Intelligent monitoring method and system for high-voltage reactive power compensation device of new energy station

Abstract

This application relates to the field of electrical equipment technology, specifically an intelligent monitoring method and system for high-voltage reactive power compensation devices in new energy power plants. Utilizing an LSTM model trained on health data, combined with control signals, cabinet internal temperature, and ambient temperature, it constructs input features to accurately predict the theoretical temperature of components under specific operating conditions. By comparing actual temperatures to generate residuals, the degree of deviation can be quantified, providing a basis for early anomaly detection. It not only analyzes the residual distribution but also integrates the temperature difference distribution characteristics between components and the residual change trend across cycles. This three-dimensional analysis of "space (temperature difference) + time (trend) + deviation (residual)" can effectively distinguish between heat dissipation system failures, sudden component damage, and gradual component aging, avoiding misjudgments based on a single indicator. This application solves the problem of delayed fault detection in existing technologies and can accurately analyze various abnormal situations, helping maintenance personnel to perform automated, full-process monitoring of reactive power compensation devices.

Description

Technical Field

[0001] This application relates to the field of electrical equipment technology, specifically to an intelligent monitoring method and system for high-voltage reactive power compensation devices in new energy power plants. Background Technology

[0002] Reactive power compensation devices for new energy power plants (such as wind farms and photovoltaic power plants) are key equipment specifically designed to solve voltage stability problems and power factor non-compliance problems caused by intermittency and fluctuation in the grid-connected operation of new energy power generation systems.

[0003] Due to the importance of reactive power compensation devices, they generally require real-time monitoring to prevent malfunctions and malfunctions. Current monitoring methods for reactive power compensation devices have significant shortcomings, for example:

[0004] (1) The fault detection is severely delayed, and the reliance on direct measurement by temperature sensors results in an average alarm time of 24 hours.

[0005] (2) It cannot distinguish between heat dissipation system deterioration (such as dust accumulation, fan failure) and individual component failure (such as IGBT damage), resulting in a high misjudgment rate;

[0006] (3) Operation and maintenance rely heavily on manual inspections (such as only once a month), which is inefficient and prone to missed inspections; fixed threshold alarm mechanisms (such as temperature > 85°C) have a false alarm rate of up to 70% under fluctuating operating conditions, resulting in a serious waste of operation and maintenance resources;

[0007] These shortcomings make existing monitoring solutions unable to meet the high reliability and real-time operation and maintenance requirements of new energy power plants, often leading to significant losses such as equipment disconnection from the grid and grid fines. There is an urgent need to upgrade from passive response to proactive prediction through intelligent means. Summary of the Invention

[0008] In view of this, the purpose of this application is to provide an intelligent monitoring method and system for high-voltage reactive power compensation devices in new energy power plants, so as to solve the problems in the background art.

[0009] To achieve the above objectives, this application adopts the following technical solution:

[0010] The intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants disclosed in this application includes the following steps:

[0011] When the controller of the reactive power compensation device outputs a control signal, the control signal, the temperature sequence inside the cabinet, and the ambient temperature outside the cabinet are obtained, and the actual temperature of multiple components inside the reactive power compensation device is obtained when the control signal ends.

[0012] Extract the value sequence of the target feature from the control signal, and construct an input feature vector based on the value sequence of the target feature, the cabinet temperature sequence, and the ambient temperature. The type of the target feature is obtained in advance through historical data analysis.

[0013] The input feature vector is input into a pre-built temperature prediction model to obtain the predicted temperature of multiple components. The temperature prediction model is built based on the operating data of a reactive power compensation device in a healthy state.

[0014] The difference between the actual temperature and the predicted temperature of multiple components is calculated to obtain a residual sequence; residual distribution features are extracted from the residual sequence; temperature difference distribution features are extracted from the actual temperature of multiple components; and residual change features are extracted based on the residual sequence of multiple maintenance cycles.

[0015] Based on the residual distribution characteristics and the temperature difference distribution characteristics, monitoring and analysis are performed to obtain heat dissipation and component damage monitoring results. Based on the residual change characteristics, monitoring and analysis are performed to obtain component aging monitoring results.

[0016] In one embodiment of this application, the method for determining the target feature includes:

[0017] Acquire control signal samples and measured temperature rise samples from multiple historical time points;

[0018] The control signal samples and measured temperature rise samples from multiple historical time points are preprocessed to obtain control signal preprocessed data and temperature rise preprocessed data. The preprocessing includes outlier removal and missing value interpolation.

[0019] Multiple feature parameters are extracted from the preprocessed control signal data, including reactive current reference value, reactive power, switching frequency, PWM duty cycle, DC bus voltage, and actual DC bus voltage.

[0020] A temperature rise sequence was constructed based on measured temperature rise samples from multiple historical time points, and multiple feature parameter sequences were constructed based on the feature values ​​of each feature parameter at multiple historical time points.

[0021] Each feature parameter sequence is normalized to obtain a normalized parameter sequence; and the correlation between each normalized parameter sequence and the temperature rise sequence is calculated. The relevance The mathematical expression is:

[0022]

[0023] In the formula, For historical time points index, For the number of historical time points, For the first Feature values ​​at each historical time point This represents the average value of the characteristic parameters. For the first Temperature rise at a historical point in time, This represents the average temperature rise.

[0024] The correlation between each feature parameter and temperature rise is compared with a preset correlation threshold, and the feature parameters that are greater than or equal to the correlation threshold are taken as target features.

[0025] In one embodiment of this application, an input feature vector is constructed based on the value sequence of the target feature, the temperature sequence inside the cabinet, and the ambient temperature, including:

[0026] The value sequence of the target feature is time-aligned with the temperature sequence inside the cabinet, and the target duration before the end time of the control command is extracted to obtain a value subsequence and a temperature subsequence inside the cabinet with fixed length.

[0027] The value subsequence and the cabinet temperature subsequence are normalized to obtain normalized value subsequence and normalized cabinet temperature subsequence.

[0028] An input feature vector is constructed based on the normalized value subsequence, the normalized cabinet temperature subsequence, and the ambient temperature.

[0029] In one embodiment of this application, the method for constructing the temperature prediction model includes:

[0030] The control signal samples, cabinet temperature sequence samples, measured temperature sequence samples of multiple components, and ambient temperature samples of the reactive power compensation device in a healthy state at multiple historical time points are obtained, wherein the cabinet temperature sequence samples are the average values ​​of the measured temperature sequence samples of multiple components.

[0031] Target feature samples are extracted from the control signal samples, and input feature samples are constructed based on the value sequence of the target feature samples, the cabinet temperature sequence samples, and the ambient temperature samples.

[0032] The measured temperature values ​​corresponding to the end time points of control signal samples are extracted from the measured temperature sequence samples of multiple components to obtain sample labels; and training data and training datasets for multiple historical time points are constructed based on the input feature samples and sample labels.

[0033] The LSTM model is trained based on the training dataset to obtain a temperature prediction model.

[0034] In one embodiment of this application, extracting residual distribution features from the residual sequence includes:

[0035] Calculate the residual mean, residual standard deviation, and maximum deviation feature of the residual sequence, wherein the maximum deviation feature is used to point to the component with the largest absolute value of the residual;

[0036] The residual distribution features are constructed based on the residual mean, the residual standard deviation, and the maximum deviation feature.

[0037] In one embodiment of this application, extracting temperature difference distribution features from the actual temperatures of multiple components includes:

[0038] Calculate the range characteristics of the actual temperature of multiple components , wherein the range feature The mathematical expression is:

[0039]

[0040] In the formula, This indicates the maximum temperature. Indicates the minimum temperature;

[0041] Obtain the coordinates of multiple components, and construct component feature vectors based on the coordinates of multiple components and the actual temperature;

[0042] Clustering the feature vectors of multiple components yields one or more feature vector clusters;

[0043] Extract the target cluster with the smallest number from one or more feature vector clusters, and calculate the ratio of the number of components within the target cluster to the total number of components. ;

[0044] Based on the range characteristics, the component number within the target cluster, and the ratio. Construct temperature difference distribution characteristics.

[0045] In one embodiment of this application, residual change features are extracted based on residual sequences from multiple maintenance cycles, including:

[0046] Obtain the target residual sequence for multiple maintenance cycles. The target residual sequence is the residual sequence generated by the first control command after maintenance is completed. The multiple maintenance cycles include the target number of maintenance cycles before the current maintenance cycle.

[0047] Linear fitting is performed on the residual trend of each component based on the target residual sequence of multiple maintenance cycles to obtain the residual change trend of each component. ;

[0048] Residual variation characteristics are constructed based on the residual variation trends of multiple components.

[0049] In one embodiment of this application, monitoring and analysis are performed based on the residual distribution characteristics and the temperature difference distribution characteristics to obtain heat dissipation and component damage monitoring results, including:

[0050] The mean of the residuals is compared with a preset residual tolerance threshold, the standard deviation of the residuals is compared with a preset standard deviation threshold, and the range feature is compared with a preset range threshold.

[0051] When the mean residual is less than or equal to a preset residual tolerance threshold, the standard deviation of the residual is less than or equal to a preset standard deviation threshold, and the value of the range characteristic is less than or equal to a preset range threshold, the heat dissipation conditions and operating status of the reactive power compensation device are determined to be normal.

[0052] When the mean residual is greater than a preset residual tolerance threshold and the standard deviation of the residual is less than or equal to a preset standard deviation threshold, it is determined that the reactive power compensation device has an overall heat dissipation abnormality.

[0053] When the residual mean is greater than a preset residual tolerance threshold and the residual standard deviation is greater than a preset standard deviation threshold, the ratio is... Compare with a preset ratio threshold;

[0054] The ratio When the ratio exceeds a preset threshold, a localized heat dissipation anomaly is determined; at the ratio... When the deviation is less than or equal to a preset proportional threshold, the maximum deviation feature is matched with the component number within the target cluster;

[0055] When the maximum deviation feature matches the component number within the target cluster, the faulty component is located based on the maximum deviation feature. Otherwise, it is further verified whether there are multiple components whose residuals are greater than a preset residual tolerance threshold. If so, it is determined that there are multiple points of local heat dissipation anomaly. If not, it is determined that there is a single point of local heat dissipation anomaly.

[0056] In one embodiment of this application, monitoring and analysis are performed based on the residual change characteristics to obtain component aging monitoring results, including:

[0057] The residual variation trends of multiple components are compared with the preset slope threshold;

[0058] When the residual change trend of any component is greater than or equal to a preset slope threshold, it is determined that the arbitrary component has an aging trend.

[0059] This application also provides an intelligent monitoring system for high-voltage reactive power compensation devices in new energy power plants, including:

[0060] The acquisition module is used to acquire the control signal, the temperature sequence inside the cabinet and the ambient temperature outside the cabinet when the controller of the reactive power compensation device outputs the control signal, and to acquire the actual temperature of multiple components inside the reactive power compensation device when the control signal ends.

[0061] The first feature extraction module is used to extract the value sequence of the target feature from the control signal, and construct an input feature vector based on the value sequence of the target feature, the cabinet temperature sequence, and the ambient temperature, wherein the type of the target feature is obtained in advance through historical data analysis;

[0062] The temperature prediction module is used to input the input feature vector into a pre-built temperature prediction model to obtain the predicted temperature of multiple components, wherein the temperature prediction model is built based on the operating data of the reactive power compensation device in a healthy state.

[0063] The second feature extraction module is used to calculate the difference between the actual temperature and the predicted temperature of multiple components to obtain a residual sequence; extract residual distribution features from the residual sequence; extract temperature difference distribution features from the actual temperature of multiple components; and extract residual change features based on the residual sequence of multiple maintenance cycles.

[0064] The monitoring and analysis module is used to perform monitoring and analysis based on the residual distribution characteristics and the temperature difference distribution characteristics to obtain heat dissipation and component damage monitoring results, and to perform monitoring and analysis based on the residual change characteristics to obtain component aging monitoring results.

[0065] The beneficial effects of this application are as follows: The intelligent monitoring method and system for high-voltage reactive power compensation devices in new energy power plants utilizes an LSTM model trained based on health data, combined with control signals, cabinet internal and ambient temperatures to construct input features, enabling accurate prediction of the theoretical temperature of components under specific operating conditions. By comparing actual temperatures to generate residuals, the degree of deviation can be quantified, providing a basis for early anomaly detection. It not only analyzes the residual distribution but also integrates the temperature difference distribution characteristics between components and the residual change trend across cycles. This three-dimensional analysis of "space (temperature difference) + time (trend) + deviation (residual)" can effectively distinguish between heat dissipation system failures, sudden component damage, and gradual component aging, avoiding misjudgments based on a single indicator. This application solves the problem of delayed fault detection in existing technologies and can accurately analyze various abnormal situations, helping maintenance personnel to perform automated, full-process monitoring of reactive power compensation devices. Attached Figure Description

[0066] The present application will be further described below with reference to the accompanying drawings and embodiments:

[0067] Figure 1 This is a structural frame diagram of a reactive power compensation device shown in one embodiment of this application;

[0068] Figure 2 This is a topology diagram of the monitoring system for the reactive power compensation device in one embodiment of this application;

[0069] Figure 3 This is a flowchart illustrating an intelligent monitoring method for a high-voltage reactive power compensation device for a new energy power station, as shown in one embodiment of this application.

[0070] Figure 4 This is a structural diagram of an intelligent monitoring system for a high-voltage reactive power compensation device for a new energy power station, shown in one embodiment of this application.

[0071] The attached diagram is labeled as follows: 210-Temperature sensor, 220-Intelligent terminal, 230-Analysis host, 240-Mobile terminal, 250-Controller. Detailed Implementation

[0072] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0073] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the layers related to this application and are not drawn according to the actual number, shape and size of the layers in the actual implementation. In the actual implementation, the form, number and proportion of each layer can be arbitrarily changed, and the layer layout may also be more complex.

[0074] Numerous details are explored in the following description to provide a more thorough explanation of embodiments of this application; however, it will be apparent to those skilled in the art that embodiments of this application may be practiced without these specific details.

[0075] Figure 1 This is a structural framework diagram of a reactive power compensation device shown in one embodiment of this application, as follows: Figure 1 As shown, Figure 1 The voltage-type bridge circuit shown is the core converter unit of a reactive power compensation device (such as a static var generator, SVG). Its basic principle is: by controlling the on and off states of power electronic switches (such as IGBTs), the DC-side voltage is inverted into an AC voltage with the same frequency, phase, and adjustable amplitude as the grid voltage.

[0076] When the AC voltage is connected to the power grid via a connecting reactor, the device mainly uses the voltage amplitude difference to regulate reactive power:

[0077] If the output voltage amplitude is greater than the grid voltage, the device provides capacitive reactive power (equivalent to a capacitor) to the grid to compensate for the inductive load;

[0078] If the output voltage amplitude is less than the grid voltage, the device absorbs inductive reactive power from the grid (equivalent to a reactor) to suppress overcompensation.

[0079] By detecting the reactive power demand of the load in real time, the control system quickly adjusts the amplitude and phase of the output voltage of the bridge circuit, so that the reactive current generated by the device cancels out the reactive current required by the load, thereby realizing dynamic and continuous compensation of reactive power, improving the power factor, and stabilizing the grid voltage.

[0080] The principles of each component in the above-mentioned reactive power compensation device structure and the necessity of monitoring them are explained below:

[0081] Power switching devices (IGBTs) are responsible for turning the main circuit on and off. During conduction, there are conduction losses (generated by saturation voltage drop), and during switching, there are switching losses (generated by voltage and current overlap). Due to their high switching frequency (typically several kilohertz), they generate the most heat and are the core component for temperature monitoring.

[0082] The freewheeling diode (connected in anti-parallel to the IGBT) conducts when the inductive load current freewheels, resulting in forward voltage drop losses; it also incurs losses during reverse recovery. Diode failure can also lead to shoot-through or open-circuit faults in the bridge arm, therefore monitoring is necessary.

[0083] DC bus capacitors are used to support DC voltage and filter ripple. They contain an equivalent series resistance, which generates Joule heating when ripple current flows through them. Furthermore, the lifespan of a capacitor is extremely sensitive to temperature. Electrolyte drying or capacity decay is usually accompanied by an abnormal temperature rise; monitoring the temperature can predict its remaining lifespan.

[0084] The AC-side connection reactor is connected between the bridge arm output and the power grid. Current flowing through the coil generates copper losses; the iron core generates hysteresis losses and eddy current losses (iron losses) in an alternating magnetic field. Overheating may lead to insulation aging or inter-turn short circuits, affecting the filtering effect. Therefore, monitoring is necessary.

[0085] Busbars and connection points are copper busbars that carry high currents. If screws become loose during installation or the contact surfaces oxidize after long-term operation, the contact resistance increases, leading to localized high temperatures. Overheating at connection points is a common source of fire hazard.

[0086] The heating characteristics and failure modes of the above-mentioned components are different. Therefore, the monitoring scheme needs to combine residual distribution characteristics (to locate abnormal components) and temperature difference distribution characteristics (to analyze uneven heat dissipation) for comprehensive diagnosis.

[0087] Figure 2 This is a topology diagram of the monitoring system for the reactive power compensation device in one embodiment of this application, as shown below. Figure 2 As shown, this application installs temperature sensors 210 at multiple component locations within the reactive power compensation device cabinet. The temperature sensors 210 are connected via wiring to an external intelligent terminal 220. The intelligent terminal 220 is externally located to prevent high temperatures from affecting its normal monitoring operation. The intelligent terminal 220 is also connected to the reactive power compensation device controller 250 to obtain control signals. Furthermore, the intelligent terminal 220 transmits data via a network to a remote analysis host 230. The analysis host 230 analyzes the monitoring results in real time and sends the results to a mobile terminal 240 carried by maintenance personnel, achieving highly timely monitoring.

[0088] Figure 3 This is a flowchart illustrating an intelligent monitoring method for a high-voltage reactive power compensation device in a new energy power plant, as shown in one embodiment of this application. Figure 3 As shown, the intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants in this application mainly includes the following steps:

[0089] S310: When the controller of the reactive power compensation device outputs a control signal, the control signal, the temperature sequence inside the cabinet, and the ambient temperature outside the cabinet are obtained, and the actual temperature of multiple components inside the reactive power compensation device is obtained when the control signal ends.

[0090] Specifically, the controller of the reactive power compensation device in this application outputs a digital command signal, which is a high-speed digital coded signal transmitted via a fieldbus (such as fiber optic, CAN bus, RS485). The controller encodes the voltage command into optical pulses and sends them to each power module via optical fiber, and the module then decodes them into PWM drive signals.

[0091] After the signal ends, the reactive power compensation device stops operating, and the internal component temperature no longer rises (or rises slightly due to thermal inertia, but can be used as a reference temperature). This temperature is used to determine whether the component temperature rise is reasonable, thereby monitoring information such as heat dissipation conditions, component status, and component lifespan.

[0092] S320, extract the value sequence of the target feature from the control signal, and construct an input feature vector based on the value sequence of the target feature, the cabinet temperature sequence, and the ambient temperature, wherein the type of the target feature is obtained in advance through historical data analysis;

[0093] In this application, considering thermal inertia and the temporal nature of temperature changes, an LSTM model is used to predict the temperature of the acquired control signals.

[0094] The control signal contains various characteristic parameters, such as reactive current reference value, reactive power, switching frequency, PWM duty cycle, DC bus voltage, and actual DC bus voltage. However, not every characteristic parameter is strongly correlated with heat generation. Therefore, before extracting features and constructing input features, it is necessary to screen out target features that are strongly correlated with heat generation. The screening process is as follows:

[0095] (1) Obtain control signal samples and measured temperature rise samples from multiple historical time points;

[0096] Specifically, during the actual operation of the reactive power compensation device, various signal data output by the controller are recorded at a fixed sampling frequency (such as per second or per millisecond), while the component temperature data at the corresponding time is also recorded. A raw dataset containing "signal-temperature rise" is constructed by aligning the timestamps.

[0097] (2) Preprocess the control signal samples and measured temperature rise samples from multiple historical time points to obtain control signal preprocessed data and temperature rise preprocessed data, wherein the preprocessing includes outlier removal and missing value interpolation.

[0098] Statistical methods (such as 3) Based on principles (box plots) or physical constraints (such as voltage not exceeding 1.2 times the rated value), identify and eliminate outliers caused by sensor malfunctions or communication interference. For a small amount of missing data, use linear interpolation, forward imputation, or the mean of adjacent time points to complete the data.

[0099] (3) Extract various feature parameters from the control signal preprocessing data, wherein the various feature parameters include reactive current reference value, reactive power, switching frequency, PWM duty cycle, DC bus voltage, and actual DC bus voltage;

[0100] From the preprocessed control signal data, specific physical quantity sequences are extracted. For example, the reactive current reference value is parsed from the internal bus of the digital controller. Calculate the switching frequency from the PWM modulation wave and duty cycle Obtain the DC bus voltage from the DC-side sampling circuit. By deconstructing complex signals into a quantifiable set of features, explicit input variables are provided for correlation analysis.

[0101] (4) Construct a temperature rise sequence based on measured temperature rise samples at multiple historical time points, and construct multiple feature parameter sequences based on the feature values ​​of each feature parameter at multiple historical time points;

[0102] Specifically, for each historical point in time Calculate the temperature rise at that moment. The definition of temperature rise can be... Arrange the temperature rise values ​​at all time points in chronological order to obtain the temperature rise sequence. .

[0103] Then, each feature parameter (such as the reactive current reference value) extracted in step (3) is arranged in the same time sequence to obtain the corresponding feature sequence. .

[0104] Serialization preserves the temporal relationship of the data, allowing subsequent calculations to examine the correspondence between features and temperature rise at the same sample points (at the same time).

[0105] (5) Normalize each feature parameter sequence to obtain a normalized parameter sequence; and calculate the correlation between each normalized parameter sequence and the temperature rise sequence. The relevance The mathematical expression is:

[0106]

[0107] In the formula, For historical time points index, For the number of historical time points, For the first Feature values ​​at each historical time point This represents the average value of the characteristic parameters. For the first Temperature rise at a historical point in time, This represents the average temperature rise.

[0108] Normalize each feature parameter sequence and temperature rise sequence (e.g., using Min-Max or Z-score) to obtain dimensionless sequences. Normalization eliminates the influence of dimensions, ensuring that the correlation in subsequent calculations is not affected by the absolute size of the features, but only reflects the correlation of the changing trend.

[0109] This application uses the Pearson correlation coefficient formula to calculate the correlation between the normalized feature sequence and the temperature rise sequence, in order to measure the degree of linear correlation between the two variables. The value range is [−1, 1]. The closer the absolute value is to 1, the stronger the linear relationship between the feature and the temperature rise (positive correlation indicates that the temperature rise increases as the feature increases, and negative correlation indicates the opposite). Essentially, it calculates whether the trends of the two variables deviating from their respective means are consistent.

[0110] (6) Compare the correlation between each feature parameter and temperature rise with the preset correlation threshold, and take the feature parameter that is greater than or equal to the correlation threshold as the target feature.

[0111] Finally, a relevance threshold is set (e.g.) ), calculate all feature parameters Compare with this threshold. Retain those that meet the criteria. Features that are greater than or equal to the threshold are used as "target features" in the subsequent temperature prediction model; features with weak correlation are removed.

[0112] Only features strongly correlated with temperature rise contribute to temperature prediction. Introducing weakly correlated or irrelevant features increases the dimensionality of the model input, introduces noise, and leads to overfitting or decreased prediction accuracy. The above process reduces the dimensionality of input features, simplifies model complexity, and reduces computational resource consumption. Eliminating irrelevant variables improves the model's robustness and prediction accuracy.

[0113] In one embodiment of this application, the final target features include a reactive current reference value and a switching frequency.

[0114] After determining the target features, the input feature vector of the model is constructed through the following process:

[0115] S321, the value sequence of the target feature is time-aligned with the temperature sequence inside the cabinet, and the target duration before the end time of the control command is extracted to obtain a value subsequence and a temperature subsequence inside the cabinet with fixed length.

[0116] Align the target feature value sequence with the cabinet temperature sequence on the time axis (e.g., through resampling or interpolation), and then truncate a fixed time period backward from the end of the control command. A fragment.

[0117] This ensures that at each time step, the feature data and temperature data reflect the physical state at the same moment, which is the foundation for subsequent time series modeling. Models such as LSTM require a fixed input dimension. Truncation of a fixed duration ensures that all samples have a consistent shape, facilitating batch training. Data before the end point is truncated because the current temperature is the result of heat accumulation over a past period, reflecting the system's thermal inertia.

[0118] In addition, if the length of the control command is insufficient, it is padded with a mask to ensure that the length is consistent.

[0119] S322, normalize the value subsequence and the cabinet temperature subsequence to obtain normalized value subsequence and normalized cabinet temperature subsequence;

[0120] The extracted value subsequence and the cabinet temperature subsequence are normalized (e.g., Min-Max or Z-score) to eliminate dimensions and facilitate subsequent processing.

[0121] S323, construct an input feature vector based on the normalized value subsequence, the normalized cabinet temperature subsequence, and the ambient temperature.

[0122] The normalized subsequence of values, the normalized subsequence of cabinet temperature, and the ambient temperature value are combined to form the final input feature vector. The final input is typically a three-dimensional tensor (number of samples, time step, feature dimension), where:

[0123] Feature dimension = number of target features (e.g., m) + 1 (cabinet temperature) + 1 (ambient temperature).

[0124] The ambient temperature is incorporated by time-step stitching: at each time step, the ambient temperature is added as a repeating feature value, which is suitable for scenarios where the ambient temperature changes gradually.

[0125] S330, The input feature vector is input into a pre-built temperature prediction model to obtain the predicted temperature of multiple components, wherein the temperature prediction model is built based on the operating data of a reactive power compensation device in a healthy state;

[0126] The method for constructing the temperature prediction model in this application is as follows:

[0127] (1) Obtain control signal samples, cabinet temperature sequence samples, measured temperature sequence samples of multiple components and ambient temperature samples of a reactive power compensation device in a healthy state at multiple historical time points, wherein the cabinet temperature sequence sample is the average value of the measured temperature sequence samples of multiple components.

[0128] The model needs to learn the input-output mapping relationship under normal operating conditions. Only by training on health data can the model accurately predict the "temperature that should be under normal conditions," and subsequent residual analysis can be used to identify anomalies.

[0129] Therefore, this application obtains historical operating data of the reactive power compensation device when it is in a healthy state (i.e., no faults, no aging, and normal heat dissipation), including control signal samples, cabinet temperature sequence samples, measured temperature sequence samples of multiple components, and ambient temperature samples. Among them, the cabinet temperature sequence sample is defined as the average value of the measured temperature sequences of multiple components.

[0130] (2) Extract target feature samples from the control signal samples, and construct input feature samples based on the value sequence of the target feature samples, the cabinet temperature sequence samples, and the ambient temperature samples;

[0131] Extract the selected target features (such as reactive current reference value, switching frequency, etc.) from the control signal sample, and construct the input feature sample consisting of the target feature value sequence, the cabinet temperature sequence, and the ambient temperature according to the methods described in S321~S323.

[0132] (3) Extract the measured temperature value corresponding to the end time point of the control signal sample from the measured temperature sequence samples of multiple components to obtain the sample label; and construct training data and training datasets of multiple historical time points based on the input feature samples and sample labels.

[0133] From the measured temperature sequences of multiple components, the temperature value corresponding to the end time of the control signal is extracted and used as the label for that sample. For Each component has a label for each sample. Dimensional vector.

[0134] (4) The LSTM model is trained based on the training dataset to obtain the temperature prediction model.

[0135] Using the pre-constructed training dataset, the LSTM model is trained through supervised learning to obtain a temperature prediction model. Specifically, mean squared error (MSE) or mean absolute error (MAE) can be used to optimize the prediction accuracy of all components simultaneously.

[0136] After obtaining the temperature prediction model, the input vector constructed earlier is fed into the model to obtain the predicted temperatures of multiple components under the assumption of a healthy state. This results in a... The predicted sequence of dimension is represented as:

[0137]

[0138] S340, calculate the difference between the actual temperature and the predicted temperature of multiple components to obtain the residual sequence. ; Residual sequence The mathematical expression is:

[0139]

[0140] From the first component to the... The residuals of each component.

[0141] Then from the residual sequence Extract residual distribution features; extract temperature difference distribution features from the actual temperatures of multiple components; and extract residual change features based on residual sequences from multiple maintenance cycles.

[0142] In this application, residual distribution characteristics, temperature difference distribution characteristics, and residual change characteristics are extracted as the basic data for the analysis of the reactive power compensation device. The extraction process is as follows:

[0143] (1) Characteristics of residual distribution

[0144] (1-1) Calculate the mean of the residuals in the residual sequence. residual standard deviation Maximum deviation characteristics The maximum deviation feature is used to point to the component with the largest absolute value of the residual;

[0145] Specifically, the formulas for calculating the residual mean, residual standard deviation, and maximum deviation characteristic are as follows:

[0146]

[0147]

[0148]

[0149] (1-2) Construct residual distribution features based on the residual mean, the residual standard deviation and the maximum deviation feature.

[0150] Final residual distribution characteristics It can be represented as:

[0151]

[0152] (2) Temperature difference distribution characteristics

[0153] (2-1) Calculate the range characteristics of the actual temperature of multiple components , wherein the range feature The mathematical expression is:

[0154]

[0155] In the formula, This indicates the maximum temperature. Indicates the minimum temperature;

[0156] (2-2) Obtain the coordinates of multiple components, and construct component feature vectors based on the coordinates of multiple components and the actual temperature;

[0157] (2-3) Cluster the feature vectors of multiple components to obtain one or more feature vector clusters;

[0158] Specifically, DBSCAN is used for clustering. Before clustering, the data needs to be normalized to naturally group components with similar features together, forming one or more feature vector clusters.

[0159] (2-4) Extract the target cluster with the smallest number from one or more feature vector clusters, and calculate the ratio of the number of components in the target cluster to the total number of components. ;

[0160] (2-5) Based on the aforementioned range characteristics and component numbers within the target cluster and the ratio Constructing temperature difference distribution characteristics, temperature difference distribution characteristics Represented as:

[0161]

[0162] (3) Characteristics of residual changes

[0163] (3-1) Obtain the target residual sequence for multiple maintenance cycles, wherein the target residual sequence is the residual sequence generated by the first control command after maintenance is completed, and the multiple maintenance cycles include the target number of maintenance cycles before the current maintenance cycle;

[0164] In this application, after each maintenance is completed by the staff, the first control command is generated, and this is taken as a stage of health status. The residual sequence at this time is collected to reflect the lifespan degradation of each component.

[0165] (3-2) Linear fitting is performed on the residual trend of each component based on the target residual sequence of multiple maintenance cycles to obtain the residual change trend of each component. ;

[0166] In this application, a linear fitting equation is constructed for each component:

[0167]

[0168] In the formula, Indicates the first The residuals of each component Indicates the maintenance time cycle. This is the intercept.

[0169] This application uses linear fitting to calculate the residual change trend of each component over a recent period. This reflects the aging status of the components.

[0170] (3-3) Construct residual change characteristics based on the residual change trends of multiple components.

[0171] S350, based on the residual distribution characteristics and the temperature difference distribution characteristics, monitoring and analysis are performed to obtain heat dissipation and component damage monitoring results, and based on the residual change characteristics, monitoring and analysis are performed to obtain component aging monitoring results.

[0172] Finally, based on the above residual distribution characteristics, temperature difference distribution characteristics, and residual variation characteristics, the following analysis is conducted, including:

[0173] S351, compare the residual mean with a preset residual tolerance threshold, compare the residual standard deviation with a preset standard deviation threshold, and compare the range feature with a preset range threshold;

[0174] Scenario 1, Normal Operation:

[0175] S352, in the residual mean Less than or equal to the preset residual tolerance threshold The residual standard deviation Less than or equal to the preset standard deviation threshold At that time, and the value of the range feature Less than or equal to the preset range threshold At that time, it was determined that the heat dissipation conditions and operating status of the reactive power compensation device were normal;

[0176] When both conditions are met: , , At that time, the residuals of all components in the instruction manual were consistent and close to zero, indicating that the heat dissipation system was working normally, with no local blockages or device malfunctions. The small temperature difference between components reflected a uniform thermal field distribution within the cabinet, with no hot spots. The actual temperature was highly consistent with the predicted value of the health model, indicating that the heat generation and dissipation under the current operating conditions were within the design expectation range, thus determining that the device was in a healthy state and required no intervention.

[0177] The above thresholds can be set based on the statistical distribution of historical health data (such as mean ± 3σ) or through expert experience.

[0178] Scenario 2: Abnormal overall heat dissipation

[0179] S353, in the residual mean Greater than the preset residual tolerance threshold And the residual standard deviation Less than or equal to the preset standard deviation threshold At that time, it was determined that the reactive power compensation device had an overall heat dissipation abnormality;

[0180] The residual mean is large, the standard deviation is small, and the temperature difference distribution is generally high but uniform. In this case, the actual temperature of all components is generally higher than the predicted value, and the deviation range is similar, indicating an abnormal overall temperature rise, but consistency between components remains. This phenomenon is usually caused by a decrease in global heat dissipation capacity, such as: cooling fan failure or reduced speed; clogged air inlet filter; or excessively high ambient temperature leading to reduced heat dissipation efficiency. Since the residual dispersion of each component is small, local faults or aging of individual components can be ruled out, thus the problem is located in the overall heat dissipation system.

[0181] S354, when the residual mean is greater than a preset residual tolerance threshold and the residual standard deviation is greater than a preset standard deviation threshold, the ratio is... Compare with a preset ratio threshold;

[0182] The large mean and standard deviation of the residuals, along with the generally high and uneven temperature distribution, make it impossible to directly determine the cause of the large mean residuals. Therefore, further analysis is required.

[0183] S3541, in the ratio When the ratio exceeds a preset threshold, it is determined that there is a local heat dissipation anomaly;

[0184] If the mean and standard deviation of the residuals are both large, and the ratio is satisfied... When the ratio exceeds a preset threshold, it indicates that some areas (such as a section of air duct or a group of capacitors) have experienced significantly higher component temperatures due to deteriorated local heat dissipation conditions, while other areas remain normal. The clustering algorithm uses component spatial coordinates and temperature values ​​to group components with high temperatures and adjacent locations into a single cluster. A larger γ indicates that the abnormal area covers a certain range and is not an isolated point. Common causes include: localized air duct blockage or uneven fan airflow distribution; localized dust accumulation on the heat sink or aging of the thermal grease; and uneven load in the area that is captured as an anomaly by the model. If it is determined to be a regional heat dissipation fault, the heat dissipation structure of that area needs to be specifically inspected.

[0185] S3542, in the ratio When the deviation is less than or equal to a preset proportional threshold, the maximum deviation feature is matched with the component number within the target cluster;

[0186] If the ratio If the ratio is less than or equal to the preset threshold, it is necessary to further determine whether it is caused by a single bad pixel or by multiple scattered inflection points.

[0187] Scene 4:

[0188] S35421, when the maximum deviation feature matches the component number within the target cluster, the faulty component is located based on the maximum deviation feature.

[0189] The residual mean is large (possibly just exceeding the threshold), the standard deviation is large, and the range is large. Temperature difference distribution clustering identifies a single high-temperature component as an isolated cluster, γ is very small, and this point happens to be the point of maximum residual. In this case, the temperature of a single component may be much higher than the predicted value, while surrounding components are normal, indicating that the component itself generates excessive heat or the heat dissipation path is interrupted. Possible causes include: thermal breakdown or increased leakage current in power devices (such as IGBTs), loose connections leading to increased contact resistance, localized heating, and a sharp increase in dielectric loss inside the capacitor.

[0190] Scene 5:

[0191] S35422, otherwise further check whether there are multiple components whose residuals are greater than the preset residual tolerance threshold. If yes, determine that there are multiple points of local heat dissipation abnormality. If no, determine that there is a single point of local heat dissipation abnormality.

[0192] The residuals have large mean, large standard deviation, and large range. The temperature difference distribution clustering shows multiple isolated high-temperature points (each forming its own cluster or noise). The γ value is small, and the maximum deviation point may not be within a target cluster, or even if it is in a cluster, there are other points exceeding the limit. Mismatch branch verification revealed that the residuals of multiple components exceeded the threshold.

[0193] The simultaneous occurrence of abnormalities in multiple non-adjacent components, and their unrelated nature, indicates the existence of multiple independent sources of failure. Possible causes include: synchronized aging of multiple components (e.g., close service life); poor contact or cold solder joints at multiple points; and batch quality issues causing simultaneous failure of multiple components. In such cases, a comprehensive investigation of all components exceeding limits is necessary, rather than addressing only the most severe one, to avoid overlooking potential problems.

[0194] In addition, this application also utilizes residual change characteristics to determine aging trends, including:

[0195] S355, compare the residual change trends of multiple components with a preset slope threshold; when the residual change trend of any component is greater than or equal to the preset slope threshold, determine that the any component has an aging trend.

[0196] The residual increases slowly over time (e.g., due to increased wear caused by component aging), and the slope remains positive, which can be used to predict the remaining lifespan.

[0197] This application presents an intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants. Utilizing an LSTM model trained on health data, and combining control signals, cabinet internal temperature, and ambient temperature to construct input features, it can accurately predict the theoretical temperature of components under specific operating conditions. By comparing actual temperatures to generate residuals, the degree of deviation can be quantified, providing a basis for early anomaly detection. It not only analyzes the residual distribution but also integrates the temperature difference distribution characteristics between components and the residual change trend across cycles. This three-dimensional analysis of "space (temperature difference) + time (trend) + deviation (residual)" can effectively distinguish between heat dissipation system failures, sudden component damage, and gradual component aging, avoiding misjudgments based on a single indicator. This application solves the problem of delayed fault detection in existing technologies and can accurately analyze various abnormal situations, helping maintenance personnel to perform automated, end-to-end monitoring of reactive power compensation devices.

[0198] like Figure 4 As shown, this application also provides an intelligent monitoring system for high-voltage reactive power compensation devices in new energy power plants, including:

[0199] The acquisition module is used to acquire the control signal, the temperature sequence inside the cabinet and the ambient temperature outside the cabinet when the controller of the reactive power compensation device outputs the control signal, and to acquire the actual temperature of multiple components inside the reactive power compensation device when the control signal ends.

[0200] The first feature extraction module is used to extract the value sequence of the target feature from the control signal, and construct an input feature vector based on the value sequence of the target feature, the cabinet temperature sequence, and the ambient temperature, wherein the type of the target feature is obtained in advance through historical data analysis;

[0201] The temperature prediction module is used to input the input feature vector into a pre-built temperature prediction model to obtain the predicted temperature of multiple components, wherein the temperature prediction model is built based on the operating data of the reactive power compensation device in a healthy state.

[0202] The second feature extraction module is used to calculate the difference between the actual temperature and the predicted temperature of multiple components to obtain a residual sequence; extract residual distribution features from the residual sequence; extract temperature difference distribution features from the actual temperature of multiple components; and extract residual change features based on the residual sequence of multiple maintenance cycles.

[0203] The monitoring and analysis module is used to perform monitoring and analysis based on the residual distribution characteristics and the temperature difference distribution characteristics to obtain heat dissipation and component damage monitoring results, and to perform monitoring and analysis based on the residual change characteristics to obtain component aging monitoring results.

[0204] This application presents an intelligent monitoring system for high-voltage reactive power compensation devices in new energy power plants. Utilizing an LSTM model trained on health data, and combining control signals, cabinet internal temperature, and ambient temperature to construct input features, it can accurately predict the theoretical temperature of components under specific operating conditions. By comparing actual temperatures to generate residuals, the degree of deviation can be quantified, providing a basis for early anomaly detection. It not only analyzes the residual distribution but also integrates the temperature difference distribution characteristics between components and the residual change trend across cycles. This three-dimensional analysis of "space (temperature difference) + time (trend) + deviation (residual)" can effectively distinguish between heat dissipation system failures, sudden component damage, and gradual component aging, avoiding misjudgments based on a single indicator. This application solves the problem of delayed fault detection in existing technologies and can accurately analyze various abnormal situations, helping maintenance personnel to perform automated, end-to-end monitoring of reactive power compensation devices.

[0205] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0206] The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory so that the terminal performs any of the methods in this embodiment.

[0207] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0208] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.

[0209] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0210] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0211] In the above embodiments, although the present application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art based on the foregoing description. The embodiments of the present application are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.

[0212] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. An intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants, characterized in that, Including the following steps: When the controller of the reactive power compensation device outputs a control signal, the control signal, the temperature sequence inside the cabinet, and the ambient temperature outside the cabinet are obtained, and the actual temperature of multiple components inside the reactive power compensation device is obtained when the control signal ends. Extract the value sequence of the target feature from the control signal, and construct an input feature vector based on the value sequence of the target feature, the cabinet temperature sequence, and the ambient temperature. The type of the target feature is obtained in advance through historical data analysis. The input feature vector is input into a pre-built temperature prediction model to obtain the predicted temperature of multiple components. The temperature prediction model is built based on the operating data of a reactive power compensation device in a healthy state. The difference between the actual temperature and the predicted temperature of multiple components is calculated to obtain a residual sequence; residual distribution features are extracted from the residual sequence; temperature difference distribution features are extracted from the actual temperature of multiple components; and residual change features are extracted based on the residual sequence of multiple maintenance cycles. Based on the residual distribution characteristics and the temperature difference distribution characteristics, monitoring and analysis are performed to obtain heat dissipation and component damage monitoring results. Based on the residual change characteristics, monitoring and analysis are performed to obtain component aging monitoring results.

2. The intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants according to claim 1, characterized in that, The method for determining the target features includes: Acquire control signal samples and measured temperature rise samples from multiple historical time points; The control signal samples and measured temperature rise samples from multiple historical time points are preprocessed to obtain control signal preprocessed data and temperature rise preprocessed data. The preprocessing includes outlier removal and missing value interpolation. Multiple feature parameters are extracted from the preprocessed control signal data, including reactive current reference value, reactive power, switching frequency, PWM duty cycle, DC bus voltage, and actual DC bus voltage. A temperature rise sequence was constructed based on measured temperature rise samples from multiple historical time points, and multiple feature parameter sequences were constructed based on the feature values ​​of each feature parameter at multiple historical time points. Each feature parameter sequence is normalized to obtain a normalized parameter sequence; and the correlation between each normalized parameter sequence and the temperature rise sequence is calculated. The relevance The mathematical expression is: In the formula, For historical time point indexing, For the number of historical time points, For the first Feature values ​​at each historical time point This represents the average value of the characteristic parameters. For the first Temperature rise at a historical point in time, This represents the average temperature rise. The correlation between each feature parameter and temperature rise is compared with a preset correlation threshold, and the feature parameters that are greater than or equal to the correlation threshold are taken as target features.

3. The intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants according to claim 1, characterized in that, An input feature vector is constructed based on the value sequence of the target feature, the temperature sequence inside the cabinet, and the ambient temperature, including: The value sequence of the target feature is time-aligned with the temperature sequence inside the cabinet, and the target duration before the end time of the control command is extracted to obtain a value subsequence and a temperature subsequence inside the cabinet with fixed length. The value subsequence and the cabinet temperature subsequence are normalized to obtain normalized value subsequence and normalized cabinet temperature subsequence. An input feature vector is constructed based on the normalized value subsequence, the normalized cabinet temperature subsequence, and the ambient temperature.

4. The intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants according to claim 1, characterized in that, The method for constructing the temperature prediction model includes: The control signal samples, cabinet temperature sequence samples, measured temperature sequence samples of multiple components, and ambient temperature samples of the reactive power compensation device in a healthy state at multiple historical time points are obtained, wherein the cabinet temperature sequence samples are the average values ​​of the measured temperature sequence samples of multiple components. Target feature samples are extracted from the control signal samples, and input feature samples are constructed based on the value sequence of the target feature samples, the cabinet temperature sequence samples, and the ambient temperature samples. The measured temperature values ​​corresponding to the end time points of control signal samples are extracted from the measured temperature sequence samples of multiple components to obtain sample labels; and training data and training datasets for multiple historical time points are constructed based on the input feature samples and sample labels. The LSTM model is trained based on the training dataset to obtain a temperature prediction model.

5. The intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants according to claim 1, characterized in that, Extracting residual distribution features from the residual sequence includes: Calculate the residual mean, residual standard deviation, and maximum deviation feature of the residual sequence, wherein the maximum deviation feature is used to point to the component with the largest absolute value of the residual; The residual distribution features are constructed based on the residual mean, the residual standard deviation, and the maximum deviation feature.

6. The intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants according to claim 5, characterized in that, Extract temperature difference distribution characteristics from the actual temperatures of multiple components, including: Calculate the range characteristics of the actual temperature of multiple components , wherein the range feature The mathematical expression is: In the formula, This indicates the maximum temperature. Indicates the minimum temperature; Obtain the coordinates of multiple components, and construct component feature vectors based on the coordinates of multiple components and the actual temperature; Clustering the feature vectors of multiple components yields one or more feature vector clusters; Extract the target cluster with the smallest number from one or more feature vector clusters, and calculate the ratio of the number of components within the target cluster to the total number of components. ; Based on the range characteristics, the component number within the target cluster, and the ratio. Construct temperature difference distribution characteristics.

7. The intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants according to claim 1, characterized in that, Residual variation features are extracted from residual sequences based on multiple maintenance cycles, including: Obtain the target residual sequence for multiple maintenance cycles. The target residual sequence is the residual sequence generated by the first control command after maintenance is completed. The multiple maintenance cycles include the target number of maintenance cycles before the current maintenance cycle. Linear fitting is performed on the residual trend of each component based on the target residual sequence of multiple maintenance cycles to obtain the residual change trend of each component. ; Residual variation characteristics are constructed based on the residual variation trends of multiple components.

8. The intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants according to claim 6, characterized in that, Based on the residual distribution characteristics and the temperature difference distribution characteristics, monitoring and analysis are performed to obtain heat dissipation and component damage monitoring results, including: The mean of the residuals is compared with a preset residual tolerance threshold, the standard deviation of the residuals is compared with a preset standard deviation threshold, and the range feature is compared with a preset range threshold. When the mean residual is less than or equal to a preset residual tolerance threshold, the standard deviation of the residual is less than or equal to a preset standard deviation threshold, and the value of the range characteristic is less than or equal to a preset range threshold, the heat dissipation conditions and operating status of the reactive power compensation device are determined to be normal. When the mean residual is greater than a preset residual tolerance threshold and the standard deviation of the residual is less than or equal to a preset standard deviation threshold, it is determined that the reactive power compensation device has an overall heat dissipation abnormality. When the residual mean is greater than a preset residual tolerance threshold and the residual standard deviation is greater than a preset standard deviation threshold, the ratio is... Compare with the preset ratio threshold; The ratio When the ratio exceeds a preset threshold, a localized heat dissipation anomaly is determined; at the ratio... When the deviation is less than or equal to a preset proportional threshold, the maximum deviation feature is matched with the component number within the target cluster; When the maximum deviation feature matches the component number within the target cluster, the faulty component is located based on the maximum deviation feature. Otherwise, it is further verified whether there are multiple components whose residuals are greater than a preset residual tolerance threshold. If so, it is determined that there are multiple points of local heat dissipation anomaly. If not, it is determined that there is a single point of local heat dissipation anomaly.

9. The intelligent monitoring method for high-voltage reactive power compensation devices in new energy power plants according to claim 7, characterized in that, Based on the residual change characteristics, monitoring and analysis are performed to obtain component aging monitoring results, including: The residual variation trends of multiple components are compared with the preset slope threshold; When the residual change trend of any component is greater than or equal to a preset slope threshold, it is determined that the arbitrary component has an aging trend.

10. An intelligent monitoring system for high-voltage reactive power compensation devices in new energy power plants, characterized in that, include: The acquisition module is used to acquire the control signal, the temperature sequence inside the cabinet and the ambient temperature outside the cabinet when the controller of the reactive power compensation device outputs the control signal, and to acquire the actual temperature of multiple components inside the reactive power compensation device when the control signal ends. The first feature extraction module is used to extract the value sequence of the target feature from the control signal, and construct an input feature vector based on the value sequence of the target feature, the cabinet temperature sequence, and the ambient temperature, wherein the type of the target feature is obtained in advance through historical data analysis; The temperature prediction module is used to input the input feature vector into a pre-built temperature prediction model to obtain the predicted temperature of multiple components, wherein the temperature prediction model is built based on the operating data of the reactive power compensation device in a healthy state. The second feature extraction module is used to calculate the difference between the actual temperature and the predicted temperature of multiple components to obtain a residual sequence; extract residual distribution features from the residual sequence; extract temperature difference distribution features from the actual temperature of multiple components; and extract residual change features based on the residual sequence of multiple maintenance cycles. The monitoring and analysis module is used to perform monitoring and analysis based on the residual distribution characteristics and the temperature difference distribution characteristics to obtain heat dissipation and component damage monitoring results, and to perform monitoring and analysis based on the residual change characteristics to obtain component aging monitoring results.

Images