Pump-controlled hydrogen compressor active fault-tolerant control method for adaptive start-stop working condition

Abstract

The application provides a pump-controlled hydrogen compressor active fault-tolerant control method for adaptive variable start-stop working conditions, relates to the technical field of hydrogen energy equipment control, and comprises the following steps: S1, collecting multi-source state data of the pump-controlled hydrogen compressor, and evaluating the degradation state of the pump-controlled hydrogen compressor; S2, adaptive adjustment of the working condition of the pump-controlled hydrogen compressor, and triggering an active fault-tolerant control signal; S3, according to the active fault-tolerant trigger signal, the degradation state and the outlet pressure, performing cooperative fault-tolerant control according to the fault-tolerant level; and S4, online self-tuning of the pump-controlled hydrogen compressor controller parameters, and active control of the pump-controlled hydrogen compressor. Through the pump-controlled hydrogen compressor degradation state online evaluator combined with a long short-term memory network and a degradation hidden Markov model, the application can identify early fault precursors and actively intervene in faults, improve the adaptive fault-tolerant performance of the pump-controlled hydrogen compressor under highly variable working conditions, reduce the cumulative damage of key components of the pump-controlled hydrogen compressor, prolong the service life of the equipment, and make the hydrogen supply pressure of the pump-controlled hydrogen compressor continuous and stable.

Description

Technical Field

[0001] This invention relates to the field of hydrogen energy equipment control technology, specifically to an active fault-tolerant control method for a pump-controlled hydrogen compressor for adaptive start-stop conditions. Background Technology

[0002] Pump-controlled hydrogen compressors are key pressurization equipment in hydrogen refueling stations and hydrogen energy systems. In actual operation, they often face start-stop conditions, such as intermittent vehicle refueling demands and fluctuations in hydrogen storage pressure. Start-stop cycles subject internal mechanical components of the pump, such as bearings, seals, and motor windings, to periodic impacts and thermal stress, accelerating fatigue degradation. Traditional control methods often focus on the dynamic response speed or steady-state pressure accuracy of a single start-stop cycle, making it difficult to simultaneously ensure long-term reliability and transient performance under start-stop conditions. Furthermore, existing fault-tolerant control strategies are mostly passive, only implementing switching or derating actions after a fault occurs or a clear alarm is triggered. This results in lag and difficulty in proactively intervening in the early stages of a fault, easily leading to cumulative damage or even sudden failure of critical components. Therefore, there is an urgent need for a pump-controlled hydrogen compressor control method that can adapt to start-stop conditions, possesses proactive fault identification and fault-tolerant capabilities, to improve equipment lifespan and hydrogen supply pressure stability. Summary of the Invention

[0003] To address the shortcomings of the existing technology, the present invention aims to provide an active fault-tolerant control method for a pump-controlled hydrogen compressor under adaptive start-stop conditions. This method utilizes online evaluation of the pump-controlled hydrogen compressor's degradation state driven by multi-source state data to identify early signs of failure. Based on real-time fault probability and start-stop frequency characteristics, it dynamically and adaptively adjusts the trigger threshold and intensity of the fault-tolerant strategy. Combining a pump body loss model with pressure composite control, it actively executes coordinated fault-tolerant actions such as redundancy switching, torque softening, leakage pre-compensation, and short-term derating. Furthermore, it smoothly switches to the fault-tolerant control mode without shutting down the system. Under highly variable operating conditions, it reduces cumulative damage to key components, extends equipment lifespan, and ensures continuous stability of the hydrogen supply pressure.

[0004] Specifically, the present invention provides an active fault-tolerant control method for a pump-controlled hydrogen compressor for adaptive start-stop conditions, comprising the following steps: S1: Collect multi-source state data of the pump-controlled hydrogen compressor and construct the multi-source state feature vector of the pump-controlled hydrogen compressor at time t. The degradation state level of the pump-controlled hydrogen compressor is evaluated; a degradation state evaluator for the pump-controlled hydrogen compressor is constructed to output the failure probability of the pump-controlled hydrogen compressor at time t. ; Count the cycle of start-up and stop events of the pump-controlled hydrogen compressor at time t. and start / stop frequency counting ; S2: Adaptive adjustment of the pump-controlled hydrogen compressor's operating conditions, based on the start-stop frequency count of the pump-controlled hydrogen compressor in S1. Divide the start / stop frequency range into three levels: Level 1, Level 2, and Level 3; set the active fault-tolerant control trigger threshold at time t. and adaptive fault-tolerant parameters at time t The probability of pump-controlled hydrogen compressor failure at time t in S1 The active fault-tolerant control signal is triggered at the appropriate time. S3: Based on the active fault-tolerant trigger signal and adaptive fault-tolerant parameters in S2 The degradation state level and multi-source state feature vector at time t in S1 Real-time outlet pressure of pump-controlled hydrogen compressor Determine the softened actual output torque control command. Based on the real-time outlet pressure of the pump-controlled hydrogen compressor at time t Temperature of the motor winding sealing cavity at time t Estimated pressure decay rate and rate of temperature change Determine the leakage flow rate and perform pre-compensation for seal leakage in the pump-controlled hydrogen compressor; switch the redundant pump set of the pump-controlled hydrogen compressor and perform collaborative fault-tolerant control; S4: Obtain the pressure deviation of the pump-controlled hydrogen compressor at time t. and rate of change The adaptive fault-tolerant parameter at time t in S2 A control model for a pump-controlled hydrogen compressor was constructed, and the online self-tuning of the pump-controlled hydrogen compressor controller parameters was performed to achieve active control of the pump-controlled hydrogen compressor.

[0005] Preferably, S1 specifically comprises: S11: Collect multi-source state data and construct... Multi-source state feature vector of pump-controlled hydrogen compressor operation at any time After normalization, the data is input into the online evaluation device for the degradation status of the pump-controlled hydrogen compressor; S12: Use a Long Short-Term Memory network to output temporal degradation features Degenerate Hidden Markov Models define multiple degenerate states. Construct a degradation state evaluator for a pump-controlled hydrogen compressor and output the failure probability of the pump-controlled hydrogen compressor at time t. ; S13: Statistical count of start / stop event cycles and start / stop frequency count for pump-controlled hydrogen compressors; start / stop event cycle count The current of the hydrogen compressor motor is controlled by the pump. From shutdown state Rise to a degenerate state judge, The current threshold is used to determine when to stop the machine. Current threshold for operation determination; start / stop frequency counting. It is determined by the number of starts and stops within the sliding time window.

[0006] Preferably, the failure probability of the pump-controlled hydrogen compressor at time t in S12 is... for: ; in, Let be the probability of failure of the pump-controlled hydrogen compressor at time t; Solve for the operator for conditional probability; The hidden state of real-time normal operation of the pump-controlled hydrogen compressor at time t; This is the time series sequence of multi-source observation features of the pump-controlled hydrogen compressor, collected from the initial time to time t. The state at time t The joint forward probability; Let be the joint forward probability at time t; This is the index of the normal degradation state of the pump-controlled hydrogen compressor; t is the time parameter.

[0007] Preferably, the joint forward probability at time t in S12 for: ; ; in, The initial joint forward probability; The initial state distribution of the pump-controlled hydrogen compressor; The initial observation probability; The state transition probability; Let be the probability of observation at time t; The device operation characteristics collected in a single data acquisition at time t; i represents the degradation state level; The device operation characteristics collected in a single initial data acquisition; Let be the joint forward probability at time t-1.

[0008] Preferably, the active fault-tolerant control trigger threshold in S2 and adaptive fault tolerance parameters Specifically: ; in, The threshold for active fault-tolerant control at time t; The baseline threshold; This is the threshold decay parameter; The adaptive fault-tolerant parameter at time t; Used as the reference strength parameter; For intensity gain parameters; t represents the start-stop frequency count of the pump-controlled hydrogen compressor at time t; e is a natural constant parameter.

[0009] Preferably, S3 specifically comprises: S31: Pressure controller outputs raw torque control command The original motor torque control command Apply a first-order low-pass filter and an S-shaped saturation limiting constraint to output a softened actual output torque control command. ; S32: Perform pre-compensation for seal leakage in the pump-controlled hydrogen compressor; input the real-time outlet pressure of the pump-controlled hydrogen compressor at time t. Temperature of the motor winding sealing cavity at time t Estimate the pressure decay rate and rate of temperature change Determine the leakage flow rate; S33: Execute short-term derating operation of the pump-controlled hydrogen compressor; input the failure probability of the pump-controlled hydrogen compressor at time t. The probability of failure of the pump-controlled hydrogen compressor at time t If the current degradation state is the third or fourth state, the derated operation will be initiated. S34: Switch the redundant pump set of the pump-controlled hydrogen compressor; the probability of pump-controlled hydrogen compressor failure at time t. The standby pump is started in advance and its speed is synchronized and pressure is pre-adjusted in a closed loop.

[0010] Preferably, the original motor torque control command in S31 Apply a first-order low-pass filter and an S-shaped saturation limiting constraint, specifically as follows: , ; ; in, The actual output torque control command is obtained after first-order low-pass filtering and shaping. This is the original motor torque reference command output by the upper-level pressure control module; The time constant of a first-order low-pass filter; The reference torque filtering time constant; For Laplace complex frequency domain operator; k is the discrete control cycle count; This is the actual output torque control command; This is the saturation limiting operation function; This is the original motor torque reference command; This is the torque softening time constant; This represents the maximum adjustment amplitude of the single-step torque. The actual output torque control command for k cycles; This is the actual output torque control command for cycle k+1.

[0011] Preferably, the pressure decay rate in S32 and rate of temperature change The leakage flow rate is determined to be: ; ; ; in, The pressure decay rate; For leakage flow; The rate of temperature change; For pressure decay related parameters; For temperature correction parameters; The real-time outlet pressure of the pump-controlled hydrogen compressor at time t; for Real-time outlet pressure of the hydrogen compressor; The real-time motor winding sealing cavity temperature of the pump-controlled hydrogen compressor at time t; for Real-time temperature of the motor winding sealing cavity in the hydrogen compressor; This represents the sampling time interval.

[0012] Preferably, in S4, the construction of the pump-controlled hydrogen compressor control model and the online self-tuning of the pump-controlled hydrogen compressor controller parameters specifically involve: ; ; in, The vector of controller parameters to be identified is updated online at time t; Let t be the regression observation vector; for The covariance matrix of the time parameter estimation error; for The controller derating parameters obtained from the previous iteration; The measured pressure at time t; Forgetting factor; For the first The control input quantity for the pump-controlled hydrogen compressor drive control model is input at the sampling time; for Real-time pressure measurement; This is the matrix transpose operator.

[0013] Preferably, the recursive least squares formula in S4 is: ; in, Let t be the gain vector. Forgetting factor; Let be the covariance matrix of the parameter estimation error at time t; It is an identity matrix.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention improves the adaptive fault-tolerant performance of the pump-controlled hydrogen compressor; the online evaluation device for the degradation state of the pump-controlled hydrogen compressor, which combines a long short-term memory network and a degenerate hidden Markov model, can identify fault precursors in advance, avoid the lag of traditional passive fault tolerance, and perform active fault intervention.

[0015] (2) The present invention can reduce the cumulative damage of key components of pump-controlled hydrogen compressor; by combining the pump body loss model and the working condition adaptive adjustment module, the fault tolerance threshold and action intensity are dynamically adjusted according to the start and stop frequency count, thereby reducing the periodic impact and thermal stress of components such as bearings, seals, and motor windings under variable start and stop conditions, and extending the overall life of the equipment.

[0016] (3) The present invention can make the hydrogen supply pressure continuously stable; by adopting pressure feedforward-feedback composite control and sealing leakage pre-compensation, during fault-tolerant actions such as redundant pump group switching, torque softening, and short-term derating operation, the outlet pressure is maintained at no less than the minimum threshold required for hydrogen supply, so that the hydrogen refueling station or hydrogen energy system can supply hydrogen stably.

[0017] (4) The present invention performs a smooth transition without stopping the machine; by online self-tuning of the controller parameters, such as fuzzy adaptive proportional integral derivative PID or recursive least squares model reference adaptive control, the normal mode is smoothly switched to the fault-tolerant control mode without interrupting operation, thus avoiding efficiency loss and secondary impact caused by shutdown.

[0018] (5) The present invention enhances the system’s adaptability to varying operating conditions; the operating condition adaptive adjustment module can dynamically adjust the trigger threshold and action intensity according to the start-stop frequency counting characteristics, making the control method applicable to start-stop scenarios of varying degrees, and has strong generalization performance and engineering practicality. Attached Figure Description

[0019] Figure 1 This is a control block diagram of the active fault-tolerant control method for a pump-controlled hydrogen compressor for adaptive start-stop conditions according to the present invention. Figure 2 This is the overall control flowchart of the pump-controlled hydrogen compressor active fault-tolerant control method of the present invention; Figure 3 This is a schematic diagram of the structure of the online evaluation device for the degradation status of the pump-controlled hydrogen compressor of the present invention; Figure 4 This is a schematic diagram of the trigger threshold curve dynamically adjusted by the adaptive adjustment module based on the start / stop frequency count in this invention. Figure 5 This is a schematic diagram of the curve showing how the adaptive adjustment module in this invention dynamically adjusts the intensity of action based on the start-stop frequency count. Figure 6 This is a flowchart illustrating the execution of the collaborative fault-tolerant action in this invention. Detailed Implementation

[0020] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0021] This invention proposes an active fault-tolerant control method for a pump-controlled hydrogen compressor with adaptive start-stop conditions, such as... Figure 1 As shown, multi-source state data of the pump-controlled hydrogen compressor are collected to evaluate its degradation state; the pump-controlled hydrogen compressor's operating conditions are adaptively adjusted to trigger an active fault-tolerant control signal; based on the active fault-tolerant trigger signal, degradation state, and outlet pressure, coordinated fault-tolerant control is executed according to the fault tolerance level; the pump-controlled hydrogen compressor controller parameters are self-tuned online to perform active control of the pump-controlled hydrogen compressor; a specific embodiment of the present invention takes a pump-controlled hydrogen compressor used in a hydrogen refueling station as an example, the controller adopts an embedded industrial computer, and the sampling period is... Active control of the pump-controlled hydrogen compressor, such as... Figure 2 The diagram shown is the overall control flowchart of an adaptive start-stop operation mode active fault-tolerant control method for a pump-controlled hydrogen compressor according to the present invention. It fully embodies the closed-loop control logic of active identification, operation mode adaptation, collaborative fault tolerance, and online self-tuning. The specific implementation steps are as follows: S1: Collect multi-source state data of the pump-controlled hydrogen compressor to evaluate its degradation state; the pump-controlled hydrogen compressor operating parameters obtained in this embodiment are shown in Table 1, which provides typical operating parameters of the pump-controlled hydrogen compressor targeted in this embodiment; among them, the rated speed of 3000 rpm and the rated outlet pressure of 45 MPa are the design operating parameters of the 45 MPa-class pump-controlled hydrogen compressor commonly used in hydrogen refueling stations; the minimum hydrogen supply pressure of 35 MPa is the minimum acceptable hydrogen supply pressure threshold for refueling vehicles at hydrogen refueling stations; the data sampling frequency of 1 kHz is matched with the control cycle of 0.001 s to obtain a complete set of multi-source state data in each control cycle; the start-stop frequency count f in the operating scenario. cyc =45 times / h represents the typical start-stop condition faced by the pump-controlled hydrogen compressor when vehicles are refueling intensively during peak hours at a hydrogen refueling station. This start-stop frequency count is within the three-level start-stop range set in S2 below, which is the high-variable operating condition scenario that this embodiment focuses on verifying.

[0022] Table 1. Statistical Table of Operating Parameters of Pump-Controlled Hydrogen Compressor

[0023] The input for this step is: the vibration acceleration of the pump body bearing at time t. Temperature of the motor winding sealing cavity at time t Real-time outlet pressure of the pump-controlled hydrogen compressor at time t Hall sensor motor current at time t Programmable Logic Controller (PLC) cumulative start / stop event loop counting The above multi-source state data, after normalization, constitute the multi-source state feature vector at time t. .

[0024] S11: During the operation of the pump-controlled hydrogen compressor, the data sampling frequency is set to 1 kHz, and the following multi-source state data are collected in real time, including: the vibration acceleration of the pump body bearing at time t. Temperature of the motor winding sealing cavity at time t Real-time outlet pressure of the pump-controlled hydrogen compressor at time t Hall sensor motor current The cyclic count of start / stop events at time t accumulated by the programmable logic controller (PLC) Build Multi-source state feature vector of pump-controlled hydrogen compressor operation at any time After normalization, the data is input into the online evaluation device for the degradation status of the pump-controlled hydrogen compressor. for The constant vibration acceleration of the pump body bearing; for Temperature of the motor winding sealing cavity at all times; for Real-time outlet pressure of the hydrogen compressor; for The Hall sensor detects the motor current at a given moment. for The cumulative count of start and stop events over time; This is a time parameter.

[0025] Table 2 shows the sensor readings of the pump-controlled hydrogen compressor obtained at t=3600s in this embodiment of the invention. These readings represent the measured values ​​and normalized results of each sensor in the pump-controlled hydrogen compressor at t=3600s. The normalization range is set according to the range of each sensor: vibration acceleration range 0-20 m / s². 2 It covers the entire range of pump-controlled hydrogen compressors from stable operation to severe wear; the temperature range is 0-150℃, covering ambient temperature to the upper limit of winding insulation withstand capability; the current range is 0-50A, covering the no-load to full-load current range; the start-stop event cycle count is 0-100 times, corresponding to the maximum number of recordable start-stop cycles per hour; the vibration acceleration at t=3600s is 7.5 m / s². 2The normalized value of 0.38 is still within the normal range, but the temperature of the motor winding sealing cavity is 72.0℃, with a normalized value of 0.48, which is significantly higher than the typical steady-state operating value of 55-60℃, indicating that the variable start-stop has caused a heat accumulation effect in the winding. The outlet pressure is 43.2 MPa, with a normalized value of 0.86, which is still higher than the minimum hydrogen supply pressure of 35 MPa, but has decreased compared to the rated value of 45 MPa, suggesting a slight leakage trend in the sealing pair. This set of normalized data constitutes a multi-source state feature vector X(t=3600)=[0.38,0.48,0.86,0.53,0.47], which serves as the input for the Long Short-Term Memory (LSTM) network and the Degenerate Hidden Markov Model (DHMM) in S12.

[0026] Table 2 Sensor Reading Statistics

[0027] Set the normalized range: Pump body bearing vibration acceleration is Temperature of the motor winding sealing cavity Pressure transmitter pressure Hall sensor motor current Start / stop event loop count Next, the multi-source state feature vector of the pump-controlled hydrogen compressor operation was obtained: .

[0028] S12: Construct a degradation state evaluator for a pump-controlled hydrogen compressor based on a long short-term memory network and a degenerate hidden Markov model.

[0029] The Long Short-Term Memory (LSTM) network contains 64 hidden units, has an input temporal length of 100 sampling points, and outputs temporal degradation features. .

[0030] The degenerate hidden Markov model sets five degenerate states. These correspond to the following states: normal operation corresponds to the first state, level one degradation corresponds to the second state, level two degradation corresponds to the third state, level three degradation corresponds to the fourth state, and pre-failure state corresponds to the fifth state. In this embodiment, when the normal degradation state is evaluated as level three degradation in S4 or pre-failure state in S5, the pump-controlled hydrogen compressor is considered to have a clear risk of failure. The transition probability matrix A and the observation probability matrix B are obtained through offline training.

[0031] At time t, the degenerate hidden Markov model calculates the failure probability of the pump-controlled hydrogen compressor at time t. In the example, the probability of being in the fifth state is specifically as follows: ; in, The failure probability of the pump-controlled hydrogen compressor at time t is the failure precursor degradation state, the fifth state. The operator for solving conditional probability represents the calculation of the posterior probability based on the previous observation sequence. The hidden state of real-time normal operation of the pump-controlled hydrogen compressor at time t; This is the time series sequence of multi-source observation features of the pump-controlled hydrogen compressor, collected from the initial time to time t. Let be the joint forward probability of being in the fifth state at time t, which corresponds to the A5 level failure precursor state of the pump-controlled hydrogen compressor under the Hidden Markov Model (DHMM). Let be the joint forward probability at time t, representing the multi-source state feature sequence that has been accumulated and observed up to time t, where the pump-controlled hydrogen compressor is actually in the j-th level of degradation at time t. The joint probability is used to recursively quantify the complete likelihood of the pump-controlled hydrogen compressor's degradation evolution from its initial normal operation to the current moment; This is the index for the normal degradation state of the pump-controlled hydrogen compressor, with values ​​j=1,2,3,4,5.

[0032] The joint forward probability in the above formula The recursive formula is: ; ; in, The initial joint forward probability; The initial state distribution of the pump-controlled hydrogen compressor represents the initial probability of each degradation state level at time t=0 when the pump-controlled hydrogen compressor is first powered on and running. Let be the state transition probability, representing the probability that, given that the pump-controlled hydrogen compressor is currently in the i-th level of degradation, it will transition to the j-th level of degradation due to the physical degradation process at the next sampling time. The initial observation probability; To determine the observation probability, a Gaussian mixture model is used, representing the multi-source state characteristics collected by sensors when the pump-controlled hydrogen compressor is in the j-th level of degradation. The conditional probability of occurrence; The device operation characteristics collected in a single data acquisition at time t; i represents the degradation state level; The device operation characteristics collected in a single initial data acquisition; Let be the joint forward probability at time t-1.

[0033] In this embodiment of the invention, a Long Short-Term Memory (LSTM) network is used to extract degradation features in a 100-step time window. These features are then fed into a fifth-state Hidden Markov Model (DHMM) for forward recursion. The evolution of the LSTM network and the degradation Hidden Markov Model (LSTM+DHMM) into normal degradation states is performed, resulting in a five-level state statistics table as shown in Table 3.

[0034] Table 3. Statistical Table of Five-Level State Probability for Pump-Controlled Hydrogen Compressor

[0035] Table 3 shows the probability distribution of the five-stage degradation state and failure probability of the pump-controlled hydrogen compressor output by the online evaluator of the Long Short-Term Memory Network and the Degenerate Hidden Markov Model (LSTM+DHMM) from t=3500s to t=3600s. The evolution process, as shown in Table 3, involves three typical stages of degradation in pump-controlled hydrogen compressors: (1) During the normal dominant stage, t=3500s: the probability of normal degradation state A1 reaches 82.5%, the probability of level 3 degradation A4 is only 0.2%, and the probability of failure precursor A5 is 0%. =0%, the pump-controlled hydrogen compressor is in a state of degradation, and there are no obvious signs of degradation in the key components.

[0036] (2) Accelerated degradation stage, t=3540s~t=3580s: With the continuous accumulation of start-stop cycles, the probability of normal operation drops sharply from 51.2% to 4.6%, the probability of secondary degradation A3 jumps from 10.8% to 40.5%, and the probability of tertiary degradation A4 rises from 2.1% to 27.3%. This stage corresponds to the physical process of accelerated wear and enhanced vibration signal trend term of the pump-controlled hydrogen compressor bearing due to periodic start-stop impacts. At t=3580s, the vibration acceleration increases from 5.2 m / s². 2 Rising to 7.1 m / s 2 The degradation feature vector extracted by LSTM shows a significant shift, and the Degradation Hidden Markov Model (DHMM) determines the degradation inflection point as t≈3580s based on this.

[0037] (3) Degradation-dominant stage, t=3600s: The probability of level 3 degradation A4 reaches 46.4%, becoming the dominant state, and the probability of failure precursor A5 rises to 13.9%, indicating that the pump-controlled hydrogen compressor has entered a high-risk operating range, and the degree of bearing wear and seal aging can no longer be ignored. At this time =13.9%, which has not yet exceeded the adaptive threshold in subsequent S2. =30.5%, so the active fault-tolerant control action will not be triggered for the time being, but the state is already in the closely monitored zone.

[0038] The data evolution in Table 3 shows that, at the start-stop frequency count f cyc With a variable start-stop cycle of 45 times / h, the pump-controlled hydrogen compressor can deteriorate from the normal operation-dominant state (normal degradation state A1=82.5%) to the level 3 degradation-dominant state (level 3 degradation A4=46.4%) within 100 seconds. The degradation rate is faster than that under continuous steady-state operation, which verifies the accelerating effect of variable start-stop cycle on the degradation of the pump-controlled hydrogen compressor.

[0039] S13: Statistical count of start / stop events and start / stop frequency of the pump-controlled hydrogen compressor. The start / stop event is set as: the motor current of the pump-controlled hydrogen compressor at time t. From shutdown state , To determine the shutdown current threshold, 5% of the rated current is used, increasing to a degradation state. , To determine the operating current threshold, 20% of the rated current is used, and the rotational speed at time t is taken as the threshold. The complete process of accelerating from zero to more than 30% of the rated speed is recorded as one start-stop event. Each time a start-stop event is detected, the PLC internally performs a start-stop event loop count at time t. Automatically increments by 1. Start / stop frequency counter. Calculated from the number of starts and stops within the sliding time window: ; in, This is a count of the start-stop frequency of the pump-controlled hydrogen compressor, in times per hour. In this embodiment, the sliding time window length is... This refers to the number of starts and stops in the past hour; for The cumulative number of starts and stops prior to the specified time.

[0040] The output of this step is: the failure probability of the pump-controlled hydrogen compressor at time t. Used for determining the triggering conditions of the formula in S2; the current degradation level of the pump-controlled hydrogen compressor, A1~A5, is used for determining the status conditions of the redundant pump group switching in S34, requiring a level 3 degradation A4 or a failure precursor A5.

[0041] S2: The pump-controlled hydrogen compressor's operating condition adaptive adjustment triggers the active fault-tolerant control signal; the input for this step is the pump-controlled hydrogen compressor failure probability at time t from S1. and the start / stop frequency count of the pump-controlled hydrogen compressor Counting based on the start-stop frequency of the pump-controlled hydrogen compressor. The start-stop frequency count of the pump-controlled hydrogen compressor is divided into three frequency ranges. The unit is times / hour, which represents the number of starts and stops in the past hour; specifically: Level 1 start / stop frequency range: times / hour.

[0042] Level 2 start / stop frequency range: times / hour.

[0043] Level 3 start / stop frequency range: times / hour.

[0044] Active fault-tolerant control trigger threshold at time t and adaptive fault-tolerant parameters at time t Specifically: ; in, The active fault-tolerant control trigger threshold at time t is adapted to the real-time start / stop frequency and dynamic updates of the pump-controlled hydrogen compressor; As a baseline threshold, in the examples ; The threshold attenuation parameter characterizes the decrease in the fault warning trigger threshold corresponding to an increase in the start / stop frequency count. In this embodiment... ; The adaptive fault-tolerant parameter at time t changes adaptively with start-up and shutdown conditions. As the reference strength parameter, ; The strength gain parameter characterizes the increase in fault tolerance strength corresponding to the increase in the start / stop frequency count. In this embodiment... ; t represents the start-stop frequency count of the pump-controlled hydrogen compressor at time t; e is a natural constant parameter.

[0045] The probability of pump-controlled hydrogen compressor failure at time t The active fault-tolerant control signal is triggered at this time. For example... Figure 4 The diagram shown illustrates the dynamic adjustment curve of the adaptive operating condition control module in this invention, based on the start / stop frequency count, to adjust the trigger threshold. Figure 5 The figure shows a curve diagram of the adaptive adjustment module for operating conditions in this invention, which dynamically adjusts the intensity of action based on the start-stop frequency count. As the start-stop frequency count increases, the fault tolerance threshold decreases and the intensity of action increases. Through the adjustment of the fault tolerance parameters for adaptive operating conditions, the module adapts to the start-stop conditions of the pump-controlled hydrogen compressor at various frequencies, balancing the reliability of equipment operation and the hydrogen supply response performance.

[0046] In this embodiment of the invention, the current start / stop frequency count Within the Level 3 start / stop frequency range, substitute the values ​​into the formula to calculate the active fault-tolerant control trigger threshold at time t. and adaptive fault-tolerant parameters at time t for: ; The probability of pump-controlled hydrogen compressor failure at t=3600s No trigger. It continued running for 10 minutes until t=4200s, during which it started and stopped approximately 8 times. Due to cumulative degradation effects, some sensor readings further deteriorated. Pump body bearing vibration acceleration at time t : 7.5→10.2m / s 2 This leads to accelerated bearing wear.

[0047] Temperature of the motor winding sealing cavity at time t Temperatures ranged from 72.0℃ to 85.5℃, with the change in start-stop cycle leading to accumulated heat dissipation.

[0048] Real-time outlet pressure of pump-controlled hydrogen compressor at time t The pressure decreased from 43.2 to 41.8 MPa, resulting in a decline in sealing performance.

[0049] The Long Short-Term Memory (LSTM) network captures the accelerated degradation trend, and the degenerate Hidden Markov Model (DHMM) is re-evaluated to determine the failure probability of the pump-controlled hydrogen compressor. .

[0050] The outputs of this step are: ① Active fault-tolerant trigger signal, used for enabling control of each sub-step in S3; ② Adaptive fault-tolerant parameters. ③ Active fault-tolerant control trigger threshold, used for calculating the original torque control command in S31; .like Figure 3 The diagram shows the process flow for multi-source data-driven degradation state evaluation and fault-tolerant triggering of a pump-controlled hydrogen compressor. The process sequentially completes multi-source state data acquisition, feature vector normalization preprocessing, LSTM temporal degradation feature extraction, degradation hidden Markov model (DHMM) state mapping, and fault probability calculation. Finally, it determines whether to trigger active fault-tolerant control by using an adaptive threshold for operating conditions.

[0051] S3: Based on the active fault-tolerance trigger signal, the current degradation state, and the exit pressure, perform coordinated fault-tolerance control according to the fault-tolerance level from low to high; the inputs for this step are: the active fault-tolerance trigger signal at time t in S2 and the adaptive fault-tolerance parameters at time t. The degradation state level and outlet pressure at time t from S1 Specifically, it includes the following steps: S31: In this embodiment of the invention, the upper pressure controller outputs the original torque control command. The torque amplitude jumps from 35Nm to 85Nm, and then the start / stop frequency count is input into S2. Used to calculate adaptive time constant .

[0052] First-order low-pass filter time constant for: ; in, The time constant of the first-order low-pass filter; The reference torque filtering time constant is denoted as .

[0053] Based on the above parameters, the motor torque control command is softened and shaped to modify the original motor torque control command. Apply a first-order low-pass filter, specifically: ; in, The actual output torque control command, after being softened and shaped by a first-order low-pass filter, is in complex frequency domain form. The original motor torque reference command output by the upper-level pressure control module is in complex frequency domain form; The time constant of a first-order low-pass filter; This is the Laplace complex frequency domain operator.

[0054] Simultaneously, the original torque control command An additional S-shaped saturation limiting constraint is applied to suppress sudden torque changes, specifically: ; Where k is the discrete control cycle count; This is the actual output torque control command; This is the saturation limiting operation function; Let n be the torque softening time constant. ; This represents the maximum adjustment amplitude of the single-step torque. ; The actual output torque control command for k cycles; This is the actual output torque control command for cycle k+1.

[0055] This step outputs the softened actual output torque control command. .

[0056] Adaptive filtering time constant The first-order low-pass filter and S-shaped rate limiting, as shown in Table 4, yielded a statistical table of trigger signals for softening the original 85 Nm step torque command. This table illustrates the softened transition process after applying a first-order low-pass filter and S-shaped rate limiting. The original step command abruptly changes from 35 Nm to 85 Nm at t=4200.000s. Direct application of this command would result in a torque change rate of approximately 1667 Nm / s, causing severe impact on the pump bearings and coupling. After adaptive filtering... After softening, the torque command transitions smoothly along an S-shaped curve, with the rise time extended to approximately 800 ms, the peak rate of change decreasing to approximately 116 Nm / s, and the torque impact rate reduced by 93.0%. The data in the table illustrate the time distribution characteristics of the softening process: an initial acceleration of 0-15s to accommodate pressure response speed; and a gradual convergence to the target value in the final 0.50~0.80s to avoid overshoot. A higher start-stop frequency count results in a larger softening time constant and a smoother transition, matching the physical characteristics of accelerated bearing impact fatigue under variable start-stop conditions.

[0057] Table 4. Trigger Signal Table for Softening Processing

[0058] Table 4 shows that after the original torque command jumped from 35 Nm to 85 Nm, the actual output torque increased smoothly under the combined effect of first-order low-pass filtering and torque change rate limitation. The torque impact rate decreased by 93.0%. The measured bearing impact load decreased from 2.4g to 0.34g, an attenuation of 85.8%.

[0059] S32: Perform pre-compensation for leaking seals on the pump-controlled hydrogen compressor; input the real-time outlet pressure of the pump-controlled hydrogen compressor at time t in S1. and the temperature of the motor winding sealing cavity Real-time estimation of pressure decay rate and rate of temperature change The leakage flow rate is determined to be: ; ; ; in, The pressure decay rate; For leakage flow; The rate of temperature change; For pressure decay related parameters; For temperature correction parameters; The real-time outlet pressure of the pump-controlled hydrogen compressor at time t; The real-time motor winding sealing cavity temperature of the pump-controlled hydrogen compressor at time t; The sampling time interval; for Real-time outlet pressure of the hydrogen compressor; for Real-time temperature of the motor winding sealing cavity of the hydrogen compressor is monitored.

[0060] In the embodiments of the present invention The feedforward compensation speed increment is determined as follows: ; in, For feedforward compensation of speed increment; The gain parameter is for leakage feedforward compensation. For leakage compensation correction index, here It reflects the nonlinear mapping relationship between leakage flow and compensation speed increment, which is jointly determined by the pump body displacement characteristics and leakage flow state.

[0061] Final speed command ,in, This is a pressure closed-loop feedback output. In this embodiment of the invention, pre-compensation for sealing leakage is performed. Before switching, the pressure drops from 41.8 MPa to 41.6 MPa within 5 seconds, and the temperature of the motor winding sealing cavity rises from 85.5°C to 86.3°C. Specifically: ; After compensation, the outlet pressure stabilized at 41.5~41.9 MPa within 10 seconds. In this example, redundant pump unit switching was performed; speed synchronization was completed in 0.1 seconds via PI control, with a deviation of 0.31% (<5%). The three-way valve switched without impact within 0.12 seconds, and the outlet pressure fluctuation was only 0.03 MPa. The current speed of the main pump... This step outputs the feedforward compensated speed increment. With pressure closed-loop feedback output The final speed command is formed by superimposing these commands.

[0062] S33: Execute short-term derating operation of the pump-controlled hydrogen compressor; input the pump-controlled hydrogen compressor failure probability at time t in S1. The probability of failure of the pump-controlled hydrogen compressor at time t When the current degradation state is in the third or fourth state, derating operation is initiated; the maximum speed limit is set as follows: ; in, This is the upper limit of the maximum speed for short-term derating operation mode; For the rated speed, ; For the depreciation parameter, .

[0063] At the same time, the real-time outlet pressure of the pump-controlled hydrogen compressor at time t If the real-time outlet pressure of the pump-controlled hydrogen compressor at time t If the rotational speed is increased to the minimum maintaining pressure, then the priority is to increase the rotational speed. In the short-term derating operation described in the example, the failure probability at time t in the alternative scenario during actual degradation is... The percentage rose continuously from 13.9% at t=3600 to 56.2% at t=4200, inevitably passing through a threshold. =30.5%. The probability of failure at time t. When the threshold is first exceeded, the time is between t=3600 and t=4200. At this point, the failure probability is approximately 35%, and the degradation state is still dominated by secondary degradation A3. The state level conditions required for redundant pump group switching in S34 have not yet been met; A4 or A5 is required. Therefore, short-term derating operation in S31 is triggered instead of redundant switching in S34. The upper limit of the speed limit for short-term derating operation mode is: ; Simultaneously monitor the real-time outlet pressure of the pump-controlled hydrogen compressor. If the pressure is below 35MPa, then pressure should be increased and maintained first. This step outputs the upper limit of the maximum speed. The overall output of this step is: the actual motor torque command executed. and speed command It is used to drive the pump unit actuator of the pump-controlled hydrogen compressor.

[0064] S34: Switch the redundant pump set of the pump-controlled hydrogen compressor; when the pump-controlled hydrogen compressor failure probability... When the current main pump is in normal degradation state 5, the standby pump is pre-started and speed synchronization and pressure closed-loop pre-adjustment are performed; when the standby pump is synchronized and the main pump outlet pressure is lower than the minimum pressure, the three-way valve is switched without impact. Measure the main pump speed Adjust the speed of the standby pump Cause synchronization deviation satisfy .

[0065] After synchronization, the three-way valve is switched smoothly, with a switching time of less than 0.2 seconds.

[0066] S4: Online self-tuning of pump-controlled hydrogen compressor controller parameters for active control of the pump-controlled hydrogen compressor. The input for this step is: pressure deviation at time t. and rate of change The adaptive fault-tolerant parameter at time t in S2 A fuzzy adaptive proportional-integral-derivative (PID) control strategy is adopted. The pressure deviation at time t is set. and rate of change As the two input variables of the fuzzy controller, after fuzzification, they are processed through 7 levels of membership functions: negative large NB, negative medium NM, negative small NS, zero ZO, positive small PS, positive medium PM, and positive large PB. These are then mapped to the fuzzy domain, and after fuzzy inference and defuzzification, the output is the proportional-integral-derivative (PID) parameter correction. , , The 7-level membership function here represents the standard fuzzy partitioning granularity of the fuzzy proportional-integral-derivative PID controller. This, along with the 5-level partitioning of the pump-controlled hydrogen compressor degradation state in S1, belongs to an independent hierarchical system for each technical aspect, and the two are unrelated in physical meaning and function. The final proportional-integral-derivative PID control parameters are: ; in, The real-time proportional control parameter at time t; Used as the baseline scaling parameter; This is the adjustment amount for the proportional parameter; The real-time integral control parameters at time t; The reference integration parameters; This is the correction amount for the integral parameter; The real-time differential control parameters at time t; Used as the reference differential parameter; This is the correction amount for the differential parameter.

[0067] In the embodiment, a fuzzy adaptive proportional-integral-derivative (PID) PID control strategy is employed. Pressure deviation at 4200s =1.4MPa, rate of change =200MPa / s. Fuzzy inference yields the proportional-integral-derivative (PID) correction value: The final result is: .

[0068] Compared to a fixed proportional-integral-derivative (PID) PID controller: the overshoot decreased from 0.62 MPa to 0.18 MPa (rate of change -71%), and the settling time decreased from 1.8 s to 0.6 s (rate of change -67%). The recursive least squares (RLS) backup path achieved parameter convergence within 0.2 s. (Baseline proportional parameters...) Baseline integral parameters Reference differential parameters Simultaneously, a recursive least squares (RLS) control model for the pump-controlled hydrogen compressor is constructed, and reference adaptive control is performed, specifically as follows: ; ; in, The vector of controller parameters to be identified is updated online at time t; For regression observation vectors; for The covariance matrix of the time parameter estimation error; for The controller derating parameters obtained from the previous iteration; The measured pressure at time t; The forgetting factor is used to apply an exponentially decaying weight to historical observation data, enabling the online identified controller parameters to quickly adapt to the drift in controlled object characteristics caused by cumulative damage from start-stop cycles in the pump-controlled hydrogen compressor. In this embodiment... ; For the first The control input is input to the pump-controlled hydrogen compressor drive control model at the sampling time.

[0069] By updating the parameters of the pump-controlled hydrogen compressor controller online, the seamless switching from normal mode to fault-tolerant mode enables the pump-controlled hydrogen compressor to smoothly switch to fault-tolerant control state without stopping the machine when the failure probability exceeds the limit under adaptive start-stop conditions. This eliminates the outlet hydrogen pressure fluctuations and mechanical shocks to the unit caused by sudden mode changes, ensuring a continuous and stable hydrogen supply to the hydrogen refueling station and reducing equipment wear and downtime maintenance costs.

[0070] In this embodiment of the invention, a recursive least squares (RLS) model with reference adaptation and an alternate path is used to obtain the regression observation vector. The recursive formula for Recursive Least Squares (RLS) is: ; in, Let t be the gain vector, used to weigh the weighting of new observations and historical parameter estimates in parameter updates; Forgetting factor; The covariance matrix of the parameter estimation error at time t is used to quantify the uncertainty of the current controller parameter identification result; This is the matrix transpose operator.

[0071] The forgetting factor λ = 0.98. Taking the first step of the recursion at t = 4200.001s as an example: ; After 200 iterations and approximately 0.2 seconds, the parameters converged to... The model output residual decreased to 0.08 MPa. The output of this step is: the proportional-integral-derivative (PID) control parameters after online self-tuning, and the controller parameter vector to be identified obtained through online RLS identification, used for updating the reference model in model reference adaptive control. S4 performs non-disruptive online updating of controller parameters through either fuzzy adaptive PID or RLS paths, ensuring that the outlet pressure fluctuation during the switch from normal operation mode to fault-tolerant control mode does not exceed ±0.2 MPa.

[0072] A comparison of the implementation effects of the embodiments of the present invention, in f cyc When the frequency was 45 times / hour, Table 5 shows the statistical results of the method comparison, which summarizes the results at f. cyc Comparison of the implementation effects of the method of the present invention and the traditional method of passive fault tolerance based on a fixed threshold under high-frequency start-stop conditions of 45 times / h: Fault identification lead time: Traditional methods trigger a fixed alarm threshold based on sensor readings, such as vibration >12 m / s. 2The fault signal is issued only after the temperature exceeds 95℃, with a lag of about 2 seconds. This invention uses an LSTM+DHMM online evaluator to capture the degradation trend in advance, and identifies the inflection point of degradation acceleration at t=3600s, which is 8.5s earlier than the actual fault occurrence, thus gaining a critical decision and execution time window for fault-tolerant actions.

[0073] Bearing impact load: Traditional methods directly apply step torque commands, and the bearing bears an impact load of 2.4g; the present invention reduces the peak impact load to 0.34g by adaptive torque softening and shaping based on start-stop frequency counting, a reduction of 85.8%, effectively reducing the cumulative contact fatigue damage rate between the bearing rolling elements and raceway under variable start-stop conditions.

[0074] Proportional-Integral-Derivative (PI) PID Overshoot: Traditional fixed-parameter PI PID can cause an outlet pressure overshoot of up to 0.62 MPa during fault-tolerant switching. This invention reduces the overshoot to 0.18 MPa by combining fuzzy adaptive PI PID with online RLS identification, a reduction of 71%, thus avoiding additional damage to the sealing surfaces and downstream pipeline valves caused by pressure overshoot.

[0075] Cumulative damage: Based on the cumulative damage of key components such as bearings, seals and windings under the same operating conditions using traditional methods as a baseline of 100%, this invention reduces the cumulative damage to 66% through collaborative fault tolerance, redundant switching to reduce single pump running time, torque softening to reduce impact amplitude, leakage pre-compensation to slow down seal degradation, and short-term derating to reduce stress level, thereby reducing the cumulative damage by 34%, directly extending the overhaul cycle and service life of the pump-controlled hydrogen compressor.

[0076] Pressure fluctuation: Traditional methods result in outlet pressure fluctuations of ±0.8 MPa; this invention, through pressure feedforward-feedback composite control and leakage pre-compensation, reduces the fluctuation range to ±0.2 MPa, a reduction of 75%, ensuring the continuous stability of hydrogen supply pressure at hydrogen refueling stations and avoiding excessive pressure fluctuations that could lead to excessive metering errors or refueling interruptions.

[0077] Number of fault-tolerant downtimes: Traditional methods result in downtime events in all three fault-tolerant processes, totaling three downtimes. This invention achieves zero downtimes through smooth switching without downtime in all fault-tolerant actions, thus improving the continuous operation performance of hydrogen refueling stations.

[0078] Table 5. Statistical table comparing the results of the two methods

[0079] In an actual operation test at a hydrogen refueling station in this embodiment of the invention, as shown in Table 5, under high-frequency operating conditions with 45 start-stop cycles per hour, the present invention, compared to the traditional proportional-integral-derivative PID passive fault tolerance, achieves an average 8.5-second earlier fault identification time; reduces cumulative damage to critical components such as bearings and seals by approximately 34%; and extends the outlet pressure fluctuation range from... Shrink to No downtime events occurred due to fault tolerance lag. Figure 6 This is an execution block diagram of the redundant pump set switching, motor torque command softening and shaping, seal leakage pre-compensation, and short-term derating operation collaborative fault-tolerant actions in this invention. When the failure probability reaches the adaptive threshold, the redundant pump set switching without impact, torque softening and shaping, seal leakage pre-compensation, and short-term derating operation intervene in stages according to their respective thresholds and are executed collaboratively in parallel to achieve active fault tolerance without shutdown, taking into account both hydrogen supply stability and equipment reliability.

[0080] The beneficial effects of the embodiments of the present invention are as follows: The embodiments of the present invention propose an active fault-tolerant control method for pump-controlled hydrogen compressors under adaptive start-stop conditions. This method can identify early signs of faults and proactively intervene in faults, thereby improving the adaptive fault-tolerant performance of the pump-controlled hydrogen compressor. By combining the pump body loss model with adaptive adjustment of operating conditions, the method reduces the periodic impact and thermal stress on components such as bearings, seals, and motor windings under variable start-stop conditions, thus extending the overall lifespan of the equipment. By adopting pressure feedforward-feedback composite control and seal leakage pre-compensation, the method maintains the outlet pressure at no lower than the minimum threshold required for hydrogen supply during fault-tolerant actions such as redundant pump group switching, torque softening, and short-term derating operation, ensuring stable hydrogen supply from hydrogen refueling stations or hydrogen energy systems. Through online self-tuning of controller parameters, the method smoothly switches from normal mode to fault-tolerant control mode, avoiding efficiency losses and secondary impacts caused by shutdowns, and achieving a smooth transition without shutdown. This method is applicable to start-stop scenarios with varying degrees of variation and has strong generalization performance and engineering practicality.

[0081] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. An active fault-tolerant control method for a pump-controlled hydrogen compressor with adaptive start-stop conditions, characterized in that: It includes: S1: Collect multi-source state data of the pump-controlled hydrogen compressor and construct the multi-source state feature vector of the pump-controlled hydrogen compressor at time t. The degradation state level of the pump-controlled hydrogen compressor is evaluated; a degradation state evaluator for the pump-controlled hydrogen compressor is constructed to output the failure probability of the pump-controlled hydrogen compressor at time t. ; Count the cycle of start-up and stop events of the pump-controlled hydrogen compressor at time t. and start / stop frequency counting ; S1 specifically refers to: S11: Collect multi-source state data and construct... Multi-source state feature vector of pump-controlled hydrogen compressor operation at any time After normalization, the data is input into the online evaluation device for the degradation status of the pump-controlled hydrogen compressor; S12: Use a Long Short-Term Memory network to output temporal degradation features Degenerate Hidden Markov Models define multiple degenerate states. Construct a degradation state evaluator for a pump-controlled hydrogen compressor and output the failure probability of the pump-controlled hydrogen compressor at time t. ; The degenerate hidden Markov model sets five degenerate states. These correspond to the following states: normal operation corresponds to the first state, first-level degradation corresponds to the second state, second-level degradation corresponds to the third state, third-level degradation corresponds to the fourth state, and failure precursors correspond to the fifth state. S13: Statistical count of start / stop event cycles and start / stop frequency count for pump-controlled hydrogen compressors; start / stop event cycle count The current of the hydrogen compressor motor is controlled by the pump. From shutdown state Rise to a degenerate state judge, The current threshold is used to determine when to stop the machine. Current threshold for operation determination; start / stop frequency counting. Determined by the number of starts and stops within the sliding time window; S2: Adaptive adjustment of the pump-controlled hydrogen compressor's operating conditions, based on the start-stop frequency count of the pump-controlled hydrogen compressor in S1. Divide the start / stop frequency range into three levels: Level 1, Level 2, and Level 3; set the active fault-tolerant control trigger threshold at time t. and adaptive fault-tolerant parameters at time t The probability of pump-controlled hydrogen compressor failure at time t in S1 The active fault-tolerant control signal is triggered at the appropriate time. S3: Based on the active fault-tolerant trigger signal and adaptive fault-tolerant parameters in S2 The degradation state level and multi-source state feature vector at time t in S1 Real-time outlet pressure of pump-controlled hydrogen compressor Determine the softened actual output torque control command. Based on the real-time outlet pressure of the pump-controlled hydrogen compressor at time t Temperature of the motor winding sealing cavity at time t Estimated pressure decay rate and rate of temperature change Determine the leakage flow rate and perform pre-compensation for seal leakage in the pump-controlled hydrogen compressor; switch the redundant pump set of the pump-controlled hydrogen compressor and perform collaborative fault-tolerant control; S3 specifically refers to: S31: Pressure controller outputs raw torque control command The original motor torque control command Apply a first-order low-pass filter and an S-shaped saturation limiting constraint to output a softened actual output torque control command. ; S32: Perform pre-compensation for seal leakage in the pump-controlled hydrogen compressor; input the real-time outlet pressure of the pump-controlled hydrogen compressor at time t. Temperature of the motor winding sealing cavity at time t Estimate the pressure decay rate and rate of temperature change Determine the leakage flow rate; S33: Execute short-term derating operation of the pump-controlled hydrogen compressor; input the failure probability of the pump-controlled hydrogen compressor at time t. The probability of failure of the pump-controlled hydrogen compressor at time t If the current degradation state is the third or fourth state, the derated operation will be initiated. S34: Switch the redundant pump group of the pump-controlled hydrogen compressor; the probability of pump-controlled hydrogen compressor failure at time t. Furthermore, when the current main pump is in normal degradation state 5, the standby pump is started in advance and speed synchronization and pressure closed-loop pre-adjustment are performed; S4: Obtain the pressure deviation of the pump-controlled hydrogen compressor at time t. and rate of change The adaptive fault-tolerant parameter at time t in S2 A control model for a pump-controlled hydrogen compressor was constructed, and the online self-tuning of the pump-controlled hydrogen compressor controller parameters was performed to achieve active control of the pump-controlled hydrogen compressor.

2. The active fault-tolerant control method for a pump-controlled hydrogen compressor with adaptive start-stop conditions according to claim 1, characterized in that: The probability of pump-controlled hydrogen compressor failure at time t in S12 for: ； in, Let be the probability of failure of the pump-controlled hydrogen compressor at time t; Solve for the operator for conditional probability; The hidden state of real-time normal operation of the pump-controlled hydrogen compressor at time t; This is the time series sequence of multi-source observation features of the pump-controlled hydrogen compressor, collected from the initial time to time t. The state at time t The joint forward probability; Let be the joint forward probability at time t; This is the index of the normal degradation state of the pump-controlled hydrogen compressor; t is the time parameter.

3. The active fault-tolerant control method for a pump-controlled hydrogen compressor with adaptive start-stop conditions according to claim 2, characterized in that: The joint forward probability at time t in S12 for: ； ； in, The initial joint forward probability; The initial state distribution of the pump-controlled hydrogen compressor; The initial observation probability; The state transition probability; Let be the probability of observation at time t; The device operation characteristics collected in a single data acquisition at time t; i represents the degradation state level; The device operation characteristics collected in a single initial data acquisition; Let be the joint forward probability at time t-1.

4. The active fault-tolerant control method for a pump-controlled hydrogen compressor with adaptive start-stop conditions according to claim 3, characterized in that: Active fault-tolerant control trigger threshold in S2 and adaptive fault tolerance parameters Specifically: ； in, The threshold for active fault-tolerant control at time t; The baseline threshold; This is the threshold decay parameter; The adaptive fault-tolerant parameter at time t; Used as the reference strength parameter; For intensity gain parameters; t represents the start-stop frequency count of the pump-controlled hydrogen compressor at time t; e is a natural constant parameter.

5. The active fault-tolerant control method for a pump-controlled hydrogen compressor with adaptive start-stop conditions according to claim 4, characterized in that: S31 provides the original motor torque control command. Apply a first-order low-pass filter and an S-shaped saturation limiting constraint, specifically as follows: ， ； ； in, The actual output torque control command is obtained after first-order low-pass filtering and shaping. This is the original motor torque reference command output by the upper-level pressure control module; The time constant of a first-order low-pass filter; The reference torque filtering time constant; For Laplace complex frequency domain operator; k is the discrete control cycle count; This is the actual output torque control command; This is the saturation limiting operation function; This is the original motor torque reference command; This is the torque softening time constant; This represents the maximum adjustment amplitude of the single-step torque. The actual output torque control command for k cycles; This is the actual output torque control command for cycle k+1.

6. The active fault-tolerant control method for a pump-controlled hydrogen compressor with adaptive start-stop conditions according to claim 5, characterized in that: Pressure decay rate in S32 and rate of temperature change The leakage flow rate is determined to be: ； ； ； in, The pressure decay rate; For leakage flow; The rate of temperature change; For pressure decay related parameters; For temperature correction parameters; The real-time outlet pressure of the pump-controlled hydrogen compressor at time t; for Real-time outlet pressure of the hydrogen compressor; The real-time motor winding sealing cavity temperature of the pump-controlled hydrogen compressor at time t; for Real-time temperature of the motor winding sealing cavity in the hydrogen compressor; This represents the sampling time interval.

7. The active fault-tolerant control method for a pump-controlled hydrogen compressor with adaptive start-stop conditions according to claim 1, characterized in that: In S4, a recursive least squares (RLS) control model for the pump-controlled hydrogen compressor is constructed, and reference adaptive control is performed, specifically as follows: ； ； in, The vector of controller parameters to be identified is updated online at time t; Let t be the regression observation vector; for The covariance matrix of the time parameter estimation error; for The controller derating parameters obtained from the previous iteration; The measured pressure at time t; Forgetting factor; For the first The control input quantity for the pump-controlled hydrogen compressor drive control model is input at the sampling time; for Real-time pressure measurement; This is the matrix transpose operator; Let t be the gain vector. Let be the covariance matrix of the parameter estimation error at time t; It is an identity matrix.

Images