A reactive power compensation device fault automatic early warning method

Abstract

The application discloses a kind of reactive power compensation device fault automatic early warning method, belong to reactive power compensation device power grid power factor correction technical field.This method collects multiple operating data, and early warning analysis is carried out to electric quantity data and operating data, and generates abnormal early warning.This method constructs and trains fault analysis diagnosis model according to historical fault data and fault case base, and inputs model after early warning information is fused to multiple-source data, obtains final diagnosis result.Finally, according to the diagnosis result, reactive power compensation strategy and disposal suggestion are retrieved from database.The application adopts multi-source data fusion and intelligent diagnosis, can realize comprehensive, depth early warning and diagnosis, significantly improves power grid power quality and fault disposal efficiency, reduces operation and maintenance cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of reactive power compensation device power factor correction, in particular to a reactive power compensation device fault automatic early warning method. BACKGROUND

[0002] The reactive power compensation device is a key equipment in the power system for improving the power factor of the power grid, reducing line loss and improving voltage quality. It provides capacitive reactive power to the power grid through the switching of capacitor banks or the adjustment of static reactive power generators, to balance the reactive power consumed by inductive loads. Real-time monitoring of the operating state of the reactive power compensation device and timely detection of potential faults are crucial for ensuring the safe, stable and economic operation of the power system. Existing reactive power compensation device fault warning methods usually rely on simple threshold monitoring of the device's own operating parameters. For example, when the current of the capacitor exceeds the set value or the temperature is too high, the system will issue an alarm. Some methods also monitor the power factor of the power grid, and trigger an alarm when the power factor is below the set lower limit. These methods are mainly based on static threshold judgment of a single data source, and use a pre-set fixed rule base to issue fault alarms.

[0003] However, the above-mentioned prior art has obvious technical defects. Since only a single data source is used for judgment, when the power grid load fluctuates normally, it may also cause the power factor to decrease temporarily or the device parameters to jump instantaneously, thus causing frequent false alarms. Conversely, for some early faults that develop slowly and have no obvious characteristics, such as slight dielectric aging of the capacitor or progressive poor contact of the connection point, the parameter changes may be within the normal range for a long time, making it difficult for existing methods to effectively detect them, resulting in missed alarms until the fault is discovered. SUMMARY

[0004] To solve the above problems, the present application provides a reactive power compensation device fault automatic early warning method, which combines multi-source data fusion analysis with a time series diagnosis model, and can realize early and accurate warning and self-adaptive disposal of the faults of the reactive power compensation device.

[0005] The above-mentioned object can be achieved by the following scheme:

[0006] The method comprises the following steps: collecting real-time operation data of a reactive power compensation device, real-time power data of a power grid, historical data, fault data and a fault case library; performing early warning analysis on the real-time power data, and obtaining power data abnormal early warning when the real-time power data of the power grid exceeds a preset threshold; performing fault early warning analysis on the real-time operation data and a key parameter change trend, and obtaining early fault hidden device fault early warning; constructing and training a fault analysis diagnosis model by using a time sequence algorithm according to the fault data and the fault case library, and obtaining a trained fault analysis diagnosis model; performing multi-source data fusion on the power data abnormal early warning and the device fault early warning, generating comprehensive fault diagnosis and inputting the comprehensive fault diagnosis into the fault analysis diagnosis model to generate a final fault diagnosis result; and according to the final fault diagnosis result, calling corresponding reactive power compensation strategies and fault disposal suggestions from a database preset with a plurality of reactive power compensation strategies and fault disposal suggestions to realize correction of a power factor in power data.

[0007] Optionally, the method further comprises: automatically adjusting operation parameters of the reactive power compensation device through a closed-loop control mechanism according to the output reactive power compensation strategy to realize correction of the power factor; and obtaining operation effect data after the operation parameters of the reactive power compensation device are adjusted, and updating the final fault diagnosis result, the executed reactive power compensation strategy and the operation effect data as new fault cases into the fault case library.

[0008] Optionally, the collecting of the real-time operation data of the reactive power compensation device, the real-time power data of the power grid, the historical data, the fault data and the fault case library comprises: the real-time operation data comprises temperature, vibration and noise during operation of the reactive power compensation device; the real-time power data of the power grid comprises reactive power, active power, a power factor, harmonic current, negative sequence current and three-phase voltage of the power grid; the historical data comprises historical operation data and historical power data of the reactive power compensation device; the fault data is historical fault of the reactive power compensation device, and comprises power data abnormal fault and device fault; and the fault case library is constructed based on analysis and mining of the fault data, and stores fault type labels, evolution rules and verified historical disposal strategies corresponding to different fault types in an associated form.

[0009] Optionally, the early warning analysis on the real-time power data comprises: setting preset thresholds of the power factor, the reactive power, the harmonic current, the negative sequence current and the three-phase voltage according to the historical data; determining whether the power factor of the power grid continuously exceeds the preset threshold or any one of the reactive power, the harmonic current, the negative sequence current and the three-phase voltage continuously exceeds a safety limit value specified by a related standard or continuously exceeds the respective preset threshold, and if yes, generating the power data abnormal early warning, wherein the early warning is used to indicate power factor abnormality or power quality decline of the power grid, and if no, continuously monitoring the power data.

[0010] Optionally, the real-time operation data is analyzed for early fault warning by a key parameter change trend to obtain an early fault hidden danger device fault warning, comprising: constructing a machine learning model for anomaly detection, unsupervised training the machine learning model with the historical operation data to obtain a normal operation baseline model; inputting the real-time operation data into the normal operation baseline model for analysis, calculating a deviation value between the real-time operation data and the normal operation baseline, wherein the deviation value is used to represent the degree of deviation of the real-time operation data from the normal state; when the deviation value continuously exceeds a normal fluctuation range, obtaining an early fault hidden danger device fault warning, the normal fluctuation range being obtained from the historical operation data and a fault case library.

[0011] Optionally, the machine learning model for anomaly detection is constructed, and the machine learning model is unsupervised trained with the historical operation data, comprising: statistically analyzing the historical operation data, and extracting statistical features of the historical operation data, wherein the statistical features include mean, variance, standard deviation and kurtosis of the operation data; unsupervised training the machine learning model with the statistical features to obtain a normal baseline model capable of calculating a deviation value between the operation data and the normal operation baseline.

[0012] Optionally, the fault analysis diagnosis model is constructed and trained by using a time series algorithm according to the fault data and the fault case library to obtain a trained fault analysis diagnosis model, comprising: constructing an initial fault analysis diagnosis model capable of learning time series dependency relationship by using a long short-term memory network algorithm; mapping the fault data and corresponding fault type labels in the fault case library to form a training sample set; iteratively training the initial fault analysis diagnosis model using the training sample set, and optimizing model parameters until its diagnosis performance converges to obtain the trained fault analysis diagnosis model.

[0013] Optionally, the power data anomaly warning and the device fault warning are fused by multi-source data to generate a comprehensive fault diagnosis and input into the fault analysis diagnosis model to generate a final fault diagnosis result, comprising: assigning corresponding confidence and priority weight to the power data anomaly warning and the device fault warning, performing weighted decision processing to generate fusion data; formatting the fusion data to generate a comprehensive fault diagnosis capable of being received by the trained fault analysis diagnosis model; inputting the comprehensive fault diagnosis into the fault analysis diagnosis model, and the fault analysis diagnosis model outputs the final fault diagnosis result.

[0014] Optionally, according to the final fault diagnosis result, corresponding reactive power compensation strategy and fault handling suggestion are retrieved from a preset database storing a plurality of reactive power compensation strategies and fault handling suggestions, including: using the fault type label contained in the final fault diagnosis result as a retrieval index; in the database storing a plurality of reactive power compensation strategies and fault handling suggestions, matching and retrieving corresponding reactive power compensation strategy and fault handling suggestion through the retrieval index; combining the reactive power compensation strategy and the fault handling suggestion into a structured handling scheme, and outputting.

[0015] Based on the same inventive concept, the application also provides an automatic early warning system for reactive power compensation device faults, which comprises: a data acquisition module for acquiring real-time operation data of a reactive power compensation device, real-time power data of a power grid, historical data, fault data and a fault case library; a warning analysis module for performing warning analysis on the real-time power data to generate power data anomaly warning, and performing key parameter change trend analysis on the real-time operation data to generate device fault warning; a model construction module for constructing and training a fault analysis and diagnosis model based on the fault data and the fault case library by using a time series algorithm to obtain a trained fault analysis and diagnosis model; a data fusion module for performing multi-source data fusion on the power data anomaly warning and the device fault warning to generate comprehensive fault diagnosis; a diagnosis analysis module for inputting the comprehensive fault diagnosis into the trained fault analysis and diagnosis model to obtain a final fault diagnosis result; and a handling output module for retrieving and outputting corresponding reactive power compensation strategy and fault handling suggestion from a preset database storing a plurality of reactive power compensation strategies and fault handling suggestions according to the final fault diagnosis result.

[0016] Compared with the prior art, the application has the following advantages:

[0017] 1. The application realizes the cooperative judgment from macro power quality fluctuation to micro device state anomaly by analyzing power grid power data and device operation data in parallel and inputting the generated double early warning signals into a diagnosis model after fusion. This multi-dimensional and multi-level analysis method effectively distinguishes external power grid disturbance and device self-fault, significantly improves the accuracy of fault early warning, and avoids false alarm or missed alarm caused by misjudgment of a single data source.

[0018] 2、The fault analysis and diagnosis model constructed by the application utilizes the deep learning ability of the time series algorithm, can autonomously explore and understand the dynamic fault evolution mode hidden behind the complex operation data. This makes the diagnosis result no longer limited to the simple "yes" or "no" judgment, but can accurately identify the specific fault type, lays a solid foundation for subsequent targeted and efficient disposal measures, realizes the intelligent upgrading from "finding problems" to "locating the root cause".

[0019] 3、The application establishes a complete closed loop from intelligent diagnosis to automatic disposal and self-optimization. Not only can the optimal reactive power compensation strategy be automatically retrieved and executed to quickly stabilize the power grid according to the accurate diagnosis result, but also the experience of each fault handling can be fed back to the fault case library for continuous learning and iteration of the model. This adaptive and self-evolutionary ability ensures that the system can maintain high efficiency and reliability for a long time and continuously adapt to changes in the power grid environment.

[0020] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the application. The objects and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0022] Figure 1 is a flowchart of a reactive power compensation device fault automatic early warning method according to an embodiment of the present application.

[0023] Figure 2 is a fault early warning logic flowchart according to an embodiment of the present application.

[0024] Figure 3 is a model training and diagnosis flowchart according to an embodiment of the present application.

[0025] Figure 4 is a structure diagram of a reactive power compensation device fault automatic early warning system according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0027] With reference to Figure 1 One embodiment of the present application proposes a reactive power compensation device fault automatic early warning method. The method combines multi-source data fusion analysis and time series diagnosis model, can realize early and accurate early warning and self-adaptive disposal of reactive power compensation device fault, significantly improves the power quality and fault disposal efficiency of the power grid, and reduces the operation and maintenance cost and the professional ability requirement of the operation personnel.

[0028] The method of the embodiment specifically comprises:

[0029] Collecting real-time operation data of the reactive power compensation device, real-time power data of the power grid, historical data, fault data and a fault case library;

[0030] Specifically, the data collection acquires physical parameters of the reactive power compensation device during operation, such as thermal state, vibration amplitude and acoustic characteristics. At the same time, the power quality parameters of the power grid are obtained from real-time monitoring, such as reactive power, harmonic content and voltage level. These data can reflect the overall power quality of the power grid. In addition, long-term historical operation data, historical fault data and a case library storing different fault type labels, evolution law and historical disposal experience of the reactive power compensation device are also obtained.

[0031] Performing early warning analysis on the real-time power data, and obtaining a power data abnormal early warning when the real-time power data of the power grid exceeds a preset threshold;

[0032] Specifically, the collected real-time power data of the power grid is continuously analyzed and compared with the preset threshold. When any data of the power factor, reactive power, harmonic content and voltage level of the power grid exceeds the safety limit value or continuously exceeds the respective preset threshold, an abnormal power data early warning signal is generated. The early warning is used to indicate that the power factor of the power grid is abnormal or the power quality is decreased. If all the power data is within the normal range, the monitoring will be continuously performed. Figure 2 As shown, the power data abnormal early warning and the equipment fault early warning in the method are intuitively displayed, and how they jointly generate comprehensive fault diagnosis through multi-source data fusion.

[0033] The real-time operation data is analyzed for early fault warning based on a key parameter change trend to obtain early fault warning of the device;

[0034] Specifically, the collected real-time operation data of the reactive power compensation device is continuously analyzed, a mathematical model capable of identifying abnormal data patterns is constructed to learn the data characteristics of the device in a normal operation state, and a mathematical baseline model is established. The real-time operation data is input into the mathematical baseline model for analysis, and the deviation value between the real-time operation data and the mathematical baseline model is calculated. When the deviation value continuously exceeds the normal fluctuation range, early fault warning of the device is generated.

[0035] According to the fault data and the fault case library, a fault analysis and diagnosis model is constructed and trained using a time series algorithm to obtain a trained fault analysis and diagnosis model.

[0036] Specifically, based on the fault data and the fault case library, an initial fault analysis and diagnosis model is constructed using a time series algorithm capable of learning dynamic dependencies in time series data. The fault data and the corresponding fault type labels in the fault case library are mapped to form a structured training sample set. The initial fault analysis and diagnosis model is iteratively trained using the training sample set, and the model parameters are continuously optimized until the diagnostic performance converges, and finally a trained model capable of accurately diagnosing faults is obtained. As shown in Figure 3 The fault data and the fault case library work together to construct and train the fault analysis and diagnosis model, and finally receive comprehensive fault diagnosis input and output diagnostic results.

[0037] The power data anomaly warning and the device fault warning are fused to generate comprehensive fault diagnosis and input into the fault analysis and diagnosis model to generate the final fault diagnosis result.

[0038] Specifically, the power data anomaly warning and the device fault warning, which are two different sources of warning information, are fused to form comprehensive fault diagnosis. The fusion process assigns appropriate weights to different warnings to generate a fusion data that represents the comprehensive degree of faults. The fusion data is input into the trained fault analysis and diagnosis model, and the model outputs the final fault diagnosis result based on the input.

[0039] According to the final fault diagnosis result, the corresponding reactive power compensation strategy and fault handling suggestion are retrieved from a pre-set database storing a plurality of reactive power compensation strategies and fault handling suggestions, and the correction of the power factor in the power data is output.

[0040] Specifically, the fault type information contained in the final fault diagnosis result is used as a retrieval index. In a database storing various reactive power compensation strategies and fault handling suggestions, the corresponding reactive power compensation strategy and fault handling suggestion are matched and called through the retrieval index. The reactive power compensation strategy and fault handling suggestion are combined into a structured handling scheme, and output through a visual interface or remote communication to guide the operation and maintenance personnel to correct the power factor, so that the power grid operation returns to normal.

[0041] The combination of multi-source data fusion analysis and time series diagnosis model can realize early and accurate early warning and self-adaptive handling of reactive power compensation device faults, significantly improve the power quality and fault handling efficiency of the power grid, and reduce the operation and maintenance cost and the professional ability requirement of the operation personnel.

[0042] Optionally, the method further comprises:

[0043] According to the output reactive power compensation strategy, the operation parameters of the reactive power compensation device are automatically adjusted through a closed-loop control mechanism to correct the power factor;

[0044] Specifically, when the reactive power compensation strategy is output, a closed-loop control unit receives the strategy and automatically sends control instructions to the reactive power compensation device according to the strategy content. These control instructions can accurately adjust the switching state of the reactive power compensation device or adjust its output power to correct the power factor of the power grid in real time. For example, if the final diagnosis result is that the capacitor group capacity is insufficient, the control unit will instruct the reactive power compensation device to switch more capacitor groups to increase the reactive power output. Conversely, if the power grid voltage is detected to be too high, the control unit will reduce the reactive power output of the reactive power compensation device by adjusting its operation parameters.

[0045] The operation effect data after adjusting the operation parameters of the reactive power compensation device is obtained, and the final fault diagnosis result, the executed reactive power compensation strategy and the operation effect data are taken as a new fault case, which is updated to the fault case library.

[0046] Specifically, real-time operation data of the reactive power compensation device and real-time power data of the power grid are continuously obtained after the execution of the reactive power compensation strategy, and these data constitute the adjusted operation effect data. The final fault diagnosis result, the executed reactive power compensation strategy and these operation effect data are associated and packaged. The packaged data is used as a new fault case for updating the fault case library. The updating process realizes self-learning and continuous optimization, so that the accuracy of fault diagnosis and handling can be continuously improved over time.

[0047] Optionally, the collection of real-time operation data of the reactive power compensation device, real-time power data of the power grid, historical data, fault data and the fault case library comprises:

[0048] The real-time operation data includes temperature, vibration and noise of the reactive power compensation device during operation;

[0049] Specifically, by arranging various sensors on the key components of the reactive power compensation device, the temperature, vibration and noise signals during equipment operation are collected in real time. These sensors can be wired or use wireless communication mode to ensure the flexibility of data collection. For example, optical fiber temperature measurement technology or wireless temperature sensor is used to obtain the temperature data of the high-voltage cable connection point and the coil; vibration sensor is used to monitor the vibration of the transformer base and the shell; sound sensor is used to monitor abnormal noise, and these data can fully reflect the physical health status of the equipment.

[0050] The real-time power data of the power grid includes reactive power, active power, power factor, harmonic current, negative sequence current and three-phase voltage of the power grid;

[0051] Specifically, by configuring multifunctional electric meters or intelligent power collection terminals at the low-voltage side outlet of the distribution transformer and the grid-connected point of the reactive power compensation device, the key power parameters of the power grid are obtained in real time. These parameters can fully reflect the power quality condition of the power grid and the compensation effect of the reactive power compensation device. The collection frequency can be adjusted according to actual needs to ensure the real-time and accuracy of the data. These power data are not only used for immediate warning, but also serve as historical data to support subsequent model training.

[0052] The historical data includes historical operation data and historical power data of the reactive power compensation device;

[0053] Specifically, by storing the real-time operation data and real-time power data in the database of the cloud platform or local server for a long time, a complete historical database is formed. These data constitute an important basis for trend analysis, baseline model training and fault mode identification, which can help to understand the long-term operation rules of the equipment, so as to more accurately judge the potential fault hidden danger. The collection frequency of historical data can be lower than that of real-time data to save storage resources.

[0054] The fault data is the historical fault of the reactive power compensation device, including power data abnormal fault and equipment fault;

[0055] Specifically, when the reactive power compensation device fails in the past, the relevant historical fault information is recorded and automatically or manually uploaded. These historical fault records include two main types of information: one is the power data abnormal fault, which records the abnormal condition of the power factor, harmonic current or voltage of the power grid at a certain time point; the other is the equipment fault, which records the fault events of the capacitor bank damage, switch cabinet tripping or temperature overload and other equipment itself. These fault data constitute the key raw materials for building the fault case library.

[0056] The fault case library is constructed based on analysis and mining of the fault data, and stores the fault type labels corresponding to different fault types, evolution rules, and verified historical treatment strategies in a correlated form.

[0057] Specifically, the collected fault data is analyzed and mined to extract the characteristics of the fault signals, and combined with expert experience and historical maintenance records. The analysis and mining process associates different types of faults, such as capacitor dielectric aging, poor contact of cable joints, or protection misoperation, with specific fault type labels and evolution rules. Ultimately, this information, together with verified historical treatment strategies, is structured and stored in the fault case library, providing a knowledge base for diagnosis and treatment.

[0058] Optionally, the early warning analysis on the real-time power data includes:

[0059] According to the historical data, set the preset threshold values of power factor, reactive power, harmonic current, negative sequence current, and three-phase voltage respectively;

[0060] Specifically, by analyzing the historical power data of the reactive power compensation device during normal operation, statistical methods are used to statistically analyze each parameter. For example, the mean and standard deviation of each parameter can be calculated, and its dynamic fluctuation range can be determined according to the criteria, which is used as the preset threshold value. These threshold values can be stored in a configuration table and called during early warning analysis.

[0061] If the power factor of the power grid continuously exceeds the preset threshold value or any of the reactive power, harmonic current, negative sequence current, and three-phase voltage exceeds the safety limit value specified by the relevant standard or continuously exceeds the respective preset threshold value, an abnormal power data early warning signal is generated, wherein the early warning signal indicates that the power factor of the power grid is abnormal or the power quality is deteriorating, otherwise, the power data is continuously monitored.

[0062] Specifically, the real-time collected power data is continuously monitored. First, it is determined whether the power factor of the power grid continuously exceeds or falls below the set threshold value within a preset time period. At the same time, it also judges each of the other key parameters such as reactive power, harmonic current, negative sequence current, and three-phase voltage, to check whether these data exceed the safety limit value specified by the national or industry standard. As long as any of the above conditions is met, an abnormal power data early warning signal will be generated. The early warning signal will be marked as power factor abnormality or power quality deterioration for subsequent analysis. If all parameters do not exceed the limit value, continuous monitoring will be performed.

[0063] Optionally, the real-time operation data and the change trend of the key parameters are analyzed for fault early warning to obtain the equipment fault early warning of early fault hidden danger, including:

[0064] constructing a machine learning model for anomaly detection, unsupervised training the machine learning model with the historical operation data to obtain a normal operation baseline model;

[0065] Specifically, a machine learning model for anomaly detection can be constructed, which can be an unsupervised learning model, such as an isolation forest model or an autoencoder model. The historical operation data of the reactive power compensation device is used for training. The training process aims to learn and identify the internal patterns and data distribution characteristics of the device in the normal operation state. The purpose of training is to establish a mathematical model that can represent the normal operation behavior of the device, i.e. the normal operation baseline model.

[0066] inputting the real-time operation data into the normal operation baseline model for analysis, calculating the deviation value between the real-time operation data and the normal operation baseline, wherein the deviation value is used to represent the degree of deviation of the real-time operation data from the normal state;

[0067] Specifically, the real-time collected reactive power compensation device operation data is continuously input into the trained normal operation baseline model. The model calculates the deviation value between the input real-time data and the normal operation baseline according to the input real-time data. The deviation value can be calculated in multiple ways, such as Euclidean distance or Mahalanobis distance. Taking Euclidean distance as an example, if the real-time operation data vector is and the normal operation baseline data vector is the calculation formula of the deviation value

[0068]

[0069] wherein, represents the dimension of the data vector, and represent the values of the real-time operation data and the normal operation baseline data in the th dimension, respectively.

[0070] When the deviation value continuously exceeds the normal fluctuation range, an early fault hidden danger device failure warning is obtained, and the normal fluctuation range is obtained from the historical operation data and the fault case library.

[0071] Specifically, the deviation value is continuously monitored and compared with a normal fluctuation range obtained from historical data analysis. The determination process of the normal fluctuation range is through statistical analysis of the historical operation data of the reactive power compensation device, and combined with the known fault modes in the fault case library, an objective threshold is determined. When the deviation value continuously exceeds the normal fluctuation range for a period of time, an early fault hidden danger device failure warning is generated.

[0072] ​​Optionally, the step of constructing a machine learning model for anomaly detection, and using the historical running data to perform unsupervised training on the machine learning model, includes:

[0073] Statistical analysis is performed on the historical operating data to extract statistical features of the historical operating data, wherein the statistical features include the mean, variance, standard deviation and kurtosis of the operating data;

[0074] Specifically, historical operating data of the reactive power compensation device is retrieved from the database, including multiple time series data such as temperature, vibration, and noise. Then, statistical analysis is performed on each type of operating data to extract its key statistical characteristics. For example, for a set of historical operating data samples... }, mean The calculation formula is:

[0075] ,

[0076] variance The calculation formula is:

[0077] ,

[0078] in, Represents the total number of samples. Indicates the first These data samples provide a comprehensive overview of the device's behavior patterns under normal operating conditions.

[0079] The statistical features are used to perform unsupervised training on the machine learning model to obtain a normal baseline model that can calculate the deviation between the running data and the normal operating baseline.

[0080] Specifically, the extracted statistical features are used as input to train an unsupervised machine learning model. This model is a clustering model, and its training objective is to cluster data points belonging to the normal state in historical operating data. The training process aims to allow the model to learn the distribution patterns of normal data points, thereby forming a mathematical model that can represent the normal operating state of the equipment, i.e., the normal baseline model. The model can calculate the deviation value between the input real-time operating data and the normal baseline.

[0081] Optionally, based on the fault data and fault case library, a fault analysis and diagnosis model is constructed and trained using a time series algorithm, resulting in a trained fault analysis and diagnosis model including:

[0082] An initial fault analysis and diagnosis model capable of learning time-series dependencies is constructed using a long short-term memory network algorithm.

[0083] Specifically, the long short-term memory network algorithm is adopted to construct an initial fault analysis and diagnosis model capable of learning and memorizing long-term time dependence in fault data. The model mainly consists of an input gate, a forgetting gate, an output gate and a memory unit, and the core is to effectively process and analyze time series data. When constructing the model, the feature sequence of the fault data is taken as the input of the model, aiming to provide an initial structure with basic learning ability for the subsequent training process.

[0084] The fault data is mapped with the corresponding fault type label in the fault case library to form a training sample set.

[0085] Specifically, the fault data is sorted and preprocessed, and is one-to-one associated with the corresponding fault type label in the fault case library. This association process maps the fault data, such as the temperature, vibration or current anomaly record of the equipment in a certain period of time, with a unique fault type label. Through this mapping, a structured training sample set is formed, each sample of which contains a fault data and a corresponding fault type label, which provides the necessary training materials for supervised learning.

[0086] The initial fault analysis and diagnosis model is iteratively trained using the training sample set, and the model parameters are optimized until its diagnostic performance converges, obtaining a trained fault analysis and diagnosis model.

[0087] Specifically, the initial fault analysis and diagnosis model is iteratively trained using the training sample set. In the training process, the training sample is input into the model for forward propagation to obtain the predicted output, and then the predicted output is compared with the real fault type label to calculate the prediction error. According to the prediction error, the back propagation algorithm is used to adjust the parameters such as weights and biases of the model. The training process will continue, and through continuous optimization of the model parameters, the diagnostic performance of the model will no longer be significantly improved until the model reaches a state of convergence, and finally a trained fault analysis and diagnosis model with accurate diagnostic ability is obtained.

[0088] Optionally, the power data anomaly warning and equipment fault warning are fused to generate a comprehensive fault diagnosis and input into the fault analysis and diagnosis model to generate the final fault diagnosis result, including:

[0089] The power data anomaly warning and the equipment fault warning are assigned with corresponding confidence and priority weights for weighted decision processing to generate fusion data.

[0090] Specifically, when receiving both power data anomaly warning and equipment failure warning, a confidence level and a priority weight are assigned to each warning. The confidence level can be determined based on the reliability of the data source and the severity of the warning, for example, the confidence level of the warning caused by instantaneous jump is low, while the confidence level of the warning caused by continuous over-limit is high. The priority weight can be pre-set according to operation and maintenance experience, for example, the power data anomaly warning directly related to power grid power quality has higher priority than the equipment failure warning. These confidence levels and priority weights are used as inputs for weighted decision processing to generate a fusion data, which can reflect the comprehensive severity and nature of the current fault.

[0091] The fusion data is formatted to generate a comprehensive fault diagnosis that can be received by the trained fault analysis and diagnosis model.

[0092] Specifically, the fusion data is formatted to conform to the input format required by the trained fault analysis and diagnosis model. The formatting process includes unified processing and coding of warning information from different sources to generate a comprehensive fault diagnosis vector that can be understood by the model. The comprehensive fault diagnosis vector is used as the input for the final diagnosis of the model.

[0093] The comprehensive fault diagnosis is input into the fault analysis and diagnosis model, and the fault analysis and diagnosis model outputs the final fault diagnosis result.

[0094] Specifically, the formatted comprehensive fault diagnosis vector is input into the trained fault analysis and diagnosis model. The model performs in-depth analysis on the input and outputs the final fault diagnosis result. The final fault diagnosis result can be a specific fault type, such as "capacitor bank aging" or "switch cabinet overheating", or a probability of fault occurrence, which provides accurate guidance for subsequent disposal.

[0095] Optionally, according to the final fault diagnosis result, the corresponding reactive power compensation strategy and fault disposal suggestion are retrieved from a pre-set database storing a plurality of reactive power compensation strategies and fault disposal suggestions.

[0096] The fault type label contained in the final fault diagnosis result is used as a retrieval index.

[0097] Specifically, when the final fault diagnosis result is output, the result will contain one or more fault type labels, such as "capacitor bank aging" or "switch cabinet overheating". The result is automatically parsed to extract these unique fault type labels, and these labels are used as indexes for retrieval in the disposal scheme database.

[0098] In the database storing a plurality of reactive power compensation strategies and fault disposal suggestions, the corresponding reactive power compensation strategy and fault disposal suggestion are retrieved and retrieved through the retrieval index matching.

[0099] Specifically, the obtained search index is used to perform a quick match in a treatment scheme database. The database is a structured storage system in which different reactive compensation strategies and fault treatment suggestions are associated with corresponding fault type tags. When an entry matching the search index is found, the associated reactive compensation strategy and fault treatment suggestion are retrieved.

[0100] The reactive compensation strategy and the fault treatment suggestion are combined into a structured treatment scheme and output.

[0101] Specifically, the retrieved reactive compensation strategy and fault treatment suggestion are integrated to form a complete and structured treatment scheme. The treatment scheme can be a step-by-step guide detailing specific treatment steps, required tools, and estimated completion time. The treatment scheme is output in the form of a visual interface, a short message, or an email, etc. to guide maintenance and repair personnel to maintain and repair the reactive compensation device.

[0102] Based on the same inventive concept, as shown in Figure 4 The present application also provides an automatic early warning system for a reactive compensation device, which comprises:

[0103] a data acquisition module for acquiring real-time operation data of the reactive compensation device, real-time power data of the power grid, historical data, fault data, and a fault case library;

[0104] a warning analysis module for performing warning analysis on the real-time power data to generate a power data anomaly warning, and performing key parameter change trend analysis on the real-time operation data to generate a device fault warning;

[0105] a model construction module for constructing and training a fault analysis and diagnosis model based on the fault data and the fault case library using a time series algorithm to obtain a trained fault analysis and diagnosis model;

[0106] a data fusion module for performing multi-source data fusion on the power data anomaly warning and the device fault warning to generate a comprehensive fault diagnosis;

[0107] a diagnosis analysis module for inputting the comprehensive fault diagnosis into the trained fault analysis and diagnosis model to obtain a final fault diagnosis result;

[0108] a treatment output module for retrieving and outputting corresponding reactive compensation strategies and fault treatment suggestions from a database pre-stored with various reactive compensation strategies and fault treatment suggestions according to the final fault diagnosis result.

[0109] Embodiment 1

[0110] To verify the feasibility of the application in practice, the application is applied to a certain large steel rolling mill. Due to the use of large motors and electric arc furnaces and other impact loads, the power factor of the power grid of the rolling mill fluctuates sharply and is often lower than the requirement of the power supply department, resulting in electricity fee penalties; at the same time, the reactive power compensation device in the factory area occasionally has sudden failures due to long-term high-load operation, causing production interruption and safety hazards. The traditional alarm method based on a fixed threshold has a high false alarm rate and cannot provide effective early warning before the failure occurs. The rolling mill hopes to use the method of the application to intelligently perform fault early warning and closed-loop management on the reactive power compensation device cluster of the main substation.

[0111] In this embodiment, the rolling mill deploys the method of the application in its reactive power compensation system. Real-time operation data of each capacitor bank is comprehensively collected, such as capacitor switching state, temperature, branch current, real-time power data such as voltage, current, and power factor of the power grid access point, and historical operation data and archived fault maintenance records of the past year are integrated to form a multi-source data set. Before inputting for analysis, all data are subjected to time alignment and normalization preprocessing. The experimental comparison period is 6 months, and the operation effect of the application is compared with the traditional management method used previously.

[0112] In November 2024, it is monitored that the real-time power factor of the power grid frequently drops during the afternoon production peak period. Under the traditional method, a fixed threshold of 0.90 will generate a large number of false alarms during this period. However, through dynamic threshold early warning analysis, the application calculates the dynamic power factor threshold of this period to be 0.88 based on the historical data of the past seven days. When the real-time power factor drops to 0.87, it is determined to be a real anomaly, and an "electricity data anomaly early warning" is generated, effectively avoiding false alarms caused by normal load fluctuations.

[0113] In early December 2024, while analyzing the trend of changes in key parameters of the real-time operation data of capacitor group No. 5, early fault hazards are identified. Through a sliding window algorithm, it is calculated that the change trend of the intrinsic temperature of the capacitor in multiple consecutive time windows is continuously a small positive value, indicating that the capacitor has an abnormal heating trend, even if its absolute temperature value has not reached the fixed threshold of 65℃ of the traditional alarm. Accordingly, a "device fault early warning" is generated.

[0114] Subsequently, the above "electricity data anomaly early warning" and "device fault early warning" are subjected to multi-source data fusion processing. Through weighted decision-making, a fusion data with a high risk score is generated, and it is integrated with temperature, current, and other time series data to form a comprehensive fault diagnosis input. The input is sent to the long short-term memory network fault analysis and diagnosis model that has been trained. The final fault diagnosis result output by the model is: "capacitor overheating and aging".

[0115] According to the diagnosis result of "capacitor overheating aging", the corresponding treatment scheme was immediately retrieved from the fault treatment suggestion database. The reactive power compensation strategy is: "immediately remove capacitor group No. 5 through control and automatically put spare capacitor group No. 8 into operation". The fault treatment suggestion is: "generate a high-level maintenance work order and notify the operation and maintenance personnel to check the cooling fan status of capacitor group No. 5 and prepare to replace the capacitor".

[0116] The closed-loop control mechanism automatically parses the reactive power compensation strategy into control signals and issues them to the device controller, completing the removal and operation of the capacitor group within 1 minute. Subsequent monitoring data shows that the power factor of the power grid quickly rises above 0.95, and the correction effect is significant. The complete fault early warning, diagnosis, treatment and effect verification data are packaged into a new case and updated to the fault case library for future incremental learning and optimization of the model.

[0117] Through 6-month data comparison, the present application shows significant advantages in early warning accuracy, diagnosis and treatment efficiency, and power quality improvement.

[0118] Table 1 Comparison table of reactive power compensation device early warning accuracy

[0119]

[0120] Table 2 Comparison table of key fault diagnosis and treatment efficiency

[0121]

[0122] Table 3 Comparison table of power quality (power factor) improvement effect

[0123]

[0124] From the data in Table 1 above, the present application achieves an average early warning accuracy of more than 90% through dynamic threshold and trend analysis, which is significantly higher than the accuracy of less than 15% of traditional methods, greatly reducing the number of false alarms and enabling operation and maintenance personnel to focus on real fault risks.

[0125] In Table 2, the rapidity and intelligence of the system in fault diagnosis and response are demonstrated. For early risks such as "capacitor overheating aging", the present application can complete diagnosis and automatic treatment within a few minutes, avoiding the production interruption and equipment damage risks caused by traditional manual troubleshooting for several hours, and achieving predictive maintenance.

[0126] The data results of Table 3 fully prove the excellent effect of the present application in improving power quality. Through fast and accurate fault disposal and closed-loop control, the system improves the average power factor from 0.89 to above 0.96, and the power factor compliance rate from about 75% to above 99%, saving a large amount of electricity fee penalties for the enterprise and guaranteeing the stable and reliable operation of the power grid.

[0127] It should be noted that the electrical connection between the above-mentioned various units does not necessarily mean the direct connection of the line, and the indirect connection mode can also be applied to the embodiments of the present application as long as the purpose of the present application is achieved. The above-mentioned is only an exemplary embodiment of the present application, and cannot limit the scope of the present application.

[0128] That is, any equivalent changes and modifications made according to the teachings of the present application are still within the scope of the present application. Other embodiments of the present application will be readily apparent to those skilled in the art upon considering the disclosure herein. The present application is intended to cover any variations, uses, or adaptive changes of the present application that follow the general principles of the present application and include common knowledge or conventional technical means in the art that are not described in the present application.

Claims

1. An automatic early warning method for faults in a reactive power compensation device, characterized in that, The method includes: Collect real-time operating data of reactive power compensation devices, real-time power data of the power grid, historical data, fault data, and a fault case database; The real-time power data is analyzed for early warning. When the real-time power data of the power grid exceeds a preset threshold, an early warning of abnormal power data is obtained. By performing fault early warning analysis on the real-time operating data and the changing trends of key parameters, early warning of equipment failure potential can be obtained. Based on the fault data and fault case library, a fault analysis and diagnosis model is constructed and trained using a time series algorithm to obtain the trained fault analysis and diagnosis model. The abnormal power data warning and equipment fault warning are fused from multiple sources to generate a comprehensive fault diagnosis, which is then input into the fault analysis and diagnosis model to generate the final fault diagnosis result. Based on the final fault diagnosis result, the corresponding reactive power compensation strategy and fault handling suggestion are retrieved from a preset database storing various reactive power compensation strategies and fault handling suggestions, and output to correct the power factor in the power data; wherein, according to the output reactive power compensation strategy, the operating parameters of the reactive power compensation device are automatically adjusted through a closed-loop control mechanism to correct the power factor; the operating effect data after adjusting the operating parameters of the reactive power compensation device is obtained, and the final fault diagnosis result, the executed reactive power compensation strategy, and the operating effect data are used as new fault cases and updated to the fault case database.

2. The automatic fault early warning method for a reactive power compensation device according to claim 1, characterized in that, The data collected includes real-time operating data of the reactive power compensation device, real-time power data of the power grid, historical data, fault data, and a fault case database. The real-time operating data includes the temperature, vibration, and noise of the reactive power compensation device during operation; The real-time power data of the power grid includes the reactive power, active power, power factor, harmonic current, negative sequence current and three-phase voltage of the power grid. The historical data includes historical operating data and historical power consumption data of the reactive power compensation device; The fault data refers to historical faults of the reactive power compensation device, including abnormal power data faults and equipment faults. The fault case library is constructed based on the analysis and mining of the fault data, and stores fault type tags, evolution patterns and verified historical handling strategies corresponding to different fault types in an associated form.

3. The automatic fault early warning method for a reactive power compensation device according to claim 2, characterized in that, The early warning analysis of the real-time power data includes: Based on the historical data, preset thresholds are set for power factor, reactive power, harmonic current, negative sequence current and three-phase voltage, respectively. If the power factor of the power grid continuously exceeds a preset threshold, or if any of the data of reactive power, harmonic current, negative sequence current, and three-phase voltage exceeds the safety limit specified by relevant standards or continuously exceeds their respective preset thresholds, an abnormal power data warning is generated. The warning is used to indicate that the power factor of the power grid is abnormal or the power quality is deteriorating. Otherwise, the power data is continuously monitored.

4. The automatic early warning method for reactive power compensation device faults according to claim 2, characterized in that, By performing fault early warning analysis on the real-time operating data and the changing trends of key parameters, early warnings of potential equipment failures are obtained, including: A machine learning model for anomaly detection is constructed, and the machine learning model is trained unsupervised using the historical operating data to obtain a normal operating baseline model. The real-time operating data is input into the normal operating baseline model for analysis, and the deviation value between the real-time operating data and the normal operating baseline is calculated. The deviation value is used to characterize the degree to which the real-time operating data deviates from the normal state. When the deviation value continues to exceed the normal fluctuation range, an early warning of potential equipment failure is obtained. The normal fluctuation range is obtained from historical operating data and a failure case database.

5. The automatic early warning method for reactive power compensation device faults according to claim 4, characterized in that, The construction of the machine learning model for anomaly detection, and the unsupervised training of the machine learning model using the historical operating data, includes: Statistical analysis is performed on the historical operating data to extract statistical features of the historical operating data, wherein the statistical features include the mean, variance, standard deviation and kurtosis of the operating data; The statistical features are used to perform unsupervised training on the machine learning model to obtain a normal baseline model that can calculate the deviation between the running data and the normal operating baseline.

6. The automatic early warning method for reactive power compensation device faults according to claim 1, characterized in that, Based on the fault data and fault case library, a fault analysis and diagnosis model is constructed and trained using a time series algorithm. The trained fault analysis and diagnosis model includes: An initial fault analysis and diagnosis model capable of learning time-series dependencies is constructed using a long short-term memory network algorithm. The fault data is mapped to the corresponding fault type labels in the fault case library to form a training sample set; The initial fault analysis and diagnosis model is iteratively trained using the training sample set. The model parameters are optimized until the diagnostic performance converges, resulting in the trained fault analysis and diagnosis model.

7. The automatic early warning method for reactive power compensation device faults according to claim 1, characterized in that, The abnormal power data warning and equipment fault warning are fused from multiple sources to generate a comprehensive fault diagnosis, which is then input into the fault analysis and diagnosis model to generate the final fault diagnosis result, including: Assign corresponding confidence levels and priority weights to the power data anomaly warning and the equipment fault warning, perform weighted decision processing, and generate fused data; The fused data is formatted to generate a comprehensive fault diagnosis that can be received by the trained fault analysis and diagnosis model. The comprehensive fault diagnosis is input into the fault analysis and diagnosis model, and the fault analysis and diagnosis model outputs the final fault diagnosis result.

8. The automatic early warning method for reactive power compensation device faults according to claim 1, characterized in that, Based on the final fault diagnosis result, the corresponding reactive power compensation strategy and fault handling suggestion are retrieved from a preset database that stores various reactive power compensation strategies and fault handling suggestions, including: The fault type tags contained in the final fault diagnosis results are used as the retrieval index; In the database that stores various reactive power compensation strategies and fault handling suggestions, the corresponding reactive power compensation strategies and fault handling suggestions are retrieved by searching the index. The reactive power compensation strategy and the fault handling suggestions are combined into a structured handling plan and then output.

Images