Fuel cell air compressor flow adaptive control method and system

Abstract

The present invention relates to a method and system for adaptive flow control of a fuel cell air compressor. The method comprises the following steps: collecting an air inlet temperature signal, an air outlet temperature signal, an air inlet pressure signal, an air outlet pressure signal, an air compressor speed signal and an air compressor output flow signal of the fuel cell air compressor, and calculating an operation characteristic matrix; performing dead zone quantization on the operation characteristic matrix, generating an operation parameter matrix, and constructing a double-layer action evaluation network structure to perform online evaluation of the operation state, and generating air compressor control evaluation parameters; establishing a state constraint function according to the air compressor control evaluation parameters, and calculating the stability constraint boundary of the fuel cell air compressor; performing recursive optimization of the air compressor control parameters, and obtaining an air compressor speed control sequence. The present invention enhances the ability of the air compressor to respond quickly to changes in operating conditions, reduces the complexity of control parameter setting, and effectively solves the problems of time-consuming parameter setting and low control accuracy in traditional control methods.

Description

Technical Field

[0001] The present invention relates to the technical field of air compressors, and in particular to a flow adaptive control method and system for a fuel cell air compressor. Background Art

[0002] Fuel cell vehicles are an important development direction of new energy vehicles, and one of the core components of their power system is the air supply system. As a key device in the air supply system, the flow control of the air compressor directly affects the power generation efficiency and system performance of the fuel cell. The existing air compressor flow control methods mainly use traditional methods such as PID control and fuzzy control. These methods often require complex iterative calculations during the control parameter setting process, and are difficult to adapt to the nonlinear characteristics and variable working conditions of the fuel cell air compressor system.

[0003] Traditional control methods have obvious deficiencies in dealing with the uncertainty of fuel cell air compressor systems. Since the air compressor system is affected by multiple factors such as temperature and pressure, its dynamic characteristics show strong coupling and strong nonlinearity, which makes it difficult for control strategies based on fixed parameters to meet the control requirements under different working conditions. At the same time, traditional control methods have limited ability to suppress system disturbances, and are prone to overshoot and oscillation when working conditions change rapidly, affecting the stability of the system. Summary of the invention

[0004] The main purpose of the present invention is to provide a fuel cell air compressor flow adaptive control method and system, the present invention enhances the air compressor's ability to respond quickly to changes in operating conditions, reduces the complexity of control parameter setting, and effectively solves the problems of time-consuming parameter setting and low control accuracy in traditional control methods.

[0005] To achieve the above object, the present invention provides a method for adaptively controlling flow of a fuel cell air compressor, comprising the following steps:

[0006] Collect the air inlet temperature signal, air outlet temperature signal, air inlet pressure signal, air outlet pressure signal, air compressor speed signal and air compressor output flow signal of the fuel cell air compressor, and calculate the operation characteristic matrix;

[0007] Quantifying the dead zone of the operating characteristic matrix to generate an operating parameter matrix, and constructing a two-layer action evaluation network structure to perform online evaluation of the operating status to generate air compressor control evaluation parameters;

[0008] Establishing a state constraint function according to the air compressor control evaluation parameter, and calculating the stability constraint boundary of the fuel cell air compressor;

[0009] The air compressor control parameters are recursively optimized according to the operating characteristic matrix and the stability constraint boundary to obtain the air compressor speed control sequence.

[0010] The present invention also provides a fuel cell air compressor flow adaptive control system, comprising:

[0011] A collection unit is used to collect the air inlet temperature signal, the air outlet temperature signal, the air inlet pressure signal, the air outlet pressure signal, the air compressor speed signal and the air compressor output flow signal of the fuel cell air compressor, and calculate the operation characteristic matrix;

[0012] An evaluation unit, used to quantify the dead zone of the operation characteristic matrix, generate an operation parameter matrix, and construct a two-layer action evaluation network structure to perform online evaluation of the operation status and generate air compressor control evaluation parameters;

[0013] A calculation unit, used to establish a state constraint function according to the air compressor control evaluation parameter, and calculate the stability constraint boundary of the fuel cell air compressor;

[0014] The recursive optimization unit is used to perform recursive optimization of air compressor control parameters according to the operating characteristic matrix and the stability constraint boundary to obtain an air compressor speed control sequence.

[0015] In summary, the technical solution provided by the present invention effectively compensates for the nonlinear characteristics of the system by introducing a generalized disturbance matrix and hysteresis quantization processing, and the method of the present invention reduces the influence of system disturbance on control performance, and suppresses signal noise through a dead zone quantization mechanism, making the control signal more stable. A two-layer action evaluation network structure is used for online evaluation of the operating state, and a cross-training mechanism of an action network and an evaluation network is combined to enhance the rapid response capability of the control system to changes in operating conditions and reduce the complexity of setting the control parameters. Based on the design of the Lyapunov state constraint function, a theoretical guarantee mechanism for system stability is established, and the stable operation of the control system under different operating conditions is ensured by calculating the stability constraint boundary. The control parameters are calculated by a recursive optimization method, and the control strategy is dynamically adjusted by an online update mechanism, which avoids the problem of repeated iterative calculations in traditional methods and ensures the convergence of the control parameters. Through a closed-loop control structure and a real-time flow deviation compensation mechanism, a complete automatic adjustment process for control parameters is established, which effectively solves the problems of time-consuming parameter setting and low control accuracy in traditional control methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of the steps of a method for adaptively controlling flow rate of a fuel cell air compressor in one embodiment of the present invention;

[0017] Figure 2 It is a structural block diagram of a fuel cell air compressor flow adaptive control system in one embodiment of the present invention.

[0018] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] Reference Figure 1 , this embodiment provides a fuel cell air compressor flow adaptive control method, comprising the following steps:

[0021] S1, collecting the air inlet temperature signal, air outlet temperature signal, air inlet pressure signal, air outlet pressure signal, air compressor speed signal and air compressor output flow signal of the fuel cell air compressor, and calculating the operation characteristic matrix;

[0022] Among them, the inlet temperature signal, outlet temperature signal, inlet pressure signal, outlet pressure signal, compressor speed signal and compressor output flow signal of the fuel cell air compressor are collected in time series. During the operation of the air compressor, the above multiple parameters are monitored in real time, and the data at different time points are recorded to form an observation data matrix. Each row of the matrix corresponds to the observation data at a time point, and each column corresponds to a specific physical quantity. At the same time, the observation data matrix is ​​range mapped, and the values ​​of different physical quantities are adjusted to a relatively uniform numerical range through normalization, standardization or other nonlinear transformation methods to reduce the influence of dimensional differences on the calculation results, and the mapped data matrix is ​​obtained. Based on the mapped data matrix, the least squares method is used to calculate to obtain the initial element value of the generalized perturbation matrix. The least squares method minimizes the sum of squares of the error through the optimal approximation method, thereby obtaining the best fitting solution for the data. In this process, each initial element value of the generalized perturbation matrix is ​​obtained by solving a series of linear or nonlinear equations, reflecting the perturbation characteristics of the fuel cell air compressor under different operating conditions. Since there are various nonlinear factors in the operation of the air compressor, such as eddy currents generated during gas compression, fluctuations in physical parameters caused by temperature changes, and the influence of air flow inertia, the generalized disturbance matrix obtained by the initial calculation is compensated nonlinearly to eliminate or reduce the error caused by linear approximation. The correction factor is introduced to correct the initial element value of the generalized disturbance matrix so that it can more accurately reflect the true dynamic characteristics of the fuel cell air compressor. The selection of the correction factor is based on historical data statistics, empirical models, or adaptive adjustment methods based on machine learning to ensure that the corrected generalized disturbance matrix is ​​more in line with the actual working conditions. Based on the matrix equation, the corrected generalized disturbance matrix and the system output matrix are solved. By constructing a mathematical model that describes the dynamic characteristics of the fuel cell air compressor and combining it with the solution algorithm, the disturbance characteristic matrix is ​​calculated to reveal the disturbance law of the air compressor under different operating conditions and the coupling relationship between various physical parameters. The disturbance characteristic matrix is ​​optimized and calculated to improve its adaptability and stability under different working conditions. The optimization calculation process is based on optimization algorithms, such as gradient descent, genetic algorithm or particle swarm optimization, to adjust the weight parameters in the disturbance characteristic matrix to minimize the disturbance effect of the system and ensure that the calculated matrix dynamically adapts to the real-time operating status of the fuel cell air compressor, and finally generates an operating characteristic matrix.

[0023] S2, quantify the dead zone of the operation characteristic matrix, generate the operation parameter matrix, and construct a two-layer action evaluation network structure to perform online evaluation of the operation status and generate the air compressor control evaluation parameters;

[0024] Specifically, an n-bit quantizer is configured for the operating characteristic matrix so as to reduce the data dimension in subsequent calculations, improve the calculation efficiency, and reduce the control instability caused by high-frequency small fluctuations. By setting an appropriate number of quantization bits, the continuous dynamic characteristic information is mapped to a discrete numerical space to obtain a quantization configuration parameter. Based on the quantization configuration parameters, a dead zone threshold is set for the dynamic characteristic information in the operating characteristic matrix, and a dead zone threshold matrix is ​​constructed. The matrix is ​​used to define the minimum perceptible change range of different physical quantities, so that small changes will not trigger the adjustment of the control signal to avoid control jitter caused by high-frequency noise. The dynamic characteristic information in the operating characteristic matrix is ​​processed by signal segmentation according to the set dead zone threshold, and the values ​​of each signal channel are classified according to the set range of the dead zone threshold so that it meets certain segmentation interval requirements, and a segmented signal matrix is ​​obtained, thereby effectively distinguishing signal changes in different amplitude ranges in data processing, making the system more sensitive to key changes, and not responding to small disturbances, thereby improving the stability of the control system. The segmented signal matrix is ​​subjected to hysteresis quantization processing, and a dead zone holding interval is set on this basis to ensure that when the signal change amplitude is within a certain range, the system will not immediately adjust the control parameters, but will update them after crossing the set holding interval, thereby reducing signal jitter and obtaining a quantization processing matrix. The quantization processing matrix is ​​subjected to signal jump suppression by a hysteresis comparator. The function of the hysteresis comparator is to make the system have a certain hysteresis in the process of signal change, that is, the state switching is triggered only when the signal change exceeds a certain threshold, effectively avoiding the oscillation phenomenon caused by the excessively high data update frequency, and ensuring the continuity of the control signal, and obtaining a smooth signal matrix. The smooth signal matrix is ​​normalized and transformed to generate an operating parameter matrix. Based on the operating parameter matrix, a two-layer action evaluation network structure is constructed to perform online evaluation of the operating status of the fuel cell air compressor. This network structure consists of two parts. The first layer of the network is used for local state analysis. Its main function is to extract features from data within a short time window and identify instantaneous dynamic changes. The second layer of the network is used for global state analysis. Its role is to combine data trends over a longer time range to evaluate the long-term stability of the fuel cell air compressor and predict future trends in operating status. Through this two-layer structure, the evaluation accuracy is effectively improved, the misjudgment caused by short-term fluctuations is reduced, and the air compressor control strategy is made more accurate and stable. Finally, the air compressor control evaluation parameters are generated to guide the adaptive control of the air compressor, so that it can dynamically adjust the speed and flow according to the real-time operating conditions, thereby maintaining the best operating state under different environmental conditions and improving the overall efficiency and stability of the fuel cell system.

[0025] The operating parameter matrix is ​​divided into an action data matrix and an evaluation data matrix. The action data matrix is ​​used to drive control decisions, while the evaluation data matrix is ​​used to evaluate the operating status of the air compressor and optimize the control strategy. An action network is constructed based on the action data matrix. The action network adopts a feedforward neural network structure and includes an input layer, a hidden layer, and an output layer. The input layer contains 5 neurons, corresponding to temperature, pressure, speed, and flow signals, respectively. These input data are the core parameters that affect the operating status of the fuel cell air compressor, and are input into the network after normalization processing to ensure the consistency of the data range and the stability of the calculation. In order to improve the nonlinear expression ability of the network, the hidden layer is designed to contain 30 neurons, and the ReLU activation function is used to ensure strong feature extraction capabilities when processing large-scale input data, while effectively preventing the gradient disappearance problem. In the output layer, the network generates a control action value based on the input data. The action value represents the optimal control strategy of the air compressor in the current state, and the output data of the action network is obtained. At the same time, the evaluation network is constructed based on the evaluation data matrix, and the output data of the action network is used as its input to evaluate the impact of the current control strategy on the operating state of the air compressor. In the evaluation network, the input layer directly receives the output data of the action network and combines the characteristic information of the evaluation data matrix to ensure the accuracy of the evaluation results. The hidden layer of the evaluation network contains 20 neurons and adopts the Tanh activation function. Unlike ReLU, Tanh can provide better gradient flow in the negative value area, so that the evaluation network has a more stable gradient update characteristic during the calculation process, thereby improving the accuracy and convergence speed of the evaluation. The output layer of the evaluation network generates an evaluation value, which is used to measure the effectiveness of the current control strategy. In order to improve the learning effect of the network, the action network and the evaluation network are cross-trained, that is, during the training process, the action network and the evaluation network influence each other, so that the action network can adapt to the feedback of the evaluation network while optimizing the control strategy, and adjust the control strategy to obtain a better control effect. The specific method of cross-training is to correct the loss function of the action network by the output value of the evaluation network, thereby guiding the learning direction of the action network, so that the control strategy improves the operating efficiency of the fuel cell air compressor and meets the requirements of safety and stability. During the cross-training process, the loss function of the two-layer action evaluation network structure is calculated and the performance of the network is evaluated to ensure the effectiveness of the training and the convergence of the model. The loss function calculation is based on the mean square error or cross entropy loss function, and the network parameters are adjusted in combination with the gradient descent optimization algorithm.Based on the network evaluation results, the action network and the evaluation network in the double-layer action evaluation network structure are updated online to ensure that the model adapts to the changing operating state of the fuel cell air compressor. The key to achieving online parameter update is to adjust the network weights through the back propagation algorithm. The back propagation algorithm calculates the gradient of the network based on the chain rule and uses optimization methods such as Adam or SGD to update the weights, thereby ensuring that the network can dynamically optimize the control strategy under different environments and finally generate the air compressor control evaluation parameters.

[0026] S3, establishing a state constraint function according to the air compressor control evaluation parameters, and calculating the stability constraint boundary of the fuel cell air compressor;

[0027] It should be noted that the positive definite symmetric matrix is ​​constructed for the air compressor control evaluation parameters to ensure that the system state matrix has good mathematical properties and is suitable for stability analysis. In mathematical modeling, the characteristics of the positive definite symmetric matrix enable it to ensure that the energy function of the state variable is always positive, and a stable solution can be obtained through eigenvalue decomposition. In the process of mapping the air compressor control evaluation parameters to the positive definite symmetric matrix, appropriate weight factors are selected to ensure the balance of the state variables in each dimension, and the matrix is ​​further optimized through quadratic function transformation to characterize the characteristics of the fuel cell air compressor under different operating conditions and form a complete state matrix. Based on the state matrix, a nonlinear integral term is designed to more accurately characterize the operating characteristics of the fuel cell air compressor. The role of the nonlinear integral term is to describe the cumulative influence of the state variable, and through the hyperbolic tangent function operation, the constraint term has a smooth transition effect in different state spaces. Since the hyperbolic tangent function has good asymptotics and differentiability, it ensures that when the state variable changes greatly, a stable numerical output is maintained, thereby avoiding the control instability caused by mutations. By combining the state matrix and nonlinear constraint terms, a Lyapunov state constraint function is constructed to ensure that the operating state of the fuel cell air compressor always tends to a stable energy domain, so that it can maintain balanced operation under different working conditions and quickly recover to a stable state when subjected to external disturbances. The time derivative of the state constraint function is calculated to analyze the change trend of the system in the time dimension. The calculation of the time derivative not only reveals the evolution law of the system under different instantaneous states, but also is used to determine whether the state meets the stability constraint conditions. When calculating the derivative results, the Euclidean norm of the state vector is introduced, and the comparison operation is used to determine whether the state variable is within the constraint range. Since the calculation of the Euclidean norm can reflect the overall amplitude change of the state vector, it is an important indicator to measure the degree to which the operating state of the air compressor deviates from the stable trajectory. By comparing the derivative calculation results with the Euclidean norm, the state constraint conditions are obtained, and based on this, it is determined whether the current operating state of the fuel cell air compressor meets the stability requirements. Based on the state constraints, the rated operating range of the fuel cell air compressor is divided to ensure that the control strategy in different working ranges is reasonably set. In the division process, appropriate stability parameters are set to ensure that a reasonable constraint range is established under different working conditions. The selection of these stability parameters is based on the operating characteristics of the fuel cell air compressor and its response under different ambient temperatures, load changes and other factors. By optimizing the stability parameters, it is ensured that the constraint range is neither too loose to reduce the control accuracy nor too strict to affect the dynamic adjustment ability of the system. Based on the constraint range, the boundary calculation is performed, and combined with the operating characteristics and actual working conditions of the system, the stability constraint boundary of the fuel cell air compressor is finally generated.

[0028] S4, recursively optimize the air compressor control parameters according to the operating characteristic matrix and the stability constraint boundary to obtain the air compressor speed control sequence.

[0029] Specifically, the control input weight matrix and the tracking error weight matrix are constructed based on the operating characteristic matrix to ensure that the control system balances the change of the control input and the deviation of the tracking target. The control input weight matrix is ​​used to describe the impact of different control inputs on the system performance, while the tracking error weight matrix is ​​used to measure the deviation between the current state of the air compressor and the target state. By reasonably setting the value of the weight matrix, the control system can reduce the error in the control process while ensuring the dynamic response speed. On this basis, the control input weight matrix and the tracking error weight matrix are substituted into the quadratic programming function to construct the optimization objective function, so that the optimization process of the control parameters is carried out in a way that minimizes the control energy consumption and error. Control input constraints and control increment constraints are imposed on the optimization objective function to ensure that the optimization results meet the actual control constraints. Among them, the control input constraints are used to ensure that the speed, pressure and flow of the air compressor will not exceed the safety range of the system, while the control increment constraints are used to prevent sudden changes in the control parameters to avoid drastic fluctuations in the system in a short period of time, affecting the stability of the fuel cell air compressor. At the same time, the stability constraint boundary is converted into a Lyapunov derivative constraint to ensure that the stability requirements of the fuel cell air compressor are met during the optimization process. The role of the Lyapunov derivative constraint is to constrain the energy function of the control system so that it always remains in a stable energy domain and converges to a stable point over time to ensure that the operating state of the air compressor does not deviate from the normal range. After the optimization constraints are applied, the entire optimization problem is converted into an optimal control problem with constraints, thereby ensuring that the solved control parameters can meet the dynamic adjustment requirements without affecting the stability of the system. The optimization objective function and the optimization constraints are input into the online recursive solver to calculate the control parameters under the discrete time series and obtain the initial sequence of control parameters. The online recursive solver gradually optimizes the control parameters at different time steps through iterative calculation to ensure that the control sequence adapts to the real-time operating state of the fuel cell air compressor. In this process, the recursive least squares algorithm, model predictive control or reinforcement learning method are used to improve the convergence speed of the calculation and ensure the robustness and adaptability of the solution results under complex operating conditions. The learning step size of the initial sequence of control parameters is set to control the speed and amplitude of parameter adjustment, and incremental constraint processing is performed in combination with the operating conditions of the fuel cell air compressor to prevent sudden changes in the control parameters from having an adverse effect on the system. The core of the incremental constraint processing is to ensure that the adjustment process of the control parameters is smooth and in line with the dynamic response characteristics of the fuel cell air compressor by setting a reasonable step size, and to generate an air compressor speed control sequence, so that the air compressor can achieve adaptive flow regulation under different load conditions, and while ensuring system stability, improve the overall energy efficiency, and ensure that the fuel cell system can operate efficiently and reliably under various operating conditions.

[0030] The air compressor speed control sequence is converted into digital-to-analog conversion, and the digital control signal is converted into an analog signal so that it can be used to drive the motor actuator of the air compressor. A high-precision digital-to-analog converter is used to ensure that the converted signal has a small quantization error and reflects the changing trend of the digital control signal. The converted analog signal is power amplified to ensure that the signal strength is sufficient to drive the motor of the air compressor. The power amplification process depends on the power drive module, which dynamically adjusts the voltage and current output according to different load requirements to ensure the stability and reliability of the air compressor drive signal. During the normal operation of the air compressor, the target air flow is calculated according to the power demand of the fuel cell. The power demand of the fuel cell depends on the load condition and the real-time working state of the stack. Therefore, when calculating the target air flow, the current load, oxygen utilization rate and operating efficiency of the fuel cell are comprehensively considered to ensure that the calculated target air flow meets the actual needs of the fuel cell. At the same time, the output flow of the air compressor is collected in real time and the flow measurement data is obtained. The flow measurement process uses a high-precision flow sensor to ensure the accuracy of data acquisition and reduce the control deviation caused by measurement errors. The difference between the flow measurement data and the target air flow is calculated to obtain the flow deviation value, which reflects the gap between the current air compressor output flow and the actual demand of the fuel cell. Based on the flow deviation value, the operating status of the air compressor is evaluated, and it is determined whether the control parameters need to be adjusted to optimize the flow output. In order to achieve adaptive control, the flow deviation value is compared with the preset control threshold to obtain the deviation judgment result. The setting of the control threshold is based on the dynamic response characteristics of the system to ensure that the system tolerates a certain degree of error in the case of a small range of deviations without triggering too frequent control adjustments, thereby improving the stability and anti-interference ability of the system. Based on the deviation judgment result, the control parameters are updated online to ensure that the control strategy is continuously optimized as the operating status of the air compressor changes. The online update process uses adaptive control algorithms, such as PID adjustment, model predictive control or reinforcement learning methods, to dynamically adjust the control parameters according to different deviation conditions to ensure that the system can maintain the optimal flow output under different working conditions. After completing the update of the control parameters, the updated control parameters are converted into speed sequences. The control parameters are mapped to speed values ​​through the air compressor characteristic curve to obtain a new speed control sequence. The air compressor characteristic curve describes the relationship between speed and flow and pressure. Through reasonable curve fitting and interpolation calculation, it is ensured that the control parameters can accurately correspond to the appropriate speed values.The new speed control sequence is input into the air compressor controller, and the air compressor speed is adjusted in a closed loop to achieve dynamic control of the air compressor output flow. The closed loop control uses a real-time feedback mechanism to enable the control system to converge to the optimal state during the continuous adjustment process, thereby ensuring that the flow output of the fuel cell air compressor is always consistent with the needs of the fuel cell, improving the system's response speed and stability, and optimizing the overall performance of the fuel cell system.

[0031] In one example, the air inlet temperature signal, the air outlet temperature signal, the air inlet pressure signal, the air outlet pressure signal, the air compressor speed signal and the air compressor output flow signal of the fuel cell air compressor are collected, and the operation characteristic matrix is ​​calculated, including:

[0032] The inlet temperature signal, outlet temperature signal, inlet pressure signal, outlet pressure signal, compressor speed signal and compressor output flow signal of the fuel cell air compressor are collected in time series to obtain an observation data matrix;

[0033] Perform range mapping on the observed data matrix to obtain a mapped data matrix, and calculate the mapped data matrix based on the least squares principle to obtain the initial element values ​​of the generalized perturbation matrix;

[0034] Perform nonlinear compensation on the initial element values ​​of the generalized disturbance matrix, and correct it by introducing a correction factor to obtain a corrected generalized disturbance matrix;

[0035] The modified generalized disturbance matrix and the system output matrix are solved based on the matrix equation to obtain the disturbance characteristic matrix, which is then optimized to generate an operating characteristic matrix of dynamic characteristic information under different working conditions.

[0036] In this example, the inlet temperature signal, outlet temperature signal, inlet pressure signal, outlet pressure signal, compressor speed signal and compressor output flow signal are collected in time series to form an observation data matrix. At time , the measured values ​​of these signals are (inlet air temperature), (Outlet temperature), (inlet pressure), (Outlet pressure), (compressor speed) and (compressor output flow), over a period of time Inside, construct the observation data matrix:

[0037]

[0038] Each row of the matrix represents the system state at a certain moment, and each column represents the time series of a specific signal. The observed data matrix is ​​range mapped to eliminate the influence of the numerical scale on subsequent calculations. The mapped data matrix is ​​realized through linear normalization transformation, that is, for any element in the matrix Normalize:

[0039]

[0040] in, and They are The minimum and maximum values ​​of the signals, the mapped data matrix is ​​recorded as The mapped data matrix is ​​calculated based on the least squares method to obtain the initial element values ​​of the generalized perturbation matrix. Assume that the perturbation state of the system is determined by an unknown perturbation vector Impact, namely:

[0041]

[0042] in, is the system characteristic matrix, is a random error term. The optimal perturbation vector is solved using the least squares method :

[0043]

[0044] Get the initial element values ​​of the generalized perturbation matrix . Perform nonlinear compensation on the initial element values. In order to optimize the accuracy of the generalized perturbation matrix, the correction factor is introduced , which is corrected by a nonlinear transformation:

[0045]

[0046] in, is the correction weight coefficient, The hyperbolic tangent function is used to smooth small perturbations to reduce numerical instability. The corrected generalized perturbation matrix Depend on It is constructed to characterize the disturbance characteristics of the system. Based on the matrix equation, the modified generalized disturbance matrix With the system output matrix Perform a solve operation to obtain the disturbance characteristic matrix , where the system output matrix is ​​defined as:

[0047]

[0048] in, and are the input coefficient matrix and bias vector of the system respectively. Using the matrix solution method, we get:

[0049]

[0050] The disturbance characteristic matrix Describe the disturbance trend of the system under different working conditions and use it to optimize the calculation to generate the final operating characteristic matrix. , the gradient descent algorithm is used for iterative adjustment to make it more accurately adapt to the dynamic characteristics of the air compressor. The optimized operating characteristic matrix Obtained through the following iterative formula:

[0051]

[0052] in, is the loss function, defined using mean square error:

[0053]

[0054] Through continuous iterative calculation, the optimized operating characteristic matrix is ​​obtained , this matrix is ​​used to describe the dynamic characteristics of the fuel cell air compressor under different working conditions.

[0055] In one example, the dead zone of the operating characteristic matrix is ​​quantified to generate an operating parameter matrix, and a two-layer action evaluation network structure is constructed to perform online evaluation of the operating status to generate air compressor control evaluation parameters, including:

[0056] Performing n-bit quantizer configuration on the operation characteristic matrix to obtain quantization configuration parameters, and performing dead zone threshold setting on dynamic characteristic information in the operation characteristic matrix based on the quantization configuration parameters to obtain a dead zone threshold matrix;

[0057] The dynamic characteristic information in the operation characteristic matrix is ​​processed by signal segmentation according to the dead zone threshold matrix to obtain a segmented signal matrix;

[0058] Performing hysteresis quantization processing on the segmented signal matrix and setting a dead zone holding interval to obtain a quantization processing matrix;

[0059] The quantization processing matrix is ​​subjected to signal jump suppression by a hysteresis comparator to obtain a smooth signal matrix, and the smooth signal matrix is ​​normalized to generate an operation parameter matrix;

[0060] Based on the operating parameter matrix, a two-layer action evaluation network structure is constructed, and the operating status of the fuel cell air compressor is evaluated online to generate the air compressor control evaluation parameters.

[0061] In this example, an n-bit quantizer is configured for the operating characteristic matrix, and the dynamic characteristic data of the fuel cell air compressor is discretized to reduce the computational complexity and improve the stability of data processing. Suppose the operating characteristic matrix is , where each element Representative The time point characteristic values, such as inlet temperature, outlet temperature, inlet pressure, outlet pressure, compressor speed and output flow. In order to quantize the matrix into n bits, according to the set quantization bit number Calculate the quantization step size:

[0062]

[0063] in, and Respectively represent The maximum and minimum values ​​of the features, and Indicates the discretization step size of the feature. By quantizing the step size, the quantization index of each data point is calculated:

[0064]

[0065] Get the quantized matrix , which represents the discrete space representation of the operating characteristic matrix after quantization. At the same time, based on the quantization configuration parameters, the dead zone threshold is set for the dynamic characteristic information in the operating characteristic matrix to avoid unnecessary disturbances to the control system caused by high-frequency small fluctuations. Dead zone threshold matrix Calculated by the following formula:

[0066]

[0067] in, is an adjustment coefficient used to control the width of the dead zone. The dynamic characteristic information in the operating characteristic matrix is ​​processed in segments according to the dead zone threshold matrix to obtain a segmented signal matrix. , calculated as follows:

[0068]

[0069] When the data change amplitude is less than the dead zone threshold, the signal remains unchanged, thereby reducing frequent control adjustments and improving system stability. Perform hysteresis quantization and set the dead zone holding interval to eliminate the jitter caused by signal jump. Hysteresis quantization introduces an upper threshold and lower threshold To control the update of the signal:

[0070]

[0071] in, is the hysteresis factor, which is used to set the hysteresis range. The calculation is as follows:

[0072]

[0073] This process ensures that when the signal changes slightly, the system will not respond immediately, but will adjust after the change exceeds the hysteresis interval, thereby avoiding frequent adjustments caused by measurement noise or small disturbances in the system and improving the robustness of the system. The signal jump is suppressed by the hysteresis comparator to obtain a smooth signal matrix The hysteresis comparator adopts a dynamic threshold strategy to smooth the change trend of the signal. The calculation formula is:

[0074]

[0075] in, is the smoothing coefficient. This method can further reduce the instability caused by data jitter. The smoothed signal matrix is ​​normalized to ensure the uniform scale of the input data. The normalization transformation formula is as follows:

[0076]

[0077] Get the operating parameter matrix , which is used as input data to construct a two-layer action evaluation network structure. Based on the running parameter matrix , a two-layer action evaluation network structure is constructed to evaluate the operating status of the fuel cell air compressor online. The two-layer network includes an action network and an evaluation network, where the input layer of the action network contains 5 neurons, corresponding to the normalized temperature. The hidden layer of the pressure, speed and flow signals contains 30 neurons and uses the ReLU activation function:

[0078]

[0079] The output layer generates control action values , which represents the optimal control strategy under the current operating state. At the same time, the input layer of the evaluation network receives the output data of the action network , and evaluate it in combination with historical operation data. The hidden layer of the evaluation network contains 20 neurons and uses the Tanh activation function:

[0080]

[0081] The output layer generates evaluation values , which is used to measure the effectiveness of the current control strategy. During the training process, the action network and the evaluation network are cross-trained so that the action network adapts to the feedback of the evaluation network, and the control strategy is optimized through loss function calculation:

[0082]

[0083] in, and are the target control action and target evaluation value respectively, is the loss weight factor. The loss function is optimized by the gradient descent algorithm, and finally the air compressor control evaluation parameters for the fuel cell air compressor are obtained to achieve precise control based on adaptive learning.

[0084] In one example, based on the operating parameter matrix, a two-layer action evaluation network structure is constructed, and the operating status of the fuel cell air compressor is evaluated online to generate air compressor control evaluation parameters, including:

[0085] The operation parameter matrix is ​​divided into an action data matrix and an evaluation data matrix;

[0086] The action network is constructed based on the action data matrix. The action network includes an input layer, a hidden layer, and an output layer. The input layer contains 5 neurons corresponding to temperature, pressure, speed, and flow signals. The hidden layer contains 30 neurons and uses the ReLU activation function. The output layer generates the control action value and obtains the action network output data.

[0087] An evaluation network is constructed based on the evaluation data matrix. The evaluation network includes an input layer, a hidden layer, and an output layer. The input layer receives the output data of the action network. The hidden layer contains 20 neurons and uses the Tanh activation function. The output layer generates an evaluation value to obtain the output data of the evaluation network.

[0088] The action network and the evaluation network are cross-trained to obtain a two-layer action evaluation network structure, and the loss function is calculated and the network performance is evaluated on the two-layer action evaluation network structure to obtain a network evaluation result;

[0089] Based on the network evaluation results, the action network and evaluation network in the two-layer action evaluation network structure are updated online, and the network weights are adjusted through the back propagation algorithm to generate the air compressor control evaluation parameters.

[0090] In this example, the parameter matrix will be run Split to distinguish the data input of control decision and state evaluation, where the action data matrix Mainly used to calculate the control strategy of the air compressor, and the evaluation data matrix It is used to evaluate the effectiveness and stability of the control strategy to ensure that the system can maintain optimal performance under different working conditions. Include Group data, each group of data consists of Features, among which and Respectively expressed as:

[0091]

[0092] in, Select key variables such as temperature, pressure, speed and flow as the input of the action network. Select additional state information, such as environmental parameters, historical data, etc., as the input of the evaluation network. Based on the action data matrix Construct an action network, which is used to generate a control strategy suitable for the current working condition based on the input system state. The action network adopts a three-layer structure, including an input layer, a hidden layer, and an output layer. The input layer contains 5 neurons, corresponding to the temperature, pressure, speed, and flow signals of the fuel cell air compressor, and the hidden layer consists of 30 neurons and uses the ReLU activation function to enhance the ability to extract nonlinear features. The specific calculation method is as follows:

[0093]

[0094] The output of the hidden layer is represented by the weight matrix and the bias term Perform the calculation:

[0095]

[0096] in, Represents the output matrix of the hidden layer. The output layer calculates the control action value based on the results of the hidden layer. :

[0097]

[0098] in, is the output layer weight matrix, is the bias term, the output Represents the optimal control strategy of the current air compressor. Based on the evaluation data matrix Construct an evaluation network, which is used to evaluate the output of the action network to determine the effectiveness of the control strategy. The input layer of the evaluation network receives the output data of the action network. , and combined with the evaluation data matrix , forming the input matrix of the evaluation network:

[0099]

[0100] in, The control strategy and system status information are combined. The hidden layer of the evaluation network contains 20 neurons and uses the Tanh activation function:

[0101]

[0102] The output of the hidden layer is calculated as follows:

[0103]

[0104] in, To evaluate the weight matrix of the network, For the bias term, the output layer of the evaluation network calculates the evaluation value :

[0105]

[0106] in, is the weight matrix of the output layer, is the bias term, the output Represents the evaluation result of the current control strategy. In order to improve the learning ability of the network, the action network and the evaluation network are cross-trained so that the action network adapts to the feedback of the evaluation network and continuously optimizes the control strategy. The core of cross-training is to construct a joint loss function , to optimize the control strategy and evaluate the accuracy:

[0107]

[0108] in, and are the target control action and target evaluation value respectively, is the loss weight factor. In order to minimize the loss function, the back propagation algorithm is used to update the network weights. During round iterations, the weight update formula is as follows:

[0109]

[0110] in, is the learning rate, the gradient The calculation is as follows:

[0111]

[0112] This formula is used to optimize the network structure so that the action network can learn better control strategies and the evaluation network can provide more accurate feedback. Based on the evaluation results of the trained network, the action network and the evaluation network in the two-layer action evaluation network structure are updated online to ensure that the network can adapt to the real-time operating status of the fuel cell air compressor. The network weights are adjusted through the back propagation algorithm so that it can dynamically adapt to different working conditions, and finally the air compressor control evaluation parameters are generated. :

[0113]

[0114] in, is the Sigmoid function:

[0115]

[0116] It is used to map the evaluation results into the control evaluation parameter range, which is used to optimize the operating status of the air compressor.

[0117] In one example, a state constraint function is established according to the air compressor control evaluation parameters, and the stability constraint boundary of the fuel cell air compressor is calculated, including:

[0118] The positive symmetric matrix of the air compressor control evaluation parameters is constructed, and the state matrix is ​​obtained through quadratic function conversion.

[0119] Based on the state matrix, nonlinear integral terms are designed, hyperbolic tangent function operations are performed on the state variables to obtain nonlinear constraint terms, and the state matrix and nonlinear constraint terms are combined to construct the Lyapunov state constraint function to obtain the state constraint function;

[0120] The time derivative of the state constraint function is calculated to obtain the derivative calculation result, and the derivative calculation result is compared with the Euclidean norm of the state vector to obtain the state constraint condition;

[0121] The rated working range of the fuel cell air compressor is divided according to the state constraint conditions, and the stability parameters are set to obtain the constraint interval. The boundary calculation is performed based on the constraint interval to generate the stability constraint boundary of the fuel cell air compressor.

[0122] In this example, based on the control evaluation parameters Generate a positive definite symmetric matrix , to ensure that the system has good mathematical properties in stability analysis. Positive definite symmetric matrix The construction method of is as follows:

[0123]

[0124] in, To control the evaluation parameter matrix, is a positive constant to ensure has strict positive definiteness, and is the identity matrix. This construction method ensures The eigenvalues ​​of are all greater than zero, thus ensuring that the system has stable characteristics. Converted to a state matrix through a quadratic function :

[0125]

[0126] in, is a reversible matrix used to transform coordinates so that the state matrix Suitable for subsequent nonlinear analysis and constraint construction. Based on the state matrix A nonlinear integral term is designed to describe the nonlinear characteristics of the system state, and a hyperbolic tangent function operation is introduced to obtain a nonlinear constraint term. The definition of the nonlinear integral term is as follows:

[0127]

[0128] This integral term is used to describe the state matrix The cumulative effect over time, and in order to enhance the constraint characteristics, the state variables Perform a hyperbolic tangent transformation to obtain nonlinear constraints:

[0129]

[0130] in, To adjust the coefficients and control the sensitivity of the nonlinear constraints, the purpose of this transformation is to maintain a high sensitivity when the state variable is small, and to stabilize in a larger range to avoid violent fluctuations. and nonlinear constraints Combination, constructing Lyapunov state constraint function :

[0131]

[0132] in, As a Lyapunov function, it is used to analyze the stability of the system. Perform time derivative calculations to obtain derivative calculation results:

[0133]

[0134] in, represents the time derivative of the state variable, and the second term This reflects the nonlinear effect of the hyperbolic tangent function. In order to further evaluate the stability of the system, the derivative calculation results are compared with the Euclidean norm of the state vector to ensure that the Lyapunov function is non-increasing in time, that is:

[0135]

[0136] in, is the stability parameter. This constraint is used to ensure that the change of the Lyapunov function does not lead to unstable behavior of the system. Based on the state constraint, the rated operating range of the fuel cell air compressor is divided, and the stability parameter is set to define the stability standard of different operating ranges. Assume that the rated operating range of the fuel cell air compressor is determined by the speed. ,pressure and flow The stability parameter Defined as:

[0137]

[0138] in, is a weight coefficient used to characterize the influence of different physical quantities on the stability of the system. Based on this stability parameter, the constraint interval is defined:

[0139]

[0140] The constraint interval is used to limit the stability boundary of the fuel cell air compressor under different operating conditions and ensure that the system does not enter the unstable area. Based on the constraint interval, the stability constraint boundary of the fuel cell air compressor is calculated. The specific calculation method is:

[0141]

[0142] in, Represents the stability boundary of the fuel cell air compressor, which is used to define the feasible domain of the control system and guide the operation strategy of the fuel cell air compressor.

[0143] In one example, the air compressor control parameters are recursively optimized based on the operating characteristic matrix and the stability constraint boundary to obtain the air compressor speed control sequence, including:

[0144] According to the operation characteristic matrix, a control input weight matrix and a tracking error weight matrix are constructed, and the control input weight matrix and the tracking error weight matrix are substituted into the quadratic programming function to obtain the optimization objective function;

[0145] Apply control input constraints and control increment constraints to the optimization objective function, and transform the stability constraint boundary into a Lyapunov derivative constraint to obtain the optimization constraint condition;

[0146] Input the optimization objective function and optimization constraints into the online recursive solver, calculate the control parameters under discrete time series, and obtain the initial sequence of control parameters;

[0147] The learning step size of the initial sequence of control parameters is set, and the incremental constraint processing is performed in combination with the operating conditions of the fuel cell air compressor to obtain the air compressor speed control sequence.

[0148] In this example, the run characteristics matrix is ​​defined , where each element Representative The time step system variables, such as inlet temperature, outlet temperature, inlet pressure, outlet pressure, compressor speed and compressor output flow. On this basis, the control input weight matrix is ​​constructed. and the tracking error weight matrix , to ensure that the control system balances the size of the control input and the tracking error of the system output during the optimization process. The control input weight matrix The definition of is in the form of a diagonal matrix to reflect the impact of different control inputs on system performance:

[0149]

[0150] in, Representative The weight coefficients of the control inputs, and the tracking error weight matrix It is used to measure the deviation between the current system output and the target output:

[0151]

[0152] in, Representatives Substitute these two weight matrices into the quadratic programming function to construct the optimization objective function:

[0153]

[0154] in, For the The control input vector for each time step is Represents system output With target value The purpose of this optimization objective function is to reduce the fluctuation of the control input and obtain a more stable control strategy while ensuring the minimum output error. Control input constraints and control increment constraints are imposed on the optimization objective function to ensure that the optimization result meets the physical constraints of the system. Among them, the control input constraints are used to ensure that the speed, pressure and flow of the air compressor do not exceed the safety range. Assume that the upper and lower limits of the control input are and , then the constraint condition is expressed as:

[0155]

[0156] At the same time, in order to avoid system oscillation caused by drastic changes in control parameters, control increment restriction conditions are imposed, namely:

[0157]

[0158] in, and In order to ensure that the solution in the optimization process meets the system stability requirements, the stability constraint boundary is converted into a Lyapunov derivative constraint. The definition is as follows:

[0159]

[0160] in, is a symmetric positive definite matrix, and the Lyapunov derivative constraint requires that:

[0161]

[0162] This constraint ensures that the Lyapunov function decreases over time, thereby ensuring the asymptotic stability of the system. The optimization objective function and optimization constraints are input into the online recursive solver to calculate the control parameters under discrete time series. The recursive solver uses the model predictive control method and recalculates the optimal control input at each time step through rolling optimization. Assume that the state transition equation of the system is:

[0163]

[0164] in, is the system state vector, and are the state matrix and input matrix of the system respectively. Then, in the model predictive control framework, by optimizing the objective function:

[0165]

[0166] And satisfy the constraints:

[0167]

[0168] Use gradient descent or second-order optimization method to solve the initial sequence of control parameters The learning step size is set for the initial sequence of control parameters, and the incremental constraint processing is performed in combination with the operating conditions of the fuel cell air compressor to ensure the smoothness of the control adjustment. The setting of the learning step size is based on the adaptive adjustment rule:

[0169]

[0170] in, is the learning step size, is the adjustment factor, and the incremental constraint processing is based on a smooth adjustment strategy:

[0171]

[0172] in, It is the smoothing coefficient, which ensures that the change of control parameters will not be too drastic, and finally obtains the air compressor speed control sequence.

[0173] In one example, the fuel cell air compressor flow adaptive control method further includes:

[0174] Performing digital-to-analog conversion on the air compressor speed control sequence to obtain a converted analog signal, and performing power amplification on the converted analog signal to obtain an air compressor drive signal;

[0175] Calculate the target air flow rate according to the power demand of the fuel cell, and collect the output flow rate of the air compressor in real time to obtain the flow measurement data;

[0176] The difference between the flow measurement data and the target air flow is calculated to obtain a flow deviation value, and the flow deviation value is compared with a preset control threshold to obtain a deviation judgment result;

[0177] Based on the deviation judgment result, the control parameters are updated online to obtain updated control parameters, and the updated control parameters are converted into speed sequences. The control parameters are mapped into speed values ​​through the air compressor characteristic curve to obtain a new speed control sequence.

[0178] The new speed control sequence is input into the air compressor controller for closed-loop control, the air compressor speed is adjusted, and the dynamic control of the air compressor output flow is completed.

[0179] In this example, the discrete digital control signal is converted into an analog signal to drive the air compressor actuator. Assume that the speed control sequence of the air compressor is ,in Indicates The digital control value of the time step is obtained by optimization calculation and converted into an analog voltage signal by a digital-to-analog converter (DAC) , the conversion relationship is expressed as:

[0180]

[0181] in, It is the DAC conversion gain, which determines the ratio of the digital signal mapped to the analog voltage. After power amplification, to ensure that the driving force is sufficient to drive the air compressor motor, a power amplifier is used to convert it into the final air compressor drive signal :

[0182]

[0183] in, It is the power amplification factor, which determines the amplification ratio of the output current. As an input signal, it drives the air compressor to follow the set speed in the control sequence. During the operation of the air compressor, the target air flow is calculated according to the power demand of the fuel cell to ensure that the fuel cell has sufficient oxygen supply under different load conditions. Assume that the instantaneous power demand of the fuel cell is , then the target air flow Calculated by empirical formula:

[0184]

[0185] in, It is the mapping coefficient between air flow and fuel cell power demand, and its value depends on the specific design of the fuel cell system and the working efficiency of the fuel cell. At the same time, in order to ensure the closed-loop regulation capability of the control system, the actual output flow of the air compressor is collected in real time. , flow measurement is obtained using a mass flow sensor:

[0186]

[0187] in, represents the transfer function of the flow sensor, The flow signal voltage measured by the sensor. and measure flow Perform error calculation to obtain the flow deviation value:

[0188]

[0189] in, Indicates the current flow error, which is used to determine whether the air compressor needs to adjust the control parameters to optimize the operating status. In order to avoid too frequent adjustments, the flow deviation is compared with the preset control threshold. For comparison:

[0190]

[0191] in, As a result of the deviation judgment, if the deviation exceeds the threshold, the control parameter update is triggered, otherwise the current control state is maintained. Based on the deviation judgment result, the control parameters are updated online to ensure that the air compressor can adapt to different operating conditions. The control parameter update adopts an adaptive adjustment method, such as incremental PID control:

[0192]

[0193] in, are proportional, integral and differential gain parameters respectively, Represents the adjustment amount of the control parameter. The updated control parameter The calculation is as follows:

[0194]

[0195] in, is the control parameter value at the previous moment. In order to ensure that the updated control parameters can be correctly mapped to the speed of the air compressor, the control parameters are mapped to the speed of the air compressor through the air compressor characteristic curve. Convert to new speed value :

[0196]

[0197] in, Represents the characteristic mapping function of the air compressor, which is obtained based on experimental data fitting, for example:

[0198]

[0199] in, is the characteristic curve fitting coefficient. After conversion, the new speed control sequence is obtained:

[0200]

[0201] This sequence is used to guide the adjustment of the operating status of the air compressor. Input the air compressor controller and adjust the air compressor speed through closed-loop control to ensure the output flow Always close to the target flow The closed-loop control system adopts a feedback adjustment mechanism to update the control parameters at each time step so that the flow error Gradually converge to zero to achieve dynamic control of the air compressor output flow.

[0202] Reference Figure 2 , this embodiment provides a fuel cell air compressor flow adaptive control system, including:

[0203] The acquisition unit 1 is used to acquire the air inlet temperature signal, the air outlet temperature signal, the air inlet pressure signal, the air outlet pressure signal, the air compressor speed signal and the air compressor output flow signal of the fuel cell air compressor, and calculate the operation characteristic matrix;

[0204] Evaluation unit 2 is used to quantify the dead zone of the operation characteristic matrix, generate the operation parameter matrix, and construct a two-layer action evaluation network structure to perform online evaluation of the operation status and generate air compressor control evaluation parameters;

[0205] The calculation unit 3 is used to establish a state constraint function according to the air compressor control evaluation parameter, and calculate the stability constraint boundary of the fuel cell air compressor;

[0206] The recursive optimization unit 4 is used to recursively optimize the air compressor control parameters according to the operation characteristic matrix and the stability constraint boundary to obtain the air compressor speed control sequence.

[0207] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0208] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, system, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, system, article or method. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, system, article or method including the element.

[0209] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A fuel cell air compressor flow adaptive control method, characterized in that: The following steps are involved: Collect the air inlet temperature signal, air outlet temperature signal, air inlet pressure signal, air outlet pressure signal, air compressor speed signal and air compressor output flow signal of the fuel cell air compressor, and calculate the operation characteristic matrix; Quantifying the dead zone of the operating characteristic matrix to generate an operating parameter matrix, and constructing a two-layer action evaluation network structure to perform online evaluation of the operating status to generate air compressor control evaluation parameters; Establishing a state constraint function according to the air compressor control evaluation parameter, and calculating the stability constraint boundary of the fuel cell air compressor; The air compressor control parameters are recursively optimized according to the operating characteristic matrix and the stability constraint boundary to obtain the air compressor speed control sequence.

2. The method for adaptively controlling flow rate of a fuel cell air compressor according to claim 1, characterized in that: The method collects the air inlet temperature signal, the air outlet temperature signal, the air inlet pressure signal, the air outlet pressure signal, the air compressor speed signal and the air compressor output flow signal of the fuel cell air compressor, and calculates the operation characteristic matrix, including: The inlet temperature signal, outlet temperature signal, inlet pressure signal, outlet pressure signal, compressor speed signal and compressor output flow signal of the fuel cell air compressor are collected in time series to obtain an observation data matrix; Performing range mapping on the observation data matrix to obtain a mapped data matrix, and calculating the mapped data matrix based on the least squares principle to obtain initial element values ​​of a generalized disturbance matrix; Performing nonlinear compensation on the initial element values ​​of the generalized disturbance matrix, and correcting it by introducing a correction factor to obtain a corrected generalized disturbance matrix; The modified generalized disturbance matrix and the system output matrix are solved based on the matrix equation to obtain the disturbance characteristic matrix, and the disturbance characteristic matrix is ​​optimized to generate an operation characteristic matrix of dynamic characteristic information under different working conditions.

3. The method for adaptively controlling flow rate of a fuel cell air compressor according to claim 1, characterized in that: The dead zone quantification of the operation characteristic matrix is ​​performed to generate an operation parameter matrix, and a two-layer action evaluation network structure is constructed to perform online evaluation of the operation status to generate air compressor control evaluation parameters, including: Performing n-bit quantizer configuration on the operation characteristic matrix to obtain quantization configuration parameters, and performing dead zone threshold setting on dynamic characteristic information in the operation characteristic matrix based on the quantization configuration parameters to obtain a dead zone threshold matrix; Performing signal segmentation processing on the dynamic characteristic information in the operation characteristic matrix according to the dead zone threshold matrix to obtain a segmented signal matrix; Performing hysteresis quantization processing on the segmented signal matrix and setting a dead zone holding interval to obtain a quantization processing matrix; The quantization processing matrix is ​​subjected to signal jump suppression by a hysteresis comparator to obtain a smooth signal matrix, and the smooth signal matrix is ​​subjected to normalization transformation to generate an operation parameter matrix; Based on the operating parameter matrix, a two-layer action evaluation network structure is constructed, and the operating state of the fuel cell air compressor is evaluated online to generate air compressor control evaluation parameters.

4. The method for adaptively controlling flow rate of a fuel cell air compressor according to claim 3, characterized in that: Based on the operating parameter matrix, a two-layer action evaluation network structure is constructed, and the operating state of the fuel cell air compressor is evaluated online to generate air compressor control evaluation parameters, including: dividing the operation parameter matrix into an action data matrix and an evaluation data matrix; An action network is constructed based on the action data matrix, wherein the action network includes an input layer, a hidden layer and an output layer, wherein the input layer includes 5 neurons corresponding to temperature, pressure, speed and flow signals, the hidden layer includes 30 neurons and adopts a ReLU activation function, and the output layer generates a control action value to obtain action network output data; An evaluation network is constructed based on the evaluation data matrix, wherein the evaluation network includes an input layer, a hidden layer and an output layer, wherein the input layer receives the output data of the action network, the hidden layer includes 20 neurons and adopts a Tanh activation function, and the output layer generates an evaluation value to obtain the output data of the evaluation network; Cross-training the action network and the evaluation network to obtain a double-layer action evaluation network structure, and performing loss function calculation and network performance evaluation on the double-layer action evaluation network structure to obtain a network evaluation result; Based on the network evaluation result, the action network and the evaluation network in the double-layer action evaluation network structure are updated with online parameters, and the network weights are adjusted through a back propagation algorithm to generate air compressor control evaluation parameters.

5. The fuel cell air compressor flow adaptive control method according to claim 1, characterized in that: The step of establishing a state constraint function according to the air compressor control evaluation parameter and calculating the stability constraint boundary of the fuel cell air compressor includes: Constructing a positive definite symmetric matrix for the air compressor control evaluation parameters, and obtaining a state matrix through quadratic function conversion; Design a nonlinear integral term based on the state matrix, perform a hyperbolic tangent function operation on the state variable to obtain a nonlinear constraint term, and construct a Lyapunov state constraint function by combining the state matrix and the nonlinear constraint term to obtain a state constraint function; Calculating the time derivative of the state constraint function to obtain a derivative calculation result, and comparing the derivative calculation result with the Euclidean norm of the state vector to obtain a state constraint condition; The rated working range of the fuel cell air compressor is divided according to the state constraint conditions, and stability parameters are set to obtain a constraint interval, and a boundary calculation is performed based on the constraint interval to generate a stability constraint boundary of the fuel cell air compressor.

6. The method for adaptive flow control of a fuel cell air compressor according to claim 1, characterized in that: The recursive optimization of the air compressor control parameters is performed according to the operating characteristic matrix and the stability constraint boundary to obtain the air compressor speed control sequence, including: Constructing a control input weight matrix and a tracking error weight matrix according to the operation characteristic matrix, and substituting the control input weight matrix and the tracking error weight matrix into a quadratic programming function to obtain an optimization objective function; Applying control input constraints and control increment constraints to the optimization objective function, and converting the stability constraint boundary into a Lyapunov derivative constraint to obtain optimization constraints; Inputting the optimization objective function and the optimization constraint conditions into an online recursive solver, calculating the control parameters under a discrete time series, and obtaining an initial sequence of control parameters; The learning step length of the initial sequence of control parameters is set, and incremental constraint processing is performed in combination with the operating conditions of the fuel cell air compressor to obtain the air compressor speed control sequence.

7. The method for adaptively controlling flow rate of a fuel cell air compressor according to claim 1, characterized in that: The fuel cell air compressor flow adaptive control method also includes: Performing digital-to-analog conversion on the air compressor speed control sequence to obtain a converted analog signal, and performing power amplification on the converted analog signal to obtain an air compressor drive signal; Calculate the target air flow rate according to the power demand of the fuel cell, and collect the output flow rate of the air compressor in real time to obtain the flow measurement data; Calculating the difference between the flow measurement data and the target air flow to obtain a flow deviation value, and comparing the flow deviation value with a preset control threshold to obtain a deviation judgment result; Based on the deviation judgment result, the control parameter is updated online to obtain the updated control parameter, and the updated control parameter is converted into a speed sequence, and the control parameter is mapped into a speed value through the air compressor characteristic curve to obtain a new speed control sequence; The new speed control sequence is input into the air compressor controller for closed-loop control, the air compressor speed is adjusted, and the dynamic control of the air compressor output flow is completed.

8. A fuel cell air compressor flow adaptive control system, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the system comprises: A collection unit is used to collect the air inlet temperature signal, the air outlet temperature signal, the air inlet pressure signal, the air outlet pressure signal, the air compressor speed signal and the air compressor output flow signal of the fuel cell air compressor, and calculate the operation characteristic matrix; An evaluation unit, used to quantify the dead zone of the operation characteristic matrix, generate an operation parameter matrix, and construct a two-layer action evaluation network structure to perform online evaluation of the operation status and generate air compressor control evaluation parameters; A calculation unit, used to establish a state constraint function according to the air compressor control evaluation parameter, and calculate the stability constraint boundary of the fuel cell air compressor; The recursive optimization unit is used to perform recursive optimization of air compressor control parameters according to the operating characteristic matrix and the stability constraint boundary to obtain an air compressor speed control sequence.

Images