Reliability Evaluation Method and System for Hydrogen-Electric Coupling System

Abstract

The present invention discloses a reliability evaluation method and system for a hydrogen-electric coupling system. The technical solution adopted by the present invention is as follows: collecting the operation data of the hydrogen-electric coupling system through equipment sensors; analyzing the participation processes of the hydrogen production system, fuel cell system, and hydrogen storage system in the hydrogen-electric coupling system, and establishing a coupling mathematical model of the hydrogen-electric coupling system; screening the reliability indexes of the hydrogen-electric coupling system; constructing a hybrid neural network integrating a convolutional neural network and a bidirectional long short-term memory network according to the mathematical model of the hydrogen-electric coupling system, and training the hybrid neural network using a training data set; performing output prediction on a test data set using the trained hybrid neural network; calculating the reliability indexes of the hydrogen production system, fuel cell system, and hydrogen storage system and judging their working states, and finally comprehensively judging the reliability of the hydrogen-electric coupling system. The present invention can effectively detect the abnormal working state of the hydrogen-electric coupling system and evaluate the operation reliability of the hydrogen-electric coupling system.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrogen energy, and specifically to a reliability evaluation method and system for a hydrogen-electric coupling system based on the fusion of convolutional neural network and bidirectional long short-term memory network. Background Technique

[0002] With the rapid development of energy technologies, renewable energy power generation technologies with high efficiency, cleanness, and low carbon have been highly regarded and widely applied in various countries around the world. It is difficult to store the energy generated by wind and solar power, and the costs of traditional electrochemical energy storage and electromagnetic energy storage are relatively high. Hydrogen energy, as a clean energy source, has the characteristics of high energy density, easy storage and transportation, etc., and has become one of the preferred solutions for the efficient utilization and green storage of wind power and photovoltaic energy. With the continuous progress of hydrogen energy technology, it provides effective technical support for the realization of hydrogen-electric coupling systems.

[0003] Hydrogen-electric coupling is an important means to achieve 100% renewable energy consumption and 100% zero-carbon power supply. Based on the major demands of the new energy industry and national strategies, hydrogen-electric coupling systems are expected to be able to design various hydrogen production and energy storage devices independently and provide relevant software services, and can achieve high-efficiency, large-capacity, and intelligent hydrogen production, thus effectively solving the problem of how to produce zero-emission green hydrogen, efficiently produce green hydrogen energy, effectively improve the energy consumption ratio of new energy power generation systems, and support the coordinated and healthy development of China's renewable energy and hydrogen fuel cell industries. However, the current research on hydrogen-electric coupling is in its infancy. In the hydrogen production, storage, supply, hydrogen addition, and fuel cell power generation links of hydrogen-electric coupling systems, there are risks such as hydrogen leakage and explosion. The stable operation of hydrogen-electric coupling systems faces huge challenges. Therefore, it is crucial to evaluate the reliability of hydrogen-electric coupling systems.

[0004] In the coupling system, the net fluctuating power of source-load imbalance is balanced by electric and hydrogen energy storage. However, due to the different characteristics of electric and hydrogen energy storage, how to adjust the participation degrees of the hydrogen production system, hydrogen storage system, and hydrogen utilization system in the coupling system and calculate the reliability indexes of each coupling subsystem is an important challenge for judging the reliable operation of hydrogen-electric coupling systems. Summary of the Invention

[0005] To solve the deficiencies in the above background technique, the present invention provides a reliability evaluation method and system for a hydrogen-electric coupling system based on the fusion of convolutional neural network and bidirectional long short-term memory network. After establishing a hydrogen-electric coupling system, a hybrid neural network fused by a convolutional neural network and a bidirectional long short-term memory network is constructed according to the established hydrogen-electric coupling system. The hybrid neural network is trained using historical experimental data, and the trained hybrid neural network is used to output and predict the experimental data. The reliability indexes of the hydrogen production system, fuel cell system, and hydrogen storage system are calculated, and the working states of each system are judged. Finally, the reliability degree of the hydrogen-electric coupling system is comprehensively judged.

[0006] To achieve the above object, a technical solution adopted by the present invention is as follows: a reliability evaluation method for a hydrogen-electric coupling system, which includes:

[0007] 1) Collect data on the operation of the hydrogen-electric coupling system through equipment sensors;

[0008] 2) Analyze the participation processes of the hydrogen production system, fuel cell system, and hydrogen storage system in the hydrogen-electric coupling system, and establish a coupling mathematical model for the hydrogen-electric coupling system, including an alkaline electrolyzer model, a hydrogen storage model, and a fuel cell model;

[0009] 3) Screen the reliability indicators of the hydrogen-electric coupling system;

[0010] 4) Construct a hybrid neural network that combines a convolutional neural network and a bidirectional long short-term memory network according to the mathematical model of the hydrogen-electric coupling system, and train the constructed hybrid neural network using a training data set;

[0011] 5) Use the trained hybrid neural network to perform output prediction on a test data set;

[0012] 6) Calculate the reliability indicators of the hydrogen production system, fuel cell system, and hydrogen storage system, determine the working states of each system, and finally comprehensively judge the reliability of the hydrogen-electric coupling system.

[0013] The training data set and the test data set constitute the historical experimental data of the hydrogen-electric coupling system, that is, the operation data of the hydrogen-electric coupling system.

[0014] Furthermore, the data on the operation of the hydrogen-electric coupling system includes the electrolyzer voltage, electrolyzer power, hydrogen storage tank pressure, and fuel cell power.

[0015] Furthermore, in step 2), the coupling mathematical model includes:

[0016] 1) Hydrogen storage model

[0017] The hydrogen storage rate is expressed as:

[0018]

[0019] where n sto (t) is the net hydrogen storage rate of the hydrogen storage tank at time t; is the hydrogen intake rate of the hydrogen storage tank at time t; is the hydrogen output rate of the hydrogen storage tank at time t;

[0020] The hydrogen storage amount is:

[0021]

[0022] where n sto (t0) is the hydrogen storage amount of the hydrogen storage tank at time t0; nsto (t0 + Δt) is the hydrogen storage amount in the hydrogen storage tank within the time of t0 + Δt; t0 represents the time point; Δt represents the time quantity;

[0023] The pressure p of the hydrogen storage tank is obtained from the gas state equation sto as:

[0024]

[0025] In the formula: T sto is the hydrogen storage temperature, V sto is the volume of the hydrogen storage tank, and R is the ideal gas constant;

[0026] Furthermore, in the step 2), the coupled mathematical model further includes:

[0027] 2) Electrolyzer model

[0028] The U-I equation of the electrolyzer is:

[0029]

[0030] In the formula: ΔG is the change in Gibbs free energy during the electrochemical reaction process, z is the number of electrons transferred per reaction; F is the Faraday constant; r1 is the electrolyte ohmic resistance constant, r2 is the first-order parameter of the electrolyte ohmic resistance; T el is the electrolyzer temperature; A cell is the area of the electrolyzer module; I el is the direct current; s1, s2, s3 are the zero-order coefficient, first-order coefficient, and second-order coefficient of the electrode overvoltage respectively, and t1, t2, t3 are the zero-order coefficient, first-order coefficient, and second-order coefficient of the electrode overvoltage coefficient respectively; U cell is the output voltage of the electrolyzer;

[0031] The hydrogen production rate n of the electrolyzer aeH2 is:

[0032]

[0033] Among them, η F The expression is:

[0034]

[0035] In the formula: η F is the Faraday efficiency; N el is the number of series-connected modules of the electrolyzer; z is the number of electrons transferred per reaction; a1, a2, a3, a4, a5 are the Faraday efficiency coefficients;

[0036] The internal heat balance equation of the electrolyzer is:

[0037]

[0038] Among them,

[0039]

[0040]

[0041] Q cool = C cw (T cw,i - T cw,o ) = hA HX LMTD(10)

[0042]

[0043]

[0044] In the formula: Q gen is the heat energy generated by the electrolytic cell; Q loss is the heat energy lost by the electrolytic cell; Q cool is the heat energy taken away by the cooling water of the electrolytic cell; C t represents the total heat capacity, ρ j , V j , c j respectively represent the density, volume and heat capacity of the components of the electrolytic cell; represents the thermal neutral voltage; E total represents the electrolytic cell voltage; T a is the ambient temperature; A stack is the outer surface area of the electrolytic cell; h is the heat transfer coefficient; T sep is the outlet temperature of the gas-liquid separator; I is the electrolytic cell current; h0 is the heat transfer coefficient at zero current density; k is the proportionality coefficient; A sep is the surface area of the gas-liquid separator; h sep , h stack are the heat transfer coefficients of the gas-liquid separator and the electrolytic cell respectively; T cw,i , T cw,o are the temperatures of the cooling water entering and leaving the electrolytic cell respectively; C cw is the heat capacity of the cooling water; LMTD is the logarithmic mean temperature difference, indicating the demand for auxiliary cooling; hA HX represents the product of the total heat transfer coefficient and area of the electrolytic cell; T cw , is the operating temperature of the electrolytic cell.

[0045] Furthermore, in step 2), the coupling mathematical model further includes:

[0046] 3) Fuel cell model

[0047] The output voltage of the fuel cell is:

[0048]

[0049] In the formula, V cell is the module output voltage; E nernst is the thermoelectric electromotive force; U act is the activation overvoltage; U ohm is the ohmic overvoltage; U conc is the concentration overvoltage; V fc is the fuel cell stack voltage; N fc is the number of series-connected cells;

[0050] The activation overvoltage is:

[0051] U act = [ξ1 + ξ2T fc + ξ3ln(C O2 ) + ξ4 × T fc × ln(I fc )] (14)

[0052] Among them,

[0053]

[0054] In the formula, ξ1, ξ2, ξ3, and ξ4 are empirical parameters; I fc is the fuel cell current; C O2 is the oxygen concentration at the cathode gas-liquid interface; P O2 is the oxygen interface partial pressure of the fuel cell; T fc is the fuel cell operating temperature;

[0055] The ohmic overvoltage is:

[0056] U ohm = I fc R ohm = I fc (r M l / A + R c ) (16)

[0057] Among them,

[0058]

[0059] In the formula, l is the fuel cell membrane thickness; r M is the fuel cell resistivity; R c is the impedance that hinders protons from passing through the proton exchange membrane; A is the effective area of the proton exchange membrane; R ohm is the ohmic resistance; λ is the water content of the proton exchange membrane;

[0060] The concentration overvoltage is:

[0061] Uconc = -B ln(1 - J / J max ) (18)

[0062] Wherein, J and J max are the current density and the maximum current density respectively; B is an equation constant determined by the operating conditions of the fuel cell.

[0063] Furthermore, in step 3), according to the established coupling mathematical model of the hydrogen - electricity coupling system, analyze the influencing factors of each system operation, and screen out the reliability indicators from them.

[0064] Furthermore, the reliability indicators include hydrogen storage rate, hydrogen storage tank pressure, hydrogen production rate, electrolyzer output voltage, heat energy taken away by the electrolyzer cooling water, heat energy generated by the electrolyzer, hydrogen tank pressure, current density, fuel cell temperature, and fuel cell oxygen interface partial pressure.

[0065] Furthermore, the specific process of step 4) is as follows:

[0066] Step 401: The one - dimensional convolutional layer performs a convolution operation on the local area of the operation data of the hydrogen - electricity coupling system, and outputs the corresponding feature map, as shown in formula (19);

[0067] c t = f(W cnn * n t + b cnn ) (19)

[0068] Wherein, W cnn represents the weight matrix of the filter in the convolution operation of the input parameters of the hydrogen - electricity coupling system; n t represents the t - th operation data information; * represents the convolution operation between the input parameters and W cnn ; b cnn represents the bias parameter in the convolution operation, f represents the activation function in the convolution operation, and c t represents the output sequence after the convolution operation;

[0069] Step 402: In the pooling layer, use max - pooling to remove redundant information in the data, compress the feature information extracted by the previous hidden layer, and simplify the complexity of the CNN - BiLSTM hybrid neural network (i.e., the hybrid neural network fused by the convolutional neural network CNN and the bidirectional long - short - term memory network BiLSTM);

[0070] Step 403: The activation function uses the scaled exponential linear unit to perform a non - linear transformation on each convolution operation, expressed as:

[0071]

[0072] Among them, α = 1.6733 and λ0 = 1.0507;

[0073] Step 404: The bidirectional long short-term memory network sets two different hidden layers, namely the forward LSTM hidden layer and the backward LSTM hidden layer. The forward LSTM hidden layer reads the operation data of the hydrogen-electricity coupling system in ascending order, outputs a sequence after convolution and max-pooling operations, and the backward LSTM hidden layer reads the operation data of the hydrogen-electricity coupling system in descending order, outputs a sequence after convolution and max-pooling operations;

[0074] Step 405: Set a dropout layer to improve the generalization ability of the CNN-BiLSTM hybrid neural network. The gray neural points in the dropout layer do not participate in network training. The output of the CNN-BiLSTM hybrid neural network with the dropout layer added is as shown in Equation (28):

[0075]

[0076] Among them, the dropout layer uses the Bernoulli function to randomly generate a vector r of 0 or 1 with probability p, so that some hidden layer nodes are inactivated; σ represents the sigmoid function in BiLSTM; W o represents the weight of the input feature in the BiLSTM training; is the forward hidden state at time t - 1; x t is the feature quantity input at time t; b o is the bias parameter in the convolution operation process;

[0077] Step 406: In the training process of the CNN-BiLSTM hybrid neural network, adopt the adaptive moment estimation optimization function to iteratively update the weights and biases of the CNN-BiLSTM hybrid neural network.

[0078] Another technical solution adopted by the present invention is: a hydrogen-electricity coupling system reliability evaluation system, which includes:

[0079] Data acquisition unit: Collect the operation data of the hydrogen-electricity coupling system through device sensors;

[0080] Coupled mathematical model establishment unit: Analyze the participation processes of the hydrogen production system, fuel cell system, and hydrogen storage system in the hydrogen-electricity coupling system, and establish a coupled mathematical model of the hydrogen-electricity coupling system, including an alkaline electrolyzer model, a hydrogen storage model, and a fuel cell model;

[0081] Reliability index screening unit: Screen the reliability indexes of the hydrogen-electricity coupling system;

[0082] Simulation model construction unit: Construct a hybrid neural network that combines a convolutional neural network and a bidirectional long short-term memory network according to the mathematical model of the hydrogen-electricity coupling system, and use the training data set to train the hybrid neural network;

[0083] Output prediction unit: Use the trained hybrid neural network to perform output prediction on the test data set;

[0084] Reliability judgment unit: Calculate the reliability indicators of the hydrogen production system, fuel cell system, and hydrogen storage system, and judge the working status of each system. Finally, comprehensively judge the reliability of the hydrogen-electricity coupling system.

[0085] The beneficial effects of the present invention are as follows: According to the established hydrogen-electricity coupling system, a hybrid neural network that combines a convolutional neural network and bidirectional long short-term fusion (i.e., the coupling mathematical model of the hydrogen-electricity coupling system) is constructed, and the coupling mathematical model is trained using the training data set; Use the trained convolutional neural network to perform output prediction on the test data set, calculate the reliability indicators of the hydrogen production system, fuel cell system, and hydrogen storage system, and judge the working status of each system, which can effectively detect the abnormal working status of the hydrogen-electricity coupling system and evaluate the operation reliability of the hydrogen-electricity coupling system. Description of the drawings

[0086] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the description of the embodiments or the prior art; Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0087] Figure 1 It is a flowchart of the method for evaluating the reliability of the hydrogen-electricity coupling system of the present invention;

[0088] Figure 2 It is a prediction process diagram of the CNN-BiLSTM hybrid neural network of the present invention ( Figure 2 The CNN-BiLSTM model in it is the CNN-BiLSTM hybrid neural network);

[0089] Figure 3 It is a schematic structural diagram of the CNN-BiLSTM hybrid neural network of the present invention. Detailed implementation manners

[0090] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0091] Embodiment 1

[0092] This embodiment is a reliability evaluation method for a hydrogen-electric coupling system based on the fusion of a convolutional neural network and bidirectional long short-term memory, as Figure 1 shown, and includes the following steps:

[0093] 1) Collect data on the operation of the hydrogen-electric coupling system through device sensors, including electrolyzer voltage, electrolyzer power, hydrogen storage tank pressure, and fuel cell power;

[0094] 2) Analyze the participation processes of the hydrogen production system, fuel cell system, and hydrogen storage system in the hydrogen-electric coupling system, and establish a coupling mathematical model of the hydrogen-electric coupling system, including an alkaline electrolyzer model, a hydrogen storage model, and a fuel cell model;

[0095] 3) Screen the reliability indicators of the hydrogen-electric coupling system;

[0096] 4) Construct a hybrid neural network fused by a convolutional neural network and a bidirectional long short-term memory network according to the mathematical model of the hydrogen-electric coupling system. Its structure is as Figure 3 shown, and use the training data set to train the constructed hybrid neural network, as Figure 2 shown;

[0097] 5) Use the trained hybrid neural network to perform output prediction on the test data set, as Figure 2 shown;

[0098] 6) Calculate the reliability indicators of the hydrogen production system, fuel cell system, and hydrogen storage system, and determine the working states of each system. Finally, comprehensively judge the reliability of the hydrogen-electric coupling system.

[0099] The training data set and the test data set constitute the historical experimental data of the hydrogen-electric coupling system, that is, the operation data of the hydrogen-electric coupling system.

[0100] In step 2), the coupling mathematical model of the hydrogen-electric coupling system is as follows:

[0101] Hydrogen storage model:

[0102] The hydrogen storage rate is:

[0103]

[0104] In the formula, n sto (t) is the net hydrogen storage rate of the hydrogen storage tank at time t; is the hydrogen intake rate of the hydrogen storage tank at time t; is the hydrogen output rate of the hydrogen storage tank at time t;

[0105] The hydrogen storage amount is:

[0106]

[0107] In the formula, n sto (t0) is the hydrogen storage amount of the hydrogen storage tank at time t0; n sto (t0 + Δt) is the hydrogen storage amount of the hydrogen storage tank within the time of t0 + Δt; t0 represents the time point; Δt represents the time quantity;

[0108] According to the gas state equation, the pressure of the hydrogen storage tank is:

[0109]

[0110] In the formula: T sto is the hydrogen storage temperature, V sto is the volume of the hydrogen storage tank.

[0111] Electrolyzer model:

[0112] The U-I equation of the electrolyzer is:

[0113]

[0114] In the formula: ΔG is the change in Gibbs free energy during the electrochemical reaction process, z is the number of electrons transferred per reaction; F is the Faraday constant; r1, r2 are the ohmic resistance parameters of the electrolyte; T el is the temperature of the electrolyzer; A cell is the area of the electrolyzer module; I el is the direct current; s1, s2, s3 are the zero-order coefficient, first-order coefficient, and second-order coefficient of the electrode overvoltage respectively, and t1, t2, t3 are the zero-order coefficient, first-order coefficient, and second-order coefficient of the electrode overvoltage coefficient respectively; U cell is the output voltage of the electrolyzer.

[0115] The hydrogen production rate of the electrolyzer is:

[0116]

[0117] Among them, η F The expression is:

[0118]

[0119] In the formula: η F is the Faraday efficiency; Nel n is the number of series-connected electrolyzer modules; z is the number of electrons transferred per reaction; a1, a2, a3, a4, and a5 are Faraday efficiency coefficients.

[0120] The internal heat balance equation of the electrolyzer is:

[0121]

[0122] Where

[0123]

[0124]

[0125] Q cool = c cw (T cw,i - T cw,o ) = hA HX LMTD(10)

[0126]

[0127]

[0128] In the formula: Q gen is the heat energy generated by the electrolyzer; Q loss is the heat energy lost by the electrolyzer; Q cool is the heat energy carried away by the cooling water of the electrolyzer; C t represents the total heat capacity, ρ j , V j , c j respectively represent the density, volume, and heat capacity of the components of the electrolyzer; represents the thermal neutral voltage. The thermal neutral voltage is 1.48 V under the environment of 298 K and 0.1 MPa. Under other temperatures and pressures, it is solved by the empirical formula; E total represents the electrolyzer voltage, which can be solved by the mechanism model or the empirical formula; T a is the ambient temperature; A stack is the outer surface area of the electrolyzer; h is the heat transfer coefficient; T sep is the outlet temperature of the gas-liquid separator; I is the electrolyzer current; h0 is the heat transfer coefficient at zero current density; k is the proportionality coefficient; A sep is the surface area of the gas-liquid separator; h sep , h stack are the heat transfer coefficients of the gas-liquid separator and the electrolyzer respectively; T cw,i , T cw,o are the temperatures of the cooling water entering and leaving the electrolyzer respectively; C cw is the heat capacity of the cooling water; LMTD is the logarithmic mean temperature difference, indicating the demand for auxiliary cooling; hAHX Represents the product of the overall heat transfer coefficient and area of the electrolytic cell; T cw , is the operating temperature of the electrolytic cell.

[0129] Fuel cell model:

[0130] The output voltage of the fuel cell is:

[0131]

[0132] Where: V cell is the output voltage; E nernst is the thermodynamic electromotive force; U act is the activation overvoltage; U ohm is the ohmic overvoltage; U conc is the concentration overvoltage; V fc is the voltage of the fuel cell stack; N fc is the number of series-connected cells.

[0133] The activation overvoltage of the fuel cell is

[0134] U act = [ξ1 + ξ2T fc + ξ3ln(C O2 ) + ξ4 × T fc × ln(I fc )] (14)

[0135] Among them,

[0136]

[0137] Where: ξ1, ξ2, ξ3, and ξ4 are empirical parameters; I fc is the fuel cell current; C O2 is the oxygen concentration at the cathode gas-liquid interface; P O2 is the oxygen interface partial pressure of the fuel cell; T fc is the operating temperature of the fuel cell.

[0138] The ohmic overvoltage of the fuel cell is:

[0139] U ohm = I fc R ohm = I fc (r M l / A + R c ) (16)

[0140] Among them,

[0141]

[0142] Where: l is the fuel cell membrane thickness; r Mis the resistivity of the fuel cell; R c is the impedance that hinders the passage of protons through the proton exchange membrane; A is the effective area of the proton exchange membrane.

[0143] The concentration overvoltage of the fuel cell is:

[0144] U conc = -Bln(1 - J / J max )(18)

[0145] Where: J and J max are the current density and the maximum current density respectively; B is an equation constant determined by the operating conditions of the fuel cell.

[0146] In step 3), according to the established coupling mathematical model of the hydrogen - electricity coupling system, analyze the influencing factors of the operation of each system:

[0147] Due to the flammable and explosive characteristics of hydrogen, the monitoring and treatment of hydrogen leakage and hydrogen discharge concentration in the hydrogen storage system are particularly important. The hydrogen storage rate and the pressure of the hydrogen storage tank are reliable indicators reflecting the safe operation of the hydrogen storage system.

[0148] The hydrogen production system is a complex non - linear system containing multiple physical fields of "machine - electricity - heat - chemistry", involving multiple physical fields such as heat transfer, electrochemical reaction, and mass transfer. The hydrogen production rate is an important energy consumption index for evaluating the performance of the electrolyzer; a short - circuit in the electrolyzer is caused by the deposition of metal impurities, making a path between the electrode and the diaphragm frame. This situation can be judged by measuring the output voltage of the electrolyzer; during the hydrogen production process, temperature is crucial for the operation of the entire electrolyzer, and the heat production is a key parameter for the internal heat balance of the electrolyzer; when the electrolyzer is operating, in order to reach the preset working temperature inside and maintain stability, a special cooling water pump is usually required to cool the heat exchanger. The heat energy Q carried away by the cooling water of the electrolyzer cool reflects the working state of the cooling system.

[0149] During the operation of the fuel cell, the proton conductivity is closely related to the membrane water content. Excessive water content inside the battery will cause flooding faults, while insufficient water content will lead to membrane dry faults. Under high current density or high - temperature conditions, the electro - drag effect is too strong, which will cause water shortage and increased ohmic impedance on the anode side of the fuel cell, resulting in increased heat generation during the operation of the fuel cell and further leading to more serious membrane dry faults; under high current density or low - temperature conditions, excessive water in the battery cannot be discharged in time due to blocking the flow channels and the pores of the diffusion layer, which will cause flooding faults in the fuel cell; the pressure of the hydrogen tank and the oxygen interface partial pressure of the fuel cell are important indicators for judging the supply status of the anode and cathode reactants. Abnormal supply of anode and cathode reactants will cause hydrogen starvation and abnormal air intake faults, which will lead to a decrease in the internal chemical reaction rate of the fuel cell and, in severe cases, reverse - polarity phenomena.

[0150] According to the above analysis, the reliability calculation indexes of the selected hydrogen - electricity coupling system are shown in Table 1:

[0151] Table 1 Reliability calculation indexes of the hydrogen - electricity coupling system

[0152]

[0153] In step 4), calculate the reliability indexes of each system of the hydrogen - electricity coupling system according to the following method, and judge the working state of each system:

[0154] Step 401: The one - dimensional convolutional layer performs a convolution operation on the local area of the operation data of the hydrogen - electricity coupling system, and outputs the corresponding feature map, as shown in Equation (19):

[0155] c t = f(W cnn *n t +b cnn ) (19)

[0156] Among them, W cnn represents the weight matrix of the filter in the convolution operation of the input parameters of the hydrogen - electricity coupling system; n t represents the t - th operation data information; * represents the convolution operation between the input parameter and W cnn ; b cnn represents the bias parameter in the convolution operation, f represents the activation function in the convolution operation, and c t represents the output sequence after the convolution operation.

[0157] Step 402: In the pooling layer, use max - pooling to remove redundant information in the data, compress the feature information extracted by the previous hidden layer, and simplify the complexity of the hybrid neural network.

[0158] Step 403: The activation function uses the scaled exponential linear unit to perform a non - linear transformation on each convolution operation, which can be expressed as

[0159]

[0160] Among them, α = 1.6733, λ0 = 1.0507.

[0161] Step 404: BiLSTM sets two different hidden layers, namely the forward LSTM hidden layer and the backward LSTM hidden layer. The forward LSTM hidden layer reads the operation data of the hydrogen - electricity coupling system in ascending order (t = 1, 2, 3,..., T), the output sequence after convolution and max - pooling operations, and the backward LSTM hidden layer reads the operation data of the hydrogen - electricity coupling system in descending order (t = T,..., 3, 2, 1), the output sequence after convolution and max - pooling operations, which are defined as follows:

[0162]

[0163]

[0164] i t = σ(W i · [h t-1 , x t + b i ) (23)

[0165] f t = σ(W i · [h t-1 , x t + b f ) (24)

[0166] C t = f t · C t-1 + i t · tanh(W C · [h t-1 , x t + b C ) (25)

[0167] o t = σ(W o · [h t-1 , x t + b o ) (26)

[0168] h t = o t · tanh(C t ) (27)

[0169] Among them, represents the forward hidden state at time t, represents the backward hidden state at time t, σ represents the sigmoid function in BiLSTM, C t represents the memory cell state retaining the decay information of the hydrogen-electric coupling system, f, i, and o are the forget gate, input gate, and output gate of the deep features of the hydrogen-electric coupling system in the BiLSTM training respectively, all b represents the bias of the deep features of the hydrogen-electric coupling system in the BiLSTM training, and all W represents the weight of the deep features of the hydrogen-electric coupling system in the BiLSTM training; h t represents the hidden state at time t; h t-1 represents the hidden state at time t - 1;

[0170] Step 405: Set a dropout layer to improve the generalization ability of the CNN-BiLSTM hybrid neural network. The gray neural points in the dropout layer do not participate in network training, which effectively reduces the interaction between hidden layer nodes and alleviates the dependence on local features during training. The output of the CNN-BiLSTM hybrid neural network with the dropout layer added is shown in Equation (28):

[0171]

[0172] Among them, the dropout layer uses the Bernoulli function to randomly generate a vector r of 0 or 1 with probability p (the set dropout value), so that some hidden layer nodes are inactivated; is the forward hidden state at time t-1.

[0173] Step 406: During the training process of the CNN-BiLSTM hybrid neural network, use the adaptive moment estimation optimization function to iteratively update the weights and biases of the CNN-BiLSTM hybrid neural network.

[0174] Embodiment 2

[0175] This embodiment provides a reliability evaluation system for a hydrogen-electric coupling system, which consists of a data acquisition unit, a coupling mathematical model establishment unit, a reliability index screening unit, a simulation model construction unit, an output prediction unit, and a reliability judgment unit.

[0176] Data acquisition unit: Collect data on the operation of the hydrogen-electric coupling system through device sensors;

[0177] Coupling mathematical model establishment unit: Analyze the participation processes of the hydrogen production system, fuel cell system, and hydrogen storage system in the hydrogen-electric coupling system, and establish a coupling mathematical model for the hydrogen-electric coupling system, including an alkaline electrolyzer model, a hydrogen storage model, and a fuel cell model;

[0178] Reliability index screening unit: Screen the reliability indexes of the hydrogen-electric coupling system;

[0179] Simulation model construction unit: Construct a hybrid neural network that combines a convolutional neural network and a bidirectional long short-term memory network according to the mathematical model of the hydrogen-electric coupling system, and use the training data set to train the hybrid neural network;

[0180] Output prediction unit: Use the trained hybrid neural network to perform output prediction on the test data set;

[0181] Reliability judgment unit: Calculate the reliability indexes of the hydrogen production system, fuel cell system, and hydrogen storage system, and determine the working states of each system. Finally, comprehensively judge the reliability of the hydrogen-electric coupling system.

[0182] The described training data set and test data set constitute the historical experimental data of the hydrogen-electricity coupling system, that is, the operating data of the hydrogen-electricity coupling system.

[0183] The data of the hydrogen-electricity coupling system operation include the electrolyzer voltage, electrolyzer power, hydrogen storage tank pressure, and fuel cell power.

[0184] In the described coupling mathematical model establishment unit, the described coupling mathematical model includes:

[0185] 1) Hydrogen storage model

[0186] The hydrogen storage rate is expressed as:

[0187]

[0188] In the formula, n sto (t) is the net hydrogen storage rate of the hydrogen storage tank at time t; is the hydrogen intake rate of the hydrogen storage tank at time t; is the hydrogen output rate of the hydrogen storage tank at time t;

[0189] The hydrogen storage amount is:

[0190]

[0191] In the formula, n sto (t0) is the hydrogen storage amount of the hydrogen storage tank at time t0; n sto (t0 + Δt) is the hydrogen storage amount of the hydrogen storage tank within the time of t0 + Δt; t0 represents the time point; Δt represents the time quantity;

[0192] According to the gas state equation, the pressure p of the hydrogen storage tank sto is:

[0193]

[0194] In the formula: T sto is the hydrogen storage temperature, V sto is the volume of the hydrogen storage tank, and R is the ideal gas constant.

[0195] 2) Electrolyzer model

[0196] The U-I equation of the electrolyzer is:

[0197]

[0198] In the formula: ΔG is the change in Gibbs free energy during the electrochemical reaction process, z is the number of electrons transferred per reaction; F is the Faraday constant; r1 is the electrolytic solution ohmic resistance constant, r2 is the first-order parameter of the electrolytic solution ohmic resistance; T el is the electrolyzer temperature; A cell is the electrolyzer module area; I elis a direct current; s1, s2, and s3 are the zero-order coefficient, first-order coefficient, and second-order coefficient of the electrode overvoltage respectively, and t1, t2, and t3 are the zero-order coefficient, first-order coefficient, and second-order coefficient of the electrode overvoltage coefficient respectively; U cell is the output voltage of the electrolytic cell.

[0199] The hydrogen production rate of the electrolytic cell is:

[0200]

[0201] where η F The expression is:

[0202]

[0203] In the formula: η F is the Faraday efficiency; N el is the number of series modules of the electrolytic cell; z is the number of electron transfers per reaction; a1, a2, a3, a4, and a5 are the Faraday efficiency coefficients;

[0204] The internal heat balance equation of the electrolytic cell is:

[0205]

[0206] where

[0207]

[0208]

[0209] Q cool = C cw (T cw,i - T cw,o ) = hA HX LMTD (10)

[0210]

[0211]

[0212] In the formula: Q gen is the heat energy generated by the electrolytic cell; Q loss is the heat energy lost by the electrolytic cell; Q cool is the heat energy taken away by the cooling water of the electrolytic cell; C t represents the total heat capacity, ρ j , V j , c j represent the density, volume, and heat capacity of the components of the electrolytic cell respectively; represents the thermal neutral voltage; E total represents the electrolytic cell voltage; T ais the ambient temperature; A stack is the outer surface area of the electrolyzer; h is the heat transfer coefficient; T sep is the outlet temperature of the gas-liquid separator; I is the electrolyzer current; h0 is the heat transfer coefficient at zero current density; k is the proportionality coefficient; A sep is the surface area of the gas-liquid separator; h sep , h stack are the heat transfer coefficients of the gas-liquid separator and the electrolyzer respectively; T cw,i , T cw,o are the temperatures at which the cooling water enters and exits the electrolyzer respectively; C cw is the heat capacity of the cooling water; LMTD is the log mean temperature difference, indicating the need for auxiliary cooling; hA HX represents the product of the total heat transfer coefficient and area of the electrolyzer; T cw , is the operating temperature of the electrolyzer.

[0213] 3) Fuel cell model

[0214] The output voltage of the fuel cell is:

[0215]

[0216] In the formula, V cell is the module output voltage; E nernst is the thermodynamic electromotive force; U act is the activation overvoltage; U ohm is the ohmic overvoltage; U conc is the concentration overvoltage; V fc is the fuel cell stack voltage; N fc is the number of series-connected cells;

[0217] The activation overvoltage is:

[0218] U act = [ξ1 + ξ2T fc + ξ3 ln(C O2 ) + ξ4 × T fc × ln(I fc )] (14)

[0219] Among them,

[0220]

[0221] In the formula, ξ1, ξ2, ξ3 and ξ4 are empirical parameters; I fc is the fuel cell current; C O2 is the oxygen concentration at the cathode gas-liquid interface; P O2 is the oxygen interface partial pressure of the fuel cell; T fc is the operating temperature of the fuel cell;

[0222] The ohmic overvoltage is:

[0223] U ohm = I fc R ohm = I fc (r M l / A + R c ) (16)

[0224] Wherein,

[0225]

[0226] In the formula, l is the fuel cell membrane thickness; r M is the fuel cell resistivity; R c is the impedance that hinders protons from passing through the proton exchange membrane; A is the effective area of the proton exchange membrane; R ohm is the ohmic resistance; λ is the water content of the proton exchange membrane;

[0227] The concentration overvoltage is:

[0228] U conc = -Bln(1 - J / J max ) (18)

[0229] In the formula, J and J max are the current density and the maximum current density respectively; B is the equation constant, which is determined by the operating conditions of the fuel cell.

[0230] In the reliability index screening unit, according to the established coupling mathematical model of the hydrogen - electricity coupling system, analyze the influencing factors of the operation of each subsystem, and screen out the reliability indexes from them.

[0231] The said reliability indexes include hydrogen storage rate, hydrogen storage tank pressure, hydrogen production rate, electrolyzer output voltage, heat energy taken away by the electrolyzer cooling water, heat energy generated by the electrolyzer, hydrogen tank pressure, current density, fuel cell temperature, and fuel cell oxygen interface partial pressure.

[0232] The specific content of the simulation model construction unit is as follows:

[0233] The one - dimensional convolutional layer performs a convolution operation on the local area of the operation data of the hydrogen - electricity coupling system and outputs the corresponding feature map, as shown in Equation (19);

[0234] c t = f(W cnn * n t + b cnn ) (19)

[0235] Wherein, W cnn represents the weight matrix of the filter in the convolution operation of the input parameters of the hydrogen - electricity coupling system; nt represents the t-th running data information; * represents the convolution operation between the input parameter and W cnn and b cnn represents the bias parameter in the convolution operation, f represents the activation function in the convolution operation, and c t represents the output sequence after the convolution operation;

[0236] In the pooling layer, max pooling is used to remove redundant information in the data, compress the feature information extracted by the previous hidden layer, and simplify the complexity of the hybrid neural network;

[0237] The activation function uses the scaled exponential linear unit to perform a non-linear transformation on each convolution operation, expressed as:

[0238]

[0239] where α = 1.6733 and λ0 = 1.0507.

[0240] The bidirectional long short-term memory network is set with two different hidden layers, namely the forward LSTM hidden layer and the backward LSTM hidden layer. The forward hidden layer reads the operation data of the hydrogen-electric coupling system in ascending order, outputs a sequence after convolution and max pooling operations, and the backward hidden layer reads the operation data of the hydrogen-electric coupling system in descending order, outputs a sequence after convolution and max pooling operations;

[0241] A dropout layer is set to improve the generalization ability of the CNN-BiLSTM hybrid neural network. The gray neural points in the dropout layer do not participate in network training. The output of the CNN-BiLSTM hybrid neural network with the dropout layer added is shown in Equation (28):

[0242]

[0243] where the dropout layer uses the Bernoulli function to randomly generate a vector r of 0 or 1 with probability p, so that some hidden layer nodes are inactivated; σ represents the sigmoid function in BiLSTM; W o represents the weight of the input feature in the BiLSTM training; is the forward hidden state at time t - 1; x t is the feature quantity input at time t; b o is the bias parameter in the convolution operation process;

[0244] During the training process of the CNN-BiLSTM hybrid neural network, the adaptive moment estimation optimization function is used to iteratively update the weights and biases of the CNN-BiLSTM hybrid neural network, and finally obtain the trained output characteristic model of the hydrogen-electric coupling system, that is, the CNN-BiLSTM hybrid neural network.

[0245] 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 foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A reliability evaluation method for a hydrogen-electric coupling system, characterized in that Including: 1) Collecting the operation data of the hydrogen-electricity coupling system through device sensors; 2) Analyzing the participation processes of the hydrogen production system, fuel cell system, and hydrogen storage system in the hydrogen-electricity coupling system, and establishing a coupling mathematical model of the hydrogen-electricity coupling system, including an alkaline electrolyzer model, a hydrogen storage model, and a fuel cell model; 3) Screening the reliability indicators of the hydrogen-electricity coupling system; 4) Constructing a hybrid neural network integrating a convolutional neural network and a bidirectional long short-term memory network according to the mathematical model of the hydrogen-electricity coupling system, and training the constructed hybrid neural network using a training data set; 5) Using the trained hybrid neural network to perform output prediction on a test data set; 6) Calculating the reliability indicators of the hydrogen production system, fuel cell system, and hydrogen storage system, and judging the working state of each system, and finally comprehensively judging the reliability of the hydrogen-electricity coupling system; The specific process of step 4) is as follows: Step 401: The one-dimensional convolutional layer performs a convolution operation on the local area of the operation data of the hydrogen-electricity coupling system, and outputs the corresponding feature map, as shown in Equation (19); c t = f(W cnn * n t + b cnn ) (19) Among them, W cnn represents the weight matrix of the filter in the convolution operation of the input parameters of the hydrogen-electric coupling system; n t represents the t-th running data information; * represents the convolution operation between the input parameters and W cnn ; b cnn represents the bias parameter in the convolution operation, f represents the activation function in the convolution operation, and c t represents the output sequence after the convolution operation; Step 402: In the pooling layer, max pooling is used to remove redundant information in the data, compress the feature information extracted by the previous hidden layer, and simplify the complexity of the hybrid neural network; Step 403: The activation function uses a scaled exponential linear unit to perform a non-linear transformation on each convolution operation, expressed as: where α = 1.6733 and λ0 = 1.0507; Step 404: The bidirectional long short-term memory network sets two different hidden layers, namely the forward LSTM hidden layer and the backward LSTM hidden layer. The forward LSTM hidden layer reads the operation data of the hydrogen-electricity coupling system in ascending order, and outputs a sequence after convolution and max pooling operations. The backward LSTM hidden layer reads the operation data of the hydrogen-electricity coupling system in descending order, and outputs a sequence after convolution and max pooling operations; Step 405: A dropout layer is set to improve the generalization ability of the CNN-BiLSTM hybrid neural network. The gray neurons in the dropout layer do not participate in network training. The output of the CNN-BiLSTM hybrid neural network with the dropout layer added is as shown in Equation (28); Among them, the dropout layer randomly generates a vector r of 0 or 1 with probability p using the Bernoulli function, so that some hidden layer nodes are inactivated; σ represents the sigmoid function in the BiLSTM; W o represents the weight of the input features in the BiLSTM training; is the forward hidden state at time t-1; x t is the feature quantity input at time t; b o is the bias parameter in the convolution operation process; Step 406: During the training process of the CNN-BiLSTM hybrid neural network, an adaptive moment estimation optimization function is used to iteratively update the weights and biases of the CNN-BiLSTM hybrid neural network.

2. The reliability evaluation method of the hydrogen-electric coupling system according to claim 1, wherein The operation data of the hydrogen-electricity coupling system includes the electrolyzer voltage, electrolyzer power, hydrogen storage tank pressure, and fuel cell power.

3. The reliability evaluation method of the hydrogen-electric coupling system according to claim 1, wherein In step 2), the coupling mathematical model includes: 1) Hydrogen storage model The hydrogen storage rate is expressed as: Where n sto (t) is the net hydrogen storage rate of the hydrogen storage tank at time t; is the hydrogen intake rate of the hydrogen storage tank at time t; is the hydrogen output rate of the hydrogen storage tank at time t; The hydrogen storage amount is: Where n sto (t0) is the hydrogen storage amount in the hydrogen storage tank at time t0; n sto (t0 + Δt) is the hydrogen storage amount in the hydrogen storage tank within the time of t0 + Δt; t0 represents the time point; Δt represents the time quantity; The pressure p of the hydrogen storage tank is obtained from the gas state equation sto as follows: Where: T sto is the hydrogen storage temperature, V sto is the volume of the hydrogen storage tank, and R is the ideal gas constant.

4. The reliability evaluation method of the hydrogen-electric coupling system according to claim 3, characterized in that In step 2), the coupling mathematical model further includes: 2) Alkaline electrolyzer model The U-I equation of the alkaline electrolyzer is: Where: ΔG is the change in Gibbs free energy during the electrochemical reaction process, z is the number of electrons transferred per reaction; F is the Faraday constant; r1 is the electrolyte ohmic resistance constant, r2 is the first-order parameter of the electrolyte ohmic resistance; T el is the electrolytic cell temperature; A cell is the area of the electrolytic cell module; I el is the direct current; s1, s2, and s3 are the zero-order term coefficient, first-order term coefficient, and second-order term coefficient of the electrode overvoltage respectively, and t1, t2, and t3 are the zero-order term coefficient, first-order term coefficient, and second-order term coefficient of the electrode overvoltage coefficient respectively; U cell is the output voltage of the electrolytic cell; Hydrogen production rate of alkaline electrolyzer is as follows: Among them, η F The expression is: Where: η F is the Faraday efficiency; N el is the number of electrolyzer series modules; z is the number of electrons transferred per reaction; a1, a2, a3, a4, and a5 are Faraday efficiency coefficients; The internal heat balance equation of the electrolyzer is: where Q cool = C cw (T cw,i - T cw,o ) = hA HX LMTD (10) Where: Q gen is the thermal energy generated by the electrolytic cell; Q loss is the thermal energy lost by the electrolytic cell; Q cool is the thermal energy carried away by the cooling water of the electrolytic cell; C t represents the total heat capacity, ρ j , V j , c j respectively represent the density, volume and heat capacity of the components of the electrolytic cell; represents the thermal neutral voltage; E total represents the electrolytic cell voltage; T a is the ambient temperature; A stack is the outer surface area of the electrolytic cell; h is the heat transfer coefficient; T sep is the outlet temperature of the gas-liquid separator; I is the electrolytic cell current; h0 is the heat transfer coefficient at zero current density; k is the proportionality coefficient; A sep is the surface area of the gas-liquid separator; h sep , h stack are the heat transfer coefficients of the gas-liquid separator and the electrolytic cell respectively; T cw,i , T cw,o are the temperatures of the cooling water entering and leaving the electrolytic cell respectively; C cw is the heat capacity of the cooling water; LMTD is the logarithmic mean temperature difference, representing the demand for auxiliary cooling; hA HX represents the product of the total heat transfer coefficient and the area of the electrolytic cell; T cw, is the operating temperature of the electrolytic cell.

5. The reliability evaluation method of the hydrogen-electric coupling system according to claim 4, wherein In step 2), the coupling mathematical model further includes: 3) Fuel cell model The output voltage of the fuel cell is: Wherein, V cell is the output voltage of the module; E nernst is the thermoelectric motive force; U act is the activation overvoltage; U ohm is the ohmic overvoltage; U conc is the concentration overvoltage; V fc is the voltage of the fuel cell stack; N fc is the number of series-connected cells; The activation overvoltage is: U act = [ξ1 + ξ2T fc + ξ3ln(C O2 ) + ξ4×T fc ×ln(I fc )] (14) where where ξ1, ξ2, ξ3, and ξ4 are empirical parameters; I fc is the fuel cell current; C O2 is the oxygen concentration at the cathode gas-liquid interface; P O2 is the oxygen interface partial pressure of the fuel cell; T fc is the operating temperature of the fuel cell; The ohmic overvoltage is: U ohm = I fc R ohm = I fc (r M l / A + R c ) (16) where where l is the thickness of the fuel cell membrane; r M is the resistivity of the fuel cell; R c is the impedance that hinders the passage of protons through the proton exchange membrane; A is the effective area of the proton exchange membrane; R ohm is the ohmic resistance; λ is the water content of the proton exchange membrane; The concentration overvoltage is: U conc = -Bln(1 - J / J max ) (18) Wherein, J and J max are the current density and the maximum current density respectively; B is the equation constant, which is determined by the operating conditions of the fuel cell.

6. The reliability evaluation method of the hydrogen-electric coupling system according to claim 1, wherein In step 3), according to the established coupling mathematical model of the hydrogen-electricity coupling system, analyze the influencing factors of the operation of each system, and screen out the reliability indicators from them.

7. The reliability evaluation method of the hydrogen-electric coupling system according to claim 6, characterized in that The reliability indicators described above include hydrogen storage rate, hydrogen storage tank pressure, hydrogen production rate, electrolyzer output voltage, heat energy carried away by the electrolyzer cooling water, heat energy generated by the electrolyzer, hydrogen gas tank pressure, current density, fuel cell temperature, and fuel cell oxygen interface partial pressure.

8. Hydrogen-electric coupling system reliability evaluation system, characterized in that including: Data acquisition unit: Collects the operation data of the hydrogen-electricity coupling system through device sensors; Coupled mathematical model establishment unit: Analyzes the participation processes of the hydrogen production system, fuel cell system, and hydrogen storage system in the hydrogen-electricity coupling system, and establishes a coupled mathematical model of the hydrogen-electricity coupling system, including an alkaline electrolyzer model, a hydrogen storage model, and a fuel cell model; Reliability indicator screening unit: Screens the reliability indicators of the hydrogen-electricity coupling system; Simulation model construction unit: Constructs a hybrid neural network integrating a convolutional neural network and a bidirectional long short-term memory network according to the mathematical model of the hydrogen-electricity coupling system, and trains the hybrid neural network using the training data set; Output prediction unit: Performs output prediction on the test data set using the trained hybrid neural network; Reliability judgment unit: Calculates the reliability indicators of the hydrogen production system, fuel cell system, and hydrogen storage system, discriminates the working states of each system, and finally comprehensively judges the reliability of the hydrogen-electricity coupling system; The specific content of the simulation model construction unit is as follows: The one-dimensional convolutional layer performs a convolutional operation on the local area of the operation data of the hydrogen-electricity coupling system and outputs the corresponding feature map, as shown in Equation (19); c t = f(W cnn * n t + b cnn ) (19) Among them, W cnn represents the weight matrix of the filter in the convolution operation of the input parameters of the hydrogen-electric coupling system; n t represents the t-th running data information; * represents the convolution operation between the input parameters and W cnn ; b cnn represents the bias parameter in the convolution operation, f represents the activation function in the convolution operation, c t represents the output sequence after the convolution operation ends; In the pooling layer, max pooling is used to remove redundant information in the data, compress the feature information extracted by the previous hidden layer, and simplify the complexity of the hybrid neural network; The activation function uses a scaled exponential linear unit to perform a non-linear transformation on each convolutional operation, expressed as: where α = 1.6733 and λ0 = 1.0507; The bidirectional long short-term memory network sets two different hidden layers, namely the forward LSTM hidden layer and the backward LSTM hidden layer. The forward LSTM hidden layer reads the operation data of the hydrogen-electricity coupling system in ascending order and outputs a sequence after convolutional and max pooling operations. The backward LSTM hidden layer reads the operation data of the hydrogen-electricity coupling system in descending order and outputs a sequence after convolutional and max pooling operations; A dropout layer is set to improve the generalization ability of the CNN-BiLSTM hybrid neural network. The gray neural points in the dropout layer do not participate in network training. The output of the CNN-BiLSTM hybrid neural network with the dropout layer added is shown in Equation (28): Among them, the dropout layer randomly generates a vector r of 0 or 1 with probability p using the Bernoulli function, thereby inactivating some hidden layer nodes; σ represents the sigmoid function in the BiLSTM; W o represents the weight of the input features in the BiLSTM training; is the forward hidden state at time t-1; x t is the feature quantity input at time t; b o is the bias parameter in the convolution operation process; During the training process of the CNN-BiLSTM hybrid neural network, the adaptive moment estimation optimization function is used to iteratively update the weights and biases of the CNN-BiLSTM hybrid neural network.

Images