Multi-band high-voltage power supply control system and normalization control and protection method thereof

Abstract

The invention discloses a multiband high-voltage power supply control system and a normalization control and protection method thereof, and relates to the technical field of power supply control, and the method comprises the steps: collecting multi-mode signals of S, C and Ku band high-voltage power supply systems and a cooling system; an SDAE deep network and an integrated learning AdaBoost framework are combined to construct a fault diagnosis model, and a fault type and confidence are obtained according to signal features; when a fault occurs, high-voltage output of a corresponding power supply unit is cut off and reported through an isolation interface, and a deep reinforcement learning agent is used for carrying out protection action intelligent decision making; a digital twinborn model of a power supply unit is constructed, an improved grey wolf optimizer is used for multi-task collaborative optimization, optimized system parameters are acquired and issued to a PLC for execution, and waveband output stability is ensured. According to the invention, high reliability, rapid response and intelligent operation and maintenance of the multi-band high-voltage power supply system are realized through deep integration of multi-source signal acquisition, intelligent fault diagnosis, adaptive protection decision and digital twinning optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of power control, and more specifically, to a multi-band high-voltage power supply control system and its normalization control and protection method. Background Art

[0002] In fields such as high-intensity electromagnetic pulse (EMP) simulation tests, radar systems, and high-energy physics experiments, the high-voltage power supply system is one of the core devices, used to drive microwave devices such as klystrons and magnetrons to generate high-power electromagnetic pulses. Traditional multi-band high-voltage power supply systems usually adopt independently designed power modules, corresponding to different bands (such as S band, C band, Ku band) respectively, resulting in a complex system structure, scattered control logic, and difficulty in achieving efficient collaborative management and fault protection.

[0003] Currently, the existing technologies have the following main problems: decentralized control and low management efficiency. High-voltage power supplies in different bands usually adopt independent control systems, resulting in inability to uniformly manage parameter adjustment, status monitoring, and protection mechanisms, increasing operation complexity and reducing system reliability. Slow response of the protection mechanism. During the operation of the high-voltage power supply, faults such as klystron arcing, titanium pump overcurrent, and cooling failure may cause serious damage to the equipment. The protection response time of the traditional solution is relatively long and cannot meet the requirements of high stability. Electromagnetic compatibility (EMC) problems. The working environment of the high-voltage power supply system is complex and vulnerable to strong electromagnetic interference. Conventional control modules do not take sufficient isolation and shielding measures, resulting in unstable signal acquisition and control instruction transmission. Insufficient coordination of the cooling system. When the multi-band high-voltage power supply operates, it generates a large amount of heat, but the existing cooling systems often operate independently and lack intelligent linkage with power control, which may affect the overall performance due to local overheating. In view of the above problems, there is an urgent need for an integrated and highly reliable multi-band high-voltage power supply control system that adopts normalization control and protection design to achieve unified management of multiple power modules, rapid fault protection, and intelligent cooling coordination to meet the stringent requirements of modern high-power microwave systems for stability, response speed, and EMC performance. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a multi-band high-voltage power supply control system and its normalization control and protection method, which realizes the high reliability, rapid response, and intelligent operation and maintenance of the multi-band high-voltage power supply system through deep integration of multi-source signal acquisition, intelligent fault diagnosis, adaptive protection decision-making, and digital twin optimization.

[0005] The first aspect of the present invention provides a multi-band high-voltage power supply control system, including: a multi-source signal acquisition and preprocessing module, an intelligent fault diagnosis module, a protection decision-making and execution module, and a digital twin and parameter optimization module; The multi-source signal acquisition and preprocessing module is responsible for acquiring multi-modal signals of the S-band high-voltage power supply system, C-band high-voltage power supply system, Ku-band high-voltage power supply system, and cooling system, preprocessing the multi-modal signals, and extracting signal features; The intelligent fault diagnosis module combines the stacked denoising autoencoder network with the integrated learning AdaBoost framework to construct a fault diagnosis model, imports the signal features into the fault diagnosis model, and outputs the fault type and confidence level; The protection decision-making and execution module trains a deep reinforcement learning agent, and uses the deep reinforcement learning agent to make intelligent decisions on protection actions based on the fault type and fault severity; The digital twin and parameter optimization module constructs a digital twin model of the power supply unit. When the multi-band high-voltage power supply system is in a fault state, it uses the digital twin model to assist in adjusting the electrical operating point parameters and thermal management parameters, performs multi-task collaborative optimization using an improved grey wolf optimizer, obtains the optimized system parameters, issues them to the PLC for execution, and monitors the feedback in real time.

[0006] In this solution, the multi-source signal acquisition and preprocessing module is specifically: Use the pre-deployed sensors in the S-band high-voltage power supply system, C-band high-voltage power supply system, Ku-band high-voltage power supply system, and cooling system to construct a cross-band sensor network, synchronously acquire multi-modal signals, perform data standardization processing on the multi-modal signals, and add time stamps and device ID identifiers; Perform digital filtering and time-domain alignment on the multi-modal signals to obtain preprocessed multi-modal signals, extract time-frequency domain features from the preprocessed multi-modal signals to construct a feature matrix, obtain clustering clusters through K-means clustering, and identify the operating conditions of the multi-band high-voltage power supply system; Use a feature extraction layer combining one-dimensional convolution and dilated convolution to perform local feature extraction on the signal segments corresponding to the operating conditions, introduce a channel attention mechanism to strengthen the key frequency band features, use a BiLSTM network to capture the temporal dependencies of the multi-modal signals within a preset time window, and calculate the cross-band feature correlations using a self-attention layer to output global features; Align the local features and global features through time-axis interpolation, and use a weighted splicing fusion strategy for feature fusion to obtain signal features.

[0007] In this solution, the intelligent fault diagnosis module is specifically: Construct weak classifiers based on the stacked denoising autoencoder network, adopt a layer-by-layer greedy training strategy for pre-training the weak classifiers, and add Gaussian noise to each hidden layer during the pre-training of the weak classifiers for anti-interference training. After the pre-training is completed, add a Softmax classifier on the top layer and use labeled data for fine-tuning of the weak classifiers; Use the AdaBoost framework to integrate the pre-trained weak classifiers. In the first round of iteration, the same weight is assigned to all training samples. During the iteration process, the sample weights are dynamically adjusted according to the classification error of the weak classifier, and the weak classifiers that do not meet the preset classification accuracy are eliminated. When the preset number of iterations is reached, the weighted voting results of each weak classifier form the final prediction. Use the test samples for verification, and generate a strong classifier after passing the verification to build a fault diagnosis model; Use the fault diagnosis model for online monitoring. Parallelly input the obtained signal features into all weak classifiers. Each weak classifier outputs a fault probability prediction value. Obtain the comprehensive score through weighted summation. Dynamically adjust the threshold according to the current working condition. When the comprehensive score exceeds the dynamic threshold, trigger a fault alarm and output the fault type and confidence level.

[0008] In this solution, in the intelligent fault diagnosis module, enhance the fault diagnosis model, specifically: After triggering a fault alarm, extract the TEO energy feature, reflected wave integral feature, and phase mutation detection feature corresponding to the electrical data in the S-band high-voltage power supply system, C-band high-voltage power supply system, and Ku-band high-voltage power supply system respectively as electrical impact features; Extract the cavity temperature rise, vibration spectrum entropy, and flow pulsation corresponding to the physical data in the C-band high-voltage power supply system, Ku-band high-voltage power supply system, and cooling system respectively as physical anomaly features; Construct fault feature fingerprints in different bands according to the electrical impact features and physical anomaly features. Use the dynamic time warping algorithm to match the fault feature fingerprints with the fingerprint templates in the fault mode library, and obtain the preliminary screening fault types through the matching results. Perform classification enhancement in the fault diagnosis model through the preliminary screening fault types.

[0009] In this solution, the protection decision and execution module is specifically: Perform feature encoding on the multi-modal signals to generate the original state vector. Combine the working condition labels output by K-means clustering and the historical state memory to perform working condition context embedding on the original state vector to construct a state space; Extract the fault type nodes and protection measure nodes according to the fault mode library and historical fault protection measures. Based on the constraint conditions, construct the edge connection relationship between the nodes according to the triggering relationship and exclusion relationship to generate a fault-measure knowledge graph. Generate the basic action set of the protection measures based on the fault-measure knowledge graph to construct an action space; Construct a deep reinforcement learning agent based on the state space and actions using an improved PPO algorithm, quantify the fault loss using a pre-established three-level fault severity evaluation system, evaluate the action cost through power loss and equipment wear, and construct a basic reward function based on the fault loss and action cost; Introduce knowledge rewards, construct a current state vector, perform similarity matching between the current state vector and fault patterns in the fault-measure knowledge graph, construct a knowledge reward function based on the similarity matching, introduce exploration rewards, calculate the Mahalanobis distance between the current state vector and historical state vectors in the circular buffer of historical states, and construct a knowledge reward function based on the Mahalanobis distance; Introduce a knowledge graph constraint term in policy update, impose a penalty gradient on actions that violate the exclusive relationship of the knowledge graph, dynamically mask non-compliant actions according to the current fault type, extract fault records based on the fault pattern library and historical fault protection measures for pre-training of the deep reinforcement learning agent, and perform online fine-tuning using the ε-greedy strategy; Use the trained deep reinforcement learning agent's basic Q-value network and knowledge Q-value network to calculate the Q-values of each action, adjust the temperature coefficient according to the working conditions through the dynamic Boltzmann strategy, and select the final action as the protection measure for the fault.

[0010] In this solution, the digital twin and parameter optimization module is specifically: Obtain the electromagnetic field distribution inside the S, C, and Ku-band klystrons and the three-dimensional temperature field distribution of the multi-band high-voltage power supply, construct physical constraints based on the electromagnetic field distribution and three-dimensional temperature field distribution, use historical multi-modal signals as the input of the physics-informed neural network, impose physical constraints in the hidden layer of the physics-informed neural network, and obtain the state prediction value of the multi-modal signal through the output layer; Verify the state prediction value of the multi-modal signal using test data, obtain the digital twin model of the pre-trained power supply unit after passing the verification, and perform online incremental training using the latest collected multi-modal signals to fine-tune the digital twin model; Real-time collect multi-modal signals, import them into the digital twin model of the power supply unit to load the state of the current multi-band high-voltage power supply, construct a grey wolf optimizer for parameter tuning, construct an objective function with the maximization of microwave power, maximization of klystron life, and minimization of cooling energy consumption as the goals, associate the objective function with the fitness function, and calculate the fitness of the grey wolves; Perform iterative optimization according to the fitness of the grey wolves. When the number of iterations reaches the maximum number of iterations or meets the iteration termination condition, output the position of the α-wolf with the highest fitness in the last iteration to generate a parameter update suggestion, use the digital twin to simulate and verify the update suggestion, compare the simulation result with the expected performance improvement, and output the update suggestion that meets the requirements.

[0011] In this solution, the Grey Wolf Optimizer is specifically as follows: Calculate the fitness of grey wolves through the fitness function, select the top three optimal solutions as the α wolf, β wolf, and δ wolf according to the fitness ranking, and record the current global optimal solution. In the iterative calculation, use the steps of surrounding, hunting, and attacking prey to update the positions of the α wolf, β wolf, δ wolf, and other grey wolves. Apply the Lévy flight strategy to the α wolf, perform Gaussian perturbation on the β wolf and δ wolf to generate a new grey wolf population, and introduce an inertia weight that changes with the number of iterations; Calculate the fitness of grey wolves in the new grey wolf population, and introduce a mutation operator that changes with the number of iterations to mutate the α wolf with the highest fitness. Compare the fitness of the α wolf before and after mutation, and use the elitist retention strategy to select a better solution to enter the next generation and guide the search direction; When the number of iterations reaches the maximum number of iterations or meets the iteration termination condition, output the position of the α wolf with the highest fitness in the last iteration to generate parameter update suggestions.

[0012] The second aspect of the present invention provides a normalization control and protection method for a multi-band high-voltage power supply, which is applied to a high-voltage power supply control system and includes the following steps: Collect multi-modal signals of the S-band high-voltage power supply system, C-band high-voltage power supply system, Ku-band high-voltage power supply system, and cooling system, preprocess the multi-modal signals, and extract signal features; Use a stacked denoising autoencoder network combined with an integrated learning AdaBoost framework to construct a fault diagnosis model, import the signal features into the fault diagnosis model, and output the fault type and confidence level; When a fault occurs, cut off the high-voltage output of the corresponding power supply unit, report it to the remote terminal through an isolation interface, train a deep reinforcement learning agent, and use the deep reinforcement learning agent to make intelligent decisions on protection actions based on the fault type and fault severity; Construct a digital twin model of the power supply unit. When the multi-band high-voltage power supply system is in a fault state, assist in adjusting the electrical operating point parameters and thermal management parameters through the digital twin model, use an improved Grey Wolf Optimizer for multi-task collaborative optimization, obtain the optimized system parameters and send them to the PLC for execution to ensure the stability of the band output.

[0013] Compared with the prior art, the beneficial effects of the present invention are: Through the deep integration of technologies such as multi-source signal acquisition, intelligent fault diagnosis, adaptive protection decision-making, and digital twin optimization, the present invention realizes the high reliability, fast response, and intelligent operation and maintenance of the multi-band high-voltage power supply system. The system uses the SDAE-AdaBoost integrated model to perform real-time analysis on multi-dimensional signals such as voltage, current, and vacuum degree of the S / C / Ku-band power supply and cooling system, and can accurately detect minor faults (such as abnormal filament current and titanium pump vacuum leakage), increasing the accuracy of fault identification. Combined with the deep reinforcement learning agent, the system can complete fault classification response within milliseconds, significantly reducing the risk of klystron damage. The normalized control and protection design constructs a digital twin model through a physics-informed neural network to realize virtual pre-play of parameter tuning, reducing the overshoot rate during the high-voltage climbing process. The improved grey wolf algorithm collaboratively optimizes multiple objectives such as power output, equipment life, and cooling energy consumption, improving the energy-saving effect of the cooling system while ensuring the preset microwave power. Brief Description of the Drawings

[0014] In order to more clearly illustrate the technical solutions in the embodiments or exemplifications of the present invention, the following will briefly introduce the drawings required for use in the description of the embodiments or exemplifications. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to these drawings shown.

[0015] Figure 1 A block diagram of a multi-band high-voltage power supply control system is shown.

[0016] Figure 2 A flowchart showing the fault diagnosis of the intelligent fault diagnosis module in the embodiment is shown; Figure 3 A flowchart showing the generation of protection decisions by the protection decision and execution module in the embodiment is shown; Figure 4 A flowchart showing a normalized control and protection method for a multi-band high-voltage power supply is shown. Detailed Embodiments

[0017] In order to be able to more clearly understand the above-mentioned objects, features, and advantages of the present invention, the present invention will be further described in detail below in conjunction with the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0018] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0019] Figure 1Shows a block diagram of a multi-band high-voltage power supply control system.

[0020] This embodiment provides a multi-band high-voltage power supply control system, including: a multi-source signal acquisition and preprocessing module 101, an intelligent fault diagnosis module 102, a protection decision and execution module 103, and a digital twin and parameter optimization module 104; The multi-source signal acquisition and preprocessing module 101 is responsible for acquiring multi-modal signals of the S-band high-voltage power supply system, C-band high-voltage power supply system, Ku-band high-voltage power supply system, and cooling system, preprocessing the multi-modal signals, and extracting signal features; The intelligent fault diagnosis module 102 combines a stacked denoising autoencoder network with an integrated learning AdaBoost framework to construct a fault diagnosis model, imports the signal features into the fault diagnosis model, and outputs the fault type and confidence level; The protection decision and execution module 103 trains a deep reinforcement learning agent, and uses the deep reinforcement learning agent to make intelligent decisions on protection actions based on the fault type and fault severity; The digital twin and parameter optimization module 104 constructs a digital twin model of the power supply unit, assists in adjusting electrical operating point parameters and thermal management parameters through the digital twin model when the multi-band high-voltage power supply system is in a fault state, uses an improved grey wolf optimizer for multi-task collaborative optimization, obtains optimized system parameters and issues them to the PLC for execution, and monitors and feeds back in real time.

[0021] It should be noted that the multi-band high-voltage power supply system includes an S-band high-voltage power supply system, a C-band high-voltage power supply system, a Ku-band high-voltage power supply system, and a cooling system. The S-band high-voltage power supply system includes: 2 sets of completely identical power supply systems (including control, titanium pump power supply, filament power supply, DC high-voltage power supply), a pulse modulator system (pulse modulator and high-voltage pulse transformer components); the C-band high-voltage power supply system includes: 2 sets of completely identical power supply systems (including control, titanium pump power supply, filament power supply, DC high-voltage power supply), a pulse modulator system (pulse modulator and high-voltage pulse transformer components); the Ku-band high-voltage power supply system includes: 2 sets of completely identical power supply systems (including control, titanium pump power supply, filament power supply, DC high-voltage power supply), a pulse modulator system (pulse modulator and high-voltage pulse transformer components); the cooling system includes: cooling control, cooling box, cooling cycle, heating device, refrigeration device, heat exchange device, and supporting sensors and links, etc. Each band's high-voltage power supply system contains two sets of power supply systems supporting klystrons. To ensure the effective operation of the high-voltage power supply system, the control of the two power supply systems is designed in a normalized manner. Therefore, the system can be divided into a control and protection module, two sets of titanium pump power supplies, two sets of filament power supplies, two sets of DC high-voltage power supplies, and two sets of pulse modulators and their pulse transformer oil tank components. All external interfaces adopt isolation measures, including digital, analog, and communication interfaces, and a shielding box is installed to ensure its electromagnetic compatibility.

[0022] Construct a cross - band sensor network using pre - deployed sensors in the S - band high - voltage power supply system, C - band high - voltage power supply system, Ku - band high - voltage power supply system and cooling system. The pre - deployed sensors include: high - voltage differential probes, Rogowski coils, capacitive vacuum gauges, electromagnetic flowmeters, temperature sensor arrays, etc. Synchronously collect multi - modal signals, including electrical parameters (cathode voltage, filament current, modulator pulse waveform) and physical parameters (titanium pump vacuum degree, coolant flow rate, tube body temperature). Standardize the data of the multi - modal signals and add time stamps and device ID tags; perform digital filtering and time - domain alignment on the multi - modal signals to obtain pre - processed multi - modal signals. Extract time - frequency domain features from the pre - processed multi - modal signals to construct a feature matrix, obtain clustering clusters through K - means clustering, identify the working conditions of the multi - band high - voltage power supply system, and realize the automatic identification of 5 typical working conditions: standby / warm - up / steady - state operation / overload / fault recovery. Each working condition adopts an exclusive feature extraction strategy (such as focusing on monitoring transient response in the overload condition). Use a feature extraction layer combining one - dimensional convolution and dilated convolution to perform local feature extraction on the signal segments corresponding to the working conditions, introduce a channel attention mechanism to strengthen key frequency band features, use a BiLSTM network to capture temporal dependencies of the multi - modal signals in a preset time window, and use a self - attention layer to calculate cross - band feature correlations to output global features; align the local features (in microseconds) and global features (in minutes) on the time axis through interpolation, and use a weighted splicing fusion strategy for feature fusion. The fusion weights are dynamically adjusted according to the working conditions to obtain signal features.

[0023] Figure 2 The flowchart of the intelligent fault diagnosis module for fault diagnosis in the embodiment is shown.

[0024] According to the embodiment of the present invention, the intelligent fault diagnosis module is specifically: S202, construct a weak classifier based on the stacked denoising auto - encoder network, adopt a layer - by - layer greedy training strategy for pre - training the weak classifier, and add Gaussian noise to each hidden layer during the pre - training of the weak classifier for anti - interference training. After the pre - training is completed, add a Softmax classifier on the top layer and fine - tune the weak classifier using labeled data; S204, use the AdaBoost framework to integrate the pre - trained weak classifiers. In the first round of iteration, all training samples are given the same weight. During the iteration process, dynamically adjust the sample weights according to the classification error of the weak classifier, eliminate the weak classifiers that do not meet the preset classification accuracy. When the preset number of iterations is reached, the weighted voting results of each weak classifier form the final prediction. Use test samples for verification. After the verification passes, generate a strong classifier to construct a fault diagnosis model; S206. Use the fault diagnosis model for online monitoring. Parallelly input the obtained signal features into all weak classifiers. Each weak classifier outputs a fault probability prediction value. Obtain a comprehensive score through weighted summation. Dynamically adjust the threshold according to the current working condition. When the comprehensive score exceeds the dynamic threshold, trigger a fault alarm and output the fault type and confidence level.

[0025] It should be noted that the stacked denoising autoencoder network is used as the basic feature extractor. Its three-layer network structure (input layer - hidden layer - output layer) obtains robustness to noise interference through the pre-training method of adding Gaussian noise. The AdaBoost framework integrates multiple denoising autoencoder network weak classifiers into a strong classifier through an iterative mechanism. The number of input layer nodes of each denoising autoencoder network corresponds to the feature dimension (such as a 50-dimensional feature vector). The number of hidden layer nodes is set to 60% of the input layer, and the ReLU activation function is adopted. The output layer generates a preliminary judgment result of the fault probability through the Sigmoid function. When training the first stacked denoising autoencoder network, Gaussian noise with a strength of 20% is added to the input data to enhance the generalization ability. Calculate the classification error rate after each round of iteration. When the classification error rate is greater than the preset threshold, automatically discard the current weak classifier. Increase the weight of misclassified samples so that the next round of stacked denoising autoencoder network pays more attention to the features of difficult samples. After N rounds of iteration, the weights of each weak classifier are calculated according to the formula . The stacked denoising autoencoder network with better classification performance obtains a higher voting weight. The decision output of the final strong classifier is the weighted summation of the outputs of each weak classifier. A reliable fault determination is achieved by setting a confidence threshold of 0.7. The first stacked denoising autoencoder network focuses on capturing microsecond-level transient features, and subsequent stacked denoising autoencoder networks gradually extract features in a longer time domain. Through the sample weight adjustment mechanism of AdaBoost, the system automatically strengthens the most sensitive feature dimensions under the current operating condition. The hidden layer outputs of each stacked denoising autoencoder network are simultaneously input into the attention module to generate a heat map of feature importance for fault tracing. The output layer contains two types of key information: fault type classification (such as sparking / overcurrent / cooling failure, etc.) and confidence score (0 - 100%). When it is detected that the THD of the S-band current exceeds the limit, the model will comprehensively consider the output results of each stacked denoising autoencoder network. If all 3 main weak classifiers determine that it is a rectifier bridge fault (confidence levels 72%, 85%, 68%), then the final determined fault probability is 82%, exceeding the threshold to trigger an early warning.

[0026] In the intelligent fault diagnosis module, after enhancing the fault diagnosis model and triggering a fault alarm, the TEO energy feature, reflected wave integral feature, and phase mutation detection feature corresponding to the electrical data are extracted from the S-band high-voltage power supply system, C-band high-voltage power supply system, and Ku-band high-voltage power supply system respectively as electrical impact features. When extracting the TEO energy feature, the original signal is filtered by a fifth-order Butterworth band-pass filter to eliminate baseline drift and high-frequency noise. The Teager energy operator is calculated using a sliding window to calculate the instantaneous energy of each sampling point, and the peak-to-peak value and standard deviation of the energy within the window are statistically calculated. Based on the energy distribution of the previous n cycles, calculate as the reference threshold. When the energy value exceeds the threshold for three consecutive windows and is accompanied by a pulse waveform distortion greater than 50 ns, it is determined as an arc discharge. When extracting the reflected wave integral feature, the reflected signal is envelope detected to extract the envelope line of the pulse waveform, and the integral energy within a single pulse period is calculated. The integral value is divided by the incident wave energy to obtain the reflection coefficient. When the reflection coefficient is greater than the preset threshold and lasts for five cycles, an alarm for abnormal waveguide connection is triggered. When extracting the phase mutation detection feature, the instantaneous phase is calculated through Hilbert transform, the phase difference is calculated, and the standard deviation of the phase differences of n normal cycles is statistically calculated to establish the phase noise background. Phase jump events are recorded based on the phase noise background. The cavity temperature rise, vibration spectrum entropy, and flow pulsation corresponding to the physical data are extracted from the C-band high-voltage power supply system, Ku-band high-voltage power supply system, and cooling system respectively as physical anomaly features. Fault feature fingerprints for different bands are constructed based on the electrical impact features and physical anomaly features. The dynamic time warping algorithm is used to match the fault feature fingerprints with the fingerprint templates in the fault mode library, and the initially screened fault types are obtained through the matching results. In the fault diagnosis model, classification enhancement is performed through the initially screened fault types to further improve the fault recognition ability of the fault diagnosis model.

[0027] Figure 3 The flowchart showing the protection decision and execution module generating a protection decision in the embodiment is presented.

[0028] According to an embodiment of the present invention, the protection decision and execution module is specifically: S302, perform feature encoding on the multi-modal signals to generate an original state vector, and perform context embedding of the working conditions on the original state vector by combining the working condition labels output by K-means clustering and the historical state memory to construct a state space; S304, extract the fault type nodes and protection measure nodes according to the fault mode library and historical fault protection measures, construct the edge connection relationship between the nodes based on the trigger relationship and exclusion relationship according to the constraint conditions to generate a fault-measure knowledge graph, generate a basic action set of protection measures based on the fault-measure knowledge graph, and construct an action space; S306. Use the improved PPO algorithm to construct a deep reinforcement learning agent based on the state space and actions. Use the pre-established three-level evaluation system for fault severity to quantify the fault losses, and evaluate the action costs through power losses and equipment wear. Construct a basic reward function according to the fault losses and action costs; S308. Introduce knowledge rewards, construct the current state vector, perform similarity matching between the current state vector and the fault modes in the fault-measure knowledge graph, construct a knowledge reward function according to the similarity matching, introduce exploration rewards, calculate the Mahalanobis distance between the current state vector and the historical state vectors in the circular buffer of historical states, and construct a knowledge reward function according to the Mahalanobis distance; S310. Introduce a knowledge graph constraint term in the policy update, impose a penalty gradient on actions that violate the exclusive relationship of the knowledge graph, and dynamically mask non-compliant actions according to the current fault type. Extract fault records according to the fault mode library and historical fault protection measures for pre-training of the deep reinforcement learning agent, and perform online fine-tuning using the ε-greedy strategy; S312. Use the trained basic Q-value network and knowledge Q-value network of the deep reinforcement learning agent to calculate the Q-values of each action, adjust the temperature coefficient according to the working conditions through the dynamic Boltzmann strategy, and select the final action as the protection measure for the fault.

[0029] It should be noted that the real-time collected electrical signals are segmented with a 0.1-second time window after being standardized. 12-dimensional time-domain features (including peak value, mean value, waveform factor, etc.) and 8-dimensional frequency-domain features (main harmonic component energy ratio) are extracted within each time window to form a 20-dimensional electrical feature vector. After the physical signals are filtered by moving average, the trend features (slope, curvature, fluctuation intensity) on a 1-minute time scale are calculated to generate a 6-dimensional physical state vector. The electrical and physical feature vectors are concatenated to form a 26-dimensional original state vector, and then it is reduced to a 15-dimensional compact representation through an autoencoder to realize multi-modal signal feature encoding. Combining the working condition labels output by K-means clustering, 3-dimensional working condition identifiers are appended in the form of one-hot encoding. Introduce the memory of the device historical state, extract 8-dimensional time-series features of the state change in the recent preset time steps through the LSTM network, and the final state space is 26-dimensional.

[0030] In the construction of the fault - measure knowledge graph, the triggering relationship connects faults and measures, and the exclusion relationship identifies measure conflicts. The preferred basic action set includes 6 discrete actions: power reduction, switching to standby power supply, adjusting cooling flow, triggering pre - charge protection, full shutdown, and maintaining the current state. Based on the improved PPO algorithm - based deep reinforcement learning agent, a three - level evaluation system for fault severity (minor 1 / moderate 3 / serious 5) is established to quantify fault losses, with the duration accurate to the millisecond level. For example, if the klystron arc - over fault (severity 5) lasts for 20 ms, the contributed loss value is 5×0.02 = 0.1. In the action cost assessment, the power loss is calculated according to the actual reduction percentage, and the equipment wear is converted through a preset life - impact coefficient. In the knowledge reward, the current state vector is matched with 20 typical fault modes in the knowledge graph (cosine similarity), and the highest matching degree is taken as the reference value. When the system takes the protection actions recommended by the knowledge graph, an additional coefficient bonus of 0.5 is given. In the exploration reward excitation, a circular buffer containing 5000 historical states is maintained. When the Mahalanobis distance between the current state and the last 1000 recorded states is greater than 3σ, it is determined as a new state. The first discovery reward for a new state is 0.2, and it decays exponentially when it reappears. A rule - checking module is added to the output layer of the policy network to impose a penalty gradient on actions that violate the exclusion relationship in the knowledge graph. For example, when both power reduction and boost operation are output simultaneously, the penalty term of 10% is added to the gradient backpropagation. Uncompliant actions are dynamically masked according to the current fault type. For example, when a cooling failure fault is detected, the boost operation action option is disabled.

[0031] In the pre - training of the deep reinforcement learning agent, fault records are extracted. Each record contains three elements: the state vector, the actions taken manually, and the final result, and oversampling is performed on a small number of types of faults for balance. The teacher network uses a 3 - layer fully - connected network, and the student network (i.e., the policy network) conducts distillation learning through the KL - divergence loss combined with the knowledge - graph compliance loss. In the online fine - tuning stage, the initial exploration probability ε of the ε - greedy policy scheduling is 0.5, which linearly decays by 0.1 after every 500 fault processes and finally maintains a basic exploration rate of 0.1. The soft - update strategy is adopted, and 0.1% of the online network parameters are updated to the target network every 1000 steps. In the dynamic Boltzmann strategy, the temperature coefficient regulation enables the direct selection of the action with the highest Q - value in case of severe faults, and the Gumbel - Softmax technique is used to achieve differentiable sampling, ensuring effective gradient backpropagation while maintaining randomness.

[0032] In a preferred embodiment, when the state recognizes that the TEO energy exceeds the limit and the vibration spectrum entropy drops by 25%, the fault mode of electrode micro-discharge is matched through knowledge retrieval, and actions of reducing the power by 20% and increasing the flow rate by 15% are generated and executed. The execution effect is that the ignition frequency drops to the safe range within 200 ms. Through the fine description of the state space and the rule constraints of the knowledge graph, deep reinforcement learning realizes rapid optimization decision-making on the premise of ensuring safety.

[0033] It should be noted that the digital twin and parameter optimization module obtains the electromagnetic field distribution in the S, C, and Ku-band klystrons and the three-dimensional temperature field distribution of the multi-band high-voltage power supply. The electromagnetic field distribution constructs a klystron electromagnetic model based on the finite element method, solves the three-dimensional Maxwell equations, and accurately simulates the electromagnetic field distribution in the S, C, and Ku bands; a three-dimensional temperature field model is established using computational fluid dynamics to simulate the heat conduction and convective heat transfer of key parts such as the tube body and the collector to construct a three-dimensional temperature field distribution. Physical constraints are constructed based on the electromagnetic field distribution and the three-dimensional temperature field distribution. The historical multi-modal signals are used as the input of the physics-informed neural network. Physical constraints are imposed in the hidden layer of the physics-informed neural network, and the state prediction value of the multi-modal signal is obtained through the output layer; the test data is used to verify the state prediction value of the multi-modal signal. After passing the verification, the digital twin model of the pre-trained power supply unit is obtained, and the latest collected multi-modal signals are used for online incremental training to fine-tune the digital twin model.

[0034] When the multi-band high-voltage power supply system is in a fault state, after temporary protection through fault protection measures, the current fault characteristics are injected into the twin body, and parameter optimization is carried out to ensure the maximum output power on the premise of ensuring that the equipment fault does not deteriorate further, so as to extend the equipment life while reducing the cooling energy consumption. Real-time collect multi-modal signals and import them into the digital twin model of the power supply unit to load the current state of the multi-band high-voltage power supply, construct a grey wolf optimizer to optimize the electrical operating point parameters and thermal management parameters. The electrical operating point parameters include the cathode high-voltage setting value, pulse modulator parameters, and the working current of the titanium pump power supply, etc. The thermal management parameters include the target flow rate of the coolant, the power setting of the refrigeration unit, and the start-stop threshold of the heater, etc. Construct an objective function with the maximization of microwave power, the maximization of the klystron life, and the minimization of the cooling energy consumption as the goals, associate the objective function with the fitness function, and calculate the fitness of the grey wolves; calculate the fitness of the grey wolves through the fitness function, select the top three optimal solutions as the α-wolf, β-wolf, and δ-wolf according to the fitness ranking, and record the current global optimal solution. In the iterative calculation, use the steps of surrounding, hunting, and attacking the prey to update the positions of the α-wolf, β-wolf, δ-wolf, and other grey wolves. Adopt the Lévy flight strategy for the α-wolf, perform Gaussian perturbation on the β-wolf and δ-wolf to generate a new grey wolf population, and introduce an inertia weight that changes with the number of iterations; calculate the fitness of the grey wolves in the new grey wolf population, and introduce a mutation operator that changes with the number of iterations to mutate the α-wolf with the highest fitness, compare the fitness of the α-wolf before and after mutation, use the elite retention strategy to select a better solution to enter the next generation and guide the search direction; when the number of iterations reaches the maximum number of iterations or meets the iteration termination condition, output the position of the α-wolf with the highest fitness in the last iteration to generate parameter update suggestions, use the digital twin to simulate and verify the update suggestions, compare the simulation results with the expected performance improvement, output the update suggestions that meet the requirements, and perform PLC execution to monitor the optimization results. Through the combination of the high-precision simulation of the digital twin and the improved optimization algorithm, the collaborative optimization of multi-objective parameters is achieved.

[0035] Figure 4 Fig. shows a flow chart of a normalization control and protection method for a multi-band high-voltage power supply.

[0036] This embodiment provides a normalization control and protection method for a multi-band high-voltage power supply, which is applied to a high-voltage power supply control system and includes the following steps: S402, collect multi-modal signals of the S-band high-voltage power supply system, C-band high-voltage power supply system, Ku-band high-voltage power supply system, and cooling system, preprocess the multi-modal signals, and extract signal features; S404, use the combination of a stacked denoising autoencoder network and an integrated learning AdaBoost framework to construct a fault diagnosis model, import the signal features into the fault diagnosis model, and output the fault type and confidence level; S406. When a fault occurs, cut off the high-voltage output of the corresponding power supply unit and report it to the remote terminal through the isolation interface. Train the deep reinforcement learning agent, and use the deep reinforcement learning agent to make intelligent decisions on protection actions based on the fault type and fault severity. S408. Build a digital twin model of the power supply unit. When the multi-band high-voltage power supply system is in a fault state, assist in adjusting the electrical operating point parameters and thermal management parameters through the digital twin model. Use the improved grey wolf optimizer for multi-task collaborative optimization, obtain the optimized system parameters and send them to the PLC for execution to ensure the stability of the band output.

[0037] This embodiment provides a computer-readable storage medium, which includes a normalization control and protection method program for a multi-band high-voltage power supply. When the normalization control and protection method program for a multi-band high-voltage power supply is executed by a processor, the steps of a normalization control and protection method for a multi-band high-voltage power supply are implemented.

[0038] In several embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described above are only illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple blocks or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed with each other can be through some interfaces, indirect coupling or communication connection of devices or units, and can be electrical, mechanical or other forms. In addition, each functional block in the embodiments of the present invention can be all integrated in one processing block, or each block can be separately used as one block, or two or more blocks can be integrated in one block; the above integrated block can be implemented in the form of hardware, or in the form of a hardware plus software functional block.

[0039] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: mobile storage devices, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical disks and other various media that can store program codes.

[0040] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A multi-band high-voltage power supply control system, characterized in that Including: Multi-source signal acquisition and preprocessing module, intelligent fault diagnosis module, protection decision-making and execution module, digital twin and parameter optimization module; The multi-source signal acquisition and preprocessing module is responsible for collecting multi-modal signals of the S-band high-voltage power supply system, C-band high-voltage power supply system, Ku-band high-voltage power supply system and cooling system, preprocessing the multi-modal signals, and extracting signal features; The intelligent fault diagnosis module combines the stacked denoising autoencoder network with the integrated learning AdaBoost framework to build a fault diagnosis model, imports the signal features into the fault diagnosis model, and outputs the fault type and confidence level; The protection decision-making and execution module trains a deep reinforcement learning agent, and uses the deep reinforcement learning agent to make intelligent decisions on protection actions based on the fault type and fault severity; The digital twin and parameter optimization module constructs a digital twin model of the power supply unit. When the multi-band high-voltage power supply system is in a fault state, it uses the digital twin model to assist in adjusting the electrical operating point parameters and thermal management parameters, uses an improved grey wolf optimizer for multi-task collaborative optimization, obtains the optimized system parameters and issues them to the PLC for execution, and monitors the feedback in real time.

2. The multi-band high-voltage power supply control system according to claim 1, wherein The multi-source signal acquisition and preprocessing module is specifically: Use the pre-deployed sensors in the S-band high-voltage power supply system, C-band high-voltage power supply system, Ku-band high-voltage power supply system and cooling system to build a cross-band sensor network, synchronously collect multi-modal signals, perform data standardization processing on the multi-modal signals, and add time stamps and device ID identifiers; Perform digital filtering and time-domain alignment on the multi-modal signals to obtain preprocessed multi-modal signals, extract time-frequency domain features from the preprocessed multi-modal signals to construct a feature matrix, obtain clustering clusters through K-means clustering, and identify the operating conditions of the multi-band high-voltage power supply system; Use a feature extraction layer that combines one-dimensional convolution and dilated convolution to perform local feature extraction on the signal segments corresponding to the operating conditions, introduce a channel attention mechanism to strengthen the key frequency band features, use a BiLSTM network to capture the temporal dependencies of the multi-modal signals in a preset time window, and use a self-attention layer to calculate the cross-band feature correlations to output global features; Align the local features and global features through time-axis interpolation, and use a weighted splicing fusion strategy for feature fusion to obtain signal features.

3. The multi-band high-voltage power supply control system according to claim 1, wherein The intelligent fault diagnosis module is specifically: Build a weak classifier based on the stacked denoising autoencoder network, adopt a layer-by-layer greedy training strategy for pre-training the weak classifier, and add Gaussian noise in each hidden layer during the pre-training of the weak classifier for anti-interference training. After the pre-training is completed, add a Softmax classifier on the top layer and use labeled data for fine-tuning the weak classifier; Use the AdaBoost framework to integrate the pre-trained weak classifiers. In the first round of iteration, the same weight is assigned to all training samples. During the iteration process, the sample weights are dynamically adjusted according to the classification error of the weak classifier, and the weak classifiers that do not meet the preset classification accuracy are eliminated. When the preset number of iterations is reached, the weighted voting results of each weak classifier are used to form the final prediction. The test samples are used for verification, and after passing the verification, a strong classifier is generated to build a fault diagnosis model; Use the fault diagnosis model for online monitoring. Parallelly input the obtained signal features into all weak classifiers. Each weak classifier outputs a fault probability prediction value, and the comprehensive score is obtained through weighted summation. The threshold is dynamically adjusted according to the current working condition. When the comprehensive score exceeds the dynamic threshold, a fault alarm is triggered, and the fault type and confidence level are output.

4. The multi-band high-voltage power supply control system according to claim 3, characterized in that, In the intelligent fault diagnosis module, enhance the fault diagnosis model, specifically: After triggering the fault alarm, extract the TEO energy feature, reflected wave integral feature, and phase mutation detection feature corresponding to the electrical data in the S-band high-voltage power supply system, C-band high-voltage power supply system, and Ku-band high-voltage power supply system respectively as the electrical impact features; Extract the cavity temperature rise, vibration spectrum entropy, and flow pulsation corresponding to the physical data in the C-band high-voltage power supply system, Ku-band high-voltage power supply system, and cooling system respectively as the physical anomaly features; Construct fault feature fingerprints in different bands according to the electrical impact features and physical anomaly features. Use the dynamic time warping algorithm to match the fault feature fingerprints with the fingerprint templates in the fault mode library, and obtain the initially screened fault types through the matching results. In the fault diagnosis model, perform classification enhancement through the initially screened fault types.

5. The multi-band high-voltage power supply control system according to claim 1, characterized in that, The protection decision and execution module is specifically: Perform feature encoding on the multi-modal signals to generate the original state vector, and perform working condition context embedding on the original state vector in combination with the working condition labels output by K-means clustering and the historical state memory to construct the state space; Extract the fault type nodes and protection measure nodes according to the fault mode library and historical fault protection measures, construct the edge connection relationship between the nodes based on the trigger relationship and exclusion relationship according to the constraint conditions, generate the fault-measure knowledge graph, and generate the basic action set of the protection measures based on the fault-measure knowledge graph to construct the action space; Use the improved PPO algorithm to construct a deep reinforcement learning agent based on the state space and actions. Use the pre-established three-level evaluation system for fault severity to quantify the fault loss, and evaluate the action cost through power loss and equipment wear. Construct the basic reward function according to the fault loss and action cost; Introduce knowledge rewards, construct the current state vector, perform similarity matching between the current state vector and the fault modes in the fault-measure knowledge graph, construct the knowledge reward function according to the similarity matching, introduce exploration rewards, calculate the Mahalanobis distance between the current state vector and the historical state vectors in the circular buffer of historical states, and construct the knowledge reward function according to the Mahalanobis distance; Introduce a knowledge graph constraint term in the policy update, impose a penalty gradient on actions that violate the exclusive relationship of the knowledge graph, and dynamically mask non-compliant actions according to the current fault type. Extract fault records based on the fault mode library and historical fault protection measures for pre-training of the deep reinforcement learning agent, and use the ε-greedy strategy for online fine-tuning; Use the trained deep reinforcement learning agent's base Q-value network and knowledge Q-value network to calculate the Q-values of each action, adjust the temperature coefficient according to the working conditions through the dynamic Boltzmann strategy, and select the final action as the protection measure for the fault.

6. The multi-band high-voltage power supply control system according to claim 1, characterized in that The protection decision and execution module, the digital twin and parameter optimization module, specifically: Obtain the electromagnetic field distribution inside the S, C, and Ku-band klystrons and the three-dimensional temperature field distribution of the multi-band high-voltage power supply. Construct physical constraints based on the electromagnetic field distribution and the three-dimensional temperature field distribution. Use the historical multi-modal signals as the input of the physics-informed neural network, impose physical constraints in the hidden layer of the physics-informed neural network, and obtain the state prediction value of the multi-modal signals through the output layer; Use the test data to verify the state prediction value of the multi-modal signals. After passing the verification, obtain the digital twin model of the pre-trained power supply unit, and use the newly collected multi-modal signals for online incremental training to fine-tune the digital twin model; Real-time collect multi-modal signals and import them into the digital twin model of the power supply unit to load the state of the current multi-band high-voltage power supply. Construct a grey wolf optimizer for parameter tuning. Construct an objective function with the maximization of microwave power, the maximization of klystron life, and the minimization of cooling energy consumption as the goals. Associate the objective function with the fitness function and calculate the fitness of the grey wolves; Perform iterative optimization according to the fitness of the grey wolves. When the number of iterations reaches the maximum number of iterations or meets the iteration termination condition, output the position of the α-wolf with the highest fitness in the last iteration to generate a parameter update suggestion. Use the digital twin to simulate and verify the update suggestion, compare the simulation result with the expected performance improvement, and output the update suggestion that meets the requirements.

7. The multi-band high-voltage power supply control system according to claim 6, characterized in that, The grey wolf optimizer, specifically: Calculate the fitness of the grey wolves through the fitness function. Select the top three optimal solutions as the α-wolf, β-wolf, and δ-wolf according to the fitness ranking, and record the current global optimal solution. In the iterative calculation, use the steps of surrounding, hunting, and attacking the prey to update the positions of the α-wolf, β-wolf, δ-wolf, and other grey wolves. Adopt the Lévy flight strategy for the α-wolf, perform Gaussian perturbation on the β-wolf and δ-wolf to generate a new grey wolf population, and introduce an inertia weight that changes with the number of iterations; Calculate the fitness of the grey wolves in the new grey wolf population, and introduce a mutation operator that changes with the number of iterations to mutate the α-wolf with the highest fitness. Compare the fitness of the α-wolf before and after mutation, and use the elitist retention strategy to select a better solution to enter the next generation and guide the search direction; When the number of iterations reaches the maximum number of iterations or meets the iteration termination condition, output the position of the α-wolf with the highest fitness in the last iteration to generate a parameter update suggestion.

8. A normalization control and protection method for a multi-band high-voltage power supply, applied to the high-voltage power supply control system described in any one of claims 1-7, characterized in that, Include the following steps: Collect multimodal signals of the S-band high-voltage power supply system, C-band high-voltage power supply system, Ku-band high-voltage power supply system and cooling system, preprocess the multimodal signals, and extract signal features; Use a stacked denoising autoencoder network combined with an integrated learning AdaBoost framework to construct a fault diagnosis model, import the signal features into the fault diagnosis model, and output the fault type and confidence level; When a fault occurs, cut off the high-voltage output of the corresponding power supply unit and report it to the remote terminal through the isolation interface, train a deep reinforcement learning agent, and use the deep reinforcement learning agent to make intelligent decisions on protection actions based on the fault type and fault severity; Build a digital twin model of the power supply unit. When the multi-band high-voltage power supply system is in a fault state, assist in adjusting the electrical operating point parameters and thermal management parameters through the digital twin model, use an improved grey wolf optimizer for multi-task collaborative optimization, obtain the optimized system parameters and send them to the PLC for execution to ensure the stability of the band output.

Images