error is minimized by optimizing the objective function, and the characteristic basis matrix and the activation matrix are updated by an iterative optimization algorithm, wherein the iterative optimization algorithm includes a multiplication update rule, and the expression of the objective function is: In the formula, is the Frobenius norm, which represents the square root of the sum of the squares of the matrix elements, and , used for non-negativity constraints on matrix elements.

[0009] Furthermore, the method of decomposing the time series data matrix by non-negative matrix decomposition to extract disturbance characteristics in the time series data matrix and outputting a blind source separation result matrix also includes: Analyzing the decomposition result of the time series data matrix by the non-negative matrix decomposition, and normalizing the activation matrix and the characteristic basis matrix respectively, so as to extract the key disturbance moment from the activation matrix based on the maximum value and distribution of the variables in the activation matrix, and extracting the key variables by comparing the column vectors of the characteristic basis matrix; Reconstructing and outputting the blind source separation result matrix based on the normalized activation matrix and the feature basis matrix; Among them, the normalized expression of the activation matrix is: In the formula, is the normalized activation matrix, represents the maximum value of each row in the activation matrix; The normalized expression of the characteristic basis matrix is: In the formula, is the normalized eigenvalue matrix, represents the column sum in the feature basis matrix; The reconstruction expression of the blind source separation result matrix is: In the formula, is the blind source separation result matrix, with a dimension of m×n, and each column of the blind source separation result matrix represents a time point The corresponding blind source characteristics contain key disturbance information.

[0010] Furthermore, the convolution operation is performed on the blind source separation result matrix through a convolutional neural network to extract the spatial pattern of the disturbance characteristics, and a spatial feature vector is generated through pooling dimensionality reduction, including: Calling the convolutional neural network to identify local patterns in the blind source separation result matrix through a convolution operation, and performing convolution calculation based on a defined convolution kernel to obtain a convolution output matrix; The output matrix after the convolution is processed by a nonlinear activation function, and the size of the matrix is ​​compressed by a pooling operation, and the pooled matrix is ​​straightened to obtain the spatial feature vector; Wherein, the expression of the spatial feature vector is: In the formula, is the spatial feature vector, and its length is the total number of elements in the matrix after pooling, p×q. is the matrix after pooling.

[0011] Further, the bidirectional GRU structure includes a forward GRU and a reverse GRU, which are used to control the transmission of information through an update gate and a reset gate, and the update gate and the reset gate are used to determine the retained spatial feature information and the forgotten spatial feature information through a Sigmoid activation function; The spatial feature vector is used as the input of the bidirectional GRU structure in the recurrent neural network to capture the time variation law of the dynamic disturbance and generate a hidden state sequence, including: The spatial feature vector is used as the input of the recurrent neural network to capture the front-end dependency relationship in the time series by recursive updating, so as to obtain a time series input matrix, wherein each moment in the time series input matrix corresponds to a vector; The time series input matrix is ​​processed respectively by the forward GRU and the reverse GRU in time order to obtain the total implicit state output by the bidirectional GRU structure, and the implicit state sequence matrix is ​​generated based on the time variation law of the total implicit state and the dynamic disturbance; The total hidden state is composed of the hidden state of the forward GRU and the hidden state of the reverse GRU, and the hidden state sequence matrix is ​​used to describe the disturbance characteristics of each time step in the time series.

[0012] Furthermore, based on the implicit state sequence, an attention mechanism is introduced into the output layer of the recurrent neural network to focus on the implicit state when the train enters the station, and an optimized disturbance feature representation is generated by weighting, including: The hidden state sequence matrix is ​​used as an input of the attention mechanism introduced by the recurrent neural network to perform a linear transformation on the hidden states in the hidden state sequence matrix and map the high-dimensional state to a hidden layer representation of the attention weight; The inner product of the hidden layer representation and the global context vector is calculated, the attention weight is calculated, and the weighted summation of the hidden state sequence matrix is ​​performed to calculate and generate the perturbation feature representation; Among them, the calculation formula of the disturbance characteristic representation is: In the formula, is the perturbation feature representation vector, is the number of time steps, is the attention weight at the t-th time step, is the implicit state vector at the tth time step, is the global context vector, is the weight matrix, is the hyperbolic tangent activation function, is the bias vector.

[0013] Furthermore, the optimized disturbance feature representation is used as the input of the fully connected layer to generate and output the corresponding energy-saving strategy category, including: Using the perturbation feature representation as an input of a fully connected layer, so as to convert the perturbation feature representation into an input of a classification unit through a linear transformation of the fully connected layer and a nonlinear transformation of an activation function; Calling the classification unit to generate corresponding energy-saving strategy categories according to the number of categories output by the fully connected layer, so as to select energy-saving strategy categories whose category probability exceeds a set threshold as the in-station energy-saving optimization control strategy based on the energy-saving strategy categories; The probability distribution generated by the energy-saving strategy category is obtained by calculating the output of the fully connected layer through the Softmax function, and its expression is: In the formula, is the probability of category c, is the output of the fully connected layer for category c, is the total number of categories.

[0014] A second aspect of the present invention discloses a terminal, including a processor and a storage medium; the characteristics are: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method described in the first aspect.

[0015] The third aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the method described in the first aspect.

[0016] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages: By obtaining the train running status, station environment data and environmental control equipment data, and preprocessing the train running status, station environment data and environmental control equipment data, a time series data matrix is ​​obtained. Then the time series data matrix is ​​decomposed by non-negative matrix decomposition to extract the disturbance characteristics in the time series data matrix, and the blind source separation result matrix is ​​output. Subsequently, the blind source separation result matrix is ​​convolved by a convolutional neural network to extract the spatial pattern of the disturbance characteristics, and a spatial feature vector is generated by pooling dimensionality reduction. After that, the spatial feature vector is used as the input of the bidirectional GRU structure in the recurrent neural network to capture the time variation law of the dynamic disturbance and generate an implicit state sequence. Based on the generated implicit state sequence, an attention mechanism is introduced in the output layer of the recurrent neural network to focus on the implicit state when the train enters the station, and the optimized disturbance feature representation is generated by weighting. Finally, the optimized disturbance feature representation is used as the input of the fully connected layer to generate and output the corresponding energy-saving strategy category. This method extracts physical disturbance characteristics through NMF, combines AI models to process nonlinear dynamic characteristics, combines CNN to capture spatial patterns, and RNN to process time series, forming a more comprehensive feature expression. In addition, by introducing the attention mechanism, key disturbance periods can be highlighted, thereby enhancing the model's ability to respond to complex disturbances, reducing the selection of artificial features, and being able to directly generate energy-saving strategies from raw data, thereby achieving more refined and real-time energy-saving optimization control. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method provided by the present invention. DETAILED DESCRIPTION

[0018] The present application is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present application.

[0019] like Figure 1 As shown, in one embodiment, a subway station environmental control and feng shui linkage energy-saving optimization control method based on deep learning includes the following steps: Step S110, obtaining the train running status, station environment data and environmental control equipment data, and preprocessing the train running status, station environment data and environmental control equipment data to obtain a time series data matrix.

[0020] Among them, the train operation status includes the train speed and the train entry time, the station environment data includes air pressure, temperature and humidity, and the environmental control equipment data includes fan speed, cold water flow and air conditioning refrigeration load.

[0021] In some embodiments, the deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method provided by the present invention obtains the train running status, station environment data and environmental control equipment data, and pre-processes the train running status, station environment data and environmental control equipment data to obtain a time series data matrix, which specifically includes the following steps: Step S111, align the in-station environment data and the environmental control equipment data according to the timestamps at the same time interval to obtain a collection data vector. The expression of the collection data vector is: In the formula, Indicates at time The collected data vector, For time The air pressure data, For time Temperature data, For time Humidity data, For time The fan speed data, For time Cold water flow data, For time Refrigeration load data.

[0022] Step S112, high-frequency noise in the collected data vector is filtered out by low-pass filtering, and signals within a preset frequency range are retained by a band-pass filter, and outliers are removed by using a triple standard deviation rule, and the removed outliers are replaced by linear interpolation.

[0023] Step S113, time window slicing is performed on the collected data vector after filtering, denoising and outlier removal, and the segmented time series is stored in a matrix form to obtain a time series data matrix. The expression of the time series data matrix is: In the formula, is the time series data matrix after segmentation, is the amount of data at the sampling point. In the time series data matrix, each row represents the time series of a variable, and each column represents any time point. All variables.

[0024] Among them, the station environment data and environmental control equipment data are data collected by sensors or equipment arranged according to the environmental control requirements of the subway station, and are used to reflect the pressure waves and thermal disturbances caused by trains entering and leaving the station.

[0025] Step S120, decomposing the time series data matrix by non-negative matrix decomposition to extract disturbance characteristics in the time series data matrix, and outputting a blind source separation result matrix.

[0026] In some embodiments, the deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method provided by the present invention decomposes the time series data matrix by non-negative matrix decomposition to extract the disturbance characteristics in the time series data matrix, and outputs the blind source separation result matrix, which specifically includes the following steps: Step S121, obtain the definition dimension of the time series data matrix, and decompose the time series data matrix by non-negative matrix decomposition based on the definition dimension, and the decomposition expression is: In the formula, is a time series data matrix with the defined dimension m×n; is the feature basis matrix with dimension m×r, which is used to represent the weights of r feature bases, and each row corresponds to m variables; is an activation matrix with a dimension of r×n, which is used to characterize the strength of each feature basis at n time points; r is the number of decomposed features and satisfies .

[0027] Step S122, minimize the matrix reconstruction error by optimizing the objective function, and update the feature basis matrix and the activation matrix by an iterative optimization algorithm, the iterative optimization algorithm includes a multiplication update rule, and the expression of the objective function is: In the formula, is the Frobenius norm, which represents the square root of the sum of the squares of the matrix elements, and , used for non-negativity constraints on matrix elements.

[0028] In some embodiments, the deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method provided by the present invention decomposes the time series data matrix by non-negative matrix decomposition to extract the disturbance characteristics in the time series data matrix, and outputs the blind source separation result matrix, which specifically also includes the following steps: Step S123, analyze the decomposition results of the time series data matrix by non-negative matrix decomposition, and normalize the activation matrix and the characteristic basis matrix respectively, so as to extract the key disturbance moments from the activation matrix based on the maximum value and distribution of the variables in the activation matrix, and extract the key variables by comparing the column vectors of the characteristic basis matrix.

[0029] Step S124, reconstructing and outputting the blind source separation result matrix based on the normalized activation matrix and the eigenvalue matrix.

[0030] Among them, the normalized expression of the activation matrix is: In the formula, is the normalized activation matrix, Represents the maximum value of each row in the activation matrix.

[0031] The normalized expression of the characteristic basis matrix is: In the formula, is the normalized eigenvalue matrix, represents the column sum in the eigenbasis matrix.

[0032] The reconstruction expression of the blind source separation result matrix is: In the formula, is the blind source separation result matrix, with a dimension of m×n. Each column of the blind source separation result matrix represents a time point The corresponding blind source characteristics contain key disturbance information.

[0033] Step S130, performing a convolution operation on the blind source separation result matrix through a convolutional neural network to extract the spatial pattern of the disturbance characteristics, and generating a spatial feature vector through pooling dimensionality reduction.

[0034] Among them, the disturbance characteristics include pressure wave intensity and thermal disturbance pattern, and the spatial pattern includes pressure wave distribution and heat transfer path in different areas of the subway station.

[0035] In some embodiments, the deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method provided by the present invention performs a convolution operation on the blind source separation result matrix through a convolutional neural network to extract the spatial pattern of the disturbance characteristics, and generates a spatial feature vector through pooling dimensionality reduction, specifically including the following steps: Step S131, calling the convolutional neural network to identify the local pattern in the blind source separation result matrix through the convolution operation, and performing convolution calculation based on the defined convolution kernel to obtain the output matrix after convolution.

[0036] Step S132, processing the output matrix after convolution through a nonlinear activation function, compressing the matrix size through a pooling operation, and straightening the pooled matrix to obtain a spatial feature vector.

[0037] Among them, the expression of the spatial eigenvector is: In the formula, is the spatial feature vector, and its length is the total number of elements in the matrix after pooling, p×q. is the matrix after pooling.

[0038] Step S140, using the spatial feature vector as the input of the bidirectional GRU structure in the recurrent neural network to capture the temporal variation law of the dynamic disturbance and generate a hidden state sequence.

[0039] In some embodiments, the deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method provided by the present invention uses the spatial feature vector as the input of the bidirectional GRU structure in the recurrent neural network to capture the time variation law of the dynamic disturbance and generate an implicit state sequence, which specifically includes the following steps: It should be noted that the bidirectional GRU structure includes a forward GRU and a reverse GRU, which are used to control the transmission of information through an update gate and a reset gate. The update gate and the reset gate are used to determine the retained spatial feature information and the forgotten spatial feature information through a Sigmoid activation function.

[0040] Step S141, using the spatial feature vector as the input of the recurrent neural network to capture the front-end dependency in the time series by recursive updating, and obtaining a time series input matrix, in which each moment in the time series input matrix corresponds to a vector.

[0041] Step S142, the time series input matrix is ​​processed respectively in time order by the forward GRU and the reverse GRU to obtain the total implicit state output by the bidirectional GRU structure, and the implicit state sequence matrix is ​​generated based on the time variation law of the total implicit state and the dynamic disturbance.

[0042] Among them, the total hidden state is composed of the hidden state of the forward GRU and the hidden state of the reverse GRU, and the hidden state sequence matrix is ​​used to describe the disturbance characteristics of each time step in the time series.

[0043] Step S150, based on the implicit state sequence, an attention mechanism is introduced into the output layer of the recurrent neural network to focus on the implicit state when the train enters the station, and an optimized disturbance feature representation is generated by weighting.

[0044] In some embodiments, the deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method provided by the present invention introduces an attention mechanism in the output layer of the recurrent neural network based on the implicit state sequence to pay attention to the implicit state when the train enters the station, and generates an optimized disturbance feature representation through weighted weighting, specifically including the following steps: Step S151, using the hidden state sequence matrix as the input of the attention mechanism introduced by the recurrent neural network to perform a linear transformation on the hidden states in the hidden state sequence matrix, and mapping the high-dimensional state to the hidden layer representation of the attention weight.

[0045] Step S152, the inner product of the hidden layer representation and the global context vector is calculated, the attention weight is calculated, and the weighted summation of the hidden state sequence matrix is ​​performed to calculate and generate the perturbation feature representation.

[0046] Among them, the calculation formula of the disturbance characteristic representation is: In the formula, is the perturbation feature representation vector, is the number of time steps, is the attention weight at the t-th time step, is the implicit state vector at the tth time step, is the global context vector, is the weight matrix, is the hyperbolic tangent activation function, is the bias vector.

[0047] Step S160, using the optimized disturbance feature representation as the input of the fully connected layer, generating and outputting the corresponding energy-saving strategy category.

[0048] In some embodiments, the deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method provided by the present invention uses the optimized disturbance feature representation as the input of the fully connected layer to generate and output the corresponding energy-saving strategy category, specifically including the following steps: Step S161, using the perturbation feature representation as the input of the fully connected layer, so as to convert the perturbation feature representation into the input of the classification unit through the linear transformation of the fully connected layer and the nonlinear transformation of the activation function.

[0049] Step S163, calling the classification unit to generate corresponding energy-saving strategy categories according to the number of categories output by the fully connected layer, so as to select energy-saving strategy categories whose category probability exceeds a set threshold as the in-station energy-saving optimization control strategy based on the energy-saving strategy categories.

[0050] Among them, the probability distribution generated by the energy-saving strategy category is obtained by calculating the output of the fully connected layer through the Softmax function, and its expression is: In the formula, is the probability of category c, is the output of the fully connected layer for category c, is the total number of categories.

[0051] The above-mentioned deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method obtains the train running status, station environment data and environmental control equipment data, and pre-processes the train running status, station environment data and environmental control equipment data to obtain a time series data matrix. Then, the time series data matrix is ​​decomposed by non-negative matrix decomposition to extract the disturbance characteristics in the time series data matrix, and the blind source separation result matrix is ​​output. Subsequently, the blind source separation result matrix is ​​convolved by a convolutional neural network to extract the spatial pattern of the disturbance characteristics, and a spatial feature vector is generated by pooling dimensionality reduction. After that, the spatial feature vector is used as the input of the bidirectional GRU structure in the recurrent neural network to capture the time variation law of dynamic disturbances and generate an implicit state sequence. Based on the generated implicit state sequence, an attention mechanism is introduced in the output layer of the recurrent neural network to pay attention to the implicit state when the train enters the station, and the optimized disturbance feature representation is generated by weighting. Finally, the optimized disturbance feature representation is used as the input of the fully connected layer to generate and output the corresponding energy-saving strategy category. This method extracts physical disturbance characteristics through NMF, combines AI models to process nonlinear dynamic characteristics, combines CNN to capture spatial patterns, and RNN to process time series to form a more comprehensive feature expression. In addition, by introducing the attention mechanism, it can highlight key disturbance periods, thereby enhancing the model's ability to respond to complex disturbances, reducing the selection of artificial features, and being able to directly generate energy-saving strategies from raw data, achieving more refined and real-time energy-saving optimization control.

[0052] In a specific embodiment, the deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method provided by the present invention includes steps 1 to 6: Step 1: Data collection and preprocessing.

[0053] Collect train operation status (such as speed, entry time), environmental data (air pressure, temperature and humidity) and environmental control equipment data (fan speed, cold water flow, air conditioning refrigeration load), filter and denoise the collected data, remove outliers and segment the time series to form a high-quality time series data matrix.

[0054] Specifically, it includes steps 1.1 to 1.4: Step 1.1, data collection equipment deployment.

[0055] According to the environmental control requirements of the subway station, sensors and equipment are arranged to collect data to reflect the pressure waves and thermal disturbances caused by the entry and exit of trains, specifically including: environmental data and environmental control equipment data.

[0056] Step 1.2: Data is synchronized with timestamp.

[0057] The raw data of each acquisition device are aligned by timestamp, and the sampling time interval is unified to △t seconds. Let the time series t=t0, t1, ..., t n , the time interval satisfies t i+1 -t i =△t, the typical value is △t=1s, so the expression of the collected data vector is: In the formula, Indicates at time The collected data vector, For time Air pressure data, unit: Pa; For time Temperature data, unit: °C; For time Humidity data, unit: % For time Fan speed data, unit RPM; For time Cold water flow data, unit: m³ / h; For time Cooling load data in kW.

[0058] Step 1.3: Data denoising and outlier processing.

[0059] Specifically, yes Perform filtering and outlier processing: Low-pass filtering: Filter out high-frequency noise and use a band-pass filter to preserve the frequency range to signal.

[0060] Outlier elimination: Use the three-times standard deviation rule. Then remove the outliers and replace them with linear interpolation. In the formula, is the sequence mean, is the series standard deviation.

[0061] Step 1.4, time series segmentation.

[0062] right The time window is sliced, each segment is L in length, corresponding to M sampling points, that is, L = M × △t, and the segmented event sequence is stored in the following matrix form: In the formula, is the time series data matrix after segmentation, is the amount of data at the sampling point. In the time series data matrix, each row represents the time series of a variable (such as air pressure, temperature, etc.), and each column represents a time point. All variables.

[0063] Step 2: Data decomposition and blind source feature extraction.

[0064] Non-negative matrix factorization (NMF) is used to decompose the preprocessed data, extract key disturbance characteristics (such as pressure wave intensity and thermal disturbance pattern), and output the blind source separation result matrix.

[0065] Specifically, it includes steps 2.1 to 2.3: Step 2.1, introduce non-negative matrix factorization (NMF).

[0066] It should be noted that non-negative matrix factorization (NMF) is a method that decomposes a non-negative matrix into the product of two non-negative matrices to extract hidden features, namely blind source separation.

[0067] Specifically, let the input be a time series data matrix , its dimension is defined as m×n, and it is decomposed into the following form through NMF: In the formula, is a time series data matrix with the defined dimension m×n; is the feature basis matrix with dimension m×r, which is used to represent the weights of r feature bases, and each row corresponds to m variables; is an activation matrix with a dimension of r×n, which is used to characterize the strength of each feature basis at n time points; r is the number of decomposed features and satisfies .

[0068] In this embodiment, the decomposition goal of NMF is to minimize the matrix reconstruction error by optimizing the following objective function: In the formula, is the Frobenius norm, which represents the square root of the sum of the squares of the matrix elements, and , used for non-negativity constraints on matrix elements.

[0069] Update by iterative optimization method (such as multiplicative update rule) and : Step 2.2, extract key disturbance features.

[0070] Analyze the NMF decomposition results and extract key perturbation characteristics from the activation function H: Each row of H represents the intensity change of a characteristic basis over time. By observing the position and distribution of the maximum value in H, the key disturbance moments, such as the pressure wave peak and the heat flow peak, are determined. Each column in the characteristic basis matrix W represents the contribution of different variables to a certain characteristic basis. By comparing the column vectors of W, variables with high contributions, such as fan speed and air pressure, are extracted.

[0071] In this embodiment, the normalization expression of the activation matrix is: In the formula, is the normalized activation matrix, represents the maximum value of each row in the activation matrix H.

[0072] The characteristic basis matrix W is normalized to highlight the contribution of the variables. The normalized expression of the characteristic basis matrix is: In the formula, is the normalized feature basis matrix, each column is normalized by column and, represents the column sum in the eigenbasis matrix.

[0073] Step 2.3, construct the blind source separation result matrix.

[0074] Combination and , reconstruct the blind source separation result matrix , the reconstruction expression of the blind source separation result matrix is: In the formula, is the blind source separation result matrix, with a dimension of m×n. Each column of the blind source separation result matrix represents a time point The corresponding blind source characteristics contain key disturbance information.

[0075] Blind source separation result matrix The meaning of the elements is expressed as: In the formula, is the i-th variable at time t j The blind source characteristic value of is the contribution value of the i-th variable in the characteristic basis r, is the characteristic basis r at time point t j The activation value of .

[0076] Step 3: spatial feature extraction.

[0077] A convolutional neural network (CNN) is used to perform convolution operations on the blind source separation result matrix to extract the spatial pattern of disturbance characteristics, such as pressure wave distribution and heat transfer path in different regions, and spatial feature vectors are generated through pooling dimensionality reduction.

[0078] Specifically, it includes steps 3.1 to 3.5: Step 3.1, introduce convolutional neural network (CNN).

[0079] Convolutional neural network (CNN) is a deep learning model used to extract spatial characteristics of data. It identifies local patterns in the blind source separation result matrix S, such as pressure wave distribution and heat transfer path, through convolution operation.

[0080] In this embodiment, represents the i-th variable at time point t j The perturbation characteristics on the input are the blind source separation result matrix S output in step 2, whose dimension is m×n: Step 3.2, convolution operation.

[0081] Define the convolution kernel K, with a size of p×q: Convolution kernel weights are It is initialized to a random value and adjusted through training.

[0082] The convolution calculation formula is: In the formula, is the output after convolution, indicating the The eigenvalues ​​at the positions, is the starting position index of the convolution window. The convolution kernel sliding step s determines the step size of each movement, and the typical value is s=1.

[0083] Step 3.3, activation function.

[0084] The matrix after convolution After being processed by a nonlinear activation function (such as ReLU) to increase the nonlinear expression ability of the model, the activation expression is: In the formula, is the activated matrix element, indicating the Nonlinear characteristics of position.

[0085] Step 3.4: Pooling dimensionality reduction.

[0086] The pooling operation extracts the significant features of the local area by compressing the matrix size and reducing the amount of calculation. The pooling window size is defined as p p ×q p , the typical value is 2×2, and the pooling operation formula takes the maximum pooling as an example: In the formula, is the matrix element after pooling, is the index within the pooling window. Pooling step length s p Determines the pooling window moving step size, the typical value is s p =2.

[0087] Step 3.5, spatial feature vector generation.

[0088] The pooled matrix Straighten to spatial eigenvector , the expression is: In the formula, is the spatial feature vector, and its length is the total number of elements in the matrix after pooling, p×q. is the matrix after pooling, is the first The final output spatial feature vector describes the main spatial pattern features extracted from the blind source separation matrix.

[0089] Step 4: Time series modeling.

[0090] The spatial feature vector is input into the bidirectional GRU structure in the recurrent neural network (RNN) to capture the temporal variation of dynamic disturbances, such as the transmission speed of pressure waves and the duration of thermal disturbances, and generate a hidden state sequence.

[0091] Specifically, it includes steps 4.1 to 4.4: Step 4.1, introduce the recurrent neural network (RNN).

[0092] Recurrent Neural Network (RNN) is a deep learning model for processing sequence data. It captures the dependencies between the previous and next time series by recursive updating. Its input is the spatial feature vector generated in step 3. , at every moment t i Corresponding to a vector x t ,Right now: in, The input matrix for the time series is T×d, where T is the number of time steps, d is the dimension of the feature vector, and x t is the input vector at time t.

[0093] Step 4.2, bidirectional GRU structure.

[0094] GRU (Gated Recurrent Unit) is an improved RNN unit that uses update gates and reset gates to control the transmission of information. The bidirectional GRU structure includes two independent GRU units: Forward GRU: Process the sequence from t=1 to t=T.

[0095] Reverse GRU: Process the sequence from t=T to t=1.

[0096] Among them, the state update formula of each GRU unit is as follows: The update gate determines how much previous information to keep, and the expression is: The reset gate determines how much previous information is forgotten, and the expression is: Candidate state: Calculate the candidate hidden state of the current time step, the expression is: The implicit state update expression is: in, , are the outputs of the update gate and the reset gate, respectively, and the range is , is the implicit state vector of the current time step, is a candidate hidden state, , , They are the weight matrix, the matrix randomly initialized, and the matrix learned through training. is the Sigmoid activation function, and the range is , is the hyperbolic tangent function, and its range is .

[0097] The final output of the bidirectional GRU is: In the formula, is the hidden state of the forward GRU, is the hidden state of the reverse GRU.

[0098] Step 4.3, capture the time variation law of dynamic disturbance.

[0099] The event change patterns of dynamic disturbances (such as pressure wave transmission and thermal disturbance duration) are captured by the implicit state sequence of the bidirectional GRU: in, , represents the implicit state at time t, is the hidden state sequence matrix, with dimension , is the hidden state dimension of the GRU unit.

[0100] Through the time dimension of H, we can capture the propagation law of disturbances over time, for example: Pressure wave transmission: Identify the intensity and propagation speed of fluctuations from the changing trend of the implicit state.

[0101] Thermal perturbation duration: The duration of the perturbation effect is determined by the degree of activation of consecutive latent states.

[0102] Step 4.4, hidden state sequence representation.

[0103] The final hidden state sequence matrix H describes the disturbance characteristics of each time step in the time series: in, is the implicit state value of the i-th dimension at the t-th time step, indicating the contribution of this dimension to the disturbance of the current time step, is the number of time steps, is the implicit state dimension of the bidirectional RGU.

[0104] Step 5: Attention mechanism optimization.

[0105] Based on the hidden state sequence, the attention mechanism is introduced in the output layer of RNN, focusing on the hidden state at key time points (such as the moment when the train enters the station), and generating optimized disturbance feature representation through weighted weighting.

[0106] Specifically, it includes steps 5.1 to 5.5: Step 5.1, introduce the attention mechanism.

[0107] The attention mechanism is a technique for optimizing sequence modeling. It highlights the hidden state contribution of key time points by calculating the importance weight of each time step in the time series. Its input is the hidden state sequence matrix H generated in step 4.

[0108] Step 5.2, calculate the weights of key time points.

[0109] First, the implicit state is transformed, that is, the implicit state Apply a linear transformation to map the high-dimensional state to a hidden representation of the attention weights: In the formula, is the hidden layer representation vector, with dimension , is the weight matrix, with dimension , is the bias vector, with dimension , is the hyperbolic tangent activation function, which is used to introduce nonlinearity.

[0110] Secondly, calculate the weight vector. With the global context vector Inner product, calculate the attention score : In the formula, is the global context vector with dimension , used to measure the correlation of time steps, is the attention weight of the t-th time step, ranging from , and all The sum of is 1.

[0111] Step 5.3, calculate the perturbation feature representation.

[0112] The perturbation feature representation is the result of weighted summation of the implicit state sequence H, highlighting the contribution of key time points. Its expression is: In the formula, is the perturbation feature representation vector, with dimension , is the implicit state vector at the tth time step, is the attention weight at the t-th time step.

[0113] Final Output Represents the optimized features of the entire time series and can capture the impact of disturbances at key time points.

[0114] Step 5.4, describe the perturbation feature representation.

[0115] Perturbation feature representation The specific form is: The i-th eigenvalue of the perturbation feature representation vector is the characteristic component obtained by time-weighted summation, such as: in, represents the i-th hidden state value at the t-th time step.

[0116] Step 5.5, summarize the attention mechanism optimization.

[0117] In the implicit state of the key time point, the key time point is weighted by the attention mechanism Improve its implicit status In the process of capturing dynamic disturbances, Weighted Hidden State , perturbation feature representation It can accurately capture the dynamic changes of different time steps in the time series.

[0118] Therefore, the final formula for attention mechanism optimization is as follows: In the formula, is the perturbation feature representation vector, is the number of time steps, is the attention weight at the t-th time step, is the implicit state vector at the tth time step, is the global context vector, is the weight matrix, is the hyperbolic tangent activation function, is the bias vector.

[0119] Step 6: Energy-saving strategy generation and classification.

[0120] The optimized feature representation is input into the fully connected layer and combined with the classification module to generate specific energy-saving strategy categories (such as adjusting fan speed and optimizing cold water supply).

[0121] Input: Optimized perturbation feature representation.

[0122] Output: Energy saving strategy category.

[0123] Specifically, it includes steps 6.1 to 6.3: Step 6.1, introduce the fully connected layer.

[0124] The fully connected layer is a neural network layer used to combine input features. Its function is to represent the optimized features. Converted into the input of the classification module, whose input is the perturbation feature representation output by step 5, with a dimension of Through the linear transformation of the fully connected layer and the nonlinear transformation of the activation function, the input features are represented as the input of the classification module, and the formula is: in, is the output of the fully connected layer, with a dimension of C, i.e. the number of categories in the classification module, is the weight matrix, with dimension , used for feature weighting, is a bias vector with dimension C.

[0125] Step 6.2, classification module.

[0126] The classification module is used to The value of generates the energy-saving strategy category, which refers to a specific control adjustment method, such as adjusting the fan speed, optimizing the chilled water supply, etc.

[0127] In this embodiment, the classification module uses the Softmax classifier to call the Softmax function to convert the fully connected layer output Transformed into probability distribution of each category: In the formula, is the probability of category c, and the range is , is the output of the fully connected layer for category c, is the total number of categories.

[0128] Then the category is determined and the category with the highest probability is selected as the final energy-saving strategy: in, is the predicted energy-saving strategy category index, To return The category index corresponding to the maximum value.

[0129] Step 6.3, energy saving strategy category definition.

[0130] Energy saving strategy category Describe the specific control adjustment behavior, such as: Category c=1: Increase fan speed.

[0131] Category c=2: Reduce fan speed.

[0132] Category c=3: Optimizing the cold water supply.

[0133] Category c=4: Adjustment of air conditioning cooling load. ...... The classification results are associated with specific control devices and executed through the device interface.

[0135] Beneficial effects: Combining mechanism and data-driven: extract physical disturbance characteristics through NMF, and combine AI model to process nonlinear dynamic characteristics.

[0136] Joint modeling of spatial-temporal features: CNN captures spatial patterns and RNN processes temporal sequences to form a comprehensive feature expression.

[0137] Attention optimization: highlight key disturbance periods and enhance the model's ability to respond to complex disturbances.

[0138] End-to-end optimization: Reduce manual feature selection and directly generate energy-saving strategies from raw data.

[0139] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0140] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0141] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0142] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0143] Various aspects of the present disclosure are described herein with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of boxes in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0144] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0145] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0146] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A subway station environmental control and feng shui linkage energy-saving optimization control method based on deep learning, characterized in that: The method comprises: Acquire train operation status, station environment data and environmental control equipment data, and pre-process the train operation status, station environment data and environmental control equipment data to obtain a time series data matrix; Decomposing the time series data matrix by non-negative matrix decomposition to extract disturbance characteristics in the time series data matrix, and outputting a blind source separation result matrix; Performing a convolution operation on the blind source separation result matrix through a convolutional neural network to extract the spatial pattern of the disturbance characteristics, and generating a spatial feature vector through pooling dimensionality reduction; The spatial feature vector is used as the input of a bidirectional GRU structure in a recurrent neural network to capture the temporal variation of the dynamic disturbance and generate a hidden state sequence; Based on the implicit state sequence, an attention mechanism is introduced into the output layer of the recurrent neural network to focus on the implicit state when the train enters the station, and an optimized disturbance feature representation is generated through weighting; The optimized perturbation feature representation is used as the input of the fully connected layer to generate and output the corresponding energy-saving strategy category; Among them, the train operation status includes the train speed and the train entry time, the station environment data includes air pressure, temperature and humidity, the environmental control equipment data includes fan speed, cold water flow and air conditioning refrigeration load, the disturbance characteristics include pressure wave intensity and thermal disturbance pattern, and the spatial pattern includes pressure wave distribution and heat transfer path in different areas of the subway station.

2. The deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method according to claim 1 is characterized in that: The station environment data and environmental control equipment data are data collected by sensors or equipment arranged according to the environmental control requirements of the subway station, and are used to reflect the pressure waves and thermal disturbances caused by trains entering and leaving the station; The train running status, station environment data and environmental control equipment data are obtained, and the train running status, station environment data and environmental control equipment data are preprocessed to obtain a time series data matrix, including: The station environment data and the environmental control equipment data are aligned according to the timestamps at the same time interval to obtain a collection data vector. The expression of the collection data vector is: ； In the formula, Indicates at time The collected data vector, For time The air pressure data, For time Temperature data, For time Humidity data, For time The fan speed data, For time Cold water flow data, For time Refrigeration load data; The high-frequency noise in the collected data vector is filtered out by low-pass filtering, the signal within a preset frequency range is retained by a band-pass filter, and outliers are removed by a triple standard deviation rule, and the removed outliers are replaced by linear interpolation; The collected data vector after filtering, denoising and outlier removal is sliced ​​by time window, and the segmented time series is stored in matrix form to obtain the time series data matrix. The expression of the time series data matrix is: ； In the formula, is the time series data matrix after segmentation, is the amount of data at the sampling point. In the time series data matrix, each row represents the time series of a variable, and each column represents any time point. All variables.

3. The deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method according to claim 2 is characterized in that: The step of decomposing the time series data matrix by non-negative matrix decomposition to extract disturbance characteristics in the time series data matrix and outputting a blind source separation result matrix comprises: Obtain the definition dimension of the time series data matrix, and decompose the time series data matrix based on the definition dimension by the non-negative matrix decomposition, the decomposition expression is: ； In the formula, is the time series data matrix, with a dimension of m×n; is the feature basis matrix with dimension m×r, which is used to represent the weights of r feature bases, and each row corresponds to m variables; is an activation matrix with a dimension of r×n, which is used to characterize the strength of each feature basis at n time points; r is the number of decomposed features and satisfies ; The matrix reconstruction error is minimized by optimizing the objective function, and the characteristic basis matrix and the activation matrix are updated by an iterative optimization algorithm, wherein the iterative optimization algorithm includes a multiplication update rule, and the expression of the objective function is: ； In the formula, is the Frobenius norm, which represents the square root of the sum of the squares of the matrix elements, and , used for non-negativity constraints on matrix elements.

4. The deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method according to claim 3 is characterized in that: The method of decomposing the time series data matrix by non-negative matrix decomposition to extract disturbance characteristics in the time series data matrix and outputting a blind source separation result matrix also includes: Analyzing the decomposition result of the time series data matrix by the non-negative matrix decomposition, and normalizing the activation matrix and the characteristic basis matrix respectively, so as to extract the key disturbance moment from the activation matrix based on the maximum value and distribution of the variables in the activation matrix, and extracting the key variables by comparing the column vectors of the characteristic basis matrix; Reconstructing and outputting the blind source separation result matrix based on the normalized activation matrix and the feature basis matrix; Among them, the normalized expression of the activation matrix is: ； In the formula, is the normalized activation matrix, represents the maximum value of each row in the activation matrix; The normalized expression of the characteristic basis matrix is: ； In the formula, is the normalized eigenvalue matrix, represents the column sum in the feature basis matrix; The reconstruction expression of the blind source separation result matrix is: ； In the formula, is the blind source separation result matrix, with a dimension of m×n, and each column of the blind source separation result matrix represents a time point The corresponding blind source characteristics contain key disturbance information.

5. The deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method according to claim 4 is characterized in that: The convolution operation is performed on the blind source separation result matrix by a convolutional neural network to extract the spatial pattern of the disturbance characteristics, and a spatial feature vector is generated by pooling and dimensionality reduction, including: Calling the convolutional neural network to identify local patterns in the blind source separation result matrix through a convolution operation, and performing convolution calculation based on a defined convolution kernel to obtain a convolution output matrix; The output matrix after the convolution is processed by a nonlinear activation function, and the size of the matrix is ​​compressed by a pooling operation, and the pooled matrix is ​​straightened to obtain the spatial feature vector; Wherein, the expression of the spatial feature vector is: ； In the formula, is the spatial feature vector, and its length is the total number of elements in the matrix after pooling, p×q. is the matrix after pooling.

6. The deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method according to claim 5 is characterized in that: The bidirectional GRU structure includes a forward GRU and a reverse GRU, which are used to control the transmission of information through an update gate and a reset gate, and the update gate and the reset gate are used to determine the retained spatial feature information and the forgotten spatial feature information through a Sigmoid activation function; The spatial feature vector is used as the input of the bidirectional GRU structure in the recurrent neural network to capture the time variation law of the dynamic disturbance and generate a hidden state sequence, including: The spatial feature vector is used as the input of the recurrent neural network to capture the front-end dependency relationship in the time series by recursive updating, so as to obtain a time series input matrix, wherein each moment in the time series input matrix corresponds to a vector; The time series input matrix is ​​processed respectively by the forward GRU and the reverse GRU in time order to obtain the total implicit state output by the bidirectional GRU structure, and the implicit state sequence matrix is ​​generated based on the time variation law of the total implicit state and the dynamic disturbance; The total hidden state is composed of the hidden state of the forward GRU and the hidden state of the reverse GRU, and the hidden state sequence matrix is ​​used to describe the disturbance characteristics of each time step in the time series.

7. The deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method according to claim 6 is characterized in that: Based on the implicit state sequence, an attention mechanism is introduced into the output layer of the recurrent neural network to focus on the implicit state when the train enters the station, and an optimized disturbance feature representation is generated by weighting, including: Using the hidden state sequence matrix as the input of the attention mechanism introduced by the recurrent neural network to perform a linear transformation on the hidden states in the hidden state sequence matrix and map the high-dimensional states to the hidden layer representation of the attention weights; The inner product of the hidden layer representation and the global context vector is calculated, the attention weight is calculated, and the weighted summation of the hidden state sequence matrix is ​​performed to calculate and generate the perturbation feature representation; Among them, the calculation formula of the disturbance characteristic representation is: ； In the formula, is the perturbation feature representation vector, is the number of time steps, is the attention weight at the t-th time step, is the implicit state vector at the tth time step, is the global context vector, is the weight matrix, is the hyperbolic tangent activation function, is the bias vector.

8. The deep learning-based subway station environmental control and feng shui linkage energy-saving optimization control method according to claim 7 is characterized in that: The optimized perturbation feature representation is used as the input of the fully connected layer to generate and output the corresponding energy-saving strategy category, including: Using the perturbation feature representation as an input of a fully connected layer, so as to convert the perturbation feature representation into an input of a classification unit through a linear transformation of the fully connected layer and a nonlinear transformation of an activation function; Calling the classification unit to generate corresponding energy-saving strategy categories according to the number of categories output by the fully connected layer, so as to select energy-saving strategy categories whose category probability exceeds a set threshold as the in-station energy-saving optimization control strategy based on the energy-saving strategy categories; The probability distribution generated by the energy-saving strategy category is obtained by calculating the output of the fully connected layer through the Softmax function, and its expression is: ； In the formula, is the probability of category c, is the output of the fully connected layer for category c, is the total number of categories.

9. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images