Operation fault early warning method, device and equipment of power station communication system and medium

Abstract

The invention discloses an operation fault early warning method, device and equipment of a power station communication system and a medium, and relates to the technical field of operation fault early warning of the power station communication system.The operation fault early warning method of the power station communication system comprises the steps that monitoring data are periodically obtained from a plurality of communication subsystems of the power station communication system; the monitoring data are input into a preset health value detection model, a system health value is obtained, and the health value detection model is a stack type auto-encoder model and is obtained through pre-training; and inputting a sequence formed by the system health value and the monitoring data along with time change into a preset fault prediction model to predict a probability value of a fault of the power station communication system within a preset time range, and triggering early warning signals of different grades based on the probability value, the fault prediction model is a long-short-term memory model and is obtained through pre-training. According to the invention, integrated health degree evaluation and fault early warning can be carried out on the power station communication system.

Description

Technical Field

[0001] This application relates to the field of early warning technology for operational faults in power plant communication systems, and in particular to methods, devices, equipment, and media for early warning of operational faults in power plant communication systems. Background Technology

[0002] The current power plant communication system consists of multiple independently managed heterogeneous subsystems, such as transmission, program-controlled switching, wireless communication, data network and power environment monitoring. Its health status monitoring generally relies on setting fixed alarm thresholds for single performance indicators in each subsystem, or applying a small number of simple association rules manually defined by experts.

[0003] However, fixed threshold alarms are prone to false alarms and missed alarms because they cannot adapt to the dynamic changes in system operation. Moreover, their alarms are isolated from each other and it is difficult to reveal the root causes of cross-system related faults. Essentially, it is a passive response mechanism for post-event remediation. On the other hand, simple rule-based methods face difficulties in maintaining the rule base, lack of flexibility, and inability to capture complex nonlinear relationships between high-dimensional data due to the complexity and variability of fault modes. Currently, existing technologies have failed to provide a health indicator that can quantitatively reflect the overall operating status of the power plant communication system, resulting in a lack of effective decision-making basis for operation and maintenance management.

[0004] In summary, how to achieve integrated health assessment and fault early warning for power plant communication systems has become a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] The main purpose of this application is to provide a method, device, equipment and medium for early warning of operational faults in a power plant communication system, aiming to achieve integrated health assessment and fault early warning of the power plant communication system.

[0006] To achieve the above objectives, this application proposes a method for early warning of operational faults in a power plant communication system. The method includes: Monitoring data is periodically acquired from multiple communication subsystems of the power plant's communication system; The monitoring data is input into a preset health value detection model to obtain the system health value. The health value detection model is a stacked autoencoder model and is obtained through pre-training. The system health value and the sequence formed by the change of the monitoring data over time are input into a preset fault prediction model to predict the probability value of the power station communication system failing within a preset time range, and trigger different levels of early warning signals based on the probability value. The fault prediction model is a long short-term memory model and is obtained through pre-training.

[0007] In one embodiment, before the step of inputting the monitoring data into a preset health value detection model to obtain the system health value, the method further includes: Historical monitoring data of the power plant communication system during the fault-free operation phase were obtained as the first training sample data. Based on the first training sample data, the stacked autoencoder model to be trained is trained with the goal of minimizing the reconstruction error between the model output data and the model input data, thereby obtaining the health value detection model.

[0008] In one embodiment, before the step of inputting the sequence formed by the changes of the system health value and the monitoring data over time into a preset fault prediction model to predict the probability value of a fault occurring in the power station communication system within a preset time range, the method further includes: A historical time series is constructed as the second training sample data, wherein each element in the historical time series represents the system health value and monitoring data at a point in time; The fault prediction model is obtained by training the long short-term memory model to be trained based on the second training sample data. The long short-term memory model includes a random deactivation layer, and the loss function used for training is the focus loss function.

[0009] In one embodiment, after the step of triggering different levels of warning signals based on the probability value, the method further includes: When the probability value exceeds the preset alarm threshold, statistical features are extracted from the monitoring data, deep features are extracted from the bottleneck layer of the health value detection model, and temporal features are extracted from the hidden state of the fault prediction model. A hybrid feature vector is constructed based on the statistical features, the deep features, and the temporal features, and the hybrid feature vector is input into a preset fault analysis model to obtain the fault analysis result. The fault analysis model is a gradient boosting tree model and is obtained through pre-training.

[0010] In one embodiment, before the step of inputting the hybrid feature vector into a preset fault analysis model to obtain the fault analysis result, the method further includes: Historical fault event data is obtained as the third training sample data, wherein the historical fault event data includes historical mixed feature vectors and corresponding fault type labels; Based on the third training sample data, the gradient boosting tree model to be trained is trained to obtain the fault analysis model.

[0011] In one embodiment, after the step of inputting the hybrid feature vector into a preset fault analysis model to obtain the fault analysis result, the method further includes: When the fault analysis result indicates that the root cause of the fault is located in the first communication subsystem of the power plant communication system, the probability value of the root cause of the fault leading to a derivative fault in the second communication subsystem is determined. Based on the probability value of derivative failures occurring in the second communication subsystem, a fault impact chain diagram across subsystems is generated, and the fault impact chain diagram is added to the fault analysis results.

[0012] In one embodiment, the step of periodically acquiring monitoring data from multiple communication subsystems of the power plant communication system includes: Initial monitoring data is periodically collected from multiple communication subsystems of the power plant's communication system; The initial monitoring data is preprocessed to obtain monitoring data, wherein the preprocessing includes data cleaning, data alignment and data standardization operations.

[0013] Furthermore, to achieve the above objectives, this application also proposes a fault early warning device for a power plant communication system, which includes: The data acquisition module is used to periodically acquire monitoring data from multiple communication subsystems of the power plant's communication system; The health value acquisition module is used to input the monitoring data into a preset health value detection model to obtain the system health value, wherein the health value detection model is a stacked autoencoder model and is obtained through pre-training; The fault early warning module is used to input the system health value and the sequence formed by the change of the monitoring data over time into a preset fault prediction model to predict the probability value of the power station communication system failing within a preset time range, and to trigger different levels of early warning signals based on the probability value. The fault prediction model is a long short-term memory model and is obtained through pre-training.

[0014] In addition, to achieve the above objectives, this application also proposes an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the power plant communication system operation fault early warning method as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above-described method for early warning of operational faults in a power plant communication system.

[0016] This application proposes a method for early warning of operational faults in a power plant communication system. In this method, monitoring data is periodically acquired from multiple communication subsystems of the power plant communication system. The monitoring data is input into a preset health value detection model to obtain a system health value. The health value detection model is a stacked autoencoder model and is obtained through pre-training. The sequence formed by the system health value and the monitoring data changing over time is input into a preset fault prediction model to predict the probability of a fault occurring in the power plant communication system within a preset time range. Based on the probability value, different levels of early warning signals are triggered. The fault prediction model is a long short-term memory model and is obtained through pre-training.

[0017] In summary, this application, by periodically acquiring monitoring data from multiple communication subsystems and using a stacked autoencoder model to calculate system health values, can comprehensively consider the operational status of each communication subsystem and achieve a quantitative assessment of the overall operational health status of the power plant communication system. Based on this, the sequence formed by the change of system health values ​​over time is input into a long short-term memory model for fault prediction, which can effectively capture dynamic changes during system operation. By predicting fault probability values ​​and triggering different levels of early warning signals, potential faults in the power plant communication system can be detected in advance, achieving integrated fault early warning for the power plant communication system. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an embodiment of the power plant communication system operation fault early warning method of this application. Figure 2 This is a schematic diagram of the fault early warning system architecture provided in Embodiment 1 of the power plant communication system operation fault early warning method of this application; Figure 3 This is a schematic diagram of the fault impact chain provided in Embodiment 2 of the power plant communication system operation fault early warning method of this application; Figure 4 This is a schematic diagram of the module structure of the power plant communication system operation fault early warning device according to an embodiment of this application; Figure 5 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the power plant communication system operation fault early warning method in this application embodiment.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] The current power plant communication system consists of multiple independently managed heterogeneous subsystems, such as transmission, program-controlled switching, wireless communication, data network and power environment monitoring. Its health status monitoring generally relies on setting fixed alarm thresholds for single performance indicators in each subsystem, or applying a small number of simple association rules manually defined by experts.

[0024] However, fixed threshold alarms are prone to false alarms and missed alarms because they cannot adapt to the dynamic changes in system operation. Moreover, their alarms are isolated from each other and it is difficult to reveal the root causes of cross-system related faults. Essentially, it is a passive response mechanism for post-event remediation. On the other hand, simple rule-based methods face difficulties in maintaining the rule base, lack of flexibility, and inability to capture complex nonlinear relationships between high-dimensional data due to the complexity and variability of fault modes. Currently, existing technologies have failed to provide a health indicator that can quantitatively reflect the overall operating status of the power plant communication system, resulting in a lack of effective decision-making basis for operation and maintenance management.

[0025] In summary, how to achieve integrated health assessment and fault early warning for power plant communication systems has become a pressing technical problem that needs to be solved in this field.

[0026] This application provides a solution that periodically acquires monitoring data from multiple communication subsystems of a power plant communication system; inputs the monitoring data into a preset health value detection model to obtain a system health value, wherein the health value detection model is a stacked autoencoder model and is obtained through pre-training; inputs the sequence formed by the system health value and the monitoring data changing over time into a preset fault prediction model to predict the probability value of a fault occurring in the power plant communication system within a preset time range, and triggers different levels of early warning signals based on the probability value, wherein the fault prediction model is a long short-term memory model and is obtained through pre-training.

[0027] In summary, the embodiments of this application, by periodically acquiring monitoring data from multiple communication subsystems and using a stacked autoencoder model to calculate system health values, can comprehensively consider the operating status of each communication subsystem and achieve a quantitative assessment of the overall operational health status of the power plant communication system. Based on this, the sequence formed by the change of system health values ​​over time is input into a long short-term memory model for fault prediction, which can effectively capture dynamic changes during system operation. By predicting fault probability values ​​and triggering different levels of early warning signals, potential faults in the power plant communication system can be detected in advance, achieving integrated fault early warning for the power plant communication system.

[0028] It should be noted that the executing entity in this embodiment can be an electronic device with data processing, network communication, and program execution functions, such as a computer, host computer, controller, etc., or an electronic device capable of performing the above functions. The following uses a fault early warning system as an example to describe this embodiment and the subsequent embodiments.

[0029] Based on this, embodiments of this application provide a method for early warning of operational faults in a power plant communication system, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the power plant communication system operation fault early warning method of this application.

[0030] In this embodiment, the method for early warning of operational faults in a power plant communication system includes steps S10 to S30: Step S10: Periodically acquire monitoring data from multiple communication subsystems of the power plant communication system; It should be noted that the power plant communication system refers to the entire dedicated communication network in power facilities such as power plants and substations, which is necessary to ensure production command, equipment monitoring, and safe and stable operation. The multiple communication subsystems of the power plant communication system specifically include the transmission equipment system, the program-controlled switching system, the wireless communication system, the data network system, and the power environment monitoring system. The monitoring data refers to the variable data related to the system's operating status, such as performance parameters, status indicators, and alarm information, collected from the above-mentioned communication subsystems.

[0031] The power plant's communication system automatically acquires operational status data periodically at set time intervals from its various independent communication subsystems. This can be achieved by deploying a unified data acquisition platform and configuring communication interfaces and protocols compatible with each subsystem. After data acquisition, preliminary integration processing can be performed to resolve inconsistencies in format, timing, and measurement between data from different sources.

[0032] In a feasible implementation scenario, a data acquisition command is triggered once according to a preset period (e.g., every 5 minutes). In response to the data acquisition command, data requests are sent to each communication subsystem in parallel, or data frames are actively pushed by each communication subsystem. The acquired raw data (i.e., initial monitoring data) is preprocessed and then converted into structured monitoring data.

[0033] In one feasible embodiment, step S10 may include steps S101-S102: Step S101: Periodically collect initial monitoring data from multiple communication subsystems of the power plant communication system; It should be noted that initial monitoring data refers to the raw, unprocessed monitoring data obtained directly from each communication subsystem.

[0034] According to the preset monitoring data collection cycle, the system automatically obtains initial monitoring data from the management interfaces of each communication subsystem. The collection process needs to adapt to the interface protocols and data formats of different communication subsystems to ensure the integrity and timeliness of the data. A data quality inspection mechanism can also be established to promptly detect collection anomalies.

[0035] In a feasible implementation scenario, the system is configured to perform a full data collection every 5 minutes. During the collection process, metadata such as data source and collection timestamp are recorded, and data points that fail to be collected are retried to ensure the integrity of the data collection.

[0036] Step S102: Preprocess the initial monitoring data to obtain monitoring data. The preprocessing includes data cleaning, data alignment and data standardization.

[0037] It should be noted that data cleaning refers to identifying and processing quality issues such as outliers and missing values ​​in data; data alignment refers to unifying data with different timestamps to the same time base; and data standardization refers to converting data with different units of measurement to a unified numerical range.

[0038] The initial monitoring data collected is preprocessed to eliminate data quality issues and resolve consistency problems among multi-source data. The preprocessing process needs to balance processing effectiveness and efficiency, controlling computational overhead while ensuring data quality. A data quality rule base can be established to automatically identify abnormal data; a time series alignment algorithm can be designed to process data with different sampling frequencies; and an appropriate standardization method can be selected to eliminate the influence of units.

[0039] In a feasible implementation scenario, the initial monitoring data is first cleaned to detect and remove outliers that significantly exceed the physical measurement range. Missing data is filled in using linear interpolation of data from previous and subsequent time points to eliminate noise and outliers from interfering with subsequent model analysis. Next, data alignment is performed, using whole minutes as the time base. For data from non-whole minutes, nearest neighbor interpolation is used to align to whole-minute timestamps, addressing the time synchronization issue of multi-source heterogeneous data. Then, data standardization is performed, using methods such as Z-Score normalization or Min-Max normalization, depending on the data distribution characteristics. This linearly maps the values ​​of various monitoring data to the [0,1] interval, making monitoring data of different properties comparable. After preprocessing, standardized monitoring data is obtained.

[0040] In addition, for data acquisition, a push-pull combination mode can be adopted. For subsystems that support data push, a listening service can be configured, while for subsystems that only support query, an active pull mode can be maintained. During data alignment, different interpolation methods can be selected for different types of data characteristics. For example, linear interpolation can be used for continuously changing temperature data, while forward padding can be used for discrete device signals.

[0041] Step S20: Input the monitoring data into the preset health value detection model to obtain the system health value. The health value detection model is a stacked autoencoder model and is obtained through pre-training. It should be noted that the health value detection model in this embodiment refers to an unsupervised learning model based on a stacked autoencoder (SAE). A stacked autoencoder is a deep neural network structure composed of multiple autoencoders stacked together, with the core consisting of an encoder and a decoder. The system health value is a continuous quantization score between 0 and 1, used to characterize the overall operational health status of the entire power station communication system, where 1 represents absolute health and 0 represents complete failure.

[0042] By analyzing the monitoring data through a pre-trained health value detection model, a quantitative score representing the overall health level of the power plant communication system is output. The core mechanism of this health value detection model is to learn the operating data pattern of the power plant communication system under normal conditions and to evaluate the system health by comparing the deviation between real-time data and the learned pattern.

[0043] In a feasible implementation scenario, preprocessed monitoring data is directly input into a deployed health value detection model. The model first performs nonlinear transformation and dimensionality reduction on the input data through multiple fully connected layers in its encoder, mapping it to a low-dimensional, dense latent feature space. This process effectively removes redundant information from the monitoring data, retaining the most critical pattern features. Then, the model's decoder, based on these latent features, reconstructs the original input data using a symmetrical network layer structure. Internally, the model calculates the difference between the reconstructed data and the original input data in real time—the reconstruction error. This reconstruction error is fed into a subsequent mapping function, such as a monotonically decreasing exponential function, which converts the reconstruction error into a system health value between 0 and 1. When the real-time data closely matches the normal pattern, the reconstruction error is small, and the health value approaches 1; when abnormal deviations occur, the error increases, and the health value decays towards 0.

[0044] Step S30: Input the sequence formed by the change of system health value and monitoring data over time into the preset fault prediction model to predict the probability value of the power station communication system failing within the preset time range, and trigger different levels of early warning signals based on the probability value. The fault prediction model is a long short-term memory model and is obtained through pre-training.

[0045] It should be noted that the fault prediction model in this embodiment is specifically a Long Short-Term Memory (LSTM) network, which is a special type of recurrent neural network that can effectively learn and remember long-term dependencies in time series; the preset time range refers to a specific time length in the future, such as 12 minutes or 1 hour in the future.

[0046] The sequence of system health values ​​and monitoring data changes over time is input into a pre-trained fault prediction model. This model is a long short-term memory model, which, after pre-training, has the ability to effectively process time series data. Through this model, the probability value of a fault occurring in the power plant communication system within a preset time range can be predicted. Then, based on the obtained probability value, different levels of early warning signals are triggered so that operation and maintenance personnel can take corresponding measures in a timely manner. The level of the early warning signal can be directly classified by comparing the probability value with the classification threshold, or more complex decision-making logic can be introduced.

[0047] In a feasible implementation scenario, the system maintains a sliding window of a fixed length, continuously incorporating the latest calculated system health values and the acquired monitoring data into the window, and removing the earliest value, thereby forming a latest time series segment. This sequence of fixed length is fed as input data into the trained fault prediction model. The model analyzes the change patterns, fluctuation periods, and downward trends in the sequence from the recent to the long-term. Finally, through the fully connected layer and activation function at its end, it outputs a scalar value between 0 and 1, which is the probability value P_fault of a fault occurring within a certain future time unit. The system has pre-set fault classification thresholds. For example, θ_warn = 0.7 and θ_critical = 0.9 are set. The system compares the real-time obtained P_fault with these thresholds. If P_fault > θ_critical, an emergency alarm is generated and issued to notify the operation and maintenance team to intervene immediately; if θ_warn < P_fault ≤ θ_critical, an early warning is issued to prompt attention to the system situation; if P_fault ≤ θ_warn, the system is considered to be operating normally.

[0048] Thus, in this embodiment, through the health value model, the system can adaptively perceive minor deviations in the overall state, overcoming the defect that the fixed threshold is insensitive to the dynamic changes of the environment. Through the fault prediction model, the system can issue early warnings dozens of minutes or even hours before a fault occurs based on the downward trend of the health degree, realizing the transformation from after-the-fact maintenance to predictive maintenance. Compared with the traditional fault warning method, this embodiment can advance the fault warning time of the power station communication system by an average of several hours and reduce meaningless alarms caused by accidental fluctuations.

[0049] In addition, in a feasible implementation manner, multiple health value detection models can be run in parallel to process data of different time scales, such as seconds, minutes, and hours. The types and quantities of monitoring data considered at different time scales can be different, so as to simultaneously capture instantaneous sudden anomalies and long-term performance decay, further improving the detection accuracy of the system operating state.

[0050] In a feasible embodiment, after step S30, steps S40 to S50 may further be included: Step S40, when the probability value exceeds the preset alarm threshold, extract statistical features from the monitoring data, extract deep features from the bottleneck layer of the health value detection model, and extract time series features from the hidden state of the fault prediction model; It should be noted that the preset alarm threshold refers to the probability threshold for triggering the fault diagnosis process set based on operation and maintenance experience, and the specific value can be set according to the actual application scenario; statistical features refer to the statistical quantities calculated from the monitoring data, such as mean, variance, extreme values, etc.; deep features refer to the low-dimensional and abstract feature representation extracted from the bottleneck layer of the stacked autoencoder; and temporal features refer to the features containing sequence dynamic information extracted from the hidden state of LSTM.

[0051] When the probability value output by the fault prediction model exceeds the preset threshold, the system automatically obtains feature information that reflects the system state from different levels and sources, including statistical features calculated from monitoring data, deep features extracted from the bottleneck layer of the health value detection model, and time-series features extracted from the hidden state of the fault prediction model.

[0052] In a feasible implementation scenario, taking a preset alarm threshold of 0.7 as an example, when the fault probability value output by the fault prediction model exceeds 0.7, the system calculates the mean, variance, maximum value, and slope of the trend for each monitoring data (or some key monitoring data) from the monitoring data of the most recent 15 minutes, forming a statistical feature vector. At the same time, the monitoring data at the current moment is input into the health value detection model. During the forward propagation of the model, the latent vector is extracted from the last layer (bottleneck layer) of the encoder as a deep feature. Meanwhile, the LSTM hidden state is extracted from the last time step of the fault prediction model to obtain a vector as a temporal feature. These three features describe the current state and historical evolution trend of the system from different perspectives.

[0053] Step S50: Construct a hybrid feature vector based on statistical features, deep features, and temporal features, and input the hybrid feature vector into a preset fault analysis model to obtain the fault analysis result. The fault analysis model is a gradient boosting tree model and is obtained through pre-training.

[0054] It should be noted that the fault analysis model refers to the machine learning model used for fault classification and root cause localization. In this embodiment, the gradient boosting tree model is used, and the fault analysis results include fault type classification and feature ranking that leads to the fault.

[0055] The extracted statistical features, deep features, and temporal features are fused to construct a unified feature representation, resulting in a hybrid feature vector. This hybrid feature vector is then input into a pre-trained fault analysis model to obtain the fault analysis results output by the model. The feature fusion methods can include normalizing the dimensions of different types of features before concatenating them, weighting the fusion based on feature importance, or dynamically adjusting the weights of different features using an attention mechanism.

[0056] In a feasible implementation scenario, the extracted statistical features, deep features, and time-series features are concatenated to obtain a hybrid feature vector. This hybrid feature vector is then input into a pre-trained fault analysis model. The fault analysis results output by the model include the fault type and the feature ranking that caused the fault. Fault types include power failure, transmission link failure, or equipment overheating. The feature ranking can list the top 10 features that contribute the most to the current fault diagnosis. Based on the feature ranking results, specific performance indicators and equipment can be traced back to achieve accurate root cause localization.

[0057] For example, to help understand the implementation process of the power plant communication system operation fault early warning method in this embodiment, please refer to... Figure 2 , Figure 2 This paper presents an overall architecture and data processing logic for a fault early warning system for power plant communication systems. In the data acquisition and preprocessing stage, initial monitoring data is obtained from communication subsystems such as the transmission system, program-controlled switching system, and power environment system. After data cleaning, alignment, and standardization, the initial monitoring data is transformed into standardized, regular monitoring data. Then, the monitoring data enters the intelligent analysis core stage. First, the health detection model trained by SAE processes the data, and its output is passed to the fault prediction model trained by LSTM. When the prediction result triggers an early warning, the fault diagnosis model trained by LightGBM (Light Gradient Boosting Machine) is activated. Finally, in the result presentation and action stage, the outputs of all models are integrated to generate a global health panel, early warning and diagnostic reports, i.e., fault analysis results, and drive the generation of maintenance work orders, achieving integrated health assessment and fault early warning for the power plant communication system.

[0058] Therefore, by constructing a hybrid feature vector, the interpretability of statistical features, the abstract representation capability of deep features, and the dynamic information of time-series features are integrated, providing a comprehensive data foundation for fault analysis. Using a gradient boosting tree model for fault classification and root cause localization not only ensures high classification accuracy but also provides good interpretability through feature importance analysis, helping maintenance personnel to quickly locate the source of the problem and significantly shorten the fault troubleshooting time.

[0059] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description and will not be repeated hereafter. Furthermore, steps A10 to A20 may be included before step S20: Step A10: Obtain historical monitoring data of the power plant communication system during the fault-free operation phase as the first training sample data; It should be noted that the fault-free operation phase refers to the period during which the power plant's communication system operates stably without any faults affecting its business functions within a historical time interval; historical monitoring data is the monitoring data collected and stored from each communication subsystem during this historical time interval.

[0060] We screened all monitoring data corresponding to time periods clearly marked as fault-free from long-term operation and maintenance data. These monitoring data were obtained by querying historical databases or extracting from archived files. During the screening process, we combined operation and maintenance work orders and alarm logs for cross-validation to ensure that the selected data was not contaminated by any known fault events. We used the historical monitoring data of these fault-free operation phases as the first training sample data for training the health value detection model.

[0061] In a feasible implementation scenario, the system accesses the historical database and checks whether any fault alarms are recorded in each 24-hour period. For time periods without alarm records, the system extracts the original monitoring data records of all communication subsystems in that period, i.e., historical monitoring data. The extracted data can be preprocessed to obtain the first training sample data for training the health value detection model.

[0062] Step A20: Based on the first training sample data, train the stacked autoencoder model to be trained with the goal of minimizing the reconstruction error between the model output data and the model input data, and obtain the health value detection model.

[0063] The first training sample data is input into the stacked autoencoder model to be trained. The model parameters are adjusted through an iterative optimization process. During training, the model learns to compress and reconstruct the input data. The model optimization objective is to minimize the difference between the reconstructed data and the original input data. Validation sample data can also be set to monitor the training process to prevent overfitting.

[0064] In a feasible implementation scenario, the first training sample data is divided into mini-batches and sequentially input into a stacked autoencoder model. For each batch of data, the model performs forward propagation, meaning the data is compressed sequentially through the encoder's multi-layer neural network and then reconstructed through the decoder's symmetric network. The reconstruction error between the reconstructed data and the original input data is then calculated as the loss for this round. Based on this loss, the training algorithm calculates the gradient through backpropagation and updates all parameters in the model. This process is iterated for tens to hundreds of cycles until the reconstruction error no longer decreases significantly, indicating that the model has fully learned the distribution characteristics of normal data. At this point, training is complete, and a usable health value detection model is obtained. The health evaluation criteria of the trained health value detection model can adapt to the specific configuration and operating characteristics of different power station communication systems, ensuring the accuracy and reliability of system health value evaluation.

[0065] In one feasible embodiment, steps B10 to B20 may be included before step S30: Step B10: Construct historical time series data as the second training sample data, where each element in the historical time series represents the system health value and monitoring data at a point in time. It should be noted that the historical time series refers to the sequence of system state data points arranged in chronological order. Each data point contains the health value at that moment and the corresponding monitoring data. The second training sample data is a labeled dataset specifically used to train the fault prediction model, in which each sample consists of a fixed-length historical time series segment and a corresponding label.

[0066] Construct a second training sample data for training the fault prediction model. Each sample data consists of a time series of T consecutive time points. The time series includes the health value at each time point and key monitoring data or full monitoring data selected from the monitoring data. Add a label to each time series. The label is defined based on whether a fault actually occurs within a certain time unit after the end of the time series segment.

[0067] In a feasible implementation scenario, the system extracts continuous time-series segments from the stored historical database using a sliding window of length T (e.g., T=60). Each segment contains data from 60 consecutive time points (e.g., one point per minute). The data vector for each time point includes the system health value at that moment, as well as several key performance indicators selected from the monitoring data, such as the total number of alarms, the CPU (Central Processing Unit) load of the core router, and backbone link traffic. Simultaneously, it checks whether the power plant communication system has generated any fault alarms within K time units (e.g., K=1 hour) after the end time point of each sequence segment. If so, the sample is marked as a positive sample; otherwise, it is marked as a negative sample. Furthermore, considering the scarcity of fault events, oversampling or adjusting the weights of the loss function can be used to address the class imbalance problem in the sample data, thereby improving the accuracy of model training.

[0068] Step B20: Train the long short-term memory model to be trained based on the second training sample data to obtain the fault prediction model. The long short-term memory model includes a random deactivation layer, and the loss function used for training is the focus loss function.

[0069] It should be noted that random deactivation layers refer to the technique of randomly ignoring some neural network units during training to prevent model overfitting; the focus loss function is an improved loss function that addresses class imbalance by adjusting the weights of easy and difficult samples.

[0070] Supervised training of the Long Short-Term Memory (LSTM) network was performed using a second training sample. During training, a random deactivation layer was introduced after the LSTM layer. A multi-layer LSTM structure was used to capture dependencies at different time scales, and an appropriate random deactivation rate, such as 0.2-0.5, was set. Focus loss was used as the optimization objective, and the model learned to identify pattern features that led to failures from historical health values ​​and monitoring data sequences.

[0071] In a feasible implementation scenario, a network structure containing two LSTM layers is constructed, with a Dropout layer (random deactivation layer) after each LSTM layer. Time series segments from the second training sample data are input into the network, and the data at each time step is processed sequentially through the LSTM layer. The hidden state at the final time step is mapped to a fault probability output through a fully connected layer. During training, a focus loss function is used, and the formula for calculating the focus loss function is: ; in, This represents the value of the focus loss function. It is the predicted probability that the sample belongs to the target category; It is a focusing parameter used to adjust the degree of weight decay; It is a balancing factor used to adjust the initial weight ratio of positive and negative samples.

[0072] Therefore, by training with labeled historical time series data, the fault prediction model can learn the complex temporal relationship between the deterioration of system health and the occurrence of faults. The focus loss function effectively solves the class imbalance problem caused by the scarcity of fault samples, while the introduction of a random deactivation layer enhances the model's generalization ability and avoids overfitting. This enables the final fault prediction model to provide accurate and reliable system fault early warning information in real-world scenarios.

[0073] In one feasible embodiment, steps C10 to C20 may be included before step S50: Step C10: Obtain historical fault event data as the third training sample data, wherein the historical fault event data includes historical mixed feature vectors and corresponding fault type labels; It should be noted that historical fault event data refers to the exact fault events and related data recorded during past operations and maintenance; fault type labels are fault classification information marked by operations and maintenance personnel according to the actual fault situation, such as "power module failure" and "transmission link interruption".

[0074] Validated failure cases are collected from the historical operation and maintenance database. For each failure case, the feature extraction process is reproduced, and labeled third training sample data is constructed to train the failure analysis model.

[0075] In a feasible implementation scenario, the system accesses the operation and maintenance knowledge base and extracts all confirmed fault event records from the past year. For each fault event, based on the timestamp of the fault occurrence, it extracts the monitoring data within 15 minutes before the fault to construct a historical hybrid feature vector. At the same time, based on the root cause analysis in the operation and maintenance report, it labels each sample with specific fault type changes, such as transmission equipment - optical power degradation, program-controlled exchange - board failure, power environment - air conditioning system abnormality, etc.

[0076] Step C20: Based on the third training sample data, train the gradient boosting tree model to be trained to obtain the fault analysis model.

[0077] It should be noted that the gradient boosting tree model is a machine learning algorithm based on decision tree ensemble, which iteratively trains multiple weak learners (decision trees) and combines them into a strong learner.

[0078] The gradient boosting tree classification model is trained using third training sample data. Through an iterative optimization process, the model is able to accurately identify fault types and evaluate feature importance.

[0079] In a feasible implementation scenario, a gradient boosting tree model is trained using third-order training sample data. Appropriate hyperparameters are set, including the number of trees, maximum tree depth, and learning rate. During training, the model sequentially constructs multiple decision trees, with each tree learning to correct the residuals of its predecessor trees. Multi-class log loss is used as the optimization objective. The resulting fault analysis model can accurately classify fault types and provide feature importance ranking, thus providing a basis for root cause localization.

[0080] Therefore, by training the gradient boosting tree model using real historical fault data, the accuracy and reliability of fault classification are ensured. The resulting fault analysis model can quickly and accurately identify fault types and reveal the root cause of faults through feature importance analysis, thereby improving operation and maintenance efficiency and problem-solving speed.

[0081] In one feasible embodiment, steps S60-S70 may be included after step S50: Step S60: When the fault analysis result indicates that the root cause of the fault is located in the first communication subsystem of the power plant communication system, determine the probability value of the root cause of the fault leading to a derivative fault in the second communication subsystem. It should be noted that the first communication subsystem refers to the subsystem in which the fault initially occurred, as identified by the fault analysis model; the second communication subsystem refers to other communication subsystems that may be affected by the fault of the first communication subsystem.

[0082] Once the fault analysis model determines that the root cause of the fault lies in a specific subsystem (i.e., the first communication subsystem), the system automatically analyzes the potential chain reactions that the fault may trigger. Based on the topological connections between subsystems, historical fault propagation records, and real-time operating status, the system assesses the risk of the fault spreading to other subsystems. Specifically, a dependency matrix between each communication subsystem can be pre-established to analyze fault propagation paths, or a graph neural network can be used to model the influence relationships between subsystems.

[0083] In a feasible implementation scenario, when the fault analysis model identifies the root cause of the fault as being located in the transmission equipment system, the system immediately initiates derivative fault analysis. First, it queries a pre-defined subsystem dependency knowledge base to determine that the subsystems directly supported by the transmission equipment system include the program-controlled switching system, the data network system, and the wireless communication system. Then, based on historical fault data statistics, it calculates that when the transmission system fails, the probability of causing the program-controlled switching system to be interrupted is 0.65, the probability of causing the data network system to be paralyzed is 0.82, and the probability of causing the wireless communication system to fail is 0.45. Simultaneously, the base probabilities are dynamically adjusted based on the real-time health values ​​of each subsystem.

[0084] Step S70: Based on the probability value of the derivative fault occurring in the second communication subsystem, generate a fault influence chain diagram across subsystems, and add the fault influence chain diagram to the fault analysis results.

[0085] It should be noted that the fault impact chain graph is a graphical representation that visually shows the propagation path and impact of faults between different subsystems. The nodes in the graph represent the various communication subsystems, and the edges represent the fault propagation relationships and their probabilities.

[0086] Based on the calculated derivative failure probability, a complete failure impact chain map is constructed to intuitively show the possible impact range and development path of the failure. The map includes the root cause of the failure, the subsystems that may be affected, the key nodes on the propagation path, and the probability of occurrence of each path. Key early warning information and handling suggestions can also be marked in the map.

[0087] In a feasible implementation scenario, the system generates a directed graph containing all five subsystem nodes. Starting from the transmission equipment system where the current root cause of the fault resides, three directed edges are drawn pointing to the program-controlled switching system, the data network system, and the wireless communication system. The calculated derivative fault probability values ​​are labeled on these edges. High-risk propagation paths with a probability exceeding 0.8 are highlighted in red, medium-risk paths with a probability between 0.5 and 0.8 are displayed in yellow, and low-risk paths with a probability below 0.5 are shown in green. A note is added next to the graph indicating that the highest-risk propagation path is currently "transmission equipment system → data network system (probability 0.91)," suggesting that the data transmission link status should be checked first.

[0088] For example, Figure 3 This paper presents a fault impact chain diagram illustrating the propagation logic of a fault among various communication subsystems. Taking a power and environmental system air conditioning system failure leading to an increase in computer room temperature as an example, the power and environmental system is the first communication subsystem to experience the fault. Its failure has a high probability of simultaneously affecting the transmission equipment system and the IT (information technology) equipment system, causing optical power fluctuations and CPU overheating and frequency reduction, respectively. Then, the anomaly of the transmission equipment system has a medium probability of causing communication interruption in the program-controlled switching system and a high probability of causing service delays in the data network system. At the same time, the anomaly of the IT equipment system can also affect the data network system and the wireless communication system, causing service delays and signal degradation. This process reveals how a single fault can form a chain reaction through inter-system dependencies, ultimately evolving into a systemic risk.

[0089] Therefore, by quantitatively assessing the risk of fault propagation between different subsystems, maintenance personnel can predict the trend of fault development and take protective measures in advance.

[0090] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the operation fault early warning method of the power plant communication system of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0091] This application also provides an operational fault early warning device for a power plant communication system. Please refer to... Figure 4 The power plant communication system's fault early warning device includes: Data acquisition module 10 is used to periodically acquire monitoring data from multiple communication subsystems of the power plant communication system; The health value acquisition module 20 is used to input monitoring data into a preset health value detection model to obtain the system health value. The health value detection model is a stacked autoencoder model and is obtained through pre-training. The fault early warning module 30 is used to input the sequence formed by the change of system health value and monitoring data over time into the preset fault prediction model, so as to predict the probability value of the power station communication system failing within a preset time range, and trigger different levels of early warning signals based on the probability value. The fault prediction model is a long short-term memory model and is obtained through pre-training.

[0092] Optionally, the power plant communication system's operation fault early warning device also includes a model training module (not shown), which is used for: Historical monitoring data of the power plant communication system during the fault-free operation phase were used as the first training sample data. Based on the first training sample data, the stacked autoencoder model to be trained is trained with the goal of minimizing the reconstruction error between the model output data and the model input data, thus obtaining the health value detection model.

[0093] Optionally, the model training module is also used for: Historical time series data are constructed as the second training sample data, where each element in the historical time series represents the system health value and monitoring data at a point in time; The Long Short-Term Memory (LSTM) model to be trained is trained based on the second training sample data to obtain the fault prediction model. The LSM model includes a random deactivation layer, and the loss function used for training is the focus loss function.

[0094] Optionally, the fault warning module 30 is also used for: When the probability value exceeds the preset alarm threshold, statistical features are extracted from the monitoring data, deep features are extracted from the bottleneck layer of the health value detection model, and time series features are extracted from the hidden state of the fault prediction model. A hybrid feature vector is constructed based on statistical features, deep features, and temporal features. The hybrid feature vector is then input into a pre-defined fault analysis model to obtain the fault analysis results. The fault analysis model is a gradient boosting tree model and is obtained through pre-training.

[0095] Optionally, the model training module is also used for: Historical fault event data is obtained as the third training sample data, which includes historical mixed feature vectors and corresponding fault type labels; Based on the third training sample data, the gradient boosting tree model to be trained is trained to obtain the fault analysis model.

[0096] Optionally, the fault warning module 30 is also used for: When the fault analysis results indicate that the root cause of the fault is located in the first communication subsystem of the power plant communication system, determine the probability value that the root cause of the fault leads to a derivative fault in the second communication subsystem. Based on the probability value of derivative failures occurring in the second communication subsystem, a fault impact chain diagram across subsystems is generated, and the fault impact chain diagram is added to the fault analysis results.

[0097] Optionally, the data acquisition module 10 is also used for: Initial monitoring data is periodically collected from multiple communication subsystems of the power plant's communication system; The initial monitoring data is preprocessed to obtain the monitoring data. The preprocessing includes data cleaning, data alignment and data standardization.

[0098] The power plant communication system operation fault early warning device provided in this application adopts the power plant communication system operation fault early warning method in the above embodiments, and can realize integrated health assessment and fault early warning of the power plant communication system. Compared with the prior art, the beneficial effects of the power plant communication system operation fault early warning device provided in this application are the same as the beneficial effects of the power plant communication system operation fault early warning method provided in the above embodiments, and other technical features in the power plant communication system operation fault early warning device are the same as the features disclosed in the power plant communication system operation fault early warning method in the above embodiments, and will not be repeated here.

[0099] This application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the power plant communication system operation fault early warning method in the above embodiment 1.

[0100] The following is for reference. Figure 5 The diagram illustrates a structural schematic of an electronic device suitable for implementing embodiments of this application. The electronic devices in these embodiments may include, but are not limited to, mobile terminals such as laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0101] like Figure 5As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. The communication device 1009 allows the electronic device to exchange data with other devices wirelessly or via wired communication. Although the diagrams show electronic devices with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented alternatively.

[0102] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0103] The electronic device provided in this application adopts the power plant communication system operation fault early warning method in the above embodiments, which can realize integrated health assessment and fault early warning of the power plant communication system. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the power plant communication system operation fault early warning method provided in the above embodiments, and other technical features of the electronic device are the same as those disclosed in the power plant communication system operation fault early warning method of the previous embodiment, and will not be repeated here.

[0104] It should be understood that the various parts disclosed in the embodiments of this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0105] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0106] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the power plant communication system operation fault early warning method in the above embodiments.

[0107] The computer-readable storage medium provided in this application embodiment may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0108] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.

[0109] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by an electronic device, the electronic device causes the electronic device to: periodically acquire monitoring data from multiple communication subsystems of the power plant communication system; input the monitoring data into a preset health value detection model to obtain a system health value, wherein the health value detection model is a stacked autoencoder model and is obtained through pre-training; input the system health value and the sequence formed by the change of monitoring data over time into a preset fault prediction model to predict the probability value of a fault occurring in the power plant communication system within a preset time range, and trigger different levels of early warning signals based on the probability value, wherein the fault prediction model is a long short-term memory model and is obtained through pre-training.

[0110] Computer program code for performing the operations of the embodiments of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0111] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0112] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0113] The readable storage medium provided in this application embodiment is a computer-readable storage medium. This medium stores computer-readable program instructions (i.e., a computer program) for executing the above-described power plant communication system operation fault early warning method, enabling integrated health assessment and fault early warning of the power plant communication system. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application embodiment are the same as the beneficial effects of the power plant communication system operation fault early warning method provided in the above embodiments, and will not be repeated here.

[0114] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for early warning of operational faults in a power plant communication system, characterized in that, The method for early warning of operational faults in the power plant communication system includes: Monitoring data is periodically acquired from multiple communication subsystems of the power plant's communication system; The monitoring data is input into a preset health value detection model to obtain the system health value. The health value detection model is a stacked autoencoder model and is obtained through pre-training. The system health value and the sequence formed by the change of the monitoring data over time are input into a preset fault prediction model to predict the probability value of the power station communication system failing within a preset time range, and trigger different levels of early warning signals based on the probability value. The fault prediction model is a long short-term memory model and is obtained through pre-training.

2. The method for early warning of operational faults in a power plant communication system as described in claim 1, characterized in that, Before the step of inputting the monitoring data into a preset health value detection model to obtain the system health value, the method further includes: Historical monitoring data of the power plant communication system during the fault-free operation phase were obtained as the first training sample data. Based on the first training sample data, the stacked autoencoder model to be trained is trained with the goal of minimizing the reconstruction error between the model output data and the model input data, thereby obtaining the health value detection model.

3. The method for early warning of operational faults in a power plant communication system as described in claim 1, characterized in that, Before the step of inputting the sequence formed by the changes of the system health value and the monitoring data over time into a preset fault prediction model to predict the probability value of the power station communication system failing within a preset time range, the method further includes: A historical time series is constructed as the second training sample data, wherein each element in the historical time series represents the system health value and monitoring data at a point in time; The fault prediction model is obtained by training the long short-term memory model to be trained based on the second training sample data. The long short-term memory model includes a random deactivation layer, and the loss function used for training is the focus loss function.

4. The method for early warning of operational faults in a power plant communication system as described in claim 1, characterized in that, After the step of triggering different levels of warning signals based on the probability value, the method further includes: When the probability value exceeds the preset alarm threshold, statistical features are extracted from the monitoring data, deep features are extracted from the bottleneck layer of the health value detection model, and temporal features are extracted from the hidden state of the fault prediction model. A hybrid feature vector is constructed based on the statistical features, the deep features, and the temporal features, and the hybrid feature vector is input into a preset fault analysis model to obtain the fault analysis result. The fault analysis model is a gradient boosting tree model and is obtained through pre-training.

5. The method for early warning of operational faults in a power plant communication system as described in claim 4, characterized in that, Before the step of inputting the hybrid feature vector into a preset fault analysis model to obtain the fault analysis result, the method further includes: Historical fault event data is obtained as the third training sample data, wherein the historical fault event data includes historical mixed feature vectors and corresponding fault type labels; Based on the third training sample data, the gradient boosting tree model to be trained is trained to obtain the fault analysis model.

6. The method for early warning of operational faults in a power plant communication system as described in claim 4, characterized in that, After the step of inputting the hybrid feature vector into a preset fault analysis model to obtain the fault analysis result, the method further includes: When the fault analysis result indicates that the root cause of the fault is located in the first communication subsystem of the power plant communication system, the probability value of the root cause of the fault leading to a derivative fault in the second communication subsystem is determined. Based on the probability value of derivative failures occurring in the second communication subsystem, a fault impact chain diagram across subsystems is generated, and the fault impact chain diagram is added to the fault analysis results.

7. The method for early warning of operational faults in a power plant communication system as described in claim 1, characterized in that, The step of periodically acquiring monitoring data from multiple communication subsystems of the power plant communication system includes: Initial monitoring data is periodically collected from multiple communication subsystems of the power plant's communication system; The initial monitoring data is preprocessed to obtain monitoring data, wherein the preprocessing includes data cleaning, data alignment and data standardization operations.

8. A fault early warning device for a power plant communication system, characterized in that, The power plant communication system's fault early warning device includes: The data acquisition module is used to periodically acquire monitoring data from multiple communication subsystems of the power plant's communication system; The health value acquisition module is used to input the monitoring data into a preset health value detection model to obtain the system health value, wherein the health value detection model is a stacked autoencoder model and is obtained through pre-training; The fault early warning module is used to input the system health value and the sequence formed by the change of the monitoring data over time into a preset fault prediction model to predict the probability value of the power station communication system failing within a preset time range, and to trigger different levels of early warning signals based on the probability value. The fault prediction model is a long short-term memory model and is obtained through pre-training.

9. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the power plant communication system operation fault early warning method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the power plant communication system operation fault early warning method as described in any one of claims 1 to 7.

Images