Management method and system based on intelligent safety inspection of new energy station

Abstract

The invention discloses a management method based on intelligent safety inspection of a new energy station, and relates to the technical field of intelligent safety inspection, and the method comprises the steps: carrying out the feature extraction and fusion through a convolutional neural network and a long-short-term memory network based on multi-modal data, and generating an equipment state report through a random forest algorithm; based on the equipment state report, constructing a risk prediction model to calculate a risk score, and generating safety inspection early warning information; based on the safety inspection early warning information, key features are extracted through a principal component analysis method, a fault prediction model is trained, and a fault prediction report is generated; according to the method, the spatial features are extracted through the convolutional neural network, the time features are extracted through the long-short-term memory network, organic fusion of the spatial features and the time features is realized in combination with feature splicing and a random forest algorithm, and the accuracy and timeliness of equipment state evaluation and fault prediction are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent safety inspection technology, and in particular to a management method and system based on intelligent safety inspection of new energy stations. Background Art

[0002] With the rapid development of new energy technologies, new energy stations have been widely used worldwide. In recent years, with the advancement of the Internet of Things, big data, and artificial intelligence, intelligent safety inspection technology has gradually become mainstream. For example, machine learning-based fault prediction models and deep learning-based anomaly detection algorithms are widely used for equipment status monitoring and risk warning.

[0003] The application of existing technologies in new energy station safety inspections still faces the following major challenges: First, the ability to integrate and process multimodal data is insufficient. Equipment operation data comes from diverse sources, and most methods simply splice data from different modalities together. This simplistic splicing approach can lead to feature redundancy or information loss, thus affecting the final fault prediction results. Second, the self-healing capabilities of existing inspection systems are limited. While some inspection systems can trigger corresponding actions based on fault prediction results, these actions are typically fixed, pre-set strategies that lack flexibility and adaptability. Summary of the Invention

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

[0005] Therefore, the present invention provides a management method based on intelligent safety inspection of new energy stations to solve the problems of insufficient multimodal data fusion and fixed self-healing strategy.

[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 a management method based on intelligent safety inspection of new energy stations, which includes collecting equipment operation data, preprocessing, generating fault labels through timestamp alignment and combining historical fault records, and finally constructing multimodal data; based on multimodal data and fault labels, feature extraction and fusion are performed through convolutional neural networks and long short-term memory networks, and a random forest algorithm is used to generate an equipment status report of the new energy station; based on the equipment status report, a risk prediction model is constructed to calculate the risk score and generate safety inspection warning information; based on the safety inspection warning information, key features are extracted and a fault prediction model is trained through principal component analysis to generate a fault prediction report; self-healing operation strategies are defined based on historical data, self-healing operations are performed based on the fault prediction report, and the fault prediction model and self-healing strategy are optimized according to the equipment operation data after self-healing.

[0008] As a preferred solution of the management method based on intelligent safety inspection of new energy stations according to the present invention, the operation data includes temperature data, humidity data, pressure data, current data, vibration data, load data and equipment image data;

[0009] The preprocessing includes data cleaning, denoising and standardization, and obtaining fault labels through timestamp alignment to construct multimodal data.

[0010] As a preferred solution of the management method based on intelligent safety inspection of new energy stations described in the present invention, wherein: based on multimodal data and fault labels, feature extraction and fusion are performed through convolutional neural networks and long short-term memory networks, and a random forest algorithm is used to generate an equipment status assessment report for the new energy station. The specific steps are as follows:

[0011] The convolutional neural network is used to perform convolution and pooling operations on the running data to extract the spatial features of the running data, and the gating mechanism of the long short-term memory network is used to extract the temporal features in the running data;

[0012] The spatial features extracted by the convolutional neural network and the temporal features extracted by the long short-term memory network are fused through the feature splicing method to generate a comprehensive feature vector;

[0013] Based on the comprehensive feature vector and fault labels, the random forest algorithm is used to generate multiple decision trees in parallel. Each tree selects the features that split the comprehensive feature vector and the critical value of the comprehensive feature vector by maximizing information gain. Finally, the classification results of all trees are aggregated through the majority voting mechanism to generate an equipment status assessment report.

[0014] As a preferred solution of the management method based on intelligent safety inspection of new energy stations described in the present invention, wherein: based on the equipment status assessment report, a risk prediction model is constructed to calculate the risk score. The specific steps are as follows:

[0015] Build a risk prediction model based on SVM and train it through comprehensive feature vectors;

[0016] The comprehensive feature vector is mapped to a high-dimensional space through the kernel function to generate the optimal classification hyperplane;

[0017] Based on the optimal classification hyperplane, the risk score p of the device is calculated by combining the decision function with the sigmoid activation function.

[0018] As a preferred solution of the management method based on intelligent safety inspection of new energy stations described in the present invention, the specific steps of generating safety inspection warning information are as follows:

[0019] Define low-risk threshold P1 and high-risk threshold P2 based on historical data;

[0020] When p < P1, it is considered that the current device is in a low-risk state, and routine inspection is required;

[0021] When P1 ≤ p < P2, it is considered that the current device is in a medium-risk state, and the inspection cycle needs to be shortened;

[0022] When p ≥ P2, it is considered that the current device is in a high-risk state, and power limitation is required and an expert needs to be contacted for review.

[0023] As a preferred solution of the management method for intelligent safety inspection of new energy power stations according to the present invention, wherein: based on the safety inspection warning information, key features are extracted by the principal component analysis method and a fault prediction model is trained to generate a fault prediction report. The specific steps are as follows.

[0024] Based on the safety inspection warning information, equipment risk scores and historical data, a time series data set is constructed according to the alignment of equipment and timestamps;

[0025] Based on the time series data set, dimensionality reduction is performed by principal component analysis to generate a low-dimensional feature matrix;

[0026] Based on the low-dimensional feature matrix, a fault prediction model is constructed through the LSTM-XGBoost joint architecture, and the low-dimensional feature matrix and fault labels are used for training. The future fault probability L is predicted through the trained fault prediction model;

[0027] According to the prediction results, a fault prediction report is generated.

[0028] As a preferred solution of the management method for intelligent safety inspection of new energy power stations according to the present invention, wherein: the self-healing operation strategy is defined based on historical data, the self-healing operation is executed based on the fault prediction report, and the fault prediction model and self-healing strategy are optimized according to the equipment operation data after self-healing. The specific steps are as follows.

[0029] Based on historical data, the state space of the reinforcement learning device, the operation action set and the multi-objective reward function are defined, and the self-healing strategy network is trained through the deep learning algorithm to generate self-healing strategies for different fault types;

[0030] Based on the self-healing strategy, the self-healing operation is executed by judging the trigger condition of the fault prediction probability;

[0031] Based on the equipment operation data after self-healing, the fault prediction model is updated through incremental learning, and the self-healing strategy is dynamically adjusted in combination with reinforcement learning.

[0032] In the second aspect, the present invention provides a management system based on intelligent safety inspection of new energy stations, including a data acquisition module, an equipment status evaluation module, an inspection warning module, a fault prediction module and a self-healing optimization module; the data acquisition module collects the operation data of the equipment, performs preprocessing, generates fault labels through timestamp alignment and combines historical fault records, and finally constructs multimodal data; the equipment status evaluation module, based on multimodal data, extracts and fuses features through convolutional neural networks and long short-term memory networks, and uses a random forest algorithm to generate an equipment status report for the new energy station; the inspection warning module, based on the equipment status report, constructs a risk prediction model to calculate the risk score and generate safety inspection warning information; the fault prediction module, based on the safety inspection warning information, extracts key features through principal component analysis and trains the fault prediction model to generate a fault prediction report; the self-healing optimization module defines the self-healing operation strategy based on historical data, performs self-healing operations based on the fault prediction report, and optimizes the fault prediction model and self-healing strategy according to the equipment operation data after self-healing.

[0033] In a third 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, any step of the management method based on intelligent safety inspection of new energy stations as described in the first aspect of the present invention is implemented.

[0034] In a fourth 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 management method based on intelligent safety inspection of new energy stations as described in the first aspect of the present invention.

[0035] The beneficial effects of this invention include: extracting spatial features through convolutional neural networks and temporal features through long-short-term memory networks, combined with feature splicing and random forest algorithms, achieving an organic fusion of spatial and temporal features, significantly improving the accuracy and timeliness of equipment status assessment and fault prediction. Combined with subsequent risk prediction, fault prediction, and self-healing operations, this invention forms a closed-loop management system from data collection to fault prediction, providing efficient and intelligent technical support for equipment maintenance and fault prevention at new energy stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1This is a flow chart of the management method based on intelligent safety inspection of new energy stations in Example 1.

[0038] Figure 2 This is a schematic diagram of the management system based on intelligent safety inspection of new energy stations in Example 1.

[0039] Figure 3 A flow chart for generating risk prediction and fault prediction reports in Example 1.

[0040] Figure 4 This is a flow chart for optimizing the self-healing operation strategy in Example 1. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0042] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. 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.

[0043] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0044] Example 1, reference Figures 1 to 4 , which is the first embodiment of the present invention, provides a management method based on intelligent safety inspection of new energy stations, including the following steps:

[0045] S1. Collect the operating data of the equipment, perform preprocessing, generate fault labels by aligning the timestamps and combining them with historical fault records, and construct multimodal data.

[0046] Operational data includes temperature data, humidity data, pressure data, current data, vibration data, load data, and equipment image data;

[0047] It should be noted that temperature data reflects temperature changes inside or outside the equipment. Abnormal temperature fluctuations may indicate equipment overheating or other potential problems. Humidity data is used to monitor the humidity level of the equipment's environment. High humidity may cause problems such as short circuits or corrosion in electrical equipment. For equipment such as hydraulic systems or compressors, pressure data is a key indicator used to detect leaks or mechanical failures. Current data records the current intensity of the equipment when it is working. Abnormal current may be a signal of equipment overload or damage to electrical components. Vibration data is used to monitor the vibration of mechanical equipment and identify potential mechanical failures. Load data shows the equipment's workload in different time periods. Excessive load may cause equipment fatigue and premature failure. Image information obtained through cameras or other imaging devices can be used to visually detect defects or abnormalities in the equipment's appearance.

[0048] Preprocessing includes data cleaning, denoising and standardization, and obtaining fault labels through timestamp alignment to construct multimodal data.

[0049] Furthermore, during the data cleaning process, for missing and abnormal operating data at new energy stations caused by environmental interference or equipment failure, weighted interpolation is performed through the correlation between the operating data of adjacent equipment and the historical operating data of the same period to fill the missing values (for example, when the photovoltaic string current data is missing, weighted interpolation is performed using the data of adjacent strings in the same array).

[0050] Furthermore, during the denoising process, for continuous operating data (such as temperature, humidity, pressure, current, vibration, etc.), a low-pass filter is used to smooth the signal and remove high-frequency noise. The low-pass filter allows signals below a certain cutoff frequency to pass through, while attenuating signals above this frequency, thereby retaining the main trend of the data and eliminating unnecessary high-frequency fluctuations. For example, in temperature data, the low-pass filter can effectively reduce small fluctuations caused by instantaneous environmental changes, making the temperature curve smoother and more stable. For more complex noise patterns, the median of the adjacent operating data points is calculated through a median filter to replace the current operating data point, which is used to remove or isolate outliers. For example, in current data, if current peaks due to electromagnetic interference appear at certain time points, the median filter can replace these outliers with the median of the adjacent points, thereby restoring the normal current curve.

[0051] Furthermore, during the normalization process, min-max scaling is used to linearly map operational data of different dimensions to a specific range. For operational data with uneven distribution or unknown upper and lower limits, such as current data and vibration data, Z-score normalization is used to transform the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. Combining min-max scaling and Z-score normalization effectively converts various types of operational data to a uniform numerical range and distribution, generating multimodal data with a uniform scale.

[0052] Furthermore, in the process of obtaining fault labels through timestamp alignment, based on the operating data and historical equipment fault records, each type of operating data is accompanied by a corresponding timestamp. The timestamp correction technology is used to align the operating data from different sources, accurately match the fault time period with the timestamp of the operating data, mark the time period and fault type of the fault, and generate a fault label; finally, the preprocessed operating data is integrated with the fault label to construct multimodal data.

[0053] S2. Based on multimodal data, feature extraction and fusion are performed through convolutional neural networks and long short-term memory networks, and the random forest algorithm is used to generate equipment status reports for new energy stations.

[0054] The convolutional neural network is used to perform convolution and pooling operations on the running data to extract the spatial features of the running data, and the gating mechanism of the long short-term memory network is used to extract the temporal features in the running data;

[0055] Furthermore, convolutional neural networks use convolutional kernels to slide over multimodal data, capturing local patterns and spatial correlations in device operating states. Nonlinear activation functions (such as ReLU) and max pooling operations enhance the expressiveness of spatial features, generating high-dimensional feature maps to represent the spatial structure of the data. Next, spatial features are converted into time series through time serialization and input into a long short-term memory (LSTM) network. The LSTM network utilizes a unique gating mechanism and memory cells to flexibly manage the information flow within the time series data. Specifically, the input gate determines whether new information is added to the memory cell at the current moment, ensuring that the latest operating data can be used to update the memory state. The forget gate selectively discards historical operating data that is no longer useful, preventing irrelevant operating data from interfering with subsequent analysis. The output gate controls which information is output from the memory cell to the next moment, thereby generating an appropriate hidden state for subsequent timestep calculations. As the core, the memory cell is responsible for storing and transmitting key information, ensuring that operating data is retained over long timescales, thereby capturing long-term dependencies in device operating states. By stacking multiple layers of LSTM memory cells, complex and abstract temporal features can be gradually extracted.

[0056] The spatial features extracted by the convolutional neural network and the temporal features extracted by the long short-term memory network are fused through the feature splicing method to generate a comprehensive feature vector;

[0057] It should be noted that, first, the spatial features extracted by the convolutional neural network (such as local patterns of equipment operating status, spatial correlation or high-dimensional feature maps) and the temporal features extracted by the long short-term memory network (such as dynamic changes, long-term dependencies or periodic trends) are aligned in dimension to ensure that they are consistent in feature representation. Then, the two types of features are fused together using feature splicing methods (such as connecting along the feature dimension) to generate a comprehensive feature vector. This process can be simply understood as merging spatial features and temporal features into a new vector by column. For example, if the dimension of the spatial feature vector generated by the convolutional neural network is 128 and the dimension of the temporal feature vector extracted by the long short-term memory network is 64, the dimension of the spliced comprehensive feature vector is 192.

[0058] Based on the comprehensive feature vector and fault labels, the random forest algorithm is used to generate multiple decision trees in parallel. Each tree selects the features that split the comprehensive feature vector and the critical value of the comprehensive feature vector by maximizing information gain. Finally, the classification results of all trees are aggregated through the majority voting mechanism to generate an equipment status assessment report.

[0059] It should be noted that the random forest algorithm performs classification by building multiple decision trees in parallel. During the construction of each tree, the optimal features and their critical values for splitting the integrated feature vector are selected by maximizing information gain. Information gain measures the influence of a specific integrated feature vector on the classification result. Selecting the spatial or temporal feature with the highest information gain as the splitting criterion for the current node can effectively improve classification accuracy. Each decision tree independently selects and splits during training, forming different subtree structures. Ultimately, the classification results of all decision trees are aggregated through a majority voting mechanism. Specifically, for each input sample, all decision trees produce a classification result (e.g., normal or faulty). The final classification result is then determined by majority voting, with the category receiving the most votes being selected as the final prediction. Based on the classification results generated by this process, an equipment status assessment report is automatically generated. This report details the equipment's health status, potential fault points, and recommended maintenance measures. For example, the integrated feature vector contains crack characteristics (spatial features) and temperature variation trends (temporal features) of a photovoltaic panel, with the fault label being "normal" or "faulty." The random forest algorithm uses multiple decision trees to classify features. For example, one tree might select the temperature feature as a splitting node, while another might select the crack feature as a splitting node. Ultimately, the classification results of all trees are combined through majority voting to determine the status of the photovoltaic panel and generate an assessment report, such as "PV panel status: Normal. Regular temperature sensor inspection recommended."

[0060] S3. Based on the device status report, construct a risk prediction model to calculate the risk score and generate safety inspection warning information.

[0061] Construct a risk prediction model based on SVM and train it with the comprehensive feature vector.

[0062] Map the comprehensive feature vector to a high-dimensional space through the kernel function to generate the optimal classification hyperplane.

[0063] It should be noted that a risk prediction model is constructed based on SVM and trained with the comprehensive feature vector as the input. SVM maps the comprehensive feature vector to a high-dimensional space through the Gaussian kernel function, making the originally linearly inseparable data linearly separable in the high-dimensional space. Then, SVM finds the optimal hyperplane in the high-dimensional space through an optimization algorithm, maximizing the distance from the sample points closest to the optimal hyperplane to the optimal hyperplane.

[0064] Based on the optimal classification hyperplane, calculate the risk score of the device through the decision function combined with the sigmoid activation function. The expression is as follows:

[0065]

[0066] where p is the risk score, α i is the weight of the i-th support vector, y i is the label of the i-th support vector, b is the bias term, h i is the comprehensive feature vector of the i-th support vector, h is the comprehensive feature vector, σ is the bandwidth parameter of the kernel function, and N is the total number of support vectors.

[0067] It should be noted that the similarity between the support vector and the comprehensive feature vector of the current device is calculated through the Gaussian kernel function, the distance between the current comprehensive feature vector and the classification hyperplane is calculated through the decision function, and finally the output of the decision function is mapped to the risk score through the sigmoid activation function, representing the probability of the device failing. When the risk score is closer to 1, it indicates a higher probability of the current device failing.

[0068] Define the low-risk threshold P1 and the high-risk threshold P2 based on historical data.

[0069] When p < P1, it is considered that the current device is in a low-risk state and requires regular inspection.

[0070] It should be noted that through historical data analysis, the upper limit P1 of the risk score when the device is in a normal state and the lower limit P2 of the risk score when the device is in a high-risk state are determined. When the probability of the device failing is less than P1, for example, the current probability of the device failing is 0.2, and the low-risk threshold P1 is 0.4, which is much lower than the set low-risk threshold. This means that the current device is operating normally without obvious signs of failure.

[0071] When P1 ≤ p < P2, it is considered that the current device is in a medium risk, and the inspection cycle needs to be shortened;

[0072] It should be noted that when the probability of the device failing is greater than or equal to the low-risk threshold P1 and less than the high-risk threshold P2, for example, the low-risk threshold P1 is 0.4, the risk threshold P2 is 0.6, and the current probability of the device failing is 0.5, which is between the high-risk and low-risk thresholds. This means that the current device has a certain risk of failure, and the inspection cycle needs to be shortened and monitoring needs to be strengthened.

[0073] When p ≥ P2, it is considered that the current device is in a high risk, and the power needs to be limited and an expert needs to be contacted for review.

[0074] It should be noted that when the probability of the device failing is greater than or equal to the high-risk threshold P2, for example, the high-risk threshold is 0.6, and the current probability of the device failing is 0.7, which is greater than the high-risk threshold. This means that the current device has a relatively high risk of failure, and the power needs to be limited immediately to avoid device damage, and an expert needs to be contacted for review and further diagnosis.

[0075] S4. Based on the safety inspection warning information, extract key features through the principal component analysis method and train a fault prediction model to generate a fault prediction report.

[0076] Based on the safety inspection warning report, the device risk score and historical data, construct a time series data set according to the alignment of the device and the timestamp;

[0077] It should be noted that first, align the data according to the device and the timestamp to ensure that the data of each device can be matched at the same time point; then, organize the aligned data in a time series to construct a time series data set. For example, each row of data includes the device, the timestamp, the risk score, the operating data (such as temperature, vibration, rotation speed, etc.) and the fault label (such as normal or faulty). This time series data set can comprehensively reflect the operating state and risk changes of the device at different time points.

[0078] Based on the time series data set, perform dimensionality reduction through principal component analysis to generate a low-dimensional feature matrix;

[0079] Furthermore, the time series features in the time series data set are standardized to eliminate the dimensional and scale differences between different time series features, ensuring that each time series feature has the same weight in the dimensionality reduction process. Then, the standardized time series feature matrix is calculated through the covariance matrix and the time series features are decomposed to obtain time series eigenvalues and time series eigenvectors. The time series eigenvalues are sorted according to their size, and the first few time series eigenvalues and their corresponding time series eigenvectors that can represent the main variability of the data are selected and determined as principal components. The standardized time series feature matrix is projected onto the principal components through principal component analysis to generate a low-dimensional feature matrix.

[0080] Based on the low-dimensional feature matrix, a fault prediction model is constructed through the LSTM-XGBoost joint architecture. The low-dimensional feature matrix and fault labels are used for training. The trained fault prediction model is used to predict the future fault probability. The expression is:

[0081]

[0082] Where L is the failure probability, m is the time series feature vector, which is extracted by processing the low-dimensional feature matrix step by time through LSTM, w is the weight vector of the time series feature m, μ is the statistical feature vector, and the mean of the low-dimensional feature matrix is calculated in the time dimension, β is the weight of μ, λ is the regularization coefficient, and T is the transpose symbol;

[0083] Generate a fault prediction report based on the prediction results.

[0084] It should be noted that the low-dimensional feature matrix is first input into the LSTM network. It is then processed time-step by time-step through a gating mechanism to extract a time series feature vector, which is used to capture the dynamic time series dependencies of the new energy station equipment (such as periodic fluctuations in vibration signals and short-term abnormal temperature changes). The statistical feature vector is then obtained by calculating the mean of the low-dimensional feature matrix along the time dimension. This is used to describe the global steady-state characteristics of the new energy station equipment (such as average energy consumption and steady-state voltage level). The time series feature vector is linearly combined with the weight vector w to quantify the contribution of dynamic features, while the statistical feature vector is linearly combined with the weight w to quantify the contribution of steady-state features. A regularization term λ is added to penalize excessive weights to suppress overfitting. Finally, the above calculation results are mapped to a fixed range using a Sigmoid activation function to obtain the failure probability.

[0085] S5. Define a self-healing operation strategy based on historical data, perform self-healing operations based on the fault prediction report, and optimize the fault prediction model and self-healing strategy based on the device operation data after self-healing.

[0086] Based on historical data, we define the reinforcement learning device state space, action set, and multi-objective reward function. We then train the self-healing strategy network using a deep learning algorithm to generate self-healing strategies for different fault types.

[0087] Based on historical data and by analyzing equipment operating parameters (such as temperature, vibration, speed, etc.) and fault records, the equipment state space is defined and the equipment state is quantified into a vector, for example, [temperature state, vibration state, fault state], providing clear input for subsequent reinforcement learning. Based on the equipment type and fault type, a set of possible self-healing operation actions is designed, such as adjusting the load, limiting power operation, starting backup equipment, and remote restart. A multi-objective reward function is designed, comprehensively considering goals such as equipment operating efficiency, fault recovery time, and resource consumption to provide optimization direction for reinforcement learning. A reinforcement learning algorithm is used to train the self-healing strategy network so that it can generate optimal operation actions based on the equipment state. Finally, corresponding self-healing strategies are generated for different fault types and stored as a rule base or model. For example, electrical faults prioritize remote restart, while mechanical faults prioritize power limitation.

[0088] Based on the self-healing strategy, the fault prediction probability triggers the condition judgment and performs the self-healing operation;

[0089] It should be noted that whether to trigger the self-healing operation is determined based on the fault prediction probability and fault type; then, the matching operation action is selected from the policy library according to the fault type, and the optimal operation is determined; then, the selected self-healing operation is executed through automated means, and the operation effect is monitored in real time; finally, the operation effect is evaluated and a report is generated to provide data support for subsequent strategy optimization.

[0090] Based on the equipment operation data after the self-healing operation, the fault prediction model is updated through incremental learning, and the self-healing strategy is dynamically adjusted in combination with reinforcement learning.

[0091] It should be noted that based on the equipment operation data after the self-healing operation, the fault prediction model is updated through incremental learning so that it can adapt to changes in the equipment status; then, the self-healing strategy is dynamically adjusted in combination with reinforcement learning to optimize the selection and execution of operation actions.

[0092] This embodiment also provides a management system based on intelligent safety inspections of new energy stations, including: a data acquisition module, an equipment status assessment module, an inspection and warning module, a fault prediction module and a self-healing optimization module; the data acquisition module collects the operating data of the equipment, performs preprocessing, generates fault labels through timestamp alignment and combines historical fault records, and constructs multimodal data; the equipment status assessment module, based on multimodal data, extracts and fuses features through convolutional neural networks and long short-term memory networks, and uses a random forest algorithm to generate an equipment status report for the new energy station; the inspection and warning module, based on the equipment status report, constructs a risk prediction model to calculate the risk score and generate safety inspection and warning information; the fault prediction module, based on the safety inspection and warning information, extracts key features through principal component analysis and trains the fault prediction model to generate a fault prediction report; the self-healing optimization module defines the self-healing operation strategy based on historical data, executes the self-healing operation based on the fault prediction report, and optimizes the fault prediction model and self-healing strategy according to the equipment operation data after self-healing.

[0093] This embodiment also provides a computer device, which is suitable for the management method based on intelligent safety inspection of new energy stations, 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 management method based on intelligent safety inspection of new energy stations proposed in the above embodiment.

[0094] The computer device may be a terminal, comprising 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 comprises 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 may 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 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 button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0095] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the management method for intelligent safety inspection based on new energy stations as proposed in the above embodiment; 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 memory, flash memory, disk or optical disk.

[0096] In summary, this invention achieves an organic fusion of spatial and temporal features by extracting spatial features using a convolutional neural network and temporal features using a long-short-term memory network, combined with feature concatenation and a random forest algorithm. This significantly improves the accuracy and timeliness of equipment status assessment and fault prediction. Combined with subsequent risk prediction, fault prediction, and self-healing operations, this invention forms a closed-loop management system from data collection to fault prediction, providing efficient and intelligent technical support for equipment maintenance and fault prevention at new energy stations.

[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. 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 management method based on intelligent safety inspection of new energy stations, characterized by: including Collect the operation data of the acquisition device, preprocess it, align it by timestamp and combine historical fault records to generate fault labels, and construct multimodal data; Based on the multimodal data, extract and fuse features through a convolutional neural network and a long short-term memory network, and use the random forest algorithm to generate a device status report for the new energy power station; Based on the device status report, construct a risk prediction model to calculate the risk score and generate a safety inspection warning message; Based on the safety inspection warning message, extract key features through the principal component analysis method and train a fault prediction model to generate a fault prediction report; Define a self-healing operation strategy based on historical data, execute the self-healing operation based on the fault prediction report, and optimize the fault prediction model and self-healing strategy according to the device operation data after self-healing.

2. The management method based on intelligent safety inspection of new energy stations according to claim 1 is characterized in that: The operation data includes temperature data, humidity data, pressure data, current data, vibration data, load data, and device image data; The preprocessing includes data cleaning, denoising, and normalization processing, and obtains fault labels through timestamp alignment to construct multimodal data.

3. The management method based on intelligent safety inspection of new energy stations according to claim 2 is characterized by: Based on the multimodal data and fault labels, extract and fuse features through a convolutional neural network and a long short-term memory network, and use the random forest algorithm to generate a device status evaluation report for the new energy power station. The specific steps are as follows Perform convolutional pooling operations on the operation data through a convolutional neural network to extract the spatial features of the operation data, and extract the temporal features in the operation data through the gating mechanism of the long short-term memory network; Fuse the spatial features extracted by the convolutional neural network and the temporal features extracted by the long short-term memory network through the feature splicing method to generate a comprehensive feature vector; Based on the comprehensive feature vector and fault labels, use the random forest algorithm to generate multiple decision trees in parallel. Each tree maximizes the information gain, selects the features and critical values of the comprehensive feature vector for splitting the comprehensive feature vector, and finally aggregates the classification results of all trees through the majority voting mechanism and generates a device status evaluation report.

4. The management method based on intelligent safety inspection of new energy stations according to claim 3 is characterized by: Based on the device status evaluation report, construct a risk prediction model to calculate the risk score. The specific steps are as follows Construct a risk prediction model based on SVM and train it through the comprehensive feature vector; Map the comprehensive feature vector to a high-dimensional space through a kernel function to generate an optimal classification hyperplane; Based on the optimal classification hyperplane, calculate the risk score p of the device through the decision function combined with the sigmoid activation function.

5. The management method based on intelligent safety inspection of new energy stations according to claim 4 is characterized in that: The steps for generating the safety inspection warning message are as follows Define a low-risk threshold P1 and a high-risk threshold P2 based on historical data; When p < P1, it is considered that the current device is in a low-risk state and requires regular inspection; When P1 ≤ p < P2, it is considered that the current device is in a medium-risk state and the inspection cycle needs to be shortened; When p ≥ P2, it is considered that the current device is in a high-risk state and power needs to be limited and an expert needs to be contacted for review.

6. The management method based on intelligent safety inspection of new energy stations according to claim 5 is characterized in that: Based on the safety inspection warning message, extract key features through the principal component analysis method and train a fault prediction model to generate a fault prediction report. The specific steps are as follows Based on the safety inspection warning report, device risk score and historical data, construct a time series dataset according to the device and timestamp alignment; Based on the time series data set, the principal component analysis is used to reduce the dimension and generate a low-dimensional feature matrix; Based on the low-dimensional feature matrix, a fault prediction model is constructed through the LSTM-XGBoost joint architecture. The low-dimensional feature matrix and fault labels are used for training. The trained fault prediction model is used to predict the future fault probability L. Generate a fault prediction report based on the prediction results.

7. The management method based on intelligent safety inspection of new energy stations according to claim 6 is characterized in that: The self-healing operation strategy is defined based on historical data, the self-healing operation is performed based on the fault prediction report, and the fault prediction model and self-healing strategy are optimized according to the equipment operation data after self-healing. The specific steps are as follows: Based on historical data, we define the reinforcement learning device state space, action set, and multi-objective reward function. We then train the self-healing strategy network using a deep learning algorithm to generate self-healing strategies for different fault types. Based on the self-healing strategy, the fault prediction probability triggers the condition judgment and performs the self-healing operation; Based on the equipment operation data after the self-healing operation, the fault prediction model is updated through incremental learning, and the self-healing strategy is dynamically adjusted in combination with reinforcement learning.

8. A management system based on intelligent safety inspection of new energy stations, based on the management method based on intelligent safety inspection of new energy stations according to any one of claims 1 to 7, characterized in that: Including data acquisition module, equipment status assessment module, inspection and warning module, fault prediction module and self-healing optimization module; The data acquisition module collects the operating data of the equipment, performs preprocessing, and generates fault labels by aligning timestamps and combining them with historical fault records, ultimately constructing multimodal data; The equipment status assessment module uses multimodal data and fault labels to extract and fuse features through convolutional neural networks and long short-term memory networks, and uses a random forest algorithm to generate equipment status reports for new energy stations. The inspection and early warning module builds a risk prediction model based on equipment status reports to calculate risk scores and generate safety inspection and early warning information; The fault prediction module, based on safety inspection warning information, extracts key features through principal component analysis and trains the fault prediction model to generate a fault prediction report; The self-healing optimization module defines the self-healing operation strategy based on historical data, performs self-healing operations based on fault prediction reports, and optimizes the fault prediction model and self-healing strategy based on the equipment operation data after self-healing.

9. 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 management method based on intelligent safety inspection of new energy stations according to any one of claims 1 to 7 are implemented.

10. 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 management method based on intelligent safety inspection of new energy stations according to any one of claims 1 to 7 are implemented.

Images