A subway station energy consumption prediction method based on hybrid deep learning algorithm

Abstract

The present invention discloses a method for predicting energy consumption of subway stations based on a hybrid deep learning algorithm, which relates to the technical field of traffic energy consumption prediction. The present invention determines the input features of the model based on mutual information, realizes feature dimension reduction, and optimizes the hyperparameters of the TCN‑BiLSTM model and decomposes the energy consumption data in combination with the Hippo optimization algorithm improved by multi-strategy fusion to obtain more accurate prediction results; the method of the present invention predicts the energy consumption of subway stations by hours, days, and months, and is applied to the energy consumption management system of subway station operations, so as to assist the subway operation department to understand the energy consumption demand and energy consumption fluctuations of different stations in advance, reasonably allocate resources, and optimize energy management.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic energy consumption prediction, and in particular to a method for predicting energy consumption of a subway station based on a hybrid deep learning algorithm. Background Art

[0002] As an efficient and convenient urban rail transit mode, subway has been widely used and developed rapidly around the world. The operation of subway system consumes a lot of energy, including electricity and water, among which electricity consumption is particularly prominent, mainly used for the operation of train traction, station lighting, ventilation and air conditioning, elevators and escalators and other equipment. The energy consumption of subway stations accounts for a considerable proportion of the energy consumption of the entire subway system.

[0003] The energy consumption of subway stations is affected by a variety of complex factors, such as the building structure and scale of the station, changes in passenger flow, the number of equipment and operating modes, seasons and climate conditions, etc., making accurate prediction of the energy consumption of subway stations a very challenging task. At present, the existing subway station energy consumption prediction methods mainly include methods based on statistical analysis, methods based on physical models, and methods based on artificial intelligence. The statistical analysis-based method predicts future energy consumption by establishing a linear regression model between energy consumption and various influencing factors. This method is simple but has limited prediction accuracy and is difficult to accurately capture the law of energy consumption changes. The physical model-based method establishes a detailed mathematical model based on the physical operating principles of each device in the subway station and predicts energy consumption through simulation calculations. This method can deeply describe the energy transfer and conversion process within the station system, but the model solution is complex and has poor universality for different stations, and is subject to many restrictions in practical applications. With the continuous development of artificial intelligence, artificial intelligence methods have been widely studied and applied in the field of energy consumption prediction, with high prediction accuracy, but it is difficult to intuitively understand the degree of influence of various factors on energy consumption, and the model has poor interpretability, which to a certain extent limits its promotion and application in practical engineering. Summary of the invention

[0004] The purpose of the present invention is to provide a subway station energy consumption prediction method and system based on a hybrid deep learning algorithm to improve the above technical problems.

[0005] In order to achieve the above-mentioned object of the invention, the embodiment of the present invention provides the following technical solutions:

[0006] A method for predicting energy consumption of subway stations based on a hybrid deep learning algorithm, characterized by comprising:

[0007] Obtain the original energy consumption data and original related influencing factor data of a subway station and perform preprocessing to obtain energy consumption data and related influencing factor data;

[0008] Use the MSIHO-VMD algorithm to perform sequence decomposition on energy consumption data and obtain subsequence decomposition data;

[0009] The hybrid deep learning algorithm is used to process the subsequence decomposition data and related influencing factor data to obtain the energy consumption prediction results of the subway station.

[0010] Furthermore, the original energy consumption data is historical energy consumption data; the original relevant influencing factor data include building factors, equipment factors, passenger flow factors, meteorological factors and time factors; the building factors include whether the subway station is a transfer station, the station layout method, the station building area, and the number of station entrances and exits; the equipment factors include the number of air conditioners, air conditioner power, the number of escalators, escalator power, the number of lighting lamps, and lighting lamp power; the passenger flow factors include the station passenger flow; the meteorological factors include the temperature and humidity of the subway station; the time factors include time point, week type, and holiday type.

[0011] Furthermore, the preprocessing includes:

[0012] use The method processes the original energy consumption data and the original related influencing factor data, determines the corresponding abnormal values ​​and removes them, and obtains the updated original energy consumption data and the updated original related influencing factor data;

[0013] The outliers are processed by using the sliding window averaging method and merged into the corresponding updated original energy consumption data and updated original related influencing factor data to obtain the initial energy consumption data and the initial related influencing factor data;

[0014] The initial energy consumption data and initial related influencing factor data are normalized to obtain energy consumption data and related influencing factor data.

[0015] Furthermore, the use of the MSIHO-VMD algorithm to perform sequence decomposition on the energy consumption data to obtain subsequence decomposition data includes:

[0016] The VMD variational mode decomposition method is used to decompose the energy consumption data to obtain simple subsequence data;

[0017] Use the simple subsequence data as the hippo population and set the initial parameters of the hippo population;

[0018] The Latin hypercube sampling method is used to initialize the hippo population, and the fitness of each hippo is calculated based on the IMF component of the simple subsequence data to determine the current optimal solution;

[0019] Use the adaptive weight strategy to iterate the exploration of each hippopotamus, generate candidate positions, calculate the corresponding fitness, and update the position of each hippopotamus;

[0020] The updated hippopotamus are optimized using the golden sine search strategy, the positions of each hippopotamus are updated, and the initial optimal solution is determined;

[0021] The initial optimal solution is optimized using the Cauchy-Gauss mutation, and the corresponding fitness is calculated based on the IMF component corresponding to the initial optimal solution to determine the global optimal solution.

[0022] Determine whether the optimal solution meets the termination condition; if so, use the global optimal solution as the subsequence decomposition data; otherwise, iterate again.

[0023] Furthermore, the hybrid deep learning algorithm adopts an energy consumption prediction model, including a MIC feature selection layer and a TCN-BiLSTM model connected in series; the TCN-BiLSTM model includes a TCN sub-model and a BiLSTM sub-model connected in series;

[0024] The hybrid deep learning algorithm includes:

[0025] The sequence decomposition data and related influencing factor data are input into the MIC feature selection layer, and the corresponding energy consumption feature data set is output;

[0026] The MSIHO algorithm is used to train the TCN-BiLSTM model to obtain a trained TCN-BiLSTM model;

[0027] The energy consumption characteristic data set is input into the optimized TCN-BiLSTM model to obtain the energy consumption prediction result of the subway station.

[0028] Furthermore, the MIC feature selection layer includes:

[0029] The subsequence decomposition data and the relevant influencing factor data are spliced ​​to obtain the initial energy consumption characteristic data set;

[0030] Calculate the mutual information value of each initial energy consumption characteristic data and arrange them in descending order to obtain the mutual information value sorting result;

[0031] Based on the mutual information value sorting results, each initial energy consumption feature data is input into the TCN-BiLSTM model in turn to calculate the corresponding MAPE index value;

[0032] Energy consumption characteristic data is selected based on the MAPE indicator value to obtain an energy consumption characteristic data set.

[0033] Furthermore, the training process of the TCN-BiLSTM model includes:

[0034] Obtaining a training energy consumption characteristic data set and an optimization energy consumption characteristic data set;

[0035] The optimized energy consumption feature data set is input into the TCN-BiLSTM model, and the MSIHO algorithm is used to determine the optimal parameter combination of the TCN-BiLSTM model;

[0036] Apply the optimal parameter combination to the TCN-BiLSTM model to obtain the optimized TCN-BiLSTM model;

[0037] Input each training energy consumption feature data in the training energy consumption feature data set into the optimized TCN-BiLSTM model to obtain the corresponding initial training energy consumption prediction result;

[0038] Each initial training energy consumption prediction result is superimposed and reconstructed to obtain the corresponding training energy consumption prediction result;

[0039] Based on the training energy consumption prediction results, calculate the corresponding evaluation indicators;

[0040] Determine whether the evaluation index meets the training threshold condition; if so, complete the training of the TCN-BiLSTM model; otherwise, retrain.

[0041] The beneficial effects of the present invention are:

[0042] The present invention determines the input features of the model based on mutual information, realizes feature dimensionality reduction, and combines the Hippo optimization algorithm improved by multi-strategy fusion to optimize the hyperparameters of the TCN-BiLSTM model and decompose the energy consumption data, and decomposes the original energy consumption data into several different simple subsequences to obtain more accurate prediction results; the method of the present invention is used to predict the energy consumption of subway stations by hours, days and months, and is applied to the energy consumption management system of subway station operations, so as to assist the subway operation department to understand the energy consumption requirements and energy consumption fluctuations of different stations in advance, reasonably allocate resources, and optimize energy management. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 A flow chart of a method in an embodiment of the present invention;

[0045] Figure 2 It is a flow chart of the improved Hippo optimization algorithm in an embodiment of the present invention;

[0046] Figure 3 This is a flow chart of the hybrid deep learning algorithm in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0048] See also Figure 1 This embodiment provides a method for predicting energy consumption of a subway station based on a hybrid deep learning algorithm, which is characterized by comprising:

[0049] S1. Obtain the original energy consumption data and original related influencing factor data of a subway station and perform preprocessing to obtain energy consumption data and related influencing factor data;

[0050] The original energy consumption data is historical energy consumption data, which is the energy consumption data of the subway station on the previous day, and the collection granularity is 1 hour. The original relevant influencing factor data includes building factors, equipment factors, passenger flow factors, meteorological factors and time factors; the building factors include whether the subway station is a transfer station, the station layout method, the station building area, and the number of station entrances and exits; the equipment factors include the number of air conditioners, air conditioner power, the number of escalators, escalator power, the number of lighting lamps, and the lighting lamp power; the passenger flow factors include the passenger flow of the station; the meteorological factors include the temperature and humidity of the subway station; the time factors include the time point, week type, and holiday type.

[0051] The pre-processing comprises:

[0052] S1-1. Utilization The method processes the original energy consumption data and the original related influencing factor data, determines the corresponding abnormal values ​​and removes them, and obtains the updated original energy consumption data and the updated original related influencing factor data; when the original energy consumption data or the original related influencing factor data is in If the data is out of the range, the data is considered abnormal data (abnormal value). represents the mean, represents the standard deviation. Abnormal data include The original energy consumption data and original related influencing factor data outside the scope.

[0053] S1-2, using the sliding window averaging method to process the outliers, and setting a fixed window by the sliding window averaging method, the size of the fixed window is 24; using the fixed window to process the outliers, and merge them into the corresponding updated original energy consumption data and the updated original related influencing factor data, to obtain the initial energy consumption data and the initial related influencing factor data;

[0054] S1-3. Since the initial energy consumption data and the initial relevant influencing factor data have different dimensions, and too large or too small sample data may make calculations difficult, increase the model calculation intensity and affect the prediction accuracy, the initial energy consumption data and the initial relevant influencing factor data are normalized to obtain energy consumption data and relevant influencing factor data.

[0055] S2. Use the MSIHO-VMD algorithm to perform sequence decomposition on the energy consumption data to obtain subsequence decomposition data; the MSIHO-VMD algorithm includes the MSIHO algorithm and the VMD algorithm, and the MSIHO algorithm is an improved Hippo optimization algorithm.

[0056] The Latin hypercube, adaptive weight strategy, golden sine search strategy and Cauchy-Gaussian mutation multi-strategy fusion are introduced to improve the Hippo optimization algorithm, so that Figure 2 As shown, the improved Hippo optimization algorithm includes:

[0057] The Latin hypercube sampling method is used to initialize the population individuals, and the interval is evenly divided into n parts. A point is randomly selected in each part as a sampling point to make the population distribution more uniform and improve the global performance of the algorithm. In the exploration stage, the global search ability is stronger when the weight of the algorithm is large at the beginning. As the number of iterations increases, the weight coefficient gradually decreases. The adaptive weight strategy can be used to conduct a fine search around the optimal solution, thereby improving the convergence speed of the algorithm. In the escape stage, the golden sine search strategy is introduced to help the hippopotamus gradually approach the optimal solution during the optimization process and speed up the optimization process. Finally, the Cauchy-Gaussian mutation is introduced to help individuals escape from the local optimal solution and find the optimal value.

[0058] Furthermore, the use of the MSIHO-VMD algorithm to perform sequence decomposition on the energy consumption data to obtain subsequence decomposition data includes:

[0059] S2-1. Decompose the energy consumption data using the VMD variational mode decomposition method to obtain simple subsequence data; the simple subsequence data includes multiple simple subsequences; each simple subsequence is a set of VMD parameters, including the mode number K and the penalty factor .

[0060] S2-2, use the simple subsequence data as the hippo population, each simple subsequence as a hippo, each river represents a candidate solution, and set the initial parameters of the hippo population;

[0061] S2-3. Use Latin hypercube sampling method to initialize the hippo population, calculate the fitness of each hippo based on the IMF component of simple subsequence data, and determine the current optimal solution; the fitness is the envelope entropy of the IMF component. In order to compare the effectiveness of the randomly initialized population and the initialized population, the population size is set to 100.

[0062] The Latin hypercube sampling formula corresponding to the Latin hypercube sampling method is:

[0063] ;

[0064] in, represents a matrix function, , They represent the upper and lower bounds of the matrix respectively. Represents the sampling result of the tth iteration, i.e., Hippo.

[0065] The formula corresponding to the envelope entropy is:

[0066] ;

[0067] ;

[0068] in, represents the envelope entropy, represents the jth envelope signal, Represents the envelope signal The normalized form of represents the sum function, N represents the number of hippopotamus population, Represents the logarithmic function with a constant base 10.

[0069] S2-4, using the adaptive weight strategy to explore and iterate each hippo, generate candidate positions and calculate the corresponding fitness, for each candidate solution, calculate the envelope entropy value of the corresponding IMF component, and update the position of each hippo;

[0070] The formula corresponding to the adaptive weight strategy is:

[0071] ;

[0072] ;

[0073] Among them, t and T represent the tth iteration and the total number of iterations respectively. , Respectively represent the updated hippopotamus position and the original hippopotamus position, represents the weight factor of the adaptive weight strategy, e represents a natural constant, Represents a random number in the interval [0, 1], Indicates the position of the dominant male hippopotamus, represents a random number in the interval [1, 2]. The dominant male hippopotamus is the individual with the lowest fitness in the population.

[0074] S2-5, optimize the updated hippopotamus using the golden sine search strategy, update the position of each hippopotamus, and determine the initial optimal solution;

[0075] The formula corresponding to the golden sine search strategy is:

[0076] ;

[0077] ;

[0078] ;

[0079] in, , They represent the positions corresponding to the tth iteration and the t+1th iteration respectively. , represents the golden ratio coefficient, , represents a constant, represents the golden section number, , represents a random number, sin represents a sine function, represents the global optimal position of the tth iteration, Indicates absolute value.

[0080] in, Controls the moving step size of the next iteration of the current hippopotamus (updated hippopotamus), , Controls the direction of the next iteration of the current hippopotamus (updated hippopotamus), .right and The initial values ​​are and , during the iteration process, and It will be dynamically adjusted according to the calculated fitness value of the hippopotamus position, that is, if the current individual fitness value is better than the global optimal fitness value, then ,otherwise , and update and .

[0081] S2-6, using Cauchy-Gauss mutation to optimize the initial optimal solution, based on the IMF component corresponding to the initial optimal solution, calculate the corresponding fitness and determine the global optimal solution;

[0082] The formula corresponding to the Cauchy-Gaussian mutation is:

[0083] ;

[0084] in, represents the position of the individual with the highest fitness after mutation, , Indicates dynamic parameters. represents the variance, It means the mean is 0 and the variance is The Cauchy distribution of It means the mean is 0 and the variance is Gaussian random quantity, represents the position of the hippopotamus before mutation. reduce, will increase, allowing the algorithm to jump out of the current stagnation, and this strategy takes into account the capabilities of local development and overall search.

[0085] S2-7. Determine whether the optimal solution meets the termination condition, that is, the current number of iterations is less than the total number of iterations; if so, use the global optimal solution as the subsequence decomposition data; otherwise, iterate again.

[0086] S3. Use a hybrid deep learning algorithm to process the subsequence decomposition data and related influencing factor data to obtain the energy consumption prediction result of the subway station.

[0087] The hybrid deep learning algorithm adopts an energy consumption prediction model, including a MIC feature selection layer and a TCN-BiLSTM model connected in series; the TCN-BiLSTM model includes a TCN sub-model and a BiLSTM sub-model connected in series;

[0088] The hybrid deep learning algorithm includes:

[0089] S3-1, input the sequence decomposition data and related influencing factor data into the MIC feature selection layer, and output the corresponding energy consumption feature data set;

[0090] The MIC feature selection layer includes:

[0091] The subsequence decomposition data and the relevant influencing factor data are spliced ​​to obtain the initial energy consumption characteristic data set;

[0092] Calculate the mutual information value of each initial energy consumption characteristic data and arrange them in descending order to obtain the mutual information value sorting result;

[0093] Based on the mutual information value sorting results, each initial energy consumption feature data is input into the TCN-BiLSTM model in turn to calculate the corresponding MAPE index value;

[0094] The energy consumption characteristic data with the smallest MAPE index value is selected to obtain the energy consumption characteristic data set.

[0095] The formula corresponding to the mutual information value is:

[0096] ;

[0097] in, , , , Represents the initial energy consumption characteristic data, Indicates initial energy consumption characteristic data and initial energy consumption characteristic data The joint probability distribution function between represents a logarithmic function with a constant as base, , Represents the initial energy consumption characteristic data and initial energy consumption characteristic data The marginal probability distribution function between .

[0098] S3-2. Use the MSIHO algorithm to train the TCN-BiLSTM model to obtain a trained TCN-BiLSTM model;

[0099] like Figure 3 As shown, the training process of the TCN-BiLSTM model includes:

[0100] S3-2-1. Obtain training energy consumption characteristic data set and optimization energy consumption characteristic data set; the training energy consumption characteristic data set and optimization energy consumption characteristic data set are divided into hourly data, daily data and monthly data. Among them, the daily data set needs to add up the energy consumption of 24 hours a day, and the monthly data set needs to add up the energy consumption of each hour of each month.

[0101] S3-2-2. Input the optimized energy consumption feature data set into the TCN-BiLSTM model, and use the MSIHO algorithm to determine the optimal parameter combination of the TCN-BiLSTM model; the parameter combination of the TCN-BiLSTM model includes learning rate, BatchSize, number of hidden layer nodes, and number of fully connected nodes.

[0102] The MSIHO algorithm uses the same method as S2-2 to S2-7 to determine the optimal parameter combination of the TCN-BiLSTM model.

[0103] S3-2-3. Apply the optimal parameter combination to the TCN-BiLSTM model to obtain the optimized TCN-BiLSTM model;

[0104] S3-2-4, input each training energy consumption feature data in the training energy consumption feature data set into the optimized TCN-BiLSTM model to obtain the corresponding initial training energy consumption prediction results, that is, [prediction result 1, prediction result 2, ..., prediction result n];

[0105] S3-2-5, superimposing and reconstructing each initial training energy consumption prediction result to obtain a corresponding training energy consumption prediction result, i.e., a predicted energy consumption sequence;

[0106] S3-2-6. Based on the training energy consumption prediction results, calculate the corresponding evaluation indicators; the evaluation indicators include mean absolute error MAE, squared mean absolute error RMAE, mean absolute percentage error MAPE and determination coefficient R 2 , the corresponding formulas are:

[0107] ;

[0108] ;

[0109] ;

[0110] ;

[0111] Where n represents the total number of samples. Indicates the actual energy consumption value collected. represents the predicted value of energy consumption, Represents the predicted value of energy consumption.

[0112] S3-2-7. Determine whether the evaluation index meets the training threshold condition; if so, complete the training of the TCN-BiLSTM model; otherwise, retrain. The smaller the RMSE and MAE values, the better the prediction performance of the model; The closer this value is to 1, the better the fit is and the closer the predicted value is to the actual observed value. Set the training threshold conditions according to actual needs.

[0113] S3-3. Input the energy consumption characteristic data set into the optimized TCN-BiLSTM model to obtain the energy consumption prediction result of the subway station.

[0114] In summary, the present invention determines the model input features based on mutual information, realizes feature dimensionality reduction, and combines the Hippo optimization algorithm improved by multi-strategy fusion to optimize the hyperparameters of the TCN-BiLSTM model and decompose the energy consumption data, and decomposes the original energy consumption data into several different simple subsequences to obtain more accurate prediction results; the method of the present invention is used to predict the hourly, daily and monthly energy consumption of subway stations, and is applied to the energy consumption management system of subway station operations, so as to assist the subway operation department to understand the energy consumption requirements and energy consumption fluctuations of different stations in advance, reasonably allocate resources, and optimize energy management.

[0115] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A subway station energy consumption prediction method based on a hybrid deep learning algorithm, characterized in that: include: Obtain the original energy consumption data and original related influencing factor data of a subway station and perform preprocessing to obtain energy consumption data and related influencing factor data; The MSIHO-VMD algorithm is used to perform sequence decomposition on the energy consumption data to obtain subsequence decomposition data; the MSIHO-VMD algorithm includes the MSIHO algorithm and the VMD algorithm, and the MSIHO algorithm is an improved Hippo optimization algorithm; The hybrid deep learning algorithm is used to process the subsequence decomposition data and related influencing factor data to obtain the energy consumption prediction result of the subway station; The method of using the MSIHO-VMD algorithm to perform sequence decomposition on the energy consumption data to obtain subsequence decomposition data includes: The VMD variational mode decomposition method is used to decompose the energy consumption data to obtain simple subsequence data; Use the simple subsequence data as the hippo population and set the initial parameters of the hippo population; The Latin hypercube sampling method is used to initialize the hippo population, and the fitness of each hippo is calculated based on the IMF component of the simple subsequence data to determine the current optimal solution; Use the adaptive weight strategy to iterate the exploration of each hippopotamus, generate candidate positions, calculate the corresponding fitness, and update the position of each hippopotamus; The updated hippopotamus are optimized using the golden sine search strategy, the positions of each hippopotamus are updated, and the initial optimal solution is determined; The initial optimal solution is optimized using the Cauchy-Gauss mutation, and the corresponding fitness is calculated based on the IMF component corresponding to the initial optimal solution to determine the global optimal solution. Determine whether the optimal solution meets the termination condition; if so, use the global optimal solution as the subsequence decomposition data; otherwise, iterate again; The hybrid deep learning algorithm adopts an energy consumption prediction model, including a MIC feature selection layer and a TCN-BiLSTM model connected in series; the TCN-BiLSTM model includes a TCN sub-model and a BiLSTM sub-model connected in series; The hybrid deep learning algorithm includes: The sequence decomposition data and related influencing factor data are input into the MIC feature selection layer, and the corresponding energy consumption feature data set is output; The MSIHO algorithm is used to train the TCN-BiLSTM model to obtain a trained TCN-BiLSTM model; The energy consumption characteristic data set is input into the optimized TCN-BiLSTM model to obtain the energy consumption prediction result of the subway station.

2. The subway station energy consumption prediction method based on hybrid deep learning algorithm according to claim 1 is characterized in that: The original energy consumption data is historical energy consumption data; the original relevant influencing factor data includes building factors, equipment factors, passenger flow factors, meteorological factors and time factors; the building factors include whether the subway station is a transfer station, the station layout method, the station building area, and the number of station entrances and exits; the equipment factors include the number of air conditioners, air conditioner power, the number of escalators, escalator power, the number of lighting lamps, and the lighting lamp power; the passenger flow factors include the station passenger flow; the meteorological factors include the temperature and humidity of the subway station; the time factors include the time point, week type, and holiday type.

3. The subway station energy consumption prediction method based on hybrid deep learning algorithm according to claim 1 is characterized in that: The pre-processing comprises: use The method processes the original energy consumption data and the original related influencing factor data, determines the corresponding abnormal values ​​and removes them, and obtains the updated original energy consumption data and the updated original related influencing factor data; The outliers are processed by using the sliding window averaging method and merged into the corresponding updated original energy consumption data and updated original related influencing factor data to obtain the initial energy consumption data and the initial related influencing factor data; The initial energy consumption data and initial related influencing factor data are normalized to obtain energy consumption data and related influencing factor data.

4. The subway station energy consumption prediction method based on hybrid deep learning algorithm according to claim 1 is characterized in that: The MIC feature selection layer includes: The subsequence decomposition data and the relevant influencing factor data are spliced ​​to obtain the initial energy consumption characteristic data set; Calculate the mutual information value of each initial energy consumption characteristic data and arrange them in descending order to obtain the mutual information value sorting result; Based on the mutual information value sorting results, each initial energy consumption feature data is input into the TCN-BiLSTM model in turn to calculate the corresponding MAPE index value; Energy consumption characteristic data is selected based on the MAPE indicator value to obtain an energy consumption characteristic data set.

5. The subway station energy consumption prediction method based on hybrid deep learning algorithm according to claim 1 is characterized in that: The training process of the TCN-BiLSTM model includes: Obtaining a training energy consumption characteristic data set and an optimization energy consumption characteristic data set; The optimized energy consumption feature data set is input into the TCN-BiLSTM model, and the MSIHO algorithm is used to determine the optimal parameter combination of the TCN-BiLSTM model; Apply the optimal parameter combination to the TCN-BiLSTM model to obtain the optimized TCN-BiLSTM model; Input each training energy consumption feature data in the training energy consumption feature data set into the optimized TCN-BiLSTM model to obtain the corresponding initial training energy consumption prediction result; Each initial training energy consumption prediction result is superimposed and reconstructed to obtain the corresponding training energy consumption prediction result; Based on the training energy consumption prediction results, calculate the corresponding evaluation indicators; Determine whether the evaluation index meets the training threshold condition; if so, complete the training of the TCN-BiLSTM model; otherwise, retrain.

Images