A wind storage integration grid performance PHM detection system and method based on distributed control technology

Abstract

The present application relates to the technical field of power system operation and control, in particular to a wind storage integration grid performance PHM detection system and method based on distributed control technology, comprising: a distributed synchronous acquisition unit; a distributed on-site data processing unit; a wind storage grid PHM evaluation unit; a distributed cooperative closed-loop control unit. The present application constructs a wind storage integration key control link physical model and a data-driven prediction model based on BP neural network, and completes fusion training. According to real-time wind speed, the double model weight ratio is dynamically adjusted, the critical point of compensation capability depletion can be calculated, and the graded fault prediction trajectory is generated. In combination with the actual response data of the wind storage integration system, the model parameters and weights can be closed-loop optimized, the fine prediction and evaluation of the wind storage integration grid performance are realized, the control loop operation state monitoring and critical point prediction are completed, and the dynamic optimization of the prediction model is realized.

Description

Technical Field

[0001] This invention relates to the field of power system operation and control technology, and more specifically, to a wind-storage integrated grid performance PHM detection system and method based on distributed control technology. Background Technology

[0002] Wind-storage integration is a core technology for the efficient consumption of new energy. Its grid connection performance directly affects the stable operation of the power system. Fault prediction and health management monitoring of this system urgently require the support of a suitable distributed monitoring technology system.

[0003] In the existing technology, relevant patents have been researched and explored in the fields of power system operation control and energy storage power station testing. For example, invention patent CN202511526769.1 discloses a method for coordinated control of active power and frequency for wind-storage combined systems, belonging to the field of power system operation and control technology. Addressing the problem of poor frequency control performance caused by the uncertainty of wind power and load, it constructs a three-layer collaborative architecture and dynamically allocates active power through scenario-based random optimization and deep reinforcement learning coordinated control, achieving the effect of smoothing frequency fluctuations and delaying the degradation of energy storage life. Another example is invention patent CN202310784510.1, which discloses a method and system for testing the grid-connected performance of energy storage power stations, involving the field of data testing technology. By collecting battery pack operating parameters to build a predictive model, it calculates the grid-connected operating capacity value and the power surplus value, sets battery pack priorities for performance testing, and improves the efficiency of energy storage power station performance testing.

[0004] Despite the design advantages of the above technical solutions, they also have the following technical defects: First, they lack the ability to make refined predictions and assessments of the grid-connected performance of integrated wind and energy storage systems, and cannot achieve critical point prediction and dynamic model optimization of the control loop. Invention patent CN202511526769.1 only focuses on the coordinated control of active power and frequency, without constructing a predictive model for the key control links of integrated wind and energy storage systems, nor does it have a design for adjusting the model according to real-time operating conditions, calculating the critical point of compensation capacity depletion, or any related schemes for model closed-loop optimization. Invention patent CN202310784510.1 can only achieve basic testing of the grid-connected performance of energy storage power stations, without a design for monitoring and predicting the operating status of the control loop, and cannot meet the core requirements of fault prediction and health management of integrated wind and energy storage grid-connected performance. Secondly, the architecture of the technical solutions cannot meet the grid connection monitoring requirements of multi-node distributed wind-storage integrated systems: the three-layer architecture of invention patent CN202511526769.1 only focuses on the generation and execution of control commands, without designing for distributed synchronous acquisition and local data processing; invention patent CN202310784510.1 is only designed for a single main body of the energy storage power station, without considering the grid connection operation characteristics of multi-node integrated wind-storage systems, and lacks a linkage design for distributed collaborative monitoring and control. Therefore, we propose a wind-storage integrated grid connection performance PHM monitoring system and method based on distributed control technology. Summary of the Invention

[0005] The purpose of this invention is to provide a wind-storage integrated grid connection performance PHM detection system and method based on distributed control technology, in order to solve the problems mentioned in the background art, such as insufficient ability to make refined predictions and evaluations of the grid connection performance of wind-storage integrated systems, inability to predict critical points of control loops and dynamically optimize models, and inability of the technical solution architecture design to match the grid connection detection requirements of multi-node distributed wind-storage integrated systems.

[0006] To address the aforementioned technical problems, one objective of this invention is to provide a wind-storage integrated grid connection performance (PHM) detection system based on distributed control technology, comprising: The distributed synchronous acquisition unit deploys distributed acquisition terminals for the grid-connected nodes of the wind-storage integrated system. It adopts distributed clock synchronization technology to synchronously acquire the grid-connected electrical parameters, operating status parameters and internal status data of the control loop of each node, and outputs the time-synchronized raw acquisition data to the distributed local data processing unit. A distributed local data processing unit is deployed at each distributed acquisition node. It performs data preprocessing, grid-connected performance feature extraction and data standardization on the raw acquired data, and outputs standardized grid-connected performance feature data and control loop status synchronization data to the wind-storage grid-connected PHM evaluation unit. The wind-storage grid-connected PHM assessment unit receives standardized grid-connected performance characteristic data and control loop status synchronization data output by the distributed local data processing unit, constructs a physical model of key control links in the wind-storage integration, constructs a data-driven prediction model and integrates physical consistency constraints into the training of the data-driven prediction model, dynamically adjusts the weight ratio of the physical model and the data-driven prediction model according to real-time operating conditions, monitors the output margin of the control loop and calculates the critical point of compensation capacity depletion, generates fault prediction trajectories, outputs grid-connected performance health status data, fault prediction data and graded intervention basis to the distributed collaborative closed-loop control unit, and optimizes the physical model parameters and data-driven prediction model weights in a closed loop based on the actual response data of the wind-storage integrated system fed back by the distributed collaborative closed-loop control unit. The distributed collaborative closed-loop control unit receives grid-connection performance health status data, fault prediction data, and graded intervention basis output by the wind-storage grid-connected PHM assessment unit. It adopts a distributed collaborative control strategy to issue local control commands to each distributed node of the integrated wind-storage system, update system control parameters and detection benchmark parameters, and feed back the actual system response data to the wind-storage grid-connected PHM assessment unit.

[0007] As a further improvement to this technical solution, the distributed synchronous acquisition unit includes a distributed acquisition terminal deployment module, a distributed clock synchronization module, a multi-parameter synchronous acquisition module, and a raw data output module, wherein: The distributed acquisition terminal deployment module deploys distributed acquisition terminals for network-connected nodes of the integrated wind and storage system. The distributed clock synchronization module uses distributed clock synchronization technology to achieve time synchronization of each distributed acquisition terminal. The multi-parameter synchronous acquisition module synchronously acquires the network electrical parameters, operating status parameters and internal status data of each node based on the time synchronization status of each distributed acquisition terminal. The raw data output module outputs the time-synchronized raw data collected synchronously to the distributed local data processing unit.

[0008] As a further improvement to this technical solution, the distributed on-site data processing unit includes a data preprocessing module, a network performance feature extraction module, a data standardization processing module, and a processed data output module, wherein: The data preprocessing module is deployed at each distributed acquisition node and performs data preprocessing operations on the raw acquisition data output by the distributed synchronous acquisition unit. The network performance feature extraction module performs network performance feature extraction operations based on the raw collected data processed by the data preprocessing module. The data standardization processing module performs data standardization processing operations based on the network performance characteristics extracted by the network performance characteristic extraction module. The processed data output module outputs the standardized grid-connected performance characteristic data and control loop status synchronization data obtained by the data standardization processing module to the wind-storage grid-connected PHM evaluation unit.

[0009] As a further improvement to this technical solution, the wind-storage grid-connected PHM evaluation unit includes a dual-drive model construction module, a weight dynamic adjustment module, a critical point prediction and trajectory generation module, a detection and evaluation output module, and a model closed-loop optimization module, wherein: The dual-drive model construction module receives standardized grid-connected performance characteristic data and control loop status synchronization data output by the distributed local data processing unit, constructs a physical model of key control links in wind-storage integration, constructs a data-driven prediction model, and incorporates physical consistency constraints into the training of the data-driven prediction model. The weight dynamic adjustment module dynamically adjusts the weight ratio between the physical model and the data-driven prediction model in the dual-drive model construction module according to the real-time operating conditions. The critical point prediction and trajectory generation module is based on the output of the dual-drive model construction module and the weight ratio of the weight dynamic adjustment module. It monitors the output margin of the control loop and calculates the critical point of compensation capability depletion to generate a fault prediction trajectory. The detection and evaluation output module, based on the output of the critical point prediction and trajectory generation module, outputs network performance health status data, fault prediction data, and hierarchical intervention basis to the distributed collaborative closed-loop control unit. The model closed-loop optimization module optimizes the physical model parameters and data-driven prediction model weights in the closed-loop optimization dual-drive model construction module based on the actual response data of the wind-storage integrated system fed back by the distributed collaborative closed-loop control unit.

[0010] As a further improvement to this technical solution, the construction and training process of the physical model and data-driven prediction model of the dual-drive model construction module includes the following steps: S31.1 Based on the design parameters and operating mechanism of the integrated wind and energy storage system, construct a physical model of the phase-locked loop, converter control loop, and energy storage system response characteristics, and output the predicted values ​​of the physical model. ; S31.2 Construct a data-driven prediction model based on a BP neural network, using standardized network performance characteristic data as input, and outputting predicted values ​​related to the control loop compensation capability. ; S31.3 Integrate the dynamic equations of the physical model as constraints into the training of the data-driven prediction model, and construct a total loss function with physical consistency constraints. ; S31.4 Minimize the total loss function using gradient descent. Update the weights and biases of the data-driven prediction model, complete the fusion training of the physical model and the data-driven prediction model, and output to the weight dynamic adjustment module.

[0011] As a further improvement to this technical solution, the weight ratio adjustment process of the weight dynamic adjustment module includes the following steps: S32.1 Obtain real-time wind speed data collected by the distributed synchronous acquisition unit. Physical model weights Weights of data-driven prediction models Satisfying constraints ; S32.2 Setting a preset first threshold Preset second threshold ;when When configuring Minimum weight , ; S32.3, when At that time, based on Linear adjustment and ; S32.4, when When configuring Maximum weight , ; S32.5, will and Output to the critical point prediction and trajectory generation module.

[0012] As a further improvement to this technical solution, the fault prediction trajectory generation process of the critical point prediction and trajectory generation module includes the following steps: S33.1 Calculate the saturation margin of the PI controller output. Collect the integral value of the phase deviation of the phase-locked loop; S33.2, Based on the fitting window Data, calculate margin change rate ; S33.3, based on With current saturation margin Calculate the critical time when the compensation capacity is exhausted. Get the remaining warning time ; S33.4, according to The system generates a graded fault prediction trajectory based on the location within the warning zone and outputs it to the detection and evaluation output module.

[0013] As a further improvement to this technical solution, the model parameter closed-loop optimization process of the model closed-loop optimization module includes the following steps: S35.1 Calculate the predicted values ​​of the physical model Compared with the actual system response value root mean square error Verify the physical model parameters ; S35.2 Setting the error threshold ,when At that time, construct the parameter optimization objective function. ; S35.3 Update the physical model parameters using gradient descent. The weights and biases of the data-driven prediction model are updated synchronously. S35.4. Feed the optimized parameters back to the dual-drive model building module to complete the collaborative closed-loop optimization of the physical model and the data-driven prediction model.

[0014] As a further improvement to this technical solution, the distributed cooperative closed-loop control unit includes a data receiving module, a cooperative control strategy execution module, a control command issuing module, a parameter updating module, and a response data feedback module, wherein: The data receiving module receives grid-connected performance health status data, fault prediction data, and tiered intervention basis output by the wind-storage grid-connected PHM assessment unit. The collaborative control strategy execution module adopts a distributed collaborative control strategy to generate local control commands; The control command issuing module issues local control commands to each distributed node; The parameter update module updates the control parameters and detection benchmark parameters; The response data feedback module feeds back the actual response data to the wind-storage grid-connected PHM evaluation unit.

[0015] The second objective of this invention is to provide a method for detecting the power grid connection performance (PHM) of an integrated wind and energy storage system based on distributed control technology. The method, based on the aforementioned distributed control technology-based integrated wind and energy storage PHM detection system, includes the following steps: S1. Distributed acquisition terminals are deployed for the grid-connected nodes of the wind-storage integrated system. Distributed clock synchronization technology is used to realize the time synchronization of each distributed acquisition terminal, and the grid-connected electrical parameters, operating status parameters and internal status data of the control loop of each node are collected synchronously, and the time-synchronized raw acquisition data is output. S2. At each distributed acquisition node, data preprocessing, network performance feature extraction and data standardization are performed sequentially on the raw acquisition data synchronized with time, and standardized network performance feature data and control loop status synchronization data are output. S3. Receive standardized grid-connected performance characteristic data and control loop status synchronization data, construct a physical model and a data-driven prediction model for key control links of wind-storage integration, integrate physical consistency constraints into the training of the data-driven prediction model and complete dual-model fusion training, dynamically adjust the weight ratio of the physical model and the data-driven prediction model according to real-time operating conditions, monitor the output margin of the control loop and calculate the critical point of compensation capacity depletion, generate fault prediction trajectory, and output grid-connected performance health status data, fault prediction data and graded intervention basis; S4. Receive network performance health status data, fault prediction data and hierarchical intervention basis, adopt a distributed collaborative control strategy to issue local control commands to each distributed node of the wind-storage integrated system, update the control parameters and detection benchmark parameters of the wind-storage integrated system, and feed back the actual response data of the wind-storage integrated system. S5. Based on the actual response data of the wind-storage integrated system, the physical model parameters and data-driven prediction model weights are optimized in a closed loop. The optimized physical model parameters and data-driven prediction model weights are then reapplied to the construction of the physical model and data-driven prediction model in the key control links of the wind-storage integrated system, as well as the process of adjusting their weight ratio, forming a closed-loop operation of detection-control-optimization.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a physical model of the key control links of wind-storage integration and a data-driven prediction model based on BP neural network and completes fusion training. It dynamically adjusts the weight ratio of the two models according to real-time wind speed, calculates the critical point of compensation capacity depletion and generates graded fault prediction trajectories. It can also combine the actual response data of the wind-storage integration system to optimize the model parameters and weights in a closed loop, realize the refined prediction and evaluation of the grid-connected performance of wind-storage integration, complete the monitoring of the operation status of the control loop and the prediction of critical points, and the dynamic optimization of the prediction model, adapting to the application requirements of grid-connected performance fault prediction and health management of wind-storage integration. 2. This invention deploys distributed acquisition terminals at grid-connected nodes in a wind-storage integrated system. It achieves synchronous acquisition of multiple parameters through distributed clock synchronization technology, completes local data processing at each distributed acquisition node, and employs a distributed collaborative control strategy to issue control commands and provide feedback on the system's actual response data. This constructs a distributed closed-loop system of detection and evaluation, collaborative control, and model optimization, adapting to the distributed grid-connection detection scenario of multi-node wind-storage integrated systems. It realizes synchronous acquisition and local processing of grid-connected nodes across the entire domain, and completes the coordinated operation of system grid-connection performance detection and distributed collaborative control. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the system framework of the present invention; Figure 2 This is a schematic diagram of the method steps of the present invention; The meanings of the labels in the diagram are as follows: 1. Distributed synchronous acquisition unit; 11. Distributed acquisition terminal deployment module; 12. Distributed clock synchronization module; 13. Multi-parameter synchronous acquisition module; 14. Raw data output module; 2. Distributed local data processing unit; 21. Data preprocessing module; 22. Network performance feature extraction module; 23. Data standardization processing module; 24. Processed data output module; 3. Wind-Storage Grid-Connected PHM Assessment Unit; 31. Dual-Drive Model Construction Module; 32. Weight Dynamic Adjustment Module; 33. Critical Point Prediction and Trajectory Generation Module; 34. Detection and Evaluation Output Module; 35. Model Closed-Loop Optimization Module; 4. Distributed collaborative closed-loop control unit; 41. Data receiving module; 42. Collaborative control strategy execution module; 43. Control command issuance module; 44. Parameter update module; 45. Response data feedback module. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, this embodiment provides a wind-storage integrated grid connection performance (PHM) detection system based on distributed control technology, including: Distributed synchronous acquisition unit 1 is configured to deploy distributed acquisition terminals at grid-connected nodes of the wind-storage integrated system (specifically, an integrated new energy power generation system that deeply integrates, coordinates, and connects wind power generation and energy storage systems to the grid, including core grid-connected nodes such as wind turbine grid connection points, energy storage system charging and discharging interfaces, wind-storage collection station busbars, grid connection gateway metering points, and collection line branch nodes). It employs distributed clock synchronization technology to synchronously acquire grid-connected electrical parameters, operating status parameters, and internal control loop status data of each node, outputting time-synchronized raw acquisition data to distributed local data processing unit 2. Distributed synchronous acquisition unit 1 includes a distributed acquisition terminal deployment module 11, a distributed clock synchronization module 12, a multi-parameter synchronous acquisition module 13, and a raw data output module 14, wherein: In this embodiment, the distributed acquisition terminal deployment module 11 deploys distributed acquisition terminals for the network-connected nodes of the integrated wind and energy storage system, determines the core acquisition nodes of the integrated wind and energy storage system connected to the network, completes terminal deployment, hardware docking and redundant networking, and builds the synchronous acquisition hardware foundation for the distributed synchronous acquisition unit 1. The specific implementation is as follows: Based on the topology of the integrated wind and energy storage system, the grid connection point on the wind turbine side, the charging and discharging interface of the energy storage system, the busbar of the wind and energy storage collection station, the metering point at the grid connection port, and the branch nodes of the collection line are identified as the core data acquisition nodes connected to the grid. Distributed data acquisition terminals are deployed at each core node according to the principle of "one terminal per point". The terminals are hard-wired to the node's electrical secondary circuit and the equipment control system's I / O interface through multi-channel universal interfaces. A dual-link redundant communication network is built for all terminals using a 5G industrial private network and fiber optic Ethernet. Each terminal is assigned a unique node number to achieve unified identification and management of the terminals.

[0020] In this embodiment, the distributed clock synchronization module 12 adopts distributed clock synchronization technology to realize time synchronization of each distributed acquisition terminal, providing a unified acquisition time reference for the multi-parameter synchronous acquisition module 13. The specific implementation is as follows: The IEEE 1588v2 precision time protocol is adopted as the distributed clock synchronization technology. Each distributed acquisition terminal has a built-in PTP slave clock module, and a PTP master clock device with Beidou + GPS dual-mode time synchronization is deployed at the wind storage collection station. The master and slave clocks exchange synchronization messages through the dual link built by the distributed acquisition terminal deployment module 11. The master clock periodically sends synchronization messages with timestamps and delay response messages. The slave clock calculates the time deviation and link transmission delay in real time and dynamically calibrates the local clock to ensure that the time of all terminals is consistent with the master clock reference time.

[0021] In this embodiment, the multi-parameter synchronous acquisition module 13 synchronously acquires the network electrical parameters, operating status parameters, and internal status data of each node based on the time synchronization status of each distributed acquisition terminal, generating time-aligned raw acquisition data. The specific implementation is as follows: Using the unified clock calibrated by the distributed clock synchronization module 12 as a reference, a collection trigger command with the same timestamp is sent to all terminals to achieve synchronous collection of multiple nodes; the collection frequency is set according to the differences in parameter characteristics, the grid-connected electrical parameters are collected at high frequency, and the operating status parameters and the internal status data of the control loop are collected according to the equipment control cycle; three types of parameters are collected: grid-connected electrical parameters, operating status parameters, and internal status data of the control loop. Each collected parameter is marked with the node number assigned by the distributed collection terminal deployment module 11 and the precise collection timestamp to achieve data time alignment.

[0022] In this embodiment, the raw data output module 14 outputs the time-synchronized raw data acquired synchronously to the distributed local data processing unit 2, ensuring the integrity and continuity of data transmission. The specific implementation is as follows: The original acquired data with identification is packaged according to the IEC61850 industrial standard. Each frame of data includes a data type identifier, node number, acquisition timestamp, parameter value, and data validity identifier. Through the dual links built by the distributed acquisition terminal deployment module 11, the packaged data is output to the corresponding distributed local data processing unit 2 according to the acquisition frequency. The original data output module 14 has a built-in local data caching mechanism. When the link is interrupted, the real-time acquired data is cached, and the data is retransmitted in the order of timestamps after the link is restored.

[0023] Distributed local data processing unit 2 is deployed at each distributed acquisition node. It performs data preprocessing, grid-connected performance feature extraction and data standardization on the raw acquired data, and outputs standardized grid-connected performance feature data and control loop status synchronization data to wind-storage grid-connected PHM evaluation unit 3. In this embodiment, the distributed on-site data processing unit 2 includes a data preprocessing module 21, a network performance feature extraction module 22, a data standardization processing module 23, and a processed data output module 24. The data preprocessing module 21 is deployed at each distributed acquisition node and performs data preprocessing operations on the raw acquisition data output by the distributed synchronous acquisition unit 1 to remove invalid data, eliminate interference noise, unify data dimensions, ensure data validity and regularity, and provide a high-quality data source for the network performance feature extraction module 22. The specific implementation is as follows: The data preprocessing module 21 is deployed at each distributed acquisition node. It receives the raw acquisition data transmitted by the raw data output module 14 of the distributed synchronous acquisition unit 1. It first removes abnormal data and duplicate acquisition data that exceed the normal operating range of the equipment by using the threshold method and the neighborhood comparison method. It then uses linear interpolation to complete a small number of missing data. The high-frequency noise in the electrical parameters of the grid is then processed using a wavelet denoising algorithm, and the random interference in the operating status parameters and the internal status data of the control loop is processed using a median filtering algorithm. Finally, the various types of data at different acquisition frequencies are resampled along the time axis, and the data are aligned according to the unified timestamp of the distributed synchronous acquisition unit 1 to form a regular preprocessed dataset.

[0024] In this embodiment, the grid connection performance feature extraction module 22 performs grid connection performance feature extraction operation based on the raw collected data processed by the data preprocessing module 21, extracts feature quantities that are strongly correlated with the grid connection performance of the wind-storage integrated system, mines effective information in the data, and provides core feature data for the data standardization processing module 23. The specific implementation is as follows: The grid-connected performance feature extraction module 22 reads the preprocessed dataset from the data preprocessing module 21 and extracts three types of grid-connected performance features based on the grid-connected operation mechanism of the integrated wind and energy storage system: first, time-domain features, including the effective values ​​of three-phase voltage / current, the average and fluctuation rate of grid frequency, the active / reactive power deviation rate, the energy storage SOC change rate, and the output amplitude of the PI controller; second, frequency-domain features, including the harmonic content of grid voltage / current, the proportion of frequency harmonic components, and the characteristic value of voltage flicker; and third, control loop features, including the phase-locked loop phase deviation rate, the converter voltage / current loop response rate, the modulation ratio change rate, and the control loop output margin related features. Meanwhile, during the extraction process, the acquisition node number and timestamp corresponding to all feature data are retained, and the original regular data of the internal state of the control loop are extracted synchronously to form a network performance feature dataset and a control loop state dataset.

[0025] In this embodiment, the data standardization processing module 23 performs data standardization processing based on the grid-related performance features extracted by the grid-related performance feature extraction module 22, eliminating the differences in the dimensions and numerical ranges of different features, and improving the accuracy of subsequent wind-storage grid-related PHM assessment unit 3 model calculation and assessment. The specific implementation is as follows: The data standardization processing module 23 receives the network performance feature dataset and control loop state dataset output by the network performance feature extraction module 22. It uses the industrially common min-max normalization algorithm to standardize the network performance feature data, linearly mapping each feature value to the [0,1] interval, preserving the physical meaning and relative change trend of the feature data. The control loop state dataset is not numerically standardized, but only its node number and timestamp integrity are verified to ensure strict synchronization between the control loop state data and the standardized network performance feature data in the node and time dimensions, ultimately forming standardized network performance feature data and control loop state synchronization data.

[0026] In this embodiment, the processed data output module 24 outputs the standardized grid-connected performance characteristic data and control loop status synchronization data obtained by the data standardization processing module 23 to the wind-storage grid-connected PHM evaluation unit 3, ensuring the synchronization, orderliness, and accuracy of data transmission. The specific implementation is as follows: After processing, the data output module 24 reads the two types of synchronous data output by the data standardization processing module 23, and integrates the data according to the acquisition node number + unified timestamp to form a synchronous data frame with one node per frame. The data frame includes node identifier, timestamp identifier, standardized feature data segment, control loop status data segment, and data validity verification segment. Meanwhile, the data frames are encapsulated using the IEC61850 industrial communication protocol, which is compatible with the distributed synchronous acquisition unit 1. Relying on the redundant communication links of each distributed acquisition node, the encapsulated synchronous data is output to the wind-storage grid-connected PHM evaluation unit 3. A built-in data verification mechanism is implemented during transmission. If the receiving end reports abnormal data, the corresponding data frame will be retransmitted immediately to ensure that the data is delivered completely.

[0027] The wind-storage grid-connected PHM assessment unit 3 receives standardized grid-connected performance characteristic data and control loop status synchronization data output from the distributed local data processing unit 2. It constructs a physical model of the key control links in the wind-storage integration, builds a data-driven prediction model, and integrates physical consistency constraints into the training of the data-driven prediction model. It dynamically adjusts the weight ratio of the physical model and the data-driven prediction model based on real-time operating conditions, monitors the output margin of the control loop and calculates the critical point of compensation capacity depletion, generates fault prediction trajectories, and outputs grid-connected performance health status data, fault prediction data, and tiered intervention basis to the distributed collaborative closed-loop control unit 4. Based on the actual response data of the wind-storage integrated system fed back by the distributed collaborative closed-loop control unit 4, it optimizes the physical model parameters and the weights of the data-driven prediction model in a closed loop. The wind-storage grid-connected PHM assessment unit 3 is the core assessment unit of the wind-storage integrated grid-connected performance PHM detection system. Its hardware is deployed on the main control server of the wind-storage aggregation station, and it achieves bidirectional data interaction with the distributed local data processing unit 2 and the distributed collaborative closed-loop control unit 4 via industrial Ethernet. The wind-storage grid-connected PHM assessment unit 3 receives standardized grid-connected performance characteristic data and control loop status synchronization data output by the distributed local data processing unit 2, constructs a physical model of key control links in the wind-storage integration, builds a data-driven prediction model and integrates physical consistency constraints into the training of the data-driven prediction model, dynamically adjusts the weight ratio of the physical model and the data-driven prediction model according to real-time operating conditions, monitors the output margin of the control loop and calculates the critical point of compensation capacity depletion, generates fault prediction trajectories, and outputs grid-connected performance health status data, fault prediction data and graded intervention basis to the distributed collaborative closed-loop control unit 4, and optimizes the physical model parameters and data-driven prediction model weights in a closed loop based on the actual response data of the wind-storage integrated system fed back by the distributed collaborative closed-loop control unit 4.

[0028] Furthermore, the wind-storage grid-connected PHM assessment unit 3 includes a dual-drive model construction module 31, a weight dynamic adjustment module 32, a critical point prediction and trajectory generation module 33, a detection and evaluation output module 34, and a model closed-loop optimization module 35, wherein: In this embodiment, the dual-drive model construction module 31 receives standardized grid-connected performance characteristic data and control loop state synchronization data output by the distributed local data processing unit 2, constructs a physical model of the key control links of wind-storage integration, constructs a data-driven prediction model, and incorporates physical consistency constraints into the training of the data-driven prediction model; the construction and training process of the physical model and data-driven prediction model of the dual-drive model construction module 31 includes the following steps: S31.1 Based on the design parameters and operating mechanism of the integrated wind and energy storage system, construct a physical model of the phase-locked loop, converter control loop, and energy storage system response characteristics, and output the predicted values ​​of the physical model. This provides a mechanistic constraint benchmark for data-driven models, specifically implemented as follows: The dual-drive model construction module 31 receives the control loop status synchronization data output by the distributed local data processing unit 2 and the data output module 24, based on the design parameters of the core control link of the wind-storage integrated system (phase-locked loop proportional coefficient). Integral coefficient Converter voltage loop proportional coefficient Integral coefficient Current loop proportionality coefficient Integral coefficient Equivalent internal resistance of energy storage battery Rated capacity (etc.), constructing three types of core physical sub-models, as follows: I. Phase-locked loop physical model: Based on the second-order generalized integral phase-locked loop (SOGI-PLL) mechanism, the sampled values ​​of the three-phase voltage of the input power grid are... First, the three-phase voltage is converted into two-phase stationary coordinate system components through Clark transformation, and then the grid voltage phase is output through phase tracking equation. ,frequency The predicted value, specifically the formula, is as follows: Clark transformation (three-phase → two-phase stationary coordinate system): ; in: express The grid voltage after Clark transformation at any given time Shaft component, in V (volts); express The sampled value of phase A voltage of the power grid at any given time, in V (volts); express The sampled value of phase B voltage of the power grid at any given time, in V (volts); express The sampled value of the C-phase voltage of the power grid at any given time, in V (volts); express The β-axis component of the grid voltage after Clark transformation at any given time, in units of V (volts).

[0029] Phase tracking and frequency calculation equations: ; in: express The phase tracking error of the phase-locked loop at any given time, expressed in rad (radians). express The phase prediction value of the grid voltage output by the phase-locked loop at any time, in rad (radians). express The rate of change of the grid voltage phase at any given moment (angular frequency), in rad / s (radians per second); This represents the rated angular frequency of the power grid, a fixed value of 2π×50rad / s, with units of rad / s (radians per second). The table represents the proportional coefficient of the phase-locked loop, which is dimensionless; represents the integral coefficient of the phase-locked loop, which is dimensionless; This represents the integration variable (time), with the unit being seconds (s). Indicates the phase error from time 0 to The integral value at time t, in units of ( ); express The predicted grid frequency output by the phase-locked loop at any given time, in Hz (Hertz). This represents a constant, which is dimensionless.

[0030] II. Physical Model of Converter Control Loop: Based on the dual closed-loop control mechanism of voltage outer loop and current inner loop, the input active power reference value... Reactive power reference value Combined with grid voltage Output converter modulation voltage Predicted values; among which , For grid voltage , The synchronous coordinate system components after Park transformation are given first with the Park transformation formula, followed by the core control equations.

[0031] Park transformation formula (two-phase stationary coordinate system → two-phase synchronous coordinate system): ; in: express The grid voltage after Park transformation at any time Axis components (synchronous coordinate system), unit is V (volts); express The cosine value of the predicted phase value of the grid voltage at any given time, which is dimensionless; express The sinusoidal value of the predicted phase value of the grid voltage at any given time, dimensionless; express The grid voltage after Park transformation at any time Axis components (synchronous coordinate system), unit is V (volt).

[0032] Core governing equations: ; in: express The voltage output of the outer loop of the converter at any time Shaft voltage reference value, in V (volts); This represents the proportional coefficient of the converter voltage loop, and is dimensionless. express Reference value of active power of converter at any time, in kW (kilowatt). express The converter outputs active power in real time, in kW (kilowatts). This represents the integral coefficient of the converter voltage loop, which is dimensionless. express The voltage output of the outer loop of the converter at any time Shaft voltage reference value, in V (volts); express Reference value of reactive power of converter at any time, in kVar (kilovar). express The real-time output reactive power of the converter is measured in kVar (kilovar). express The current output of the converter inner loop at all times Shaft current reference value, in A (amperes); This represents the proportional coefficient of the converter current loop, and is dimensionless. This represents the integral coefficient of the converter current loop, which is dimensionless. express The current output of the converter inner loop at all times Shaft current reference value, in A (amperes); express Moment converter modulation voltage Axial component prediction, in V (volts). express Moment converter modulation voltage Axial component prediction, in V (volts). express The voltage output of the outer loop of the converter at any time Shaft voltage reference value, in V (volts); express The voltage output of the outer loop of the converter at any time Shaft voltage reference value, in V (volts).

[0033] III. Physical Model of Energy Storage System Response Characteristics: Based on the Thevenin equivalent circuit model, input energy storage charging and discharging power reference value Output energy storage Terminal voltage The predicted value, the core formula is as follows: ; in: express The predicted state of charge of the energy storage system at any given time is dimensionless and ranges from 0 to 1. express The initial moment is the state of charge of the energy storage system, dimensionless, with a value range of 0 to 1; This indicates the initial time of integration, in seconds (s). This indicates the rated capacity of the energy storage battery, measured in Ah (ampere-hours). express Real-time energy storage charging and discharging power, in kW (kilowatt). express The terminal voltage of the energy storage system at any given time, in V (volts). express Predicted terminal voltage of the energy storage system at any given time, in V (volts). This indicates the open-circuit voltage of the energy storage battery, measured in volts (V). This represents the equivalent internal resistance of the energy storage battery, measured in Ω (ohms). express The energy storage charging and discharging current at any time, measured in amperes (A). express The actual charging and discharging power of the energy storage at any given time, in kW (kilowatts).

[0034] Furthermore, the dual-drive model construction module 31 integrates the output results of the above three types of physical sub-models to obtain the comprehensive prediction value of the physical model. The calculation formula is as follows: ; in: express The vector of comprehensive predicted values ​​from the physical model at any given time, with dimensions of 6×1 and no units; This represents the transpose operation of a vector, and is dimensionless.

[0035] S31.2 Construct a data-driven prediction model based on a BP neural network. Using standardized network performance characteristic data as input, mine the potential patterns in actual operation data and output predicted values ​​related to the compensation capability of the control loop. To compensate for the insufficient adaptation of the physical model to the dynamic operating conditions of the integrated wind and energy storage system, the specific implementation is as follows: The dual-drive model construction module 31 receives standardized network performance characteristic data output by the distributed local data processing unit 2's processed data output module 24, and constructs a three-layer BP neural network structure data-driven prediction model, including an input layer, a hidden layer, and an output layer. Input layer: The number of neurons is consistent with the dimension of the standardized network performance feature data, and the input vector is... ,in For the first Standardized grid-connected performance characteristics (such as effective voltage, frequency, active power, reactive power, harmonic content, and energy storage SOC).

[0036] Hidden layer: Employs ReLU activation function, number of neurons According to empirical formulas Determine and implement the nonlinear mapping of features.

[0037] Output layer: 1 neuron; outputs the predicted value related to the compensation capability of the control loop. .

[0038] The formula for calculating the forward propagation of a BP neural network is: ; ; in: express The output vector of the hidden layer of the BP neural network at time 1 is dimensionless, and its dimension is the same as the number of neurons in the hidden layer. This represents the activation function of the hidden layer in a BP neural network. In this embodiment, the ReLU activation function is used, and its expression is: Unitless, used to implement nonlinear mapping of input features; This represents the weight matrix from the input layer to the hidden layer of the BP neural network. It is dimensionless and its dimension is the number of neurons in the hidden layer × the dimension of the input layer features. express The input vector of the BP neural network at time step is the standardized network performance feature data output by the distributed local data processing unit 2. It is dimensionless and its dimension is consistent with the number of standardized network performance features. This represents the bias vector from the input layer to the hidden layer of a BP neural network. It is dimensionless and its dimension is the same as the number of neurons in the hidden layer. express The predicted values ​​related to the control loop compensation capability output by the time-data-driven prediction model are dimensionless, ranging from 0 to 1, and characterize the strength of the control loop compensation capability. This represents the weight matrix from the hidden layer to the output layer of the BP neural network. It is dimensionless and its dimension is the number of neurons in the output layer × the number of neurons in the hidden layer. This represents the bias vector from the hidden layer to the output layer of a BP neural network. It is dimensionless and has the same dimension as the number of neurons in the output layer. The dimension representing standardized network performance characteristic data, a positive integer, without units; Represents the number of neurons in the hidden layer of a BP neural network; a positive integer with no unit.

[0039] S31.3 Integrate the dynamic equations of the physical model as constraints into the training of the data-driven prediction model, and construct a total loss function with physical consistency constraints. The actual value in the control loop status synchronization data output by the processed data output module 24 Based on the physical model predictions output by S31.1, S31.2 output data drives the prediction model's predicted values Construct a total loss function that includes a data fitting loss term and a physical consistency constraint loss term. The calculation formula is: ; ； ; in: This represents the total loss function with physical consistency constraints. It is dimensionless and takes values ​​≥ 0. The smaller the value, the better the prediction effect of the data-driven prediction model and the closer it is to the physical mechanism. The data fitting loss term of the data-driven prediction model is the mean square error between the predicted and actual values. It is dimensionless, takes a value ≥ 0, and characterizes the degree of fit between the data-driven prediction model and the actual operating data. The weight coefficient of the physical consistency constraint loss term is dimensionless and ranges from (0,1). It is used to balance the proportion of the data fitting loss term and the physical consistency constraint loss term in the total loss function. The physical consistency constraint loss term of the data-driven prediction model is the mean square error between the predicted value and the predicted value of the physical model. It is dimensionless, takes a value ≥ 0, and characterizes the degree of fit between the output of the data-driven prediction model and the physical mechanism. The number of samples for a single training iteration of the data-driven prediction model is a positive integer with no unit, determined by the dual-drive model construction module 31 based on the historical operating data of the wind-storage integrated system. The sequence number of the training sample is a positive integer with no unit, ranging from 1 to n, used to iterate through all samples in a single training iteration. Indicates the first The actual value of the control loop compensation capability corresponding to each training sample is dimensionless and ranges from 0 to 1. It is calculated from the control loop state synchronization data output by the distributed local data processing unit 2. Indicates the first The predicted value of the control loop compensation capability output by the data-driven prediction model corresponding to each training sample is dimensionless and ranges from 0 to 1. Indicates the first The predicted control loop compensation capability output by the physical model for each training sample is dimensionless, ranging from 0 to 1, and is derived from the predicted value of the physical model in S31.1. Obtained through conversion; This represents the sum of squared errors between the predicted and actual values ​​of the data-driven prediction model across all training samples, and is dimensionless. This represents the sum of squared errors between the data-driven prediction model's predictions and the physical model's predictions across all training samples, and is dimensionless.

[0040] Furthermore, The specific conversion method is as follows: Comprehensive prediction vectors from the 6×1 physics model In the process, the core control state variables are extracted: phase-locked loop phase tracking error. Converter modulation voltage deviation Energy storage SOC deviation ; The control loop compensation capability value is calculated using a weighted fusion method: ,in, , , The preset weighting coefficients are used to balance the impact of different control links on the overall compensation capability. Those skilled in the art can adjust them according to the control priority of the wind storage system. right After normalization, the values ​​are mapped to the [0,1] interval to obtain the final physical model output control loop compensation capability value. .

[0041] S31.4 Minimize the total loss function using gradient descent. Update the weights and biases of the data-driven prediction model, complete the fusion training of the physical model and the data-driven prediction model, and output to the weight dynamic adjustment module 32. The specific implementation is as follows: The dual-drive model construction module 31 uses the industrially common gradient descent method as the model optimization algorithm, with the total loss function... The minimum value is the optimization objective, and the weight matrix of the BP neural network constructed in S31.2 is used as the optimization objective. , ) and bias vector ( , The weights and biases are updated iteratively using the following general update formula: ; ; in: Indicates the first The updated weight matrix of the data-driven prediction model after the next iteration includes the weight matrix from the input layer to the hidden layer. Weight matrix from hidden layer to output layer Dimensionless, with dimensions consistent with the corresponding weight matrix; Indicates the first The weight matrix of the data-driven prediction model at the next iteration, including the weight matrix from the input layer to the hidden layer. Weight matrix from hidden layer to output layer Dimensionless, with dimensions consistent with the corresponding weight matrix; Indicates the number of model iterations, a non-negative integer, without units, and the initial value. (i.e., initial weights / biases); The learning rate of gradient descent is dimensionless and ranges from (0,1). It is used to control the step size of weight / bias updates in each iteration. Indicates the first In the next iteration, the total loss function For the weight matrix First-order partial derivatives, dimensionless, dimensionless Consistency, characterizing the total loss function With weight matrix The rate of change; Indicates the first The updated bias vector of the data-driven prediction model after the next iteration includes the bias vectors from the input layer to the hidden layer. Bias vector from hidden layer to output layer Dimensionless, with dimensions consistent with the corresponding bias vector; Indicates the first The bias vector of the data-driven prediction model at each iteration includes the bias vectors from the input layer to the hidden layer. Bias vector from hidden layer to output layer Dimensionless, with dimensions consistent with the corresponding bias vector; Indicates the first In the next iteration, the total loss function For bias vector First-order partial derivatives, dimensionless, dimensionless Consistency, characterizing the total loss function With bias vector The rate of change.

[0042] Furthermore, the dual-drive model construction module 31 sets the iteration termination condition to stop training when any of the following conditions are met: The number of iterations has reached the preset maximum number. ; The difference in total loss function between two adjacent iterations .

[0043] After training is terminated, the weights and biases of the data-driven prediction model are updated to optimal values. The physical model, as a mechanism constraint, is deeply integrated into the data-driven prediction model, forming a dual-drive fusion model that combines physical mechanism support and dynamic data adaptation. The dual-drive model construction module 31 outputs the fusion model completely to the weight dynamic adjustment module 32, providing a basis for the subsequent dynamic allocation of model weights.

[0044] In this embodiment, the weight dynamic adjustment module 32 dynamically adjusts the weight ratio of the physical model and the data-driven prediction model in the dual-drive model construction module 31 according to the real-time operating conditions; the weight ratio adjustment process of the weight dynamic adjustment module 32 includes the following steps: S32.1 Obtain real-time wind speed data collected by distributed synchronous acquisition unit 1. Physical model weights Weights of data-driven prediction models Satisfying constraints This provides a basis for subsequent weight adjustments, and the specific implementation is as follows: The weight dynamic adjustment module 32 acquires real-time wind speed data collected by the distributed synchronous acquisition unit 1. And clarify the weights of the physical model. Weights of data-driven prediction models The constraint relationship is calculated using the following formula: ; in: express The real-time wind speed data, in m / s (meters per second), is collected by distributed synchronous acquisition unit 1. Represents the weights of the physical model, dimensionless, with a value range of [value range missing]. ; This represents the weights of the data-driven prediction model; they are dimensionless and range from [value range missing]. .

[0045] S32.2 Setting a preset first threshold Preset second threshold ;when When configuring Minimum weight , The purpose of this step is to configure the physical model as the minimum weight and the data-driven prediction model as the corresponding weight under low wind speed conditions, adapting to scenarios where the operating conditions are stable under low wind speeds and the physical model can accurately describe them. Specifically, the weight dynamic adjustment module 32 sets a preset first threshold. Preset second threshold (in ); when real-time wind speed Configure physical model weights Minimum weight Data-driven prediction model weights The calculation formula is: ; in: This indicates the preset first wind speed threshold, in m / s (meters per second), used to distinguish between high wind speed and medium wind speed operating conditions. This indicates a preset second wind speed threshold, in m / s (meters per second), used to distinguish between medium and low wind speed operating conditions, and must meet the following conditions: ; This represents the minimum weight of the physical model. It is dimensionless and ranges from [0,1). It is preset by the system based on the reliability of the physical model under low wind speed conditions.

[0046] S32.3, when At that time, based on Linear adjustment and The purpose of this step is to linearly adjust the weights of the physical model and the data-driven prediction model based on real-time wind speed under medium wind speed conditions, so as to achieve a smooth transition of weights as the operating conditions change and avoid prediction fluctuations caused by abrupt changes in weights. The specific implementation is as follows: When the real-time wind speed meets At that time, the weight dynamic adjustment module 32 adjusts the weight based on the real-time wind speed. Linear adjustment of physical model weights The calculation formula is: ; Data-driven prediction model weights The calculation formula is: ; in: This represents the maximum weight of the physical model, is dimensionless, and ranges from (0,1]. It is preset by the system based on the reliability of the physical model under high wind speed conditions, and satisfies the following conditions: ; This represents the normalized proportion of real-time wind speed within the medium wind speed range. It is dimensionless and its value ranges from [0,1]. Represents the weights of the physical model, dimensionless, with a value range of [value range missing]. It changes linearly with real-time wind speed; This represents the weights of the data-driven prediction model; they are dimensionless and range from [value range missing]. , Linear change.

[0047] S32.4, when When configuring Maximum weight , The purpose of this step is to configure the physical model as the most weighted and the data-driven prediction model as the corresponding weight under high wind speed conditions, adapting to scenarios with complex operating conditions under high wind speeds where the physical model can still provide core mechanism constraints. The specific implementation is as follows: When real-time wind speed At that time, the weight dynamic adjustment module 32 configures the weights of the physical model. Maximum weight Data-driven prediction model weights The calculation formula is: ; in: This represents the maximum weight of the physical model; it is dimensionless and ranges from (0,1).

[0048] S32.5, will and The output is sent to the critical point prediction and trajectory generation module 33 to provide weighting criteria for subsequent fusion prediction and fault prediction. The specific implementation is as follows: The weight dynamic adjustment module 32 will adjust the weights of the physical model. Weights of data-driven prediction models The weight data is output to the critical point prediction and trajectory generation module 33 through the industrial communication protocol. The output weight data includes weight values, corresponding timestamps and working condition identifiers, ensuring that the critical point prediction and trajectory generation module 33 can accurately obtain the weight ratio under the corresponding working condition for the fusion prediction calculation of the dual-drive model.

[0049] In this embodiment, the critical point prediction and trajectory generation module 33, based on the output of the dual-drive model construction module 31 and the weight ratio of the weight dynamic adjustment module 32, monitors the output margin of the control loop and calculates the critical point of compensation capability depletion to generate a fault prediction trajectory. The fault prediction trajectory generation process of the critical point prediction and trajectory generation module 33 includes the following steps: S33.1 Calculate the saturation margin of the PI controller output. The phase deviation integral value of the phase-locked loop is collected to provide basic data for margin change rate calculation and critical point prediction. The specific implementation is as follows: The critical point prediction and trajectory generation module 33 calculates the saturation margin of the PI controller output based on the physical model prediction value output by the dual-drive model construction module 31 and the data-driven prediction model prediction value, combined with the weight ratio output by the weight dynamic adjustment module 32. The calculation formula is: ; Simultaneously, the phase deviation integral value of the phase-locked loop is collected. The calculation formula is: ; in: express The saturation margin of the PI controller output at any given time is dimensionless and ranges from [0, ... The smaller the margin, the closer the control loop's compensation capability is to exhaustion; This represents the maximum saturation threshold of the PI controller output. It is dimensionless and is determined by the output limiting parameter of the system's PI controller. express The real-time output value of the PI controller at any given time is dimensionless and is calculated from the fusion prediction results of the dual-drive model construction module 31. express The integral value of the phase deviation of the phase-locked loop at any given time, in units of ( This characterizes the cumulative effect of phase-locked loop phase tracking error; express The phase deviation of the phase-locked loop at any given time, in rad (radians), is defined as in S31.1; This indicates the phase deviation of the phase-locked loop from time 0 to... The integral value at time t, in units of ( ).

[0050] Furthermore, The specific fusion prediction formula is the weighted sum formula: ; in: for The weights of the physical model at any given time are output by the weight dynamic adjustment module 32. for Time-based data drives model weights; Output the control loop compensation capability value for the converted physical model; The predicted value of the control loop compensation capability output by the data-driven model (BP neural network); The final prediction result of the dual-drive fusion model is used to calculate the output value of the PI controller. ,Right now ,in This is the rated output value of the PI controller.

[0051] S33.2, Based on the fitting window Data, calculate margin change rate This provides a slope basis for calculating the critical point time, and the specific implementation is as follows: The critical point prediction and trajectory generation module 33 selects a length of... A sliding fit window, the window contains arrive Moment Group The data was linearly fitted using the least squares method within the window, and the slope of the fitted line is the margin of change. The calculation formula is: ; in: This represents the margin rate of change within the fitting window, dimensionless per unit time. This indicates that the margin decreases over time. This indicates that the margin increases over time; This represents the length of the sliding fitting window; it is a positive integer with no unit and is preset by the system based on the data sampling frequency and prediction accuracy requirements. Indicates the first [number]th [unit] within the fitting window The timestamps of each data point are in seconds (s); Indicates the first [number]th [unit] within the fitting window The saturation margin value corresponding to each data point is dimensionless and ranges from [value missing]. ; This represents the sum of the products of the timestamps within the window and their corresponding margin values, expressed in seconds (s). This represents the sum of all timestamps within the window, expressed in seconds (s). This represents the sum of all margin values ​​within the window, and is dimensionless. This represents the sum of squares of timestamps within the window, expressed in seconds (s²).

[0052] S33.3, based on With current saturation margin Calculate the critical time when the compensation capacity is exhausted. Get the remaining warning time This provides a time basis for tiered early warning, and is implemented as follows: The critical point prediction and trajectory generation module 33 uses the current time... saturation margin Starting from this point, assume the margin changes according to the fitted rate of change. The linear decline occurs when the margin reaches zero, which is the critical point where the compensation capacity is exhausted. The critical point time is... The calculation formula is: ; Remaining warning time The calculation formula is: ; like This indicates that the margin has not decreased or has increased. It has no practical significance and is classified as a risk with no critical point. when hour, There is a risk of reaching a critical point.

[0053] S33.4, according to Within the warning zone, a graded fault prediction trajectory is generated and output to the detection and evaluation output module 34 to achieve graded warning. The specific implementation is as follows: The critical point prediction and trajectory generation module 33 presets three-level early warning intervals: Level 1 Warning (High Risk): ; Level 2 Warning (Medium Risk): ; Level 3 Warning (Low Risk): ; in, The preset warning time threshold (in seconds).

[0054] Based on the rate of change of margin and current margin , generate from arrive The fault prediction trajectory, the trajectory equation is: ; according to The warning zone is marked, the warning level corresponding to the trajectory is marked, and the trajectory data (including timestamp, margin value, and warning level) is output to the detection and evaluation output module 34.

[0055] in: This represents the maximum remaining warning time threshold for Level 1 (high risk) warnings, expressed in seconds (s), and is preset by the system based on high-risk warning requirements. This indicates the maximum remaining warning time threshold for a Level 2 (medium risk) alert, expressed in seconds (s). This represents the maximum remaining warning time threshold for Level 3 (low risk) alerts, expressed in seconds (s). Indicating fault prediction trajectory The margin prediction value at time 1, dimensionless, with a range of values ​​of 1. ; This represents the time variable in the fault prediction trajectory, with units of seconds (s) and a value range of [value range missing]. ; Warning Level: Level 1 Warning (High Risk) Level II warning (medium risk) corresponds to Level 3 warning (low risk) corresponds to Risk-free response .

[0056] In this embodiment, the detection and evaluation output module 34, based on the output of the critical point prediction and trajectory generation module 33, outputs network performance health status data, fault prediction data, and hierarchical intervention basis to the distributed collaborative closed-loop control unit 4, providing accurate and executable evaluation results for the system's closed-loop control decision-making. The specific implementation is as follows: The data receiving and integration detection and evaluation output module 34 receives the fault prediction trajectory data, early warning level identifier and critical point time information output by the critical point prediction and trajectory generation module 33. At the same time, it associates the physical model prediction value and data-driven prediction model prediction value output by the dual-drive model construction module 31, as well as the weight ratio data output by the weight dynamic adjustment module 32, and integrates them to form a complete network performance evaluation dataset.

[0057] The network performance health status is determined based on the margin change trend and early warning level in the fault prediction trajectory. The detection and evaluation output module 34 determines the network performance health status. When the warning level is no risk, the person is considered to be in a healthy state. When the warning level is Level 3 (low risk), it is determined to be a sub-healthy state; When the warning level is Level II (medium risk), it is determined to be in a warning state; When the warning level is Level 1 (high risk), it is determined to be a critical fault state.

[0058] Furthermore, the tiered intervention criteria are generated based on health status and remaining warning time. Generate corresponding tiered intervention criteria: Health status: No intervention, maintain current operating strategy; Sub-health state: It is recommended to monitor changes in margin and regularly verify the model accuracy; Warning status: It is recommended to adjust the active / reactive power reference values ​​and optimize control parameters; Critical fault condition: It is recommended to immediately activate the emergency control strategy and switch to the backup control mode.

[0059] In addition, during the data standardization and encapsulation process, the output detection and evaluation output module 34 integrates the network performance health status data, fault prediction data (including trajectory data, critical point time, early warning level), and graded intervention basis, and encapsulates them in accordance with the IEC61850 industrial communication protocol. The encapsulated data frame contains fields such as node identifier, timestamp, health status code, fault prediction trajectory segment, and intervention suggestion code, and is output to the distributed collaborative closed-loop control unit 4 via industrial Ethernet to ensure that the control unit can directly parse and execute the intervention strategy.

[0060] In this embodiment, the model closed-loop optimization module 35 optimizes the physical model parameters and data-driven prediction model weights in the dual-drive model construction module 31 based on the actual response data of the wind-storage integrated system fed back by the distributed collaborative closed-loop control unit 4. The model parameter closed-loop optimization process of the model closed-loop optimization module 35 includes the following steps: S35.1 Calculate the predicted values ​​of the physical model Compared with the actual system response value root mean square error Verify the physical model parameters This provides a basis for subsequent parameter optimization, and the specific implementation is as follows: Model closed-loop optimization module 35 receives actual response data of the wind-storage integrated system from the distributed collaborative closed-loop control unit 4. Simultaneously, the physical model prediction values ​​output by the dual-drive model building module 31 are obtained. Calculate the root mean square error The calculation formula is: ; in: , represents the root mean square error between the physical model's predicted value and the system's actual response value. It is dimensionless and takes a value ≥ 0. The smaller the value, the higher the accuracy of the physical model's prediction. The number of validation samples is a positive integer with no unit, and is determined by the model closed-loop optimization module 35 based on the actual amount of feedback response data. This represents the verification sample number, a positive integer with no unit, and a value range of 1 to n. Indicates the first The physical model prediction value corresponding to each validation sample is dimensionless and defined as in S31.1. Indicates the first The actual system response value corresponding to each verification sample is dimensionless and is calculated from the actual operating data fed back by the distributed collaborative closed-loop control unit 4. This represents the physical model parameter vector, which includes core control parameters such as phase-locked loop and converter control loop. Its dimensions are consistent with the number of physical model parameters. This represents the sum of squared errors between the physical model predictions and the actual responses across all validation samples, and is dimensionless.

[0061] pass Value verification of physical model parameters (Including the phase-locked loop proportional coefficient) Integral coefficient Converter voltage loop proportional coefficient Integral coefficient Current loop proportionality coefficient The effectiveness of (etc.).

[0062] S35.2 Setting the error threshold ,when At that time, construct the parameter optimization objective function. This provides optimization directions for subsequent parameter updates, and the specific implementation is as follows: Model closed-loop optimization module 35 sets error threshold (Dimensionless, value ≥ 0), when the calculated value is... At that time, determine the current physical model parameters. If the accuracy requirement is not met, construct a parameter optimization objective function. The calculation formula is: ; in: The error threshold representing the prediction accuracy of the physical model is dimensionless and takes a value ≥ 0. It is preset by the system based on the model accuracy requirements. This represents the objective function for parameter optimization. It is dimensionless and takes a value ≥ 0. The smaller the value, the higher the prediction accuracy of the model after parameter optimization and the more stable the parameters. Indicates the first Each validation sample corresponds to a parameter-based The physical model predictions are dimensionless. This represents the regularization term weight coefficient, which is dimensionless and ranges from (0,1). It is used to balance prediction accuracy and parameter stability. This represents the parameter regularization term, used to prevent parameter overfitting. It is dimensionless; this embodiment uses L2 regularization, and the calculation formula is as follows: , representing the square norm of the parameter vector; in the formula Represents the parameter vector The L2 norm squared, which is dimensionless, is used to constrain the range of parameter values ​​and prevent overfitting.

[0063] S35.3 Update the physical model parameters using gradient descent. The weights and biases of the data-driven prediction model are updated synchronously to achieve collaborative optimization of the two models. The specific implementation is as follows: The model closed-loop optimization module 35 uses gradient descent as the parameter optimization algorithm, with the objective function... The minimum value is the optimization objective, and the physical model parameters are... Perform iterative updates, with the update formula as follows: ; Simultaneously, the weight matrix of the data-driven prediction model is updated. , With bias vector , The updated formula is: ; ; in: Indicates the first The updated physical model parameter vector after the next iteration has dimensions and Consistent; Indicates the first The physical model parameter vector at the next iteration has dimensions and Consistent; Indicates the number of iterations for the parameter, a non-negative integer, without units, and the initial value. ; This represents the learning rate of gradient descent, is dimensionless, and ranges from (0,1). It is used to control the step size of parameter updates. Describe the objective function For parameter vectors First-order partial derivatives, dimension and Consistency characterizes the rate at which the objective function changes with parameters; Indicates the first The updated weight matrix of the data-driven prediction model after the next iteration includes , Dimensions and Consistent; Indicates the first The weight matrix of the data-driven prediction model at the next iteration includes , Dimensions and Consistent; Indicates the first The data-driven prediction model bias vector updated in the next iteration includes , Dimensions and Consistent; Indicates the first The bias vector of the data-driven prediction model at the next iteration includes , Dimensions and Consistent; Describe the objective function For the weight matrix First-order partial derivatives, dimension and Consistent; Describe the objective function For bias vector First-order partial derivatives, dimension and Consistent.

[0064] S35.4. The optimized parameters are fed back to the dual-drive model construction module 31 to complete the collaborative closed-loop optimization of the physical model and the data-driven prediction model, thereby improving the prediction accuracy and adaptability of the model under all operating conditions. The specific implementation is as follows: Iteration termination determination: After each iteration, the model closed-loop optimization module 35 checks whether the preset iteration termination condition is met. Condition 1: The number of iterations reaches the preset maximum number of optimizations. ; Condition 2: The objective function for parameter optimization in two consecutive iterations The difference .

[0065] When any condition is met, the iteration stops, and the parameters obtained at this point are the optimal optimization parameters.

[0066] Specifically, the model closed-loop optimization module 35 extracts the optimal parameters at the end of the iteration, including the physical model parameter vector. Weight matrix of data-driven prediction model With bias vector These parameters are encapsulated in a preset binary stream format, and include a parameter version identifier, timestamp, and parameter verification code to ensure that the dual-drive model construction module 31 can directly parse and verify the validity and integrity of the parameters.

[0067] Specifically, the model closed-loop optimization module 35 feeds back the encapsulated optimization parameters to the dual-drive model building module 31 through the inter-module industrial Ethernet communication interface, and synchronously transmits parameter verification codes for the dual-drive model building module 31 to verify that the parameter transmission is distortion-free.

[0068] Furthermore, after receiving and verifying the optimized parameters, the dual-drive model construction module 31 performs a collaborative update: Using the optimized physical model parameter vector Update the core parameters of the physical sub-models for the phase-locked loop, converter control loop, and energy storage system response characteristics in S31.1, and reconstruct the predicted values ​​of the physical model. The computational logic; Using the optimized weight matrix With bias vector Update the weights and biases of the BP neural network data-driven prediction model in S31.2, and reconstruct the predicted values ​​of the data-driven prediction model. The computational logic; Based on the updated physical model and data-driven prediction model, the fusion training of S31.3 and S31.4 is re-executed to generate a new dual-drive fusion model, which is then output to the weight dynamic adjustment module 32 to complete the collaborative closed-loop optimization process.

[0069] in: Indicates the first The objective function value for parameter optimization in the next iteration is dimensionless. This represents the convergence threshold of the objective function for parameter optimization. It is dimensionless and ranges from (0,1), and is preset by the system according to the optimization accuracy requirements. This represents the optimized physical model parameter vector, with dimensions consistent with the number of physical model parameters, including core control parameters such as phase-locked loop and converter control loop; This represents the weight matrix from the input layer to the hidden layer of the optimized BP neural network, with dimension 1. ( This represents the number of neurons in the hidden layer. (where the input layer feature dimension is the dimension), which is dimensionless; This represents the weight matrix from the hidden layer to the output layer of the optimized BP neural network, with dimension 1. Dimensionless; This represents the optimized bias vector from the input layer to the hidden layer of the BP neural network, with dimension . Dimensionless; This represents the bias vector from the hidden layer to the output layer of the optimized BP neural network. It has a dimension of 1×1 and is dimensionless.

[0070] The distributed collaborative closed-loop control unit 4 receives grid-connected performance health status data, fault prediction data, and graded intervention basis output by the wind-storage grid-connected PHM assessment unit 3. It adopts a distributed collaborative control strategy to issue local control commands to each distributed node of the integrated wind-storage system, update system control parameters and detection benchmark parameters, and feed back the actual system response data to the wind-storage grid-connected PHM assessment unit 3.

[0071] In this embodiment, the distributed cooperative closed-loop control unit 4 includes a data receiving module 41, a cooperative control strategy execution module 42, a control command issuing module 43, a parameter updating module 44, and a response data feedback module 45, wherein: The data receiving module 41 receives the grid connection performance health status data, fault prediction data, and graded intervention basis output by the wind-storage grid connection PHM assessment unit 3. It completes data verification, anomaly handling, and standardized parsing, providing complete, distortion-free, and directly callable input data for subsequent control strategy execution and instruction generation, ensuring the accuracy of the control logic. The specific implementation is as follows: Communication link establishment: A stable point-to-point link is established with the detection and evaluation output module 34 via industrial Ethernet and the IEC61850-9-2 standard communication protocol. The communication baud rate is 100Mbps and the latency is ≤50ms, ensuring real-time data transmission.

[0072] Complete data reception: We receive three types of core data without omission: network performance health status data (four types of status identifiers and judgment criteria, corresponding to status codes: 00-healthy, 01-sub-healthy, 10-warning, 11-critical fault); and fault prediction data (fault prediction trajectory, critical point time, and remaining warning time). (1) Level 3 early warning level indicator; (2) Basis for tiered intervention (intervention suggestion code, parameter adjustment range, emergency response indicator).

[0073] Dual data verification: frame structure verification (verifies frame header 0xAA55, frame tail 0x55AA, node identifier CCU04 and data length; if invalid, discard and record); CRC32 integrity verification; if verification fails, retransmit (maximum 3 times); if retransmission fails, trigger an alarm (alarm code 0x0001) and log it.

[0074] Parsing and standardization: Parse the verified data by field, remove redundant fields, and convert it into binary data that the system can recognize; add millisecond-level timestamps, encapsulate it into an internal general data structure, and transmit it synchronously to the collaborative control strategy execution module 42, while caching it locally for 24 hours (automatically overwritten if the time limit is exceeded).

[0075] The collaborative control strategy execution module 42 adopts a distributed collaborative control strategy to generate local control commands, complete command collaboration verification, avoid conflicts, and ensure the command's relevance, collaboration, and executability. The specific implementation is as follows: Strategy matching: The control strategy is automatically matched based on the health status, as follows: Health status: Standard collaborative strategy, maintain stability, only generate data acquisition instructions; Sub-healthy state: Lightweight strategy, fine-tuning inverter and energy storage parameters, and strengthening monitoring frequency; Early warning status: Precise strategy, based on fault prediction trajectory to optimize power allocation and delay the depletion of compensation capacity; Critical fault state: Emergency strategy, switch to standby mode, limit the power of the faulty node, and prevent the fault from escalating.

[0076] Instruction generation: Generate custom instructions based on node type, as follows: Wind turbine converter: active / reactive power regulation, pitch angle regulation, phase-locked loop parameter fine-tuning (±10%); Energy storage converter: charging and discharging power regulation, SOC threshold adjustment (20%~80%), and terminal voltage stabilization (±5% of rated voltage); Grid connection point: Detection frequency adjustment (1~10Hz), harmonic monitoring start, abnormal alarm.

[0077] Collaborative verification: A consensus algorithm is used to verify the coordination of instructions, eliminate conflicts such as power allocation between wind turbines and energy storage, and after adjustment to the optimal value, the instructions are synchronously transmitted to the control instruction issuing module 43 and the parameter update module 44.

[0078] The control command distribution module 43 distributes local control commands to each distributed node, accurately and in real-time according to node priority, ensuring reliable and timely command transmission, recording the entire process, and achieving traceability of command execution. The specific implementation is as follows: Node matching and channel establishment: Extract node identifiers (wind turbines WT01~WTn, energy storage ESS01~ESSn, grid connection point PCC01) from the instructions and match the communication addresses; establish a one-to-one sending channel using the Modbus / TCP protocol, with a baud rate of 50Mbps and a latency of ≤30ms.

[0079] Priority delivery: Priorities are divided according to "critical fault nodes > core nodes > auxiliary nodes", and nodes of the same priority are delivered synchronously; the instruction carries an identifier, timestamp, and execution deadline (100ms for normal, 50ms for emergency), and receives node confirmation frames in real time and records the reception status.

[0080] Reliability assurance: If no acknowledgment is received within a timeout period, retransmission will be performed (interval of 20ms, maximum of 3 times). If retransmission fails, an alarm (0x0002) will be triggered, the log will be recorded, and the regeneration instruction will be sent to the collaborative control strategy execution module 42. The execution result of the receiving node will be synchronously fed back to the collaborative control strategy execution module 42.

[0081] Log recording: Records instruction identifier, issuance time, node, content, priority, reception / execution status, retransmission count, and alarm information, stored for 7 days, and supports retrieval and traceability.

[0082] The parameter update module 44 updates the control parameters and detection benchmark parameters, completing parameter verification, synchronization, and backup to ensure that the parameters match the control commands and achieve uniformity in system control and detection standards. The specific implementation is as follows: Parameter classification and extraction: Two types of parameters to be updated are extracted from the control commands. One type is the system control parameters issued to each distributed node, including the core regulation coefficients and operating thresholds of the phase-locked loop, converter, energy storage and wind turbine. The other type is the detection benchmark parameters synchronized to the wind-storage grid-connected PHM evaluation unit 3, including control margin threshold, early warning judgment threshold, data sampling benchmark and fault detection threshold.

[0083] Parameter update and rationality verification: Send parameter update instructions to the corresponding distributed nodes and evaluation units, and verify the updated parameters within the preset value range; if the verification is successful, confirm the update and record the parameter change log; if the verification fails, reject the update, trigger a parameter anomaly alarm and feed it back to the collaborative control strategy execution module 42 to regenerate the adjustment instructions.

[0084] Parameter synchronization and backup: Verify the consistency of parameters across the entire system to ensure that the parameters of each node and evaluation unit are consistent; perform versioned backup of the original parameters before the update to support quick rollback to the historical optimal parameters when the system malfunctions.

[0085] The response data feedback module 45 feeds back the actual response data to the wind-storage grid-connected PHM evaluation unit 3, providing real and reliable measured data support for the collaborative optimization of the dual-drive model. The specific implementation is as follows: Actual response data acquisition: Through industrial Ethernet and Modbus / TCP protocol, core operating data of wind turbine converters, energy storage converters and grid connection point metering nodes are acquired at an adaptive frequency of 1~10Hz, covering power, voltage, frequency, state of charge, phase deviation and node operating status code, with acquisition delay ≤50ms.

[0086] Data preprocessing: The raw collected data is processed sequentially to remove outliers, perform first-order low-pass filtering for noise reduction, and normalize the data to eliminate data noise and dimensional differences, ensuring that the data meets the requirements for optimized use of the model.

[0087] Data standardization encapsulation: Preprocessed data is encapsulated according to the IEC61850-9-2 protocol. The data frame includes a frame header, node identifier, millisecond-level timestamp, valid data segment and CRC32 check code, so as to achieve unified data format and verifiability.

[0088] Feedback transmission and reliability assurance: The encapsulated data is fed back to the model closed-loop optimization module using the TCP protocol. An acknowledgment-based transmission mechanism is used. If no acknowledgment frame is received within the timeout period, the data will be automatically retransmitted, with a maximum of 3 retransmissions. If the retransmission fails, the log will be logged and an alarm will be triggered to ensure the integrity and reliability of data transmission.

[0089] like Figure 2 As shown, this embodiment also provides a wind-storage integrated grid connection performance PHM detection method based on distributed control technology. The wind-storage integrated grid connection performance PHM detection system based on the above-mentioned distributed control technology includes the following steps: S1. Distributed acquisition terminals are deployed for the grid-connected nodes of the wind-storage integrated system. Distributed clock synchronization technology is used to realize the time synchronization of each distributed acquisition terminal, and the grid-connected electrical parameters, operating status parameters and internal status data of the control loop of each node are collected synchronously, and the time-synchronized raw acquisition data is output. S2. At each distributed acquisition node, data preprocessing, network performance feature extraction and data standardization are performed sequentially on the raw acquisition data synchronized with time, and standardized network performance feature data and control loop status synchronization data are output. S3. Receive standardized grid-connected performance characteristic data and control loop status synchronization data, construct a physical model and a data-driven prediction model for key control links of wind-storage integration, integrate physical consistency constraints into the training of the data-driven prediction model and complete dual-model fusion training, dynamically adjust the weight ratio of the physical model and the data-driven prediction model according to real-time operating conditions, monitor the output margin of the control loop and calculate the critical point of compensation capacity depletion, generate fault prediction trajectory, and output grid-connected performance health status data, fault prediction data and graded intervention basis; S4. Receive network performance health status data, fault prediction data and hierarchical intervention basis, adopt a distributed collaborative control strategy to issue local control commands to each distributed node of the wind-storage integrated system, update the control parameters and detection benchmark parameters of the wind-storage integrated system, and feed back the actual response data of the wind-storage integrated system. S5. Based on the actual response data of the wind-storage integrated system, the physical model parameters and data-driven prediction model weights are optimized in a closed loop. The optimized physical model parameters and data-driven prediction model weights are then reapplied to the construction of the physical model and data-driven prediction model in the key control links of the wind-storage integrated system, as well as the process of adjusting their weight ratio, forming a closed-loop operation of detection-control-optimization.

[0090] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0091] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A wind-storage integrated grid connection performance (PHM) testing system based on distributed control technology, characterized in that, include: Distributed synchronous acquisition unit (1) deploys distributed acquisition terminals for the grid-connected nodes of the wind-storage integrated system, adopts distributed clock synchronization technology, synchronously acquires the grid-connected electrical parameters, operating status parameters and internal status data of the control loop of each node, and outputs the time-synchronized raw acquisition data to the distributed local data processing unit (2). Distributed local data processing unit (2), which is deployed at each distributed acquisition node, performs data preprocessing, network performance feature extraction and data standardization processing on the original acquired data, and outputs standardized network performance feature data and control loop status synchronization data to wind-storage network PHM evaluation unit (3). The wind-storage grid-connected PHM assessment unit (3) receives standardized grid-connected performance characteristic data and control loop status synchronization data output by the distributed local data processing unit (2), constructs a physical model of key control links of wind-storage integration, constructs a data-driven prediction model and integrates physical consistency constraints into the training of the data-driven prediction model, dynamically adjusts the weight ratio of the physical model and the data-driven prediction model according to the real-time operating conditions, monitors the output margin of the control loop and calculates the critical point of compensation capacity depletion, generates fault prediction trajectory, outputs grid-connected performance health status data, fault prediction data and graded intervention basis to the distributed collaborative closed-loop control unit (4), and optimizes the physical model parameters and data-driven prediction model weights in the closed loop according to the actual response data of the wind-storage integrated system fed back by the distributed collaborative closed-loop control unit (4). The distributed collaborative closed-loop control unit (4) receives the grid-connected performance health status data, fault prediction data and graded intervention basis output by the wind-storage grid-connected PHM evaluation unit (3), adopts a distributed collaborative control strategy, issues local control commands to each distributed node of the wind-storage integrated system, updates the system control parameters and detection benchmark parameters, and feeds back the actual system response data to the wind-storage grid-connected PHM evaluation unit (3).

2. The wind-storage integrated grid connection performance PHM testing system based on distributed control technology according to claim 1, characterized in that, The distributed synchronous acquisition unit (1) includes a distributed acquisition terminal deployment module (11), a distributed clock synchronization module (12), a multi-parameter synchronous acquisition module (13), and a raw data output module (14), wherein: The distributed acquisition terminal deployment module (11) deploys distributed acquisition terminals for the network nodes of the integrated wind and storage system. The distributed clock synchronization module (12) adopts distributed clock synchronization technology to realize time synchronization of each distributed acquisition terminal; The multi-parameter synchronous acquisition module (13) synchronously acquires the network electrical parameters, operating status parameters and internal status data of each node based on the time synchronization status of each distributed acquisition terminal. The raw data output module (14) outputs the time-synchronized raw data obtained by synchronous acquisition to the distributed local data processing unit (2).

3. The wind-storage integrated grid connection performance PHM testing system based on distributed control technology according to claim 2, characterized in that, The distributed local data processing unit (2) includes a data preprocessing module (21), a network performance feature extraction module (22), a data standardization processing module (23), and a processed data output module (24), wherein: The data preprocessing module (21) is deployed at each distributed acquisition node and performs data preprocessing operations on the raw acquisition data output by the distributed synchronous acquisition unit (1). The network performance feature extraction module (22) performs network performance feature extraction operation based on the raw collected data processed by the data preprocessing module (21); The data standardization processing module (23) performs data standardization processing operations based on the network performance characteristics extracted by the network performance characteristic extraction module (22); The processed data output module (24) outputs the standardized grid-connected performance characteristic data and control loop status synchronization data obtained by the data standardization processing module (23) to the wind-storage grid-connected PHM evaluation unit (3).

4. The wind-storage integrated grid connection performance PHM testing system based on distributed control technology according to claim 3, characterized in that, The wind-storage grid-connected PHM evaluation unit (3) includes a dual-drive model construction module (31), a weight dynamic adjustment module (32), a critical point prediction and trajectory generation module (33), a detection and evaluation output module (34), and a model closed-loop optimization module (35), wherein: The dual-drive model construction module (31) receives the standardized network performance characteristic data and control loop status synchronization data output by the distributed local data processing unit (2), constructs a physical model of the key control link of wind-storage integration, constructs a data-driven prediction model, and integrates physical consistency constraints into the training of the data-driven prediction model. The weight dynamic adjustment module (32) dynamically adjusts the weight ratio of the physical model and the data-driven prediction model in the dual-drive model construction module (31) according to the real-time operating conditions. The critical point prediction and trajectory generation module (33) monitors the output margin of the control loop and calculates the critical point of compensation capability depletion based on the output of the dual-drive model construction module (31) and the weight ratio of the weight dynamic adjustment module (32), and generates a fault prediction trajectory. The detection and evaluation output module (34) outputs network performance health status data, fault prediction data and graded intervention basis to the distributed collaborative closed-loop control unit (4) based on the output of the critical point prediction and trajectory generation module (33). The model closed-loop optimization module (35) optimizes the physical model parameters and data-driven prediction model weights in the closed-loop optimization dual-drive model construction module (31) based on the actual response data of the wind-storage integrated system fed back by the distributed collaborative closed-loop control unit (4).

5. The wind-storage integrated grid connection performance PHM testing system based on distributed control technology according to claim 4, characterized in that, The physical model and data-driven prediction model construction and training process of the dual-drive model construction module (31) includes the following steps: S31.1 Based on the design parameters and operating mechanism of the integrated wind and energy storage system, construct a physical model of the phase-locked loop, converter control loop, and energy storage system response characteristics, and output the predicted values ​​of the physical model. ; S31.2 Construct a data-driven prediction model based on a BP neural network, using standardized network performance characteristic data as input, and outputting predicted values ​​related to the control loop compensation capability. ; S31.3 Integrate the dynamic equations of the physical model as constraints into the training of the data-driven prediction model, and construct a total loss function with physical consistency constraints. ; S31.4 Minimize the total loss function using gradient descent. Update the weights and biases of the data-driven prediction model, complete the fusion training of the physical model and the data-driven prediction model, and output to the weight dynamic adjustment module (32).

6. The wind-storage integrated grid connection performance PHM testing system based on distributed control technology according to claim 5, characterized in that, The weight ratio adjustment process of the weight dynamic adjustment module (32) includes the following steps: S32.1 Obtain real-time wind speed data collected by the distributed synchronous acquisition unit (1) Physical model weights Weights of data-driven prediction models Satisfying constraints ; S32.2 Setting a preset first threshold Preset second threshold ;when When configuring Minimum weight , ; S32.3, when At that time, based on Linear adjustment and ; S32.4, when When configuring Maximum weight , ; S32.5, will and Output to the critical point prediction and trajectory generation module (33).

7. The wind-storage integrated grid connection performance PHM testing system based on distributed control technology according to claim 6, characterized in that, The fault prediction trajectory generation process of the critical point prediction and trajectory generation module (33) includes the following steps: S33.1 Calculate the saturation margin of the PI controller output. Collect the integral value of the phase deviation of the phase-locked loop; S33.2, Based on the fitting window Data, calculate margin change rate ; S33.3, based on With current saturation margin Calculate the critical point time when the compensation capacity is exhausted. Receive the remaining warning time ; S33.4, according to The warning interval is generated, and the graded fault prediction trajectory is output to the detection and evaluation output module (34).

8. The wind-storage integrated grid connection performance PHM testing system based on distributed control technology according to claim 7, characterized in that, The model parameter closed-loop optimization process of the model closed-loop optimization module (35) includes the following steps: S35.1 Calculate the predicted values ​​of the physical model Compared with the actual system response value root mean square error Verify the physical model parameters ; S35.2 Setting the error threshold ,when At that time, construct the parameter optimization objective function. ; S35.3 Update the physical model parameters using gradient descent. The weights and biases of the data-driven prediction model are updated synchronously. S35.

4. Feed the optimized parameters back to the dual-drive model construction module (31) to complete the collaborative closed-loop optimization of the physical model and the data-driven prediction model.

9. The wind-storage integrated grid connection performance PHM testing system based on distributed control technology according to claim 8, characterized in that, The distributed collaborative closed-loop control unit (4) includes a data receiving module (41), a collaborative control strategy execution module (42), a control command issuing module (43), a parameter updating module (44), and a response data feedback module (45), wherein: The data receiving module (41) receives the grid-connected performance health status data, fault prediction data and graded intervention basis output by the wind-storage grid-connected PHM assessment unit (3); The collaborative control strategy execution module (42) adopts a distributed collaborative control strategy to generate local control commands; The control command issuing module (43) issues local control commands to each distributed node; The parameter update module (44) updates the control parameters and detection benchmark parameters; The response data feedback module (45) feeds back the actual response data to the wind-storage grid-connected PHM evaluation unit (3).

10. A method for detecting the grid connection performance (PHM) of an integrated wind and energy storage system based on distributed control technology, wherein the method is based on the integrated wind and energy storage grid connection performance (PHM) system based on distributed control technology as described in any one of claims 1-9, characterized in that... Includes the following steps: S1. Distributed acquisition terminals are deployed for the grid-connected nodes of the wind-storage integrated system. Distributed clock synchronization technology is used to realize the time synchronization of each distributed acquisition terminal, and the grid-connected electrical parameters, operating status parameters and internal status data of the control loop of each node are collected synchronously, and the time-synchronized raw acquisition data is output. S2. At each distributed acquisition node, data preprocessing, network performance feature extraction and data standardization are performed sequentially on the raw acquisition data synchronized with time, and standardized network performance feature data and control loop status synchronization data are output. S3. Receive standardized grid-connected performance characteristic data and control loop status synchronization data, construct a physical model and a data-driven prediction model for key control links of wind-storage integration, integrate physical consistency constraints into the training of the data-driven prediction model and complete dual-model fusion training, dynamically adjust the weight ratio of the physical model and the data-driven prediction model according to real-time operating conditions, monitor the output margin of the control loop and calculate the critical point of compensation capacity depletion, generate fault prediction trajectory, and output grid-connected performance health status data, fault prediction data and graded intervention basis. S4. Receive network performance health status data, fault prediction data and hierarchical intervention basis, adopt a distributed collaborative control strategy to issue local control commands to each distributed node of the wind-storage integrated system, update the control parameters and detection benchmark parameters of the wind-storage integrated system, and feed back the actual response data of the wind-storage integrated system. S5. Based on the actual response data of the wind-storage integrated system, the physical model parameters and data-driven prediction model weights are optimized in a closed loop. The optimized physical model parameters and data-driven prediction model weights are then reapplied to the construction of the physical model and data-driven prediction model in the key control links of the wind-storage integrated system, as well as the process of adjusting their weight ratio, forming a closed-loop operation of detection-control-optimization.

Images