SVG valve hall cooling efficiency evaluation method and system based on probabilistic graph model

Abstract

The invention discloses an SVG valve hall cooling efficiency evaluation method and system based on a probabilistic graph model, and the method comprises the steps: obtaining original time sequence data which is obtained through the collection of a cooling system multi-parameter monitoring sensor group disposed in an SVG valve hall in continuous T sampling periods; preprocessing the original time series data to obtain a credible time series data set; constructing a Bayesian network topological structure comprising three-level nodes of an environment layer, a component layer and an efficiency layer and causal dependence edges, and optimizing parameters of the Bayesian network topological structure by adopting a maximum likelihood estimation method to form a dynamic Bayesian network model after parameter calibration; and the credible time sequence data set is used as an evidence variable to be input into the Bayesian network model after parameter calibration, calculation is carried out through a belief propagation reasoning algorithm, a final control instruction set is generated through probability weighted scoring processing, the final control instruction set is fed back to a valve group monitoring system, and early warning and automatic load reduction are achieved. The problems of large evaluation deviation and early warning lag in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of power system monitoring and evaluation, specifically to a method and system for evaluating the cooling efficiency of SVG valve halls based on a probabilistic graphical model. Background Technology

[0002] Currently, in the field of SVG valve hall operation monitoring in high-voltage direct current transmission systems, cooling efficiency assessment is a crucial link in ensuring the safe and stable operation of equipment. As SVG valve hall capacity develops towards higher power, the heat generation of IGBT modules increases significantly. Cooling systems need to cope with the coupled effects of multiple factors such as environmental fluctuations, component aging, and load changes, continuously increasing the requirements for the real-time and accurate efficiency assessment. However, existing cooling efficiency assessment technologies still mainly rely on traditional steady-state thermal resistance models and single-threshold alarms: steady-state thermal resistance models are based on fixed thermal conductivity coefficients and can only reflect heat dissipation capacity under static operating conditions, failing to capture dynamic parameter fluctuations in cooling systems caused by diurnal temperature differences and airborne dust in scenarios such as suburban industrial parks; single-threshold alarms achieve risk warnings by setting fixed thresholds for parameters such as IGBT junction temperature and fan speed (e.g., alarming when IGBT junction temperature exceeds 85℃), without considering the nonlinear correlation between parameters (e.g., the coupling effect of micro-blockage in pipes and decrease in fan speed), resulting in a 15%-20% deviation in the assessment results.

[0003] To compensate for the above deficiencies, some technologies attempt to introduce simple linear regression models to correct the thermal resistance coefficient, or increase the number of sensors to increase the data acquisition density. However, linear regression models cannot handle the complex causal relationships between parameters, and additional sensors can only increase the amount of data but do not solve the problem of data "fragmentation". They still cannot achieve accurate quantitative assessment of cooling efficiency. Especially in hidden fault scenarios such as "dust accumulation causing micro-blockage in pipes" and "fan bearing wear causing speed fluctuations", existing technologies often take more than 72 hours to detect the risk of a sudden drop in efficiency, which is difficult to meet the early warning requirements of SVG valve halls.

[0004] In view of the above, this application is hereby submitted. Summary of the Invention

[0005] The technical problem this invention aims to solve is that existing cooling efficiency assessment methods cannot achieve accurate quantitative assessment of cooling efficiency and suffer from delayed early warning, making it difficult to meet the early warning requirements of SVG valve halls. The purpose of this invention is to provide a cooling efficiency assessment method and system for SVG valve halls based on a probabilistic graphical model, providing accurate data support for the operation and maintenance decisions of SVG valve halls and effectively addressing the core pain points of existing technologies, such as large assessment bias and delayed early warning.

[0006] This invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention provides a method for evaluating the cooling efficiency of an SVG valve hall based on a probabilistic graphical model, the method comprising:

[0008] The raw time-series data was acquired by a multi-parameter monitoring sensor group of the cooling system arranged in the SVG valve hall over a continuous T sampling period.

[0009] The original time-series data is preprocessed to obtain a reliable time-series dataset; the preprocessing includes outlier removal, dust interference drift compensation, and standardization.

[0010] A Bayesian network topology structure containing three levels of nodes (environment layer, component layer, and efficiency layer) and causal dependency edges is constructed. The maximum likelihood estimation method is used to optimize the parameters of the Bayesian network topology structure, forming a parameter-calibrated dynamic Bayesian network model.

[0011] The Bayesian network model, calibrated by using a credible time-series dataset as input parameters for evidence variables, is calculated using a belief propagation inference algorithm and then processed with probability-weighted scoring to generate the final control command set. This final control command set is then fed back to the valve group monitoring system to achieve early warning and automatic load reduction.

[0012] Furthermore, the raw time series data includes raw temperature time series data, raw time series data of micro-differential pressure leakage, raw time series data of corrosion status, and raw time series data of fan current.

[0013] Further, the raw time-series data is obtained, including:

[0014] Within the SVG valve hall, in the IGBT module area, cooling pipe loop, and fan compartment, a multi-parameter monitoring sensor group for the cooling system is deployed. This multi-parameter monitoring sensor group includes multiple fluorescent fiber optic temperature sensors, high-frequency micro-differential pressure leakage sensors, electrochemical corrosion probes, and fan current Hall sensors.

[0015] The IGBT junction temperature and the inlet ambient temperature are collected by a fluorescent fiber optic temperature sensor to form raw time-series temperature data.

[0016] The pressure difference between the inlet and outlet of the cooling pipe is collected by a high-frequency micro differential pressure leakage sensor to form raw time-series data of micro differential pressure leakage.

[0017] The corrosion current density of the inner wall of the pipeline is collected by an electrochemical corrosion probe to form the original time series data of the corrosion state.

[0018] The fan operating current is collected by a fan current Hall sensor, and the fan speed is calculated by combining the current and speed conversion coefficient specified in the fan nameplate parameters, thus forming the original time series data of the fan current.

[0019] The original time-series data of temperature, micro-differential pressure leakage, corrosion status, and fan current were aligned according to the sampling time to construct an original data matrix, which was then used as the original time-series data.

[0020] Furthermore, the original time-series data is preprocessed to obtain a reliable time-series dataset, including:

[0021] An outlier removal method based on sliding window entropy rate is used to eliminate interference fluctuations in the original time series data, resulting in the first data matrix;

[0022] The first data matrix is ​​subjected to dust interference drift compensation processing to obtain the compensated parameter values; the outliers are removed from the compensated parameter values ​​by combining the Laida criterion and the missing values ​​are filled in based on linear interpolation to obtain the second data matrix;

[0023] The Z-score standardization method is used to standardize the parameters in the second data matrix to obtain a standardized data matrix, which is a reliable time series dataset.

[0024] Furthermore, the steps for obtaining the first data matrix and the second data matrix are as follows:

[0025] Calculate the Shannon entropy rate of each column parameter in the original time series data within each sliding window;

[0026] If the Shannon entropy rate of the parameter at the sampling time is lower than the preset threshold, the parameter value at that time is determined to be an outlier, and the outlier value is replaced by the median of the parameter in the corresponding sliding window to obtain the first data matrix.

[0027] Based on the first data matrix, compensation calculations were performed on three parameters—air inlet ambient temperature, inlet and outlet pressure difference of cooling pipe, and corrosion current density of pipe inner wall—based on the compensation factor, and the compensated parameters were obtained.

[0028] Based on the Laida criterion, the mean and standard deviation of each parameter after compensation are calculated. Outliers with a difference between the value and the mean greater than a preset multiple of the standard deviation are removed, and missing values ​​are filled in using linear interpolation to obtain the second data matrix.

[0029] Furthermore, a Bayesian network topology is constructed, comprising three levels of nodes (environment layer, component layer, and efficiency layer) and causal dependency edges. The maximum likelihood estimation method is then used to optimize the Bayesian network topology parameters, resulting in a parameter-calibrated dynamic Bayesian network model, including:

[0030] Based on a reliable time-series dataset and combined with the correlation patterns between cooling system failure modes and parameters in historical fault data, a Bayesian network topology containing three levels of nodes (environment layer, component layer, and efficiency layer) and causal dependency edges is constructed through training using a structure learning algorithm.

[0031] A set of latent variables associated with nodes in the SVG valve hall cooling system is introduced. Based on expert knowledge, the conditional probability relationship between the latent variables and each observed node is set, an initial node conditional probability table is constructed, the probability value of each node under different combinations of parent node states is determined, and the node conditional probability table is initialized.

[0032] Using a reliable time series dataset as observation data, the parameters in the initial node conditional probability table are optimized using the maximum likelihood estimation method.

[0033] Based on the initial Bayesian network topology and combined with the quantification patterns in historical data, the conditional probability table of the cooling efficiency node is constrained and optimized to obtain the dynamic Bayesian network model after parameter calibration.

[0034] Furthermore, the calibrated Bayesian network model, using a trusted time-series dataset as input parameters for evidence variables, is computed through a belief propagation inference algorithm and processed with probability-weighted scoring to generate the final control command set. This final control command set is then fed back to the valve group monitoring system to achieve early warning and automatic load reduction, including:

[0035] The Bayesian network model, calibrated by using a reliable time-series dataset as input parameters for evidence variables, is computed using a belief propagation inference algorithm. Simultaneously, based on the fusion iteration of forward and backward probabilistic inference, the failure probabilities of each component in the cooling system, the predicted IGBT junction temperature, and the cooling efficiency for the next sampling cycle are obtained. posterior probability distribution ;

[0036] The failure probability of each component in the comprehensive cooling system, the predicted value of the IGBT junction temperature, and the cooling efficiency in the next sampling cycle are all considered. posterior probability distribution Based on the cooling efficiency evaluation threshold and early warning threshold, and after probability weighted scoring, the SVG valve hall cooling efficiency is classified into levels and the key influencing factors are output.

[0037] If the predicted cooling efficiency is lower than the warning threshold, the closed-loop control process of SVG reactive power command reduction is triggered, the final control command set is generated and fed back to the valve group monitoring system to realize early warning and automatic load reduction.

[0038] Among them, forward probability inference is based on reliable time series data at time t and before, combined with the state transition probability of dynamic Bayesian network, to calculate the prior probability distribution of each node at time t+1, so as to predict the trend of cooling efficiency change; backward probability inference is based on real-time reliable data at time t+1, and reversely corrects the probability distribution at time t and before.

[0039] Secondly, the present invention provides an SVG valve hall cooling efficiency evaluation system based on a probabilistic graphical model, the system comprising:

[0040] The acquisition unit is used to acquire raw time-series data, which is collected by a multi-parameter monitoring sensor group of the cooling system arranged in the SVG valve hall over a continuous T sampling period.

[0041] The preprocessing unit is used to preprocess the raw time-series data to obtain a reliable time-series dataset; the preprocessing includes outlier removal, dust interference drift compensation, and standardization.

[0042] The model building unit is used to construct a Bayesian network topology that includes three levels of nodes: environment layer, component layer, and efficiency layer, as well as causal dependency edges. The maximum likelihood estimation method is used to optimize the parameters of the Bayesian network topology to form a parameter-calibrated dynamic Bayesian network model.

[0043] The calculation and evaluation unit is used to take the credible time series dataset as the input parameter of the calibrated Bayesian network model as evidence variable, perform calculations through the belief propagation inference algorithm, and generate the final control command set after probability weighted scoring processing; and feed the final control command set back to the valve group monitoring system to realize early warning and automatic load reduction.

[0044] Thirdly, the present invention also provides an SVG valve hall cooling efficiency evaluation device based on a probabilistic graphical model, comprising:

[0045] Memory, used to store computer programs;

[0046] A processor for executing a computer program to implement the SVG valve hall cooling efficiency evaluation method based on a probabilistic graphical model as described in any one of claims 1 to 7.

[0047] Fourthly, the present invention also provides a readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for evaluating the cooling efficiency of an SVG valve hall based on a probabilistic graphical model.

[0048] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0049] This invention presents a method and system for evaluating the cooling efficiency of SVG valve halls based on a probabilistic graphical model. The invention includes acquiring raw time-series data from a multi-parameter monitoring sensor group deployed within the SVG valve hall over T consecutive sampling periods; eliminating interference fluctuations using an outlier removal method based on sliding window entropy rate; simultaneously performing dust interference drift compensation processing to obtain a reliable time-series dataset; constructing a Bayesian network topology containing three levels of nodes (environment layer, component layer, and efficiency layer) and causal dependency edges; forming a parameter-calibrated dynamic Bayesian network model; performing calculations using a belief propagation inference algorithm; and generating a final control command set through probability-weighted scoring processing, which is then fed back to the valve group monitoring system to achieve early warning and automatic load reduction. This invention achieves integrated "evaluation-early warning-control," significantly reducing the risk of downtime due to cooling failure; it provides accurate data support for the operation and maintenance decisions of SVG valve halls, effectively addressing the core pain points of existing technologies such as large evaluation bias and delayed early warning. Attached Figure Description

[0050] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a flowchart of the SVG valve hall cooling efficiency evaluation method based on a probabilistic graphical model according to the present invention;

[0052] Figure 2 This is a block diagram of the SVG valve hall cooling efficiency evaluation system based on a probabilistic graphical model according to the present invention.

[0053] Figure 3 This is a schematic diagram of the SVG valve hall cooling efficiency evaluation device based on a probabilistic graphical model according to the present invention.

[0054] In the diagram: 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0056] This invention designs a method and system for evaluating the cooling efficiency of SVG valve halls based on a probabilistic graphical model. It collects core parameters such as IGBT junction temperature, pipe pressure difference, corrosion current density, and fan current through a multi-parameter monitoring sensor group of the cooling system. After sliding window entropy rate anomaly removal, dust interference drift compensation, and Z-score normalization, a high-quality, reliable time-series dataset is constructed. Subsequently, based on this dataset and combined with the "parameter-efficiency" quantification pattern in historical fault data (e.g., 30μm dust causing a 40% decrease in heat dissipation efficiency), a PC structure learning algorithm is used to construct an "environment layer-component layer-efficiency layer" model. The system employs a three-level Bayesian network topology and optimizes the node conditional probability table using maximum likelihood estimation. It introduces latent variables such as diurnal temperature difference and micro-blockage in pipelines to capture hidden influencing factors. Finally, it utilizes forward-backward probability inference of the dynamic Bayesian network to achieve real-time calculation of the failure probability of cooling system components and the posterior probability distribution of cooling efficiency. Based on the 5% quantile threshold, it triggers closed-loop control, which can provide early warning of cooling efficiency anomalies up to 72 hours in advance. At the same time, it outputs key influencing factors, providing accurate data support for the operation and maintenance decisions of SVG valve halls, effectively solving the core pain points of large evaluation bias and delayed early warning in existing technologies.

[0057] Specifically, by arranging a multi-parameter monitoring sensor group for the cooling system in the core area of ​​the SVG valve hall, raw data from all dimensions such as IGBT junction temperature, pipeline pressure difference, corrosion current density, and fan current are collected. The data is then processed by sliding window entropy rate anomaly removal, dust interference drift compensation, and Z-score standardization to construct a reliable time-series dataset. Compared to existing technologies that only collect a single temperature parameter, this method improves data coverage and provides high-quality input for subsequent evaluations, avoiding evaluation biases caused by incomplete or distorted data. Based on a reliable time-series dataset and combined with historical failure quantification patterns, a three-level topology of "environment layer - component layer - efficiency layer" is constructed using a PC structure learning algorithm. The node conditional probability table is optimized using maximum likelihood estimation, and latent variables such as diurnal temperature difference and pipe micro-blockage are introduced to address the shortcomings of existing steady-state thermal resistance models in capturing multi-factor coupling. This reduces the deviation between the Bayesian network's probabilistic description of the cooling system state and the actual correlation, while the dynamic characteristics adapt to the slow efficiency degradation pattern, avoiding the "instantaneous jump" misjudgment of static models. Reliable time-series data is used as evidence variables input into the calibrated dynamic Bayesian network. Through the fusion of forward inference (predicting efficiency trends) and backward inference (correcting historical probabilities), the component failure probability and the posterior distribution of cooling efficiency are calculated, providing accurate decision-making basis for operation and maintenance, avoiding blind repairs. This achieves integrated "evaluation-early warning-control," automatically reducing load before a sudden drop in efficiency, allowing the SVG to... The risk of shutdown due to cooling failure in the valve hall is reduced, ensuring the stable operation of the high-voltage direct current transmission system.

[0058] Example 1

[0059] like Figure 1 As shown, the present invention provides a method for evaluating the cooling efficiency of an SVG valve hall based on a probabilistic graphical model. This method includes:

[0060] Step 1: Obtain raw time series data. The raw time series data is collected by the multi-parameter monitoring sensor group of the cooling system arranged in the SVG valve hall within T consecutive sampling periods. The raw time series data includes raw temperature time series data, raw time series data of micro-differential pressure leakage, raw time series data of corrosion status, and raw time series data of fan current.

[0061] In this embodiment, obtaining the original time-series data includes:

[0062] Step 11: In the three core monitoring areas of the SVG valve hall—the IGBT module area, the cooling pipe loop, and the fan nacelle—arrange a multi-parameter monitoring sensor group for the cooling system. The multi-parameter monitoring sensor group for the cooling system includes multiple fluorescent fiber optic temperature sensors, high-frequency micro-differential pressure leakage sensors, electrochemical corrosion probes, and fan current Hall sensors.

[0063] The sampling period of each parameter monitoring sensor was set to 10 seconds, and 7200 sampling periods were continuously collected to cover a continuous monitoring duration of 12 hours, thus determining the time range and spatial coverage of the original time series data collection.

[0064] Step 12: Collect IGBT junction temperature and inlet ambient temperature using a fluorescent fiber optic temperature sensor to generate raw temperature time-series data; collect pressure difference between the inlet and outlet of the cooling pipe using a high-frequency micro-differential pressure leakage sensor to generate raw micro-differential pressure leakage time-series data; collect corrosion current density on the inner wall of the pipe using an electrochemical corrosion probe to generate raw corrosion state time-series data; collect fan operating current using a fan current Hall sensor and calculate the fan speed using the current-speed conversion factor specified in the fan nameplate parameters to generate raw fan current time-series data; thus completing the classification and collection of four types of core parameters.

[0065] Step 13: Align the original time-series data of temperature, micro-differential pressure leakage, corrosion status, and fan current according to the sampling time, construct an original data matrix, and use it as the original time-series data. This original data matrix has 7200 rows and 6 columns, corresponding to IGBT junction temperature, inlet ambient temperature, cooling pipe inlet and outlet pressure difference, pipe inner wall corrosion current density, fan operating current, and fan speed, respectively. Each element in the matrix represents the monitored value of the corresponding parameter at the corresponding sampling time, thus obtaining a structured original data matrix.

[0066] It should be noted that a multi-parameter monitoring group, including fluorescent fiber optic temperature sensors, high-frequency micro-differential pressure leakage sensors, electrochemical corrosion probes, and fan current Hall sensors, was deployed in the three core areas of the SVG valve hall: the IGBT module area, the cooling pipe loop, and the fan nacelle. A 10-second sampling period was set, and 7200 consecutive cycles (covering 12 hours) were collected. This clearly defined the spatiotemporal range of data collection, accurately covered the key state areas of the cooling system, and adapted to the SVG load fluctuation cycle of "high during the day and low at night." This not only avoided information loss due to area omissions but also balanced "dynamic capture" and "data volume control" through the 10-second sampling period, providing comprehensive and efficient basic data for subsequent analysis.

[0067] In step S12, a fluorescent fiber optic temperature sensor collects the IGBT junction temperature and the inlet air temperature to form raw time-series temperature data; a high-frequency micro-differential pressure sensor collects the pressure difference between the pipe inlet and outlet to obtain raw data on micro-differential pressure leakage; an electrochemical corrosion probe collects the corrosion current density on the inner wall of the pipe to form raw data on corrosion status; and a fan current Hall sensor collects the fan operating current, and calculates the fan speed using the current-speed conversion coefficient specified on the fan nameplate to form raw fan current data. The system collects three core dimensions of parameters for the cooling system: "thermal status," "loop health," and "component performance." Furthermore, the conversion from "electrical parameters" to "heat dissipation capacity parameters" is achieved through current-speed conversion, avoiding the limitations of single parameters and providing multi-dimensional data support for subsequent evaluation of cooling efficiency.

[0068] In step S13, the four types of raw time-series data are aligned according to the sampling time to construct a raw data matrix with 7200 rows (number of sampling periods) and 6 columns (corresponding to IGBT junction temperature, inlet temperature, pipeline pressure difference, corrosion current density, fan current, and fan speed). The matrix elements are the parameter monitoring values ​​at the corresponding time. Integrating the scattered time-series data into a structured raw data matrix achieves unified data management, avoids timestamp misalignment problems in subsequent processing, and provides standardized input for preprocessing steps such as outlier removal and standardization, thereby improving the efficiency of subsequent processes.

[0069] Step 2: Preprocess the original time series data to obtain a reliable time series dataset; preprocessing includes outlier removal, dust interference drift compensation, and standardization.

[0070] In this embodiment, step 2 specifically includes:

[0071] Step 21: An outlier removal method based on sliding window entropy rate is used to eliminate interference fluctuations in the original time series data, resulting in the first data matrix; the specific steps are as follows:

[0072] Define a sliding window length of 50 to cover a 500-second data window. Calculate the Shannon entropy rate of each column parameter in the original data matrix (i.e., the original time series data) within each sliding window. The formula for calculating the Shannon entropy rate is:

[0073]

[0074] In the formula, Let L be the Shannon entropy rate of the monitored parameters in column j within the w-th sliding window, and L be the length of the sliding window. Let be the probability density of the parameters in the j-th column within the w-th sliding window.

[0075] The probability density is calculated using the kernel density estimation method, and the Shannon entropy rate threshold is set to 0.8 times the average Shannon entropy rate of all sliding windows. If the Shannon entropy rate of the parameter at the sampling time is lower than the threshold, the parameter value at that time is determined to be an outlier, and the outlier value is replaced with the median of the parameter in the corresponding sliding window to obtain the first data matrix after preliminary processing.

[0076] Step 22: Perform dust interference drift compensation processing on the first data matrix to obtain the compensated parameter values; use the Laida criterion to remove outliers from the compensated parameter values ​​and fill in missing values ​​based on linear interpolation to obtain the second data matrix; specifically including:

[0077] A dust interference compensation factor is introduced to address parameters such as the ambient temperature at the air inlet, the pressure difference between the inlet and outlet of the cooling pipes, and the corrosion current density on the inner wall of the pipes. The calculation formula for the compensation factor is as follows:

[0078]

[0079] In the formula, Let be the dust interference compensation factor at time t. Let be the pressure difference between the inlet and outlet of the cooling pipe at time t. The baseline pressure difference is used as the pipeline clean condition. The compensation factor is used to calculate the compensation for three types of parameters to obtain the compensated parameter values. Combined with the Laida criterion, the mean and standard deviation of each parameter after compensation are calculated. Outliers with a difference between the value and the mean greater than 3 times the standard deviation are removed. The missing values ​​are filled in by linear interpolation to obtain the second data matrix after secondary processing.

[0080] Step 23: The Z-score standardization method is used to standardize the parameters in the second data matrix to obtain the standardized data matrix, which is the reliable time series dataset.

[0081] Specifically, the credible time series dataset includes standardized credible time series data of temperature, credible time series data of micro-differential pressure leakage, credible time series data of corrosion status, and credible time series data of wind turbine current. These are represented as standardized matrices and denoted as observational evidence for Bayesian network modeling.

[0082] A standardized data matrix is ​​a reliable time series dataset. ,in:

[0083] The reliable time-series temperature data at time t is the standardized result obtained after processing the original temperature data.

[0084] The reliable time-series data of micro-differential pressure leakage at time t is a standardized result obtained after processing the original pipeline inlet and outlet differential pressure data, reflecting the leakage and blockage related status of the cooling pipeline circuit;

[0085] The reliable time-series data of corrosion status at time t is the standardized result obtained after processing the original corrosion current density data of the inner wall of the pipeline, which characterizes the degree of corrosion of the cooling pipeline.

[0086] The reliable time-series data of the wind turbine current at time t is the standardized result obtained after processing the original wind turbine operating current and the converted fan speed data, reflecting the wind turbine operating status; T is the total number of sampling periods, that is, the total number of data acquisitions.

[0087] It's important to note that the sliding window entropy rate essentially identifies outliers that deviate from normal data patterns by calculating the uncertainty (Shannon entropy rate) of data within a sliding time window. In the SVG valve hall scenario, raw time-series data is susceptible to transient interference. For example, short-term power fluctuations in the IGBT module can cause a transient spike of 1-2°C in the junction temperature (not actual heat accumulation), and the current can experience transient pulses of 3-5 times the rated value during fan start-up and shutdown. If such interference is not removed, subsequent models will misjudge the cooling efficiency as abnormal. Under normal operating conditions, the IGBT junction temperature entropy rate is stable between 1.2 and 1.5 (data fluctuation pattern). When transient interference occurs, the entropy rate drops sharply to below 0.8 (data is concentrated in spikes or pulse values, uncertainty is reduced).

[0088] Among them, dust drift compensation corrects systematic errors caused by environmental factors, reducing the measurement deviations of inlet temperature, pipe pressure difference, and corrosion current density to within ±0.2℃, ±1Pa, and ±1μA / cm², respectively, ensuring data authenticity. The second elimination of the Laida criterion is to eliminate extreme anomalies caused by sensor failures. Combined with interpolation completion, the data integrity is improved to over 99%, avoiding the impact of missing data on model training. The final reliable time series dataset has an anomaly rate of less than 1%, a systematic error of less than 0.5%, and uniform dimensions, providing high-quality input for subsequent Bayesian network modeling, which is a key prerequisite for achieving accurate evaluation of cooling efficiency.

[0089] Step 3: Construct a Bayesian network topology containing three levels of nodes (environment layer, component layer, and efficiency layer) and causal dependency edges, and use the maximum likelihood estimation method to optimize the parameters of the Bayesian network topology to form a parameter-calibrated dynamic Bayesian network model.

[0090] In this embodiment, step 3 specifically includes:

[0091] Step 31: Based on the reliable time series dataset and combined with the correlation patterns between cooling system failure modes and parameters in historical fault data, a Bayesian network topology containing three levels of nodes (environment layer, component layer, and efficiency layer) and causal dependency edges is constructed through training using a structure learning algorithm.

[0092] Specifically, in step 31, the standardized data matrix (i.e., the reliable time series dataset) is input into the PC structure learning algorithm. By combining the correlation patterns between the three failure modes of the cooling system—fan jamming, coolant embolism, and pipe corrosion leakage—and various monitoring parameters in the historical fault data, a three-level node topology structure including an environmental layer, a component layer, and an efficiency layer is constructed. The PC algorithm is used to calculate the conditional independence relationship between nodes, determine the causal dependency edges between nodes, and form the initial Bayesian network topology structure.

[0093] It should be noted that the environmental layer node includes standardized inlet ambient temperature and humidity fluctuation latent variables, the component layer node includes standardized fan operating current, fan speed, cooling pipe inlet and outlet pressure difference, pipe inner wall corrosion current density, pipe micro-blockage latent variable, and fan bearing wear latent variable, and the efficiency layer node includes standardized IGBT junction temperature and cooling efficiency.

[0094] Step 32: Introduce the latent variable set of the SVG valve hall cooling system nodes, which includes latent variables of diurnal temperature difference, latent variables of pipe micro-blockage, latent variables of humidity fluctuation, and latent variables of fan bearing wear; Based on expert knowledge, set the conditional probability relationship between the latent variables and each observation node, construct the initial node conditional probability table, determine the probability value of each node under different parent node state combinations, and complete the initialization of the node conditional probability table.

[0095] Step 33: Using the credible time series dataset as observation data, the parameters in the initial node conditional probability table are optimized using the maximum likelihood estimation method;

[0096] Specifically, step 33 transforms the standardized data matrix (i.e., the reliable time series dataset) into observed data and uses the maximum likelihood estimation method to optimize the parameters in the initial node conditional probability table. The objective function of the maximum likelihood estimation is calculated as follows:

[0097]

[0098] In the formula, This is the optimized set of conditional probability distribution parameters. Let T be the set of conditional probability distribution parameters to be optimized, and T be the number of sampling periods. Let G be the row vector of the normalized matrix at time t, and G be the initial Bayesian network topology.

[0099] It should be noted that the parameter correlations in the SVG cooling system are nonlinear and time-dependent (e.g., a change in fan speed for 10 seconds affects the IGBT junction temperature). However, when modeling with Bayesian networks, time-series data is typically transformed into a "static sample-state correlation" model (each sampling period is treated as an independent sample, ignoring short-term time lags, as the 500s sliding window has already smoothed out dynamic fluctuations). In this case, the 7200 sampling periods of the normalized matrix can be regarded as 7200 independent and identically distributed observation samples, which conforms to the basic assumption of "independent and identically distributed samples" in MLE, providing a data foundation for parameter optimization.

[0100] MLE requires calculating probabilities based on discrete state observation data, therefore the continuous values ​​of the standardized matrix must first be partitioned into intervals (discretized): This is combined with the normal distribution characteristics of the standardized parameters (99.73% of the data fall within...). Using the "3σ interval subdivision method": Interval 1 (low deviation): The corresponding standardized value is 1 standard deviation below the mean; Interval 2 (normal deviation): The range is within ±1 standard deviation of the mean (core normal range); Interval 3 (high deviation): (1,3], corresponding to standardized values ​​that are 1 standard deviation above the mean.

[0101] Understandably, through calibration with 7200 real-world running samples, the average absolute error of the probability values ​​in CPT decreased from the initial 0.15 (expert experience bias) to below 0.03, making the Bayesian network's probabilistic description of the cooling system's state more realistic. The Laplace smoothing and incremental update strategies address the pain points of "imbalanced samples" and "high real-time requirements" in industrial scenarios, enabling parameter optimization methods to be directly deployed to the SVG online monitoring system without additional hardware upgrades.

[0102] Step 34: Based on the initial Bayesian network topology and combined with the quantization patterns in historical data, the conditional probability table of the cooling efficiency node is constrained and optimized to obtain the dynamic Bayesian network model after parameter calibration.

[0103] Specifically, step 34 optimizes the conditional probability table of cooling efficiency nodes and the temporal transition probability table of the dynamic Bayesian network based on the initial Bayesian network topology and combined with the quantification of historical data, to obtain the parameter-calibrated DBN model, thus solving the defects of the initial CPT relying on expert experience and not quantifying key influence patterns.

[0104] Step 4: The credible time series dataset is used as the input parameter of the calibrated Bayesian network model as evidence variable. The model is then calculated using the belief propagation inference algorithm and processed by probability weighted scoring to generate the final control command set. The final control command set is then fed back to the valve group monitoring system to achieve early warning and automatic load reduction.

[0105] In this embodiment, step 4 specifically includes:

[0106] Step 41: Using the credible time-series dataset as the input parameter of the calibrated Bayesian network model as evidence variables, calculations are performed using the belief propagation inference algorithm. Simultaneously, based on the fusion iteration of forward and backward probability inference, the failure probabilities of each component of the cooling system, the predicted IGBT junction temperature, and the cooling efficiency for the next sampling cycle are obtained. posterior probability distribution ;

[0107] Among them, forward probability inference is based on reliable time series data at time t and before, combined with the state transition probability of dynamic Bayesian network, to calculate the prior probability distribution of each node at time t+1, so as to predict the trend of cooling efficiency change; backward probability inference is based on real-time reliable data at time t+1, and reversely corrects the probability distribution at time t and before.

[0108] Understandably, in step 41, based on all data at time t and before in the standardized data matrix obtained in the above steps, and combined with the state transition probabilities in the obtained parameter-calibrated dynamic Bayesian network model, the forward algorithm is used to calculate the prior probability distribution of each node at time t+1. The state transition probabilities are determined by the time-dependent parameters in the parameter-calibrated model. The forward algorithm calculates the probability distribution of each node at each time point through recursion to obtain the prior probability distribution of the cooling efficiency node at time t+1, thereby achieving the prediction of the cooling efficiency change trend.

[0109] After acquiring the real-time standardized data at time t+1, based on the backward transition probabilities in the obtained parameter-calibrated dynamic Bayesian network model, a backward algorithm is used to correct the probability distribution of all nodes at time t and earlier. The backward transition probabilities are determined by the time-dependent parameters in the parameter-calibrated model. Through a message passing mechanism, the real-time observation data at time t+1 is transformed into "messages" and sent back to each node at time t, updating the failure probabilities of each component node, such as the probability that the latent variable of fan bearing wear is in a "severe" state and the probability that the latent variable of pipe micro-blockage is in a "blocked" state. Combining the obtained prior probability distribution of each node at time t+1 with the obtained corrected probability distribution, the posterior probability distribution of each node at time t+1 is calculated by fusing them using the Bayesian formula. The posterior probability distribution of the set of failure probabilities of each component in the cooling system, the predicted value of IGBT junction temperature, and the cooling efficiency of the next sampling period are output through fusion calculation. The predicted value of IGBT junction temperature is the sum of the products of each discrete value of IGBT junction temperature node and its corresponding posterior probability.

[0110] Step 42: Combine the failure probability of each component in the comprehensive cooling system, the predicted value of the IGBT junction temperature, and the cooling efficiency in the next sampling cycle. posterior probability distribution The system references the cooling efficiency evaluation threshold and uses the 5th percentile of cooling efficiency as the early warning threshold. After probability-weighted scoring, the cooling efficiency of the SVG valve hall is classified into levels and key influencing factors are output. If the predicted cooling efficiency is lower than the early warning threshold, the closed-loop control process of SVG reactive power command reduction is triggered, the final control command set is generated and fed back to the valve group monitoring system, so as to realize early warning and automatic load reduction.

[0111] Understandably, in step 42, the cooling efficiency of the SVG valve hall is divided into four levels, including "excellent, good, warning, and failure".

[0112] In this step, the cooling efficiency evaluation threshold is referenced. Within this threshold, the normalized IGBT junction temperature corresponding to the normal operating temperature is no greater than 0.3, and the cooling efficiency value corresponding to the 80% derating warning is no greater than 0.8. The posterior probability distribution of the cooling efficiency obtained in step S400 is weighted and scored: "Excellent" corresponds to 1.0, "Good" to 0.8, "Warning" to 0.6, and "Failure" to 0.4. Cooling efficiency levels are determined based on the scoring results: a score of 0.9 or higher is "Excellent," a score of 0.7 or higher but less than 0.9 is "Good," a score of 0.5 or higher but less than 0.7 is "Warning," and a score less than 0.5 is "Failure." Simultaneously, key influencing factors are determined based on the failure probability of each component. For example, when the failure probability of the micro-blockage variable in the pipeline is the highest, the key influencing factor is "micro-blockage in the pipeline."

[0113] Sort all cooling efficiency values ​​corresponding to the posterior probability distribution of cooling efficiency obtained in the above steps in ascending order to obtain an ordered set. Calculate the 5th percentile of this ordered set. The 5th percentile is determined by taking the value at the 5th position after rounding up in the ordered set. That is, only 5% of the cooling efficiency values ​​are lower than this quantile. This quantile is set as the cooling efficiency warning threshold. Calculate the expected predicted value of the obtained posterior probability distribution of cooling efficiency. This expected predicted value is the sum of the products of each discrete value of cooling efficiency and its corresponding posterior probability. If this expected predicted value is lower than the warning threshold, trigger the closed-loop control of SVG reactive power command reduction. Calculate the reduction coefficient, and generate reactive power adjustment commands and fan speed adjustment commands based on the reduction coefficient. The reactive power adjustment command is the current reactive power output capacity multiplied by... The fan speed adjustment command is the current fan speed multiplied by 1. The final control instruction set containing these two instructions is fed back to the valve group monitoring system. Combined with the downward trend of the expected predicted value of cooling efficiency over multiple cycles, it enables early warning and automatic load reduction 72 hours in advance.

[0114] Example 2

[0115] like Figure 2 As shown, the difference between this embodiment and Embodiment 1 is that this embodiment provides an SVG valve hall cooling efficiency evaluation system based on a probabilistic graphical model, which corresponds one-to-one with the SVG valve hall cooling efficiency evaluation method based on a probabilistic graphical model in Embodiment 1; the system includes:

[0116] The acquisition unit is used to acquire raw time-series data, which is collected by a multi-parameter monitoring sensor group of the cooling system arranged in the SVG valve hall over a continuous T sampling period.

[0117] The preprocessing unit is used to preprocess the raw time-series data to obtain a reliable time-series dataset; the preprocessing includes outlier removal, dust interference drift compensation, and standardization.

[0118] The model building unit is used to construct a Bayesian network topology that includes three levels of nodes: environment layer, component layer, and efficiency layer, as well as causal dependency edges. The maximum likelihood estimation method is used to optimize the parameters of the Bayesian network topology to form a parameter-calibrated dynamic Bayesian network model.

[0119] The calculation and evaluation unit is used to take the credible time series dataset as the input parameter of the calibrated Bayesian network model as evidence variable, perform calculations through the belief propagation inference algorithm, and generate the final control command set after probability weighted scoring processing; and feed the final control command set back to the valve group monitoring system to realize early warning and automatic load reduction.

[0120] Specifically, the acquisition unit includes:

[0121] Sub-unit defined: This unit is used to deploy a multi-parameter monitoring sensor group for the cooling system in three core monitoring areas within the SVG valve hall: the IGBT module area, the cooling pipe loop, and the fan nacelle. The multi-parameter monitoring sensor group for the cooling system includes multiple fluorescent fiber optic temperature sensors, high-frequency micro-differential pressure leakage sensors, electrochemical corrosion probes, and fan current Hall sensors. The sampling period is set to 10 seconds, and 7200 sampling periods are continuously collected to cover a 12-hour continuous monitoring duration. The time range and spatial coverage of the raw data collection are determined.

[0122] The classification and acquisition subunit is used to collect IGBT junction temperature and inlet ambient temperature through a fluorescent fiber optic temperature sensor to form raw temperature time-series data; to collect pressure difference between the inlet and outlet of the cooling pipe through a high-frequency micro-differential pressure leakage sensor to form raw micro-differential pressure leakage time-series data; to collect corrosion current density on the inner wall of the pipe through an electrochemical corrosion probe to form raw corrosion status time-series data; and to collect fan operating current through a fan current Hall sensor and convert it to fan speed by combining it with the current-speed conversion coefficient specified in the fan nameplate parameters to form raw fan current time-series data, thereby completing the classification and acquisition of four types of core parameters.

[0123] The first construction subunit is used to align the original time-series data of temperature, micro-differential pressure leakage, corrosion status, and fan current according to the sampling time, and construct the original data matrix. The matrix has 7200 rows and 6 columns, corresponding to IGBT junction temperature, inlet ambient temperature, cooling pipe inlet and outlet pressure difference, pipe inner wall corrosion current density, fan operating current, and fan speed, respectively. Each element in the matrix represents the monitored value of the corresponding parameter at the corresponding sampling time, thus obtaining a structured original data matrix.

[0124] Specifically, the preprocessing unit includes:

[0125] Define a sub-cell: used to define a sliding window length of 50 to cover a 500-second data window, and calculate the Shannon entropy rate of each column parameter in the original data matrix within each sliding window. The formula for calculating the Shannon entropy rate is:

[0126]

[0127] In the formula, Let L be the probability density of the parameters in the j-th column within the w-th sliding window, and L be the length of the sliding window. Let be the probability density of the parameters in the j-th column within the w-th sliding window;

[0128] Decision Subunit: Used to calculate the probability density using kernel density estimation, and set the Shannon entropy threshold to 0.8 times the average Shannon entropy of all sliding windows. If the Shannon entropy of the parameter at the sampling time is lower than the threshold, the parameter value at that time is determined to be an outlier, and the outlier is replaced with the median of the parameter in the corresponding sliding window to obtain the data matrix after preliminary processing.

[0129] The calculation subunit is used to introduce a dust interference compensation factor for parameters such as the ambient temperature at the air inlet, the pressure difference between the inlet and outlet of the cooling pipe, and the corrosion current density of the inner wall of the pipe. The calculation formula for the compensation factor is as follows:

[0130]

[0131] In the formula, Let be the dust interference compensation factor at time t. Let be the pressure difference between the inlet and outlet of the cooling pipe at time t. The baseline pressure difference is used as the pipeline clean condition. The compensation factor is used to calculate the compensation for three types of parameters to obtain the compensated parameter values. Combined with the Laida criterion, the mean and standard deviation of each parameter after compensation are calculated. Outliers with a difference between the value and the mean greater than 3 times the standard deviation are removed. The missing values ​​are filled in by linear interpolation to obtain the data matrix after secondary processing.

[0132] Processing subunit: Used to perform Z-score normalization on all column parameters in the secondary processed data matrix, and obtain a reliable time series dataset through normalization. The reliable time series dataset includes normalized reliable time series data of temperature, reliable time series data of micro-differential pressure leakage, reliable time series data of corrosion status, and reliable time series data of wind turbine current. It is represented as a normalized matrix and recorded as the observation evidence for Bayesian network modeling.

[0133] Specifically, the model building unit includes:

[0134] The second construction subunit is used to input the standardized matrix as an input to the PC structure learning algorithm. It combines the correlation between the three failure modes of the cooling system (fan jamming, coolant embolism, and pipe corrosion leakage) and the monitoring parameters in the historical fault data to construct a three-level node topology structure including the environment layer, component layer, and efficiency layer. The PC algorithm is used to calculate the conditional independence relationship between nodes, determine the causal dependency edges between nodes, and form the initial Bayesian network topology structure.

[0135] The third construction subunit is used to introduce the latent variable set associated with the nodes of the SVG valve hall cooling system. The latent variable set includes latent variables of diurnal temperature difference, latent variables of pipe micro-blockage, latent variables of humidity fluctuation, and latent variables of fan bearing wear. Based on expert knowledge, the conditional probability relationship between the latent variables and each observation node is set, an initial node conditional probability table is constructed, the probability value of each node under different combinations of parent node states is determined, and the initialization of the node conditional probability table is completed.

[0136] Optimization sub-unit: Used to transform the standardized matrix into observed data and optimize the parameters in the initial nodal conditional probability table using the maximum likelihood estimation method. The objective function of the maximum likelihood estimation is calculated as follows:

[0137]

[0138] In the formula, This is the optimized set of conditional probability distribution parameters. Let T be the set of conditional probability distribution parameters to be optimized, and T be the number of sampling periods. Let G be the row vector of the normalized matrix at time t, and G be the initial Bayesian network topology.

[0139] Obtaining sub-units: Based on the initial Bayesian network topology and combined with the quantization patterns in historical data, the conditional probability table of the cooling efficiency node is constrained and optimized, and finally the dynamic Bayesian network model after parameter calibration is obtained.

[0140] The execution process of each unit can be carried out according to the steps of the SVG valve hall cooling efficiency evaluation method based on the probabilistic graphical model in Example 1, and will not be described in detail in this example.

[0141] Example 3

[0142] Corresponding to the above method embodiments, this embodiment also provides an SVG valve hall cooling efficiency evaluation device based on a probabilistic graphical model. The SVG valve hall cooling efficiency evaluation device based on a probabilistic graphical model described below can be referred to in correspondence with the SVG valve hall cooling efficiency evaluation method based on a probabilistic graphical model described above.

[0143] Figure 3 This is a block diagram illustrating an SVG valve hall cooling efficiency evaluation device 800 based on a probabilistic graphical model, according to an exemplary embodiment. Figure 3 As shown, the SVG valve hall cooling efficiency evaluation device 800 based on a probabilistic graphical model includes a processor 801 and a memory 802. The SVG valve hall cooling efficiency evaluation device 800 also includes one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0144] The processor 801 controls the overall operation of the SVG valve hall cooling efficiency evaluation device 800 based on the probabilistic graphical model to complete all or part of the steps in the aforementioned SVG valve hall cooling efficiency evaluation method based on the probabilistic graphical model. The memory 802 stores various types of data to support the operation of the SVG valve hall cooling efficiency evaluation device 800. This data may include, for example, instructions for any application or method operating on the SVG valve hall cooling efficiency evaluation device 800, as well as application-related data such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as keyboards, mice, or buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the probabilistic graphical model-based SVG valve hall cooling efficiency evaluation device 800 and other devices. Wireless communication includes, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, or an NFC module.

[0145] In an exemplary embodiment, the SVG valve hall cooling efficiency evaluation device 800 based on the probabilistic graphical model can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described probabilistic graphical model-based SVG valve hall cooling efficiency evaluation method.

[0146] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided. When executed by a processor, these program instructions implement the steps of the above-described probabilistic graphical model-based SVG valve hall cooling efficiency evaluation method. For example, the computer-readable storage medium may be the memory 802 including the program instructions, which may be executed by the processor 801 of the probabilistic graphical model-based SVG valve hall cooling efficiency evaluation device 800 to complete the above-described probabilistic graphical model-based SVG valve hall cooling efficiency evaluation method.

[0147] Example 4

[0148] Corresponding to the above method embodiments, this embodiment also provides a readable storage medium. The readable storage medium described below can be referred to in conjunction with the above-described method for evaluating the cooling efficiency of SVG valve hall based on a probabilistic graphical model.

[0149] A computer program is stored on a readable storage medium, and when the computer program is executed by a processor, it implements the steps of the SVG valve hall cooling efficiency evaluation method based on a probabilistic graphical model as described in the above method embodiments.

[0150] Specifically, the readable storage medium can be a USB flash drive, external hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, or any other readable storage medium capable of storing program code.

[0151] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the cooling efficiency of an SVG valve hall based on a probabilistic graphical model, characterized in that, The method includes: Acquire raw time-series data, which is collected by a multi-parameter monitoring sensor group of the cooling system arranged in the SVG valve hall over T consecutive sampling periods; The original time-series data is preprocessed to obtain a reliable time-series dataset; the preprocessing includes outlier removal, dust interference drift compensation, and standardization. A Bayesian network topology structure containing three levels of nodes (environment layer, component layer, and efficiency layer) and causal dependency edges is constructed. The maximum likelihood estimation method is used to optimize the parameters of the Bayesian network topology structure, forming a parameter-calibrated dynamic Bayesian network model. The trusted time-series dataset is used as an evidence variable and input into the parameter-calibrated Bayesian network model. The model is then calculated using a belief propagation inference algorithm and subjected to probability-weighted scoring to generate a final control instruction set. This final control instruction set is then fed back to the valve group monitoring system to achieve early warning and automatic load reduction.

2. The method for evaluating the cooling efficiency of an SVG valve hall based on a probabilistic graphical model according to claim 1, characterized in that, The raw time series data includes raw temperature time series data, raw time series data of micro-differential pressure leakage, raw time series data of corrosion status, and raw time series data of fan current.

3. The method for evaluating the cooling efficiency of SVG valve hall based on a probabilistic graphical model according to claim 2, characterized in that, Obtain raw time series data, including: In the three monitoring areas of the SVG valve hall—the IGBT module area, the cooling pipe loop, and the fan compartment—a multi-parameter monitoring sensor group for the cooling system is arranged. The multi-parameter monitoring sensor group for the cooling system includes multiple fluorescent fiber optic temperature sensors, high-frequency micro-differential pressure leakage sensors, electrochemical corrosion probes, and fan current Hall sensors. The IGBT junction temperature and the inlet ambient temperature are collected by a fluorescent fiber optic temperature sensor to form raw time-series temperature data. The pressure difference between the inlet and outlet of the cooling pipe is collected by a high-frequency micro differential pressure leakage sensor to form raw time-series data of micro differential pressure leakage. The corrosion current density of the inner wall of the pipeline is collected by an electrochemical corrosion probe to form the original time series data of the corrosion state. The fan operating current is collected by a fan current Hall sensor, and the fan speed is calculated by combining the current and speed conversion coefficient specified in the fan nameplate parameters, thus forming the original time series data of the fan current. The original time-series data of temperature, micro-differential pressure leakage, corrosion status, and fan current are aligned according to the sampling time to construct an original data matrix, which is then used as the original time-series data.

4. The method for evaluating the cooling efficiency of an SVG valve hall based on a probabilistic graphical model according to claim 1, characterized in that, The original time-series data is preprocessed to obtain a reliable time-series dataset, including: An outlier removal method based on sliding window entropy rate is used to eliminate interference fluctuations in the original time series data, resulting in a first data matrix; The first data matrix is ​​subjected to dust interference drift compensation processing to obtain the compensated parameter values; outliers are removed from the compensated parameter values ​​by combining the Laida criterion and missing values ​​are filled in based on linear interpolation to obtain the second data matrix; The parameters in the second data matrix are standardized using the Z-score standardization method to obtain a standardized data matrix, which is a reliable time series dataset.

5. The method for evaluating the cooling efficiency of an SVG valve hall based on a probabilistic graphical model according to claim 4, characterized in that, The steps for obtaining the first data matrix and the second data matrix are as follows: Calculate the Shannon entropy rate of each column parameter in the original time series data within each sliding window; If the Shannon entropy rate of the parameter at the sampling time is lower than the preset threshold, the parameter value at that time is determined to be an outlier, and the outlier value is replaced by the median of the parameter in the corresponding sliding window to obtain the first data matrix. Based on the first data matrix, the compensation calculation is performed on three parameters—air inlet ambient temperature, inlet and outlet pressure difference of cooling pipe, and corrosion current density of pipe inner wall—based on the compensation factor, and the compensated parameters are obtained. Based on the Laida criterion, the mean and standard deviation of each parameter after compensation are calculated. Outliers with a difference between the value and the mean greater than a preset multiple of the standard deviation are removed, and missing values ​​are filled in using linear interpolation to obtain the second data matrix.

6. The method for evaluating the cooling efficiency of an SVG valve hall based on a probabilistic graphical model according to claim 1, characterized in that, A Bayesian network topology is constructed, comprising three levels of nodes (environment layer, component layer, and efficiency layer) and causal dependency edges. The maximum likelihood estimation method is then used to optimize the Bayesian network topology parameters, resulting in a parameter-calibrated dynamic Bayesian network model, including: Based on the trusted time series dataset and combined with the correlation patterns between cooling system failure modes and parameters in historical fault data, a Bayesian network topology containing three levels of nodes (environment layer, component layer, and efficiency layer) and causal dependency edges is constructed through training using a structure learning algorithm. A set of latent variables associated with nodes in the SVG valve hall cooling system is introduced. Based on expert knowledge, the conditional probability relationship between the latent variables and each observed node is set, an initial node conditional probability table is constructed, the probability value of each node under different combinations of parent node states is determined, and the node conditional probability table is initialized. Using the trusted time series dataset as observation data, the parameters in the initial node conditional probability table are optimized using the maximum likelihood estimation method; Based on the initial Bayesian network topology and combined with the quantification patterns in historical data, the conditional probability table of the cooling efficiency node is constrained and optimized to obtain the dynamic Bayesian network model after parameter calibration.

7. The method for evaluating the cooling efficiency of an SVG valve hall based on a probabilistic graphical model according to claim 1, characterized in that, The trusted time series dataset is used as evidence variables and input into the parameter-calibrated Bayesian network model. The model is then calculated using a belief propagation inference algorithm and subjected to probability-weighted scoring to generate the final control instruction set. The final control command set is then fed back to the valve group monitoring system to achieve early warning and automatic load reduction, including: The trusted time-series dataset is used as evidence variables and input into the parameter-calibrated Bayesian network model. The model is then computed using a belief propagation inference algorithm. Simultaneously, based on the fusion iteration of forward and backward probability inference, the failure probabilities of each component in the cooling system, the predicted IGBT junction temperature, and the cooling efficiency for the next sampling cycle are obtained. posterior probability distribution ; Based on the failure probabilities of each component in the cooling system, the predicted IGBT junction temperature, and the cooling efficiency in the next sampling cycle... posterior probability distribution Based on the cooling efficiency evaluation threshold and early warning threshold, and after probability weighted scoring, the SVG valve hall cooling efficiency is classified into levels and the key influencing factors are output. If the predicted cooling efficiency is lower than the warning threshold, the closed-loop control process of SVG reactive power command reduction is triggered, the final control command set is generated and fed back to the valve group monitoring system to realize early warning and automatic load reduction. Among them, forward probability inference is based on reliable time series data at time t and before, combined with the state transition probability of dynamic Bayesian network, to calculate the prior probability distribution of each node at time t+1, so as to predict the trend of cooling efficiency change; backward probability inference is based on real-time reliable data at time t+1, and reversely corrects the probability distribution at time t and before.

8. An SVG valve hall cooling efficiency evaluation system based on a probabilistic graphical model, characterized in that, The system includes: The acquisition unit is used to acquire raw time-series data, which is collected by a multi-parameter monitoring sensor group of the cooling system arranged in the SVG valve hall over a continuous T sampling period. The preprocessing unit is used to preprocess the original time-series data to obtain a reliable time-series dataset; the preprocessing includes outlier removal, dust interference drift compensation, and standardization. The model building unit is used to construct a Bayesian network topology that includes three levels of nodes: environment layer, component layer, and efficiency layer, as well as causal dependency edges. The maximum likelihood estimation method is used to optimize the parameters of the Bayesian network topology to form a parameter-calibrated dynamic Bayesian network model. The calculation and evaluation unit is used to input the credible time series dataset as evidence variables into the parameter-calibrated Bayesian network model, perform calculations through the belief propagation inference algorithm, and generate the final control instruction set after probability weighted scoring processing; and feed the final control instruction set back to the valve group monitoring system to realize early warning and automatic load reduction.

9. An SVG valve hall cooling efficiency evaluation device based on a probabilistic graphical model, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the SVG valve hall cooling efficiency evaluation method based on a probabilistic graphical model as described in any one of claims 1 to 7 when executing the computer program.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program that, when executed by a processor, implements the SVG valve hall cooling efficiency evaluation method based on a probabilistic graphical model as described in any one of claims 1 to 7.

Images