A knowledge graph-based method for energy system fault prediction

Abstract

The present invention discloses a method for predicting energy system faults based on knowledge graphs, which relates to the technical field of graph neural networks, including collecting real-time data of new energy and preprocessing the real-time data; constructing a dynamic graph model based on the preprocessed data; identifying abnormal data using a graph neural network based on the output result of the dynamic graph model; analyzing the identified abnormal data again using the dynamic graph model to predict faults; obtaining dynamically optimized fault judgment results based on historical data and real-time data through an adaptive fault threshold adjustment algorithm; the present invention estimates action values ​​through two independent Q networks, effectively alleviating the over-estimation problem in traditional Q learning, thereby improving the stability and accuracy of decision-making.

Description

Technical Field

[0001] The present invention relates to the field of graph neural network technology, and in particular to an energy system fault prediction method based on knowledge graph. Background Art

[0002] With the widespread application of new energy systems, it is particularly important to ensure their stable operation and efficient maintenance; traditional fault prediction methods mainly rely on rule-based systems, statistical models and simple machine learning algorithms; these methods show certain effectiveness in processing static data, but have limitations when facing complex and changeable energy systems; for example, rule-based methods are difficult to adapt to dynamic changes in the system, while statistical models often require a large amount of historical data for training and are sensitive to outliers; in recent years, graph neural networks, as an emerging data processing technology, have shown great potential in processing structured data; GNN can capture the state patterns of nodes and their neighboring nodes, identify potential abnormal behaviors, and provide new ideas for fault prediction in energy systems.

[0003] However, existing GNN-based fault prediction methods still have shortcomings; first, most methods fail to make full use of time series data and ignore the characteristics of system status changing over time, resulting in limited prediction accuracy; second, existing methods usually lack effective adaptive mechanisms and cannot dynamically adjust fault thresholds according to real-time data, thus affecting the accuracy and practicality of the prediction results; in addition, existing technologies have low computational efficiency when processing large-scale data sets and have high requirements for real-time data preprocessing, which increases the difficulty of practical applications. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an energy system fault prediction method based on knowledge graph to solve the problem of computational efficiency of large-scale data sets.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides an energy system fault prediction method based on a knowledge graph, which includes collecting real-time data of new energy and preprocessing the real-time data; constructing a dynamic graph model based on the preprocessed data; using a graph neural network to identify abnormal data based on the output results of the dynamic graph model; using the dynamic graph model again to analyze the identified abnormal data and predict faults; and obtaining a dynamically optimized fault judgment result based on historical data and real-time data through an adaptive fault threshold adjustment algorithm.

[0008] As a preferred solution of the energy system fault prediction method based on knowledge graph described in the present invention, wherein: the real-time data includes performance data, alarm data and log files;

[0009] The performance data includes voltage data, current data and temperature data;

[0010] The alarm data includes alarm data, historical alarm data and fault data.

[0011] As a preferred solution of the energy system fault prediction method based on knowledge graph described in the present invention, the preprocessing includes data cleaning, removing outliers, supplementing missing information and standardization processing.

[0012] As a preferred solution of the energy system fault prediction method based on knowledge graph described in the present invention, the specific steps of constructing the dynamic graph model are as follows:

[0013] Extract device names from log files using entity recognition (NER);

[0014] Identify device names, use dependency parsing to understand the sentence structure in log files, and determine the association between device nodes;

[0015] Define each device node’s static attributes as device type and installation location, and its dynamic attributes as operating status and real-time performance data;

[0016] Define the device-to-device edge attributes as connection type and connection strength;

[0017] Introduce the time dimension and add timestamps to the dynamic attributes of each device node to form time series data;

[0018] Merge historical fault data with clean data sets into historical fault information data sets, and mark known equipment faults;

[0019] Use the NetworkX graphics library to build an initial static graph;

[0020] Based on the static graph, time series data is introduced to build a dynamic graph, and the node and edge attributes of each device are updated over time.

[0021] As a preferred solution of the energy system fault prediction method based on knowledge graph described in the present invention, the specific steps of using graph neural network to identify abnormal data are as follows:

[0022] The graph neural network model GNN is used to capture the status of nodes and their neighboring nodes and identify potential abnormal behaviors;

[0023] Based on the dynamic graph model, the clean data set is used as input to train the graph neural network GNN model;

[0024] Select the graph convolutional network GCN as the architecture of the graph neural network model GNN;

[0025] Define the parameters of the graph neural network model GNN;

[0026] Input the node attributes, edge attributes and time series data in the dynamic graph model into the graph neural network model GNN;

[0027] The graph neural network model GNN gradually aggregates node and neighbor node information through multi-layer graph convolution and learns node status;

[0028] Use supervised learning method to define the mean square error (MSE) loss function, minimize the mean square error (MSE) loss function through back propagation algorithm and optimizer Adam, and update model parameters;

[0029] Input the clean data set into the trained graph neural network model GNN, and output the predicted data state value;

[0030] Compare the predicted data state value of the graph neural network model GNN with the real-time data, and calculate the difference by the absolute difference;

[0031] Set a threshold. If the difference exceeds the set threshold, it is judged as abnormal data.

[0032] As a preferred solution of the energy system fault prediction method based on knowledge graph described in the present invention, the specific steps of analyzing and predicting faults by using a dynamic graph model are as follows:

[0033] By matching the detected abnormal data with the equipment faults in the dynamic graph model, all nodes and edges of the abnormal data are determined;

[0034] Combine time series data, node attributes, and edge attributes in dynamic graphs for comprehensive analysis;

[0035] Use Bayesian reasoning to determine causal relationships between abnormal data and identify potential equipment failures;

[0036] Predict specific equipment failures based on the attributes of abnormal data and historical failure data;

[0037] Based on the time series data in the dynamic graph model, the specific time point when equipment failure occurs is predicted.

[0038] As a preferred solution of the energy system fault prediction method based on knowledge graph described in the present invention, wherein: the dynamically optimized fault judgment result is obtained by the adaptive fault threshold adjustment algorithm, and the specific steps are as follows:

[0039] Based on the historical fault information data set, the initial fault judgment threshold is set as T0;

[0040] Adopting dual Q learning reinforcement learning algorithm for adaptive fault threshold adjustment;

[0041] Define the new energy system state as S t , adjust the threshold action to A t ;

[0042] Define the reward function as R(S t , A t );

[0043] The device fault threshold is adjusted by adaptive threshold, and the expression is:

[0044] T t+1 =T t +ΔT·sign(R(S t , A t ));

[0045] Among them, T t+1 is the failure threshold at the next time point t+1, T t is the fault threshold at the current time point t, ΔT is the fixed step size for adjusting the fault threshold each time, and sign is the sign function.

[0046] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the energy system fault prediction method based on knowledge graph as described in the first aspect of the present invention.

[0047] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the energy system fault prediction method based on knowledge graph as described in the first aspect of the present invention.

[0048] The beneficial effects of the present invention are as follows: by adopting the dual Q learning reinforcement learning algorithm to perform adaptive fault threshold adjustment, the dynamic optimization capability of fault prediction is significantly improved; the dual Q learning algorithm estimates the action value through two independent Q networks, effectively alleviating the over-estimation problem in traditional Q learning, thereby improving the stability and accuracy of decision-making; in the new energy system fault prediction, the algorithm can dynamically adjust the fault judgment threshold according to real-time data and historical data, making fault detection more sensitive and adaptive. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0050] Figure 1 This is a flow chart of the energy system fault prediction method based on knowledge graph in Example 1.

[0051] Figure 2 This is a schematic diagram of predicting failure in Example 1. DETAILED DESCRIPTION

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0053] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0054] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0055] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides an energy system fault prediction method based on knowledge graph, comprising the following steps:

[0056] S1. Collect new energy real-time data and pre-process the real-time data.

[0057] S1.1. Furthermore, performance data, alarm data, historical alarm data, fault data and log files are collected through voltage sensors, current sensors, temperature sensors, SCADA supervisory control equipment and PLCs programmable logic controllers.

[0058] It should be noted that the voltage sensor is used to monitor the voltage level in the new energy system. The data type is real-time voltage value. The acquisition frequency is in seconds or milliseconds, depending on the requirements of the new energy system. The data format is numerical data and the unit is volt (V).

[0059] The current sensor is used to monitor the current intensity in the new energy system. The data type is real-time current value. The acquisition frequency is in seconds or milliseconds, depending on the requirements of the new energy system. The data format is numerical data and the unit is ampere (A).

[0060] Temperature sensors are used to monitor temperature changes in new energy systems. The data type is real-time temperature values, and the acquisition frequency is in seconds or minutes, depending on the needs of the new energy system. The data format is numerical data in degrees Celsius (℃) or Fahrenheit (℉). SCADA supervisory control equipment is used to remotely monitor and control industrial processes, collect and record large amounts of real-time data, and the data types include performance data, alarm data, historical alarm data, fault data, and log files.

[0061] Performance data includes voltage, current, power, frequency, etc. Alarm data is the alarm information generated when the new energy system is abnormal or fails, and historical alarm data is the past alarm records; fault data is detailed fault information, including fault time, location, type, etc.; log files are new energy system operation logs, which record system operation status and events; the collection frequency is seconds or minutes, depending on the new energy system configuration; the data format is structured data, such as CSV, JSON, etc.

[0062] PLCs programmable logic controllers are used for automated control and data acquisition, especially in industrial environments; data types include performance data, alarm data, historical alarm data, fault data and log files; performance data includes switch status, sensor readings, actuator status, etc.; alarm data is the alarm information generated when the new energy system is abnormal or fails; historical alarm data is the alarm record in the past; fault data is detailed fault information, including fault time, location, type, etc.

[0063] The log file is the new energy system operation log, which records the system operation status and events; the collection frequency is in seconds or milliseconds, depending on the new energy system configuration; the data format is structured data, such as CSV, JSON, etc.

[0064] Specifically, collecting real-time data on new energy includes the following steps:

[0065] Install voltage sensors, current sensors and temperature sensors at key locations to ensure coverage of all important parts;

[0066] Configure the sensor's acquisition frequency and data transmission method;

[0067] Set up data collection points for SCADA supervisory control equipment to ensure comprehensive monitoring of key parameters;

[0068] Configure alarm and logging functions to ensure that abnormal situations can be captured in time;

[0069] Program PLC to collect necessary performance data, alarm data and fault data;

[0070] Set up data transmission protocols to ensure that data can be reliably transmitted to the central server;

[0071] Transmit data collected by sensors, SCADA and PLC to the central server through communication protocols;

[0072] Store the data in a database to facilitate subsequent data preprocessing and analysis;

[0073] Regularly check the completeness and accuracy of data to ensure data quality;

[0074] Mark and process abnormal data to ensure data reliability.

[0075] S1.2. Clean the real-time data, remove incomplete data, redundant data and correct outliers in the data;

[0076] Convert semi-structured data into structured data and supplement missing information;

[0077] Use Z-score to standardize the real-time data to obtain a clean data set.

[0078] It should be noted that by collecting real-time data on new energy through a variety of devices, the comprehensiveness and real-time nature of the data are achieved, a rich and reliable data source is provided, the real-time data is cleaned, incomplete data, redundant data are removed, and outliers are corrected, the quality and consistency of the data are improved, noise and errors are reduced, semi-structured data are converted into structured data, the availability and integrity of the data are enhanced, the data processing process is simplified, and the efficiency of data processing is improved. The real-time data is standardized using Z-score, which eliminates the dimensional differences between different features, improves the convergence speed and prediction accuracy of the model, and enhances the robustness of the model.

[0079] S2. Build a dynamic graph model based on the preprocessed data.

[0080] Furthermore, we use entity recognition NER to extract device names from log files as dynamic graph nodes;

[0081] Identify device names, use dependency syntax analysis to understand the sentence structure in log files, and determine the association between device nodes. It should be noted that dependency syntax is an important technology in natural language processing (NLP), which is used to analyze the dependency relationship between words in a sentence and reveal the grammatical structure and semantic relationship of a sentence by establishing direct connections between words.

[0082] Define each device node's static attributes as device type and installation location, and the node's dynamic attributes as operating status and real-time performance data. It should be noted that static attributes provide basic information about the device, and dynamic attributes reflect the device's real-time status;

[0083] Define the device-to-device edge attributes as connection type and connection strength;

[0084] Introduce the time dimension, add corresponding timestamps to the dynamic attributes of each device node, form time series data, monitor the device status in real time, and detect abnormal changes in time;

[0085] Merge historical fault data with clean data sets into historical fault information data sets, and mark known equipment faults;

[0086] Use the NetworkX graphics library to build an initial static graph; it should be noted that NetworkX is a graphics library written in Python for creating, manipulating, and studying the structure, dynamics, and functions of complex networks;

[0087] Based on the static graph, time series data is introduced to construct a dynamic graph;

[0088] As time changes, the node and edge attributes of each device are updated, and the graph structure of the node status changing over time is captured through the temporal graph convolutional network T-GCN;

[0089] Use the NetworkX graphics library to build an initial static graph;

[0090] Based on the static graph, time series data is introduced to build a dynamic graph, and the node and edge attributes of each device are updated over time;

[0091] It should be noted that the temporal graph convolutional network T-GCN is used to capture the graph structure of node states changing over time.

[0092] It should be noted that by using named entity recognition (NER) to automatically extract device names from log files, the workload of manual annotation is reduced, the data processing efficiency is improved, and the correctness of nodes is ensured; dependency syntactic analysis is used to understand the sentence structure in the log files, and the complex relationship between devices is mined, providing rich edge attributes and enhancing the understanding of device interaction patterns; static and dynamic attributes of each device node are defined to comprehensively describe the status and characteristics of the device, providing multi-dimensional information support for data analysis and fault prediction; by defining the connection type and connection strength between devices, the relationship between devices is quantified, which is helpful for analyzing the fault propagation path; the time dimension is introduced to form time series data, which realizes real-time monitoring of device status and timely detection of abnormal changes; combining historical fault data with the current data set, the fault mode and experience are accumulated, and known faults are quickly identified; NetworkX is used to construct an initialized static graph, providing intuitive graph visualization function; time series data is introduced on the basis of the static graph to construct a dynamic graph, and the time series graph convolutional network T-GCN effectively learns the characteristics of node status changing over time, which improves the model's prediction ability and dynamic adjustment ability.

[0093] S3. Based on the output results of the dynamic graph model, use the graph neural network to identify abnormal data.

[0094] Furthermore, the graph neural network model GNN is used to capture the state patterns of nodes and their neighboring nodes and identify potential abnormal behaviors;

[0095] Based on the dynamic graph model, the clean data set is used as input to train the graph neural network GNN model. It should be noted that the clean data set includes node attributes, edge attributes and time series data;

[0096] Select the graph convolutional network GCN as the architecture of the graph neural network model GNN. It should be noted that the parameters of the graph neural network model GNN are defined, such as the number of layers, the number of hidden units in each layer, the activation function, the optimizer, the learning rate, etc. The number of layers is 2-3, the number of hidden units in each layer is 64 or 128, the activation function is ReLU, the optimizer is Adam, and the learning rate is 0.001;

[0097] Input the node attributes, edge attributes and time series data in the dynamic graph model into the graph neural network model GNN;

[0098] The graph neural network model GNN gradually aggregates node and neighbor node information through multi-layer graph convolution and learns node status patterns;

[0099] Use supervised learning method to define the mean square error (MSE) loss function, minimize the mean square error (MSE) loss function through back propagation algorithm and optimizer Adam, and update model parameters;

[0100] Input the clean data set into the trained graph neural network model GNN, and output the predicted data state value;

[0101] Compare the predicted data state value of the graph neural network model GNN with the real-time data, and calculate the difference by the absolute difference;

[0102] Set a threshold. If the difference exceeds the set threshold, it is judged as abnormal data.

[0103] It should be noted that the threshold is usually set through historical data and statistical analysis, and can be a fixed value or dynamically adjusted; for example, the threshold can be set by calculating the mean and standard deviation of historical data, or the experience of domain experts can be used to determine a reasonable threshold range.

[0104] It should be noted that the operating status includes: working status, performance indicators, alarm status, fault status and other status. The working status includes normal operation of the equipment, equipment standby, equipment shutdown and equipment maintenance status. The performance indicators include equipment voltage level, equipment current level, equipment operating temperature and equipment power output.

[0105] The alarm status includes no alarm, slight alarm and serious alarm. The fault status includes no fault, potential fault and confirmed fault. Other status includes load, working efficiency and response time.

[0106] To determine the current status of a device, it is usually necessary to combine a variety of data, the most common of which are:

[0107] Use voltage sensors, current sensors, and temperature sensors to collect equipment performance data in real time; these data can be used to determine the equipment's working status, performance indicators, and alarm status;

[0108] Extract the operation records of the equipment from the log files, including operation logs such as startup, stop, and maintenance; use natural language processing technology (such as entity recognition NER and dependency syntax analysis) to extract key information from the log files and determine the operation status of the equipment;

[0109] Monitor the alarm information of the new energy system and obtain the alarm status of the equipment in real time; determine whether the equipment has potential faults or confirmed faults based on the severity of the alarm information;

[0110] Combine historical data with current data to predict the current status of the equipment through data analysis and machine learning methods; for example, use time series analysis methods to predict the performance indicators and failure risks of equipment;

[0111] Leverage the knowledge and experience of domain experts to develop rules and thresholds to determine the current status of a device. For example, if the temperature of a device exceeds a certain threshold and there is a serious alarm message, it is determined that the device may be at risk of overheating.

[0112] It should be noted that by adopting the graph neural network GNN model, especially the graph convolutional network GCN architecture, the state pattern of the node and its neighboring nodes is captured, and the identification of potential abnormal behaviors is realized; based on the dynamic graph model, the clean data set is used as input to train the GNN model, and the node and neighboring node information is gradually aggregated through multi-layer graph convolution to learn the node state pattern; using the supervised learning method, the mean square error MSE loss function is defined, and the loss function is minimized through the back propagation algorithm and the optimizer Adam, the model parameters are updated, and the prediction accuracy of the model is improved; the clean data set is input into the trained GNN model, the predicted data state value is output, and compared with the actual data, and the difference is calculated by the absolute difference; a threshold is set, and when the difference exceeds the threshold, it is judged as abnormal data; not only the accuracy and real-time performance of fault detection are improved, but also the robustness and reliability of the system are enhanced, the false alarm and missed alarm rates are reduced, and the efficient and reliable operation of the energy system is ensured.

[0113] S4. By identifying abnormal data, the dynamic graph model is used again to analyze and predict faults.

[0114] Furthermore, the use of dynamic graph models to analyze and predict faults includes the following specific steps:

[0115] By matching the detected abnormal data with the equipment faults in the dynamic graph model, all nodes and edges of the abnormal data are determined;

[0116] Combine time series data, node attributes, and edge attributes in dynamic graphs for comprehensive analysis;

[0117] Use Bayesian reasoning to determine causal relationships between abnormal data and identify potential equipment failures;

[0118] Predict specific equipment failures based on the attributes of abnormal data and historical fault data. It should be noted that by analyzing the causal relationship graph, it is possible to determine which equipment or sensor state changes are most likely to cause the abnormal data, thereby identifying potential faulty equipment. Bayesian reasoning focuses on the causal relationship between abnormal data, that is, which equipment or sensor state changes cause the abnormal data, which helps to locate the root cause of the fault. It should be noted that by comparing the characteristics of the current abnormal data with the historical fault data, similar situations can be found, and based on this, specific equipment failure types and possible causes can be predicted;

[0119] Based on the attributes of the abnormal data and historical fault data, more emphasis is placed on predicting the specific fault type and possible causes. This step relies on pattern matching in historical fault data to determine the fault type corresponding to the current abnormal data.

[0120] Based on the time series data in the dynamic graph model, the specific time point of equipment failure is predicted. It should be noted that based on the failure, specific preventive measures to prevent equipment failure are provided, and the impact of equipment failure is evaluated.

[0121] It should be noted that by detecting abnormal data and matching it with equipment faults in the dynamic graph model, the abnormal nodes and edges are accurately located, which improves the efficiency of fault diagnosis; a comprehensive analysis is conducted by combining time series data, node attributes, and edge attributes to enhance the understanding of abnormal situations; Bayesian reasoning is used to determine causal relationships and improve the accuracy of fault identification; specific equipment failures are predicted based on historical fault data to achieve early warning. Time series data is used to predict the time of failure and provide a basis for maintenance plans; specific preventive measures are provided and the impact of failures is evaluated to reduce the risk of failures, ensure efficient and reliable operation of the system, significantly improve operation and maintenance efficiency, and reduce maintenance costs.

[0122] S5. According to historical data and real-time data, a dynamically optimized fault judgment result is obtained through an adaptive fault threshold adjustment algorithm.

[0123] Furthermore, based on the historical fault information data set, the initial fault judgment threshold is set as T0. It should be noted that the initial fault judgment threshold is the starting point of the adaptive adjustment, which can provide a reasonable initial threshold based on the experience of historical data;

[0124] Adopting dual Q learning reinforcement learning algorithm for adaptive fault threshold adjustment;

[0125] Define the new energy system state as S t , adjust the threshold action to A t ,It should be explained that the new energy system status includes real-time ,performance data, historical performance data, fault data, equipment status, time information and ,environmental information;

[0126] Adjusting the threshold action includes increasing the threshold, decreasing the threshold, or keeping it unchanged;

[0127] Define the reward function as R(S t , A t );

[0128] The device fault threshold is adjusted by adaptive threshold, and the expression is:

[0129] T t+1 =T t +ΔT·sign(R(S t, A t ));

[0130] Among them, T t+1 is the failure threshold at the next time point t+1, T t is the fault threshold at the current time point t, ΔT is the fixed step size for adjusting the fault threshold each time, and sign is the sign function, which is used to determine the adjustment direction of the fault threshold T according to the positive or negative value of the reward function.

[0131] It should be noted that the accuracy and adaptability of fault detection are significantly improved through the adaptive fault threshold adjustment algorithm based on historical data and real-time data; the initial threshold is set by historical data to provide a reasonable starting point; the double Q learning reinforcement learning algorithm is used to dynamically adjust the fault threshold, which improves the stability and accuracy of the system; the system state and adjustment actions are defined; through adaptive dynamic adjustment, the optimal threshold is gradually approached to reduce false alarms and missed alarms.

[0132] This embodiment also provides a computer device, which is suitable for the case of an energy system fault prediction method based on a knowledge graph, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the energy system fault prediction method based on a knowledge graph as proposed in the above embodiment.

[0133] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0134] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for predicting energy system faults based on knowledge graphs as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0135] In summary, the present invention significantly improves the dynamic optimization capability of fault prediction by adopting a dual Q learning reinforcement learning algorithm for adaptive fault threshold adjustment; the dual Q learning algorithm estimates the action value through two independent Q networks, effectively alleviating the over-estimation problem in traditional Q learning, thereby improving the stability and accuracy of decision-making; in the new energy system fault prediction, the algorithm can dynamically adjust the fault judgment threshold according to real-time data and historical data, making fault detection more sensitive and adaptive.

[0136] Example 2, referring to Table 1, is the second example of the present invention. In order to further verify the technical solution of the present invention, experimental simulation data of an energy system fault prediction method based on a knowledge graph is provided.

[0137] First, real-time data of the new energy system is collected through a variety of devices and preprocessed, including cleaning data, removing incomplete and redundant data, correcting outliers, converting semi-structured data into structured data, and standardizing the real-time data using the Z-score method.

[0138] Secondly, a dynamic graph model is built based on the preprocessed data. Named Entity Recognition (NER) technology is used to extract device names as nodes from log files, and the relationship between device nodes is determined through dependency syntax analysis. The static and dynamic attributes of each device node are defined, and the edge attributes are defined. The time dimension is introduced to form time series data, and the NetworkX graphics library is used to build an initialized static graph. Based on the static graph, time series data is introduced to build a dynamic graph, and the graph structure of node status changing over time is captured through the temporal graph convolutional network (T-GCN).

[0139] Next, use the graph neural network GNN model to identify abnormal data; select the graph convolution network GCN as the architecture of GNN, and define the model parameters; input the node attributes, edge attributes and time series data in the dynamic graph model into GNN, and gradually aggregate the node and neighbor node information through multi-layer graph convolution to learn the node state mode. Use the supervised learning method to define the mean square error MSE loss function, minimize the loss function through the back propagation algorithm and optimizer Adam, and update the model parameters; input the clean data set into the trained GNN model, output the predicted data state value, and compare it with the actual data, and calculate the difference through the absolute difference; set the threshold, if the difference exceeds the set threshold, it is judged as abnormal data.

[0140] Finally, by matching the detected abnormal data with the equipment failure in the dynamic graph model, all nodes and edges of the abnormal data are determined; a comprehensive analysis is performed on the time series data, node attributes, and edge attributes in the dynamic graph, and Bayesian reasoning is used to determine the causal relationship between the abnormal data and identify equipment failures; based on the attributes of the abnormal data and historical failure data, specific equipment failures are predicted, and the specific time point of equipment failure is predicted based on the time series data; specific preventive measures to prevent equipment failures are provided, and the impact of equipment failures is evaluated. In addition, based on historical data and real-time data, the fault judgment results are dynamically optimized through an adaptive fault threshold adjustment algorithm to ensure efficient and reliable operation of the system.

[0141] The details are shown in Table 1 below:

[0142] Table 1 Comparison of fault prediction performance of new energy systems

[0143]

[0144] Through the above data comparison, it can be seen that the present invention has significant improvements in multiple key performance indicators; the present invention adopts multi-sensor acquisition and Z-score standardization methods to improve the stability of voltage and current, achieving improvements of 58.33% and 62.5%, which means that the present invention can more effectively reduce the fluctuation of voltage and current, thereby improving overall stability and reliability;

[0145] In terms of fault detection accuracy, the present invention uses graph neural network GNN and temporal graph convolutional network T-GCN for fault prediction, and the accuracy is increased from 80% to 95%, an overall improvement of 18.75%. It not only improves the accuracy of fault detection, but also enhances the early warning capability of potential problems, allowing operation and maintenance personnel to take measures earlier to avoid the occurrence of faults.

[0146] Finally, in terms of false alarm rate, the graph neural network GNN and the temporal graph convolutional network T-GCN were also used for detection. The false alarm rate was reduced from 15% to 5%, a reduction of 66.67%. This not only enables more accurate identification of real faults, but also significantly reduces false alarms, thereby improving overall reliability and operation and maintenance efficiency.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting energy system faults based on knowledge graph, characterized by: include, Collect new energy real-time data and pre-process the real-time data; Based on the preprocessed data, a dynamic graph model is constructed, specifically, Extract device names from log files using entity recognition (NER); Identify device names, use dependency parsing to understand the sentence structure in log files, and determine the association between device nodes; Define each device node’s static attributes as device type and installation location, and its dynamic attributes as operating status and real-time performance data; Define the device-to-device edge attributes as connection type and connection strength; Introduce the time dimension and add timestamps to the dynamic attributes of each device node to form time series data; Merge historical fault data with clean data sets into historical fault information data sets, and mark known equipment faults; Use the NetworkX graphics library to build an initial static graph; Based on the static graph, time series data is introduced to build a dynamic graph, and the node and edge attributes of each device are updated over time; Based on the output results of the dynamic graph model, the graph neural network is used to identify abnormal data. Specifically, The graph neural network model GNN is used to capture the status of nodes and their neighboring nodes and identify potential abnormal behaviors; Based on the dynamic graph model, the clean data set is used as input to train the graph neural network GNN model; Select the graph convolutional network GCN as the architecture of the graph neural network model GNN; Input the node attributes, edge attributes and time series data in the dynamic graph model into the graph neural network model GNN; The graph neural network model GNN gradually aggregates node and neighbor node information through multi-layer graph convolution and learns node status; Use supervised learning method to define the mean square error (MSE) loss function, minimize the mean square error (MSE) loss function through back propagation algorithm and optimizer Adam, and update model parameters; Input the clean data set into the trained graph neural network model GNN, and output the predicted data state value; Compare the predicted data state value of the graph neural network model GNN with the real-time data, and calculate the difference by the absolute difference; Set a threshold. If the difference exceeds the set threshold, it is considered abnormal data. By identifying abnormal data, the dynamic graph model is used again to analyze and predict faults; According to historical data and real-time data, a dynamically optimized fault judgment result is obtained through an adaptive fault threshold adjustment algorithm.

2. The energy system fault prediction method based on knowledge graph according to claim 1, characterized in that: The real-time data includes performance data, alarm data and log files; The performance data includes voltage data, current data and temperature data; The alarm data includes alarm data, historical alarm data and fault data.

3. The energy system fault prediction method based on knowledge graph according to claim 2, characterized in that: The preprocessing includes data cleaning, removing outliers, supplementing missing information and standardization.

4. The energy system fault prediction method based on knowledge graph according to claim 3 is characterized in that: The specific steps of using the dynamic graph model to analyze and predict faults are as follows: By matching the detected abnormal data with the equipment faults in the dynamic graph model, all nodes and edges of the abnormal data are determined; Combine time series data, node attributes, and edge attributes in dynamic graphs for comprehensive analysis; Use Bayesian reasoning to determine causal relationships between abnormal data and identify potential equipment failures; Predict specific equipment failures based on the attributes of abnormal data and historical failure data; Based on the time series data in the dynamic graph model, the specific time point when equipment failure occurs is predicted.

5. The energy system fault prediction method based on knowledge graph according to claim 4, characterized in that: The adaptive fault threshold adjustment algorithm is used to obtain the dynamically optimized fault judgment result, and the specific steps are as follows: Based on the historical fault information data set, the initial fault judgment threshold is set as T0; Adopting dual Q learning reinforcement learning algorithm for adaptive fault threshold adjustment; Define the new energy system state as S t , adjust the threshold action to A t ; Define the reward function as R(S t , A t ); The device fault threshold is adjusted by adaptive threshold, and the expression is: T t+1 =T t +ΔT·sign(R(S t ,A t )); Among them, T t+1 is the failure threshold at the next time point t+1, T t is the fault threshold at the current time point t, ΔT is the fixed step size for adjusting the fault threshold each time, and sign is the sign function.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the energy system fault prediction method based on knowledge graph described in any one of claims 1 to 5 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the energy system fault prediction method based on knowledge graph described in any one of claims 1 to 5 are implemented.

Images