Substation operation state evaluation method and system based on comparative learning algorithm

Abstract

The invention discloses a transformer substation operation state evaluation method and system based on a comparative learning algorithm, and the method comprises the steps: obtaining the operation state historical data of transformer substation equipment, and constructing a transformer substation historical operation state data set; taking the state characteristic index of the primary equipment as a substation state evaluation index, and constructing an operation state characteristic set; determining operation state levels of all samples; establishing an initial contrast learning model based on a neural network model, extracting feature vectors of all samples, generating positive and negative sample pairs from the samples of the same and different operation state levels, constructing a contrast loss function according to the distance of the feature vectors and the positive and negative of the sample pairs, and training a substation state evaluation contrast learning model; and monitoring the operation state of primary equipment in the transformer substation in real time, inputting the trained model, and outputting the operation state grade of the transformer substation. Limited samples can be fully utilized, good evaluation performance is shown under the conditions of small samples and unbalanced samples, and the operation and maintenance work efficiency of the transformer substation can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of substation operation status evaluation and processing, and more specifically, relates to a substation operation status evaluation method and system based on a contrastive learning algorithm. Background Art

[0002] With the continuous growth of electricity demand and the increasing complexity of power systems, the reliable operation of power systems is particularly important. As a key node in the power system, the operating status of substations directly affects the stability and reliability of the entire power system. The diagnosis and evaluation of substation operation failures is crucial to ensure the stable operation of the power system and is also an important means to ensure the safety of the power system. Through real-time monitoring and data analysis of key substation equipment, abnormal conditions and potential fault hazards can be discovered in time, so that targeted preventive measures can be taken to avoid equipment damage and power outages, reduce economic losses, and improve power supply reliability and safety.

[0003] At present, substation operation status assessment mainly relies on regular maintenance and manual inspections, which is not only time-consuming and labor-intensive, but also cannot reflect the operation status of the equipment in real time, and has lags and subjectivity. In recent years, with the development of the Internet of Things, big data and artificial intelligence technologies, data-driven substation status assessment methods have gradually become a research hotspot.

[0004] In the prior art, there is a fuzzy logic method based on Q-Voc slope and SOH rule base to monitor the operating status of substation batteries. In the prior art, there is a substation secondary equipment status assessment method based on grey theory and cloud model, which improves the traditional single indicator assessment method. In the prior art, there is a substation whole-station monitoring and diagnosis system based on SpringBoot architecture, and a diagnosis expert library for substation maintenance faults has been established. However, the prior art still has many limitations in data processing, feature extraction, and model accuracy, especially in how to effectively use multi-source heterogeneous data for accurate status assessment, which still needs in-depth research. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the present invention provides a substation operation status evaluation method and system based on a contrastive learning algorithm, which accurately evaluates the overall operation status of the substation primary equipment under the conditions of small samples and sample imbalance, promptly discovers abnormal conditions of various equipment during the operation of the substation, and reduces the human resources required in the operation and maintenance of the substation.

[0006] The present invention adopts the following technical solution.

[0007] A first aspect of the present invention provides a method for evaluating substation operation status based on a contrastive learning algorithm, comprising:

[0008] Obtain the historical operating status data of substation equipment and construct a historical operating status data set of the substation; use the status characteristic indicators of primary equipment as substation status evaluation indicators to construct an operating status characteristic set;

[0009] According to the operating status feature set, the operating status level of the substation is divided, and the operating status level of all samples in the substation historical operating status data set is determined;

[0010] An initial contrastive learning model is established based on a neural network model. The feature vectors of each sample in the substation historical operating status data set are extracted. According to the set rules, positive sample pairs are generated from samples with the same operating status level, and negative sample pairs are generated from samples with different operating status levels. A contrastive loss function is constructed based on the distance between the feature vectors corresponding to the two samples in each sample pair and the positive and negative values ​​of the sample pairs to train the substation status assessment contrastive learning model.

[0011] Monitor the operating status of primary equipment in the substation in real time, input the real-time operating status data of the equipment into the trained substation status assessment comparative learning model, and output the substation operating status level.

[0012] Preferably, the status characteristic indicators of the primary equipment include:

[0013] Transformer status characteristic indicators, including: transformer oil temperature, oil level, oil color spectrum and transformer partial discharge evaluation indicators;

[0014] Status characteristic indicators of high-voltage switchgear, including: density and pressure of insulating gas, switch partial discharge frequency, mechanical displacement, contact acceleration and main circuit bus temperature;

[0015] The status characteristic indicators of the lightning arrester include: temperature, humidity, total current and resistive current.

[0016] Preferably, the divided operating status levels of the substation include four levels: normal state, caution state, abnormal state and severe state.

[0017] Preferably, training the substation state assessment comparative learning model further includes:

[0018] The original data in the substation historical operation status dataset is preprocessed and divided into a training set and a test set;

[0019] Perform feature extraction on the preprocessed training set to obtain the feature vector of each sample in the training set;

[0020] Set a label for each sample pair, and calculate the contrast loss function based on the Euclidean distance between the feature vectors corresponding to the two samples in the sample pair and the label value;

[0021] The test set is used to check whether the operation feature matching of the comparative learning model is correct. If so, the final trained substation status assessment comparative learning model is generated. Otherwise, a new operation status feature set is selected to retrain the model.

[0022] Preferably, the preprocessing includes: cleaning outliers and erroneous measurements, data scale normalization and data enhancement, wherein data enhancement includes adding random noise, filter smoothing and time series slicing.

[0023] Preferably, the data enhancement further includes: performing sample balancing processing on the substation historical operation status data set to balance the number of each type of samples in the data set.

[0024] Preferably, positive sample pairs and negative sample pairs are generated based on the following rules:

[0025] The positive sample pairs are composed of data samples from the same device in normal operating state, and the negative samples are composed of data samples from the same device in normal state and fault state, or data samples between different devices.

[0026] Preferably, the contrast loss function is expressed as follows:

[0027] L(x i ,x j ,y)=y·D(x i ,x j ) 2 +(1-y)·max(0,mD(x i ,x j )) 2

[0028] Where:

[0029] D(x i ,x j ) represents the sample pair (x i , x j ) between ;

[0030] m is the boundary threshold;

[0031] y is the label.

[0032] Preferably, using the test set to check whether the running feature matching of the contrast learning model is correct includes:

[0033] The test samples are matched with the training samples to form new sample pairs, and the similarity between the new sample pairs and the known sample pairs is calculated. If all new sample pairs have known positive sample pairs or known negative sample pairs with a similarity less than a preset threshold, the match is considered correct, otherwise it is considered incorrect.

[0034] A second aspect of the present invention provides a substation operation status assessment system based on a contrastive learning algorithm, and the substation operation status assessment method based on a contrastive learning algorithm is executed, comprising:

[0035] The feature selection module is used to obtain the historical operating status data of the substation equipment, select the substation status evaluation indicators, and construct the substation historical operating status data set and operating status feature set;

[0036] A state classification module is used to classify the operation state levels of samples in the substation historical operation state data set;

[0037] The comparison model module is used to extract the feature vector of each sample in the substation historical operation status data set, generate positive sample pairs and negative sample pairs, calculate the comparison loss function, and generate a comparison learning model for substation status assessment;

[0038] The result output module is used to obtain the real-time operating status data of the substation equipment and output the substation operating status level.

[0039] Compared with the prior art, the beneficial effects of the present invention include at least:

[0040] (1) The present invention uses a contrastive learning algorithm to construct positive and negative sample pairs of substation equipment and perform feature extraction, thereby enhancing the ability to distinguish between normal and abnormal states of substations, and especially improving the model's ability to recognize abnormal states with a small number of samples.

[0041] (2) The present invention does not rely on manual labeling. It extracts useful information from a small part of unlabeled monitoring data. It is particularly effective in small sample scenarios. It can make full use of limited samples and extract valuable information from the intrinsic structure of the data, thereby improving the generalization ability and robustness of the model.

[0042] (3) The comparative learning model in the present invention has the ability to enhance features under unbalanced sample conditions, exhibits good evaluation performance, has the advantages of high accuracy and fast operation speed, and can effectively improve the efficiency of substation operation and maintenance.

[0043] (4) The present invention utilizes the contrast loss function value to optimize the contrast learning model, which further improves the accuracy and efficiency of fault detection and early warning, and has practical value for ensuring the reliable operation of substations and reducing unexpected downtime. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a structural diagram of a substation monitoring system based on the IEC-61850 protocol provided in accordance with an embodiment of the present invention;

[0045] Figure 2 is a schematic diagram of a contrastive learning algorithm provided in accordance with an embodiment of the present invention;

[0046] Figure 3 It is a flow chart of a method for evaluating substation operation status based on a contrastive learning algorithm provided in accordance with an embodiment of the present invention;

[0047] Figure 4 It is a schematic diagram of a chaotic matrix of evaluation results provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0049] The present invention uses self-supervised learning technology to effectively utilize unlabeled monitoring data, divide the data into pairs for learning, and extract feature representations of normal and abnormal conditions of equipment operation to achieve accurate assessment of fault conditions.

[0050] like Figure 3 As shown, embodiment 1 of the present invention provides a method for evaluating the operation status of a substation based on a contrastive learning algorithm, comprising the following steps:

[0051] Step 1: Obtain the historical operating status data of substation equipment, build a substation historical operating status data set, select the status characteristic indicators of primary equipment as substation status evaluation indicators, and build an operating status feature set.

[0052] In the substation operation status assessment, accurate selection of evaluation indicators is the key to ensure the reliability and effectiveness of the evaluation results. The selection of evaluation indicators should be based on the operating characteristics and possible failure modes of substation equipment, covering all aspects of equipment operation to fully reflect the health status of the substation.

[0053] like Figure 1 As shown in the figure, substations are usually equipped with substation monitoring systems. According to the IEC61850 protocol, the substation monitoring system structure mainly includes the process layer, the interval layer and the station control layer. The process layer is composed of the data sampling network of each device monitoring, among which the substation primary equipment operation monitoring system mainly includes transformers, high-voltage switchgear, lightning arresters and other electrical equipment operation status monitoring modules. The status monitoring data of substation electrical equipment is transmitted based on the IEC61850 protocol to realize the digital transmission of data information, and the status monitoring information is analyzed and displayed on the monitoring platform.

[0054] In a preferred but non-limiting embodiment of the present invention, according to the structure of the substation primary equipment operation monitoring system, the key operating status of the substation primary equipment is selected as the evaluation index, including the status characteristic indicators of three types of electrical equipment: transformer, high-voltage switch and lightning arrester. Step 1 specifically includes:

[0055] Step 1.1, obtain the historical operating status data of transformers, high-voltage switches and lightning arresters from the status monitoring data of substation electrical equipment, and construct the substation historical operating status data set. Among them, the historical operating data of substation primary equipment is monitored by multiple sensors, which record key parameters including but not limited to the temperature, voltage, current, etc. of the equipment.

[0056] Step 1.2, select the state characteristic indicators of the transformer, taking the oil-immersed transformer as an example, including indicators such as transformer oil temperature, oil level, oil color spectrum and transformer partial discharge.

[0057] Transformer oil is the main material in the transformer insulation system, used for cooling and insulation. Its temperature and oil level directly affect the cooling effect and insulation performance of the transformer. Too high oil temperature will accelerate the deterioration of oil quality and reduce insulation performance. Too low oil level will affect the cooling effect and increase the risk of overheating. Therefore, monitoring the transformer oil temperature and oil level can prevent transformer overheating and insulation failure and ensure its stable operation.

[0058] Oil chromatography can reflect the electrical and thermal faults inside the transformer by detecting the dissolved gas components in the transformer oil, such as hydrogen, methane, etc. The changes in different gas components can detect early transformer faults and take preventive measures to avoid serious faults. The partial discharge phenomenon of the transformer reflects its insulation state, and frequent partial discharge may lead to insulation breakdown. By monitoring partial discharge, insulation degradation problems can be discovered, and maintenance can be carried out in time to prevent insulation breakdown failures.

[0059] The partial discharge of a transformer refers to the partial discharge phenomenon of the internal or external insulating material of the transformer. The evaluation indicators include the discharge intensity, current amplitude, change trend, displacement trend, etc.

[0060] Step 1.3, select the state characteristic indicators of the high-voltage switchgear, including the density and pressure of the insulating gas, the switch partial discharge frequency, the mechanical displacement, the contact acceleration, and the main circuit bus temperature.

[0061] Taking SF6 high-voltage circuit breakers as an example, SF6 gas is used as an insulating and arc-extinguishing medium. Its temperature, humidity, density and pressure directly affect its insulation performance and arc-extinguishing effect. Too high temperature will accelerate the decomposition of SF6, too high humidity will reduce the insulation strength, and too low density and pressure will weaken the insulation performance and increase the risk of failure. Therefore, monitoring these parameters of SF6 gas can effectively prevent switch failures and ensure its safe operation.

[0062] Partial discharge of high-voltage switches is an early sign of insulation degradation, and frequent partial discharge may lead to insulation breakdown. By monitoring partial discharge, insulation problems can be discovered early, and maintenance measures can be taken in time to prevent equipment failure.

[0063] Mechanical displacement and contact acceleration reflect the mechanical operating status of the switch. Abnormal displacement and acceleration may cause switch operation failure. Monitoring these indicators can prevent mechanical failures and ensure the normal operation of the switch.

[0064] The main circuit busbar temperature is an important indicator of the switch's electrical performance. Excessive current and high temperature can cause the switch to overload and overheat, increasing the risk of equipment damage. By monitoring the main circuit current and temperature, overload and overheating problems can be discovered in a timely manner, and corresponding measures can be taken to ensure the safe operation of the switch.

[0065] Step 1.4, select the state characteristic indicators of the lightning arrester, including temperature, humidity, total current and resistive current.

[0066] The temperature and humidity of the arrester are important parameters for its normal operation. Abnormal temperature and humidity may affect the performance of the arrester. High temperature and high humidity environment may cause the degradation of the arrester insulation material and the decline of electrical performance. Therefore, monitoring the temperature and humidity of the arrester to ensure that it is within the safe range can prevent performance degradation and failure.

[0067] The total current and resistive current of the arrester reflect its discharge capacity and internal resistance performance during lightning strikes or overvoltage. Abnormal total current may indicate aging or damage of the arrester, and abnormal resistive current may indicate degradation or failure inside the arrester. By monitoring the total current and resistive current of the arrester, performance changes of the arrester can be discovered, and replacement or maintenance can be performed in a timely manner to ensure the effective protection of the arrester.

[0068] In an exemplary but non-limiting implementation of the present invention, as shown in Table 1, Table 1 lists the various state characteristic indicators of three types of equipment in the substation, namely transformers, high-voltage switches and lightning arresters, and the data type of each state indicator, and the data type includes continuous values ​​and discrete values.

[0069] Table 1 Substation status characteristic indicators

[0070]

[0071]

[0072] Step 2: Classify the operating status of the substation according to the operating status feature set of step 1, and determine the operating status level of all samples in the substation historical operating status data set.

[0073] In an exemplary but non-limiting embodiment of the present invention, the operating state of the substation is divided into four state levels, namely normal state, attention state, abnormal state and serious state, as shown in Table 2. According to the importance of the state characteristic indicators, some indicators are selected as important state quantities, and the standard limit value and the over-limit degree evaluation rules (slight over-limit and serious over-limit) of each state quantity are set. According to the situation where the state characteristic indicators are close to or exceed the standard limit value and the importance of the state characteristic indicators, the substation operating state level can be determined. The standard limit value can refer to the specified value in the standard or be set based on experience.

[0074] Table 2 Classification of substation operation status

[0075]

[0076] It is understandable that by classifying the substation operating status into levels and making each level correspond to different equipment operating conditions, it can help operation and maintenance personnel to understand the specific conditions of the equipment more intuitively and clearly, thereby formulating more effective maintenance strategies.

[0077] Step 3: Establish an initial contrastive learning model based on the neural network model, extract the feature vector of each sample in the substation historical operation status data set in step 1, and generate positive sample pairs and negative sample pairs according to the set rules. According to the distance between the feature vectors corresponding to the two samples in each sample pair and the positive and negative of the sample pair, construct a contrast loss function to train the substation status evaluation contrastive learning model.

[0078] Contrastive Learning is a self-supervised learning method that aims to learn data representation by constructing similar and dissimilar sample pairs. The core idea of ​​this method is to maximize the similarity of similar sample pairs and minimize the similarity of dissimilar sample pairs, so that the model can learn effective feature representation without a large amount of labeled data. Contrastive learning has achieved remarkable results in computer vision, natural language processing and other fields, and is also one of the important research directions of self-supervised learning.

[0079] For unlabeled small sample data, the contrastive learning algorithm optimizes the model's feature representation by constructing positive and negative sample pairs and calculating the contrastive loss function, making the feature distance of similar sample pairs as small as possible and the feature distance of dissimilar sample pairs as large as possible. Figure 2 As shown, by performing data enhancement on the input original data, several contrast samples are obtained, feature extraction is performed on each contrast sample, the feature vector of the contrast sample is obtained, the contrast loss is calculated, the effective features are learned, the model is continuously optimized, and finally a trained contrast loss model is obtained.

[0080] In a preferred but non-limiting embodiment of the present invention, step 3 specifically comprises:

[0081] Step 3.1, preprocess the original data in the substation historical operation status dataset in step 1, and divide it into a training set and a test set for the comparative learning model.

[0082] Since the collected raw data usually contains noise and potential outliers, data preprocessing can improve the generalization ability of the model.

[0083] Further preferably, the preprocessing includes: cleaning outliers and erroneous measurements, data scale normalization, and data enhancement.

[0084] For example, the raw data collected in the substation monitoring system can be enhanced in the following different ways to generate similar sample pairs:

[0085] (1) Adding random noise to the original data can simulate the noise interference that may occur in actual operation and enhance the robustness of the model. The calculation formula is as follows:

[0086] x noise =x+ε (1)

[0087] Where:

[0088] x is the original data, ε is the random noise that follows a certain distribution, and Gaussian noise is generally chosen, with a mean of 0 and a variance of σ 2 .

[0089] (2) Smooth the original data through filters to generate new analysis samples. Commonly used filters include mean filters and Gaussian filters.

[0090] The mean filter formula is as follows:

[0091]

[0092] Among them, x(i) is the original data and k is the window size.

[0093] The Gaussian filter formula is as follows:

[0094]

[0095] Where G(j) is the Gaussian kernel function, defined as formula 4:

[0096]

[0097] Where σ is the standard deviation. By smoothing the original data with a Gaussian filter, high-frequency noise can be removed and low-frequency signals can be retained, thereby generating smoother sample data.

[0098] (3) Time series slicing: Slicing the original data according to the sliding window of the time series can generate multiple similar comparison samples, and these samples can be processed and analyzed independently.

[0099] Assume that the time series data is x={x1,x2,x3,…,x T}, the window size is W, the step size is S, then the time series slice is represented by x slice (t) = {x t ,x t+1 ,…,x t+W-1}. Where T represents the total time length, t=1,1+S,1+2S,….

[0100] It is worth noting that the historical data on substation operating status usually have serious sample imbalance problems, and the number of data in each substation operating status category in the dataset varies greatly.

[0101] In a more preferred but non-limiting embodiment of the present invention, the synthetic minority over-sampling technique (SMOTE) is used to perform sample balancing on the substation historical operating status data set, and new synthetic samples are generated between the minority class samples to alleviate the sample imbalance problem of the historical data of the substation operating status. SMOTE can balance the number of samples of each type in the data set, thereby improving the model's ability to recognize the minority class, which is convenient for feature extraction of the subsequent comparative learning model.

[0102] Step 3.2, use the neural network as the initial contrastive learning model to extract features from the preprocessed training set in step 3.1, obtain the feature vector of each sample in the training set, and learn the operation mode and potential nonlinear relationship of the substation equipment from the time series data of each sample. These learning results will be used to support the subsequent optimization training process of the contrastive learning model.

[0103] Further preferably, in the contrastive learning framework, feature extraction is implemented by a neural network with shared weights to ensure that the features extracted from different input samples are comparable. This configuration allows the model to perform the same processing steps on all input data, regardless of their source or category, ensuring that all samples are evaluated in the same feature space, thereby effectively comparing the similarities and differences between different samples.

[0104] Neural networks with shared weights typically consist of multiple layers, each of which performs a specific transformation to extract information from the input data. These networks include convolutional layers, activation layers, pooling layers, dropout layers, and fully connected layers, each of which processes the input data, gradually refining and compressing information, and ultimately generating highly abstract feature representations.

[0105] The convolution layer extracts local features by performing convolution operations on the image. Convolution is implemented through a series of learnable filters, each of which focuses on capturing a specific pattern or feature of the input data.

[0106] The activation layer usually uses nonlinear functions to increase the nonlinear processing capabilities of the network and help the network learn complex data patterns.

[0107] The pooling layer is used to reduce the spatial dimension of the data and improve the robustness of the features. Through the pooling operation (usually maximum pooling or average pooling), the sensitivity to position is reduced while maintaining important feature information.

[0108] The Dropout layer is used to prevent the model from overfitting and enhance the generalization ability of the model by randomly discarding some neurons.

[0109] At the end of the network, the fully connected layer combines all the features extracted by the previous layers to generate a global feature vector, which contains comprehensive information about the input data.

[0110] It can be understood that through these continuous processing layers, the shared weight neural network can transform the input substation equipment operating status historical data into expressive feature vectors, which can accurately describe the key attributes of the data. In the subsequent optimization training process of the contrastive learning model, these feature vectors are used to adjust the network weights by calculating the Euclidean distance between sample pairs and optimizing a specific loss function to make samples from the same category closer and samples from different categories more distant. The use of the contrastive learning model to extract the features of the substation status data not only improves the model's ability to discriminate the normal and faulty state features of the substation, but also enhances the model's generalization ability to new and unseen samples.

[0111] In step 3.3, based on artificially set rules, samples are selected from samples with the same operating status level to form positive sample pairs, and samples are selected from samples with different operating status levels to form negative sample pairs, and labels are set. The contrast loss function is calculated according to the Euclidean distance and label value between the feature vectors corresponding to the two samples in the sample pair, and the substation status assessment contrast learning model is trained.

[0112] In the process of building a contrastive learning model, the core task is to form and effectively use sample pairs, including positive sample pairs, i.e. similar sample pairs, and negative sample pairs, i.e. dissimilar sample pairs. The training of the contrastive learning model learns effective feature representation by building positive and negative sample pairs, calculating the contrastive loss function value, and measuring the difference between similar sample pairs and dissimilar sample pairs.

[0113] In a further preferred embodiment of the present invention, step 3.3 specifically includes:

[0114] Step 3.3.1, using data samples from the same device in normal operation to form positive sample pairs, and using data samples from the same device in normal and fault states, or data samples between different devices to form negative samples, to generate positive and negative sample pairs for training the substation state assessment comparative learning model;

[0115] It can be understood that the positive and negative sample pairs generated by this rule can also generate a large amount of training data for comparative learning for the small sample situation in which a certain device has little historical data in the substation equipment operation status data and the unbalanced sample situation in which a certain type of data of a certain device is large but the rest of the data is small. It can make full use of multi-source heterogeneous data to extract the characteristics of different operating states of the substation, thereby accurately evaluating the operating status of the substation.

[0116] Step 3.3.2, for each sample pair (x i , x j )Set the label y. Specifically, when the sample pair is a positive sample, set the label y=1, and when the sample pair is a negative sample, set the label y=0, so as to accurately distinguish between positive and negative samples according to the characteristics of the data.

[0117] Step 3.3.3, calculate the contrast loss function for each iteration, adjust the weight parameters of the model according to the size of the loss, and use the back propagation algorithm to calculate the gradient of the contrast loss function.

[0118] More preferably, the contrast loss function measures the difference between similar sample pairs and dissimilar sample pairs by calculating the Euclidean distance between each sample pair, and optimizes the feature representation of the model by minimizing the distance between similar sample pairs and maximizing the distance between dissimilar sample pairs. The contrast loss function is defined as Formula 5:

[0119] L(x i ,x j ,y)=y·D(x i ,x j ) 2 +(1-y)·max(0,mD(x i ,x j )) 2 (5)

[0120] Where:

[0121] D(x i ,x j ) represents the Euclidean distance between sample pairs, and m is the boundary threshold.

[0122] Step 3.3.4, based on the contrast loss calculation results, use the optimization algorithm to iteratively adjust the model parameters according to the calculated gradient to optimize the model. Through multiple iterations, the model weights are continuously updated until the contrast loss function reaches the preset minimum value and the model training is completed.

[0123] It is understandable that the contrast mechanism of the contrastive learning model is particularly effective in small sample scenarios, because the contrastive learning algorithm does not rely on a large amount of labeled data, makes full use of limited samples, and extracts valuable information from the intrinsic structure of the data, thereby improving the generalization ability and robustness of the model.

[0124] In step 3.4, use the test set to check whether the operation feature matching of the comparative learning model is correct. If so, generate the final trained substation status assessment comparative learning model. Otherwise, select a new operation status feature set and return to step 3.3 to retrain the model.

[0125] Specifically, the operating characteristics are key feature vectors extracted from the historical operating data of the equipment through the comparative learning model, including the operating status parameters of the equipment such as temperature, voltage, and current. After model training, the operating characteristics of the test data are matched with the trained feature samples to form new sample pairs, and the similarity between the new sample pairs and the known positive and negative sample pairs is calculated. If all new sample pairs formed by the test data have known positive sample pairs or known negative sample pairs with similarity less than the preset threshold, the match is considered correct; otherwise, the match is considered wrong, that is, the current state feature indicators of the comparative learning model cannot distinguish the normal or faulty state of the substation. Therefore, the standard for whether the match is correct is based on the similarity calculation of the feature vectors and the preset judgment threshold.

[0126] Step 4: monitor the operating status of the primary equipment in the substation in real time, input the real-time operating status data of the equipment into the substation status evaluation and comparison learning model in step 3, and use the trained substation status evaluation and comparison learning model to evaluate the substation operating status to obtain the substation operating status level.

[0127] Specifically, the contrastive learning model trained in step 3 is used to analyze real-time data and identify the current operating status of the substation. The contrastive learning model determines whether the substation is in a normal state, a warning state, an abnormal state, or a serious state based on the feature vector. The output of the model provides real-time feedback on the status of substation equipment, enabling operation and maintenance personnel to take preventive measures or perform repairs when necessary.

[0128] Embodiment 2 of the present invention provides a substation operation status assessment system based on a contrastive learning algorithm, which runs the substation operation status assessment method based on a contrastive learning algorithm as described in Embodiment 1, including:

[0129] The feature selection module is used to obtain the historical operating status data of the substation equipment, select the substation status evaluation indicators, and construct the substation historical operating status data set and operating status feature set;

[0130] A state classification module is used to classify the operation state levels of samples in the substation historical operation state data set;

[0131] The comparison model module is used to extract the feature vector of each sample in the substation historical operation status data set, generate positive sample pairs and negative sample pairs, calculate the comparison loss function, and generate a comparison learning model for substation status assessment;

[0132] The result output module is used to obtain the real-time operating status data of the substation equipment and output the substation operating status level.

[0133] In order to more clearly demonstrate the technical solution of the present invention and the beneficial technical effects brought about by it, the present invention is further described below through a specific embodiment.

[0134] A total of 600 pieces of equipment historical operation data with labels were collected from the station control layer database system of a substation, covering 12 months of equipment operation time. All collected data types are numeric, and the corresponding labels are character types. The application scenario is the operation status data in the actual substation. The data set includes 400 pieces of normal status data, 100 pieces of warning status data, 80 pieces of abnormal status data, and 20 pieces of fault status data.

[0135] To ensure the consistency and effectiveness of data processing, all data were normalized before training. The specific examples are shown in Table 3. The specific normalization strategy is: for continuous indicators, such as oil temperature and ambient humidity, the normalized values ​​are floating point numbers falling within the interval [0,1]; for discrete indicators, such as appearance damage, the normalized values ​​are discrete values ​​falling within the interval [0,1]. Table 3 selects two examples of key feature data of substations with normal and attention status, respectively, where the attention value is the boundary value that distinguishes the indicator status type.

[0136] Table 3 Examples of some operating status data

[0137]

[0138] Due to the imbalance of substation operation data, in the model training process, this embodiment uses the SMOTE algorithm to perform sample balancing on the original data in the substation historical operation status data set. After processing, the number of samples of the four categories of correct, attention, abnormal, and fault is consistent, among which the attention state, abnormal state, and fault state data are supplemented to 400 cases. This embodiment divides the substation historical operation status data set that has been balanced to obtain a training set and a test set, of which 60% is the training set and 40% is the test set.

[0139] The neural network with parameter settings shown in Table 4 is used to extract features from the preprocessed data samples to obtain the feature vectors of each sample in the training set.

[0140] Table 4 Neural network parameter settings

[0141]

[0142] Use the feature vector to construct positive and negative sample pairs and train the contrastive learning model. Use the trained contrastive learning model to test the test set. The state evaluation results are as follows: Figure 4 As shown in Table 5, the evaluation accuracy of the normal state is 94%, the evaluation accuracy of the attention state is 90%, the evaluation accuracy of the abnormal state is 84%, and the evaluation accuracy of the fault state is 82%. It can be seen that the average accuracy of the substation operation status evaluation method based on the contrastive learning algorithm is 92%.

[0143] Table 5 Comparison of learning model performance

[0144]

[0145] The precision in Table 5 indicates the probability that the samples identified as a certain operating state actually meet the evaluation results. The calculation formula is as shown in Formula 6:

[0146]

[0147] Where:

[0148] TP represents the number of samples that are correctly identified, and FP represents the number of samples that are incorrectly identified among the samples identified as this state.

[0149] Recall is the probability of being correctly identified in a sample of a certain running state, and the calculation formula is as shown in Formula 7:

[0150]

[0151] Where:

[0152] P represents the total number of samples that are actually in this state.

[0153] The F1 score represents the harmonic mean of precision and recall, combining the advantages of the two indicators to comprehensively evaluate the model performance. The calculation formula is as shown in Formula 8:

[0154]

[0155] In this embodiment, the evaluation performance of XGBoost, decision tree and random forest models is compared with the comparison learning model, where the evaluation time is the total time taken by each model to complete the evaluation of 400 running test data. The comparison results are shown in Table 6.

[0156] Table 6 Comparison of evaluation performance of different models

[0157]

[0158] As can be seen from Table 6, the decision tree model has a simple structure and performs well in terms of evaluation time, but its evaluation accuracy is significantly lower than that of other models. The XGBoost model has good evaluation performance with an accuracy of 96%, but its evaluation efficiency is average and the overall evaluation takes a long time. The random forest model has a higher evaluation efficiency than the above two models, with an evaluation accuracy of 87% and an evaluation time of 21s. However, according to the evaluation results of the contrastive learning model, the contrastive learning model has an accuracy of 90% and an evaluation time of only 33s. It can still maintain good operating efficiency with a high accuracy. Therefore, the evaluation performance comparison verifies that the evaluation performance of the substation operation status evaluation method based on contrastive learning is better than that of the other three models.

[0159] Compared with the prior art, the beneficial effects of the present invention include at least:

[0160] (1) The present invention uses a contrastive learning algorithm to construct positive and negative sample pairs of substation equipment and perform feature extraction, thereby enhancing the ability to distinguish between normal and abnormal states of substations, and especially improving the model's ability to recognize abnormal states with a small number of samples.

[0161] (2) The present invention does not rely on manual labeling. It extracts useful information from a small part of unlabeled monitoring data. It is particularly effective in small sample scenarios. It can make full use of limited samples and extract valuable information from the intrinsic structure of the data, thereby improving the generalization ability and robustness of the model.

[0162] (3) The comparative learning model in the present invention has the ability to enhance features under unbalanced sample conditions, exhibits good evaluation performance, has the advantages of high accuracy and fast operation speed, and can effectively improve the efficiency of substation operation and maintenance.

[0163] (4) The present invention utilizes the contrast loss function value to optimize the contrast learning model, which further improves the accuracy and efficiency of fault detection and early warning, and has practical value for ensuring the reliable operation of substations and reducing unexpected downtime.

[0164] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A substation operation status assessment method based on a contrastive learning algorithm, the substation includes a primary device, and the primary device includes: Transformer, high voltage switchgear and lightning arrester, characterized by comprising: Obtain the historical operating status data of substation equipment and construct a historical operating status data set of the substation; use the status characteristic indicators of primary equipment as substation status evaluation indicators to construct an operating status characteristic set; According to the operating status feature set, the operating status level of the substation is divided, and the operating status level of all samples in the substation historical operating status data set is determined; An initial contrastive learning model is established based on a neural network model. The feature vectors of each sample in the substation historical operating status data set are extracted. According to the set rules, positive sample pairs are generated from samples with the same operating status level, and negative sample pairs are generated from samples with different operating status levels. A contrastive loss function is constructed based on the distance between the feature vectors corresponding to the two samples in each sample pair and the positive and negative values ​​of the sample pairs to train the substation status assessment contrastive learning model. Monitor the operating status of primary equipment in the substation in real time, input the real-time operating status data of the equipment into the trained substation status assessment comparative learning model, and output the substation operating status level.

2. The substation operation status assessment method based on contrastive learning algorithm according to claim 1, characterized in that: The status characteristic indicators of primary equipment include: Transformer status characteristic indicators, including: transformer oil temperature, oil level, oil color spectrum and transformer partial discharge evaluation indicators; Status characteristic indicators of high-voltage switchgear, including: density and pressure of insulating gas, switch partial discharge frequency, mechanical displacement, contact acceleration and main circuit bus temperature; The status characteristic indicators of the lightning arrester include: temperature, humidity, total current and resistive current.

3. The substation operation status assessment method based on contrastive learning algorithm according to claim 1, characterized in that: The substation operation status levels are divided into four levels: normal state, caution state, abnormal state and severe state.

4. The substation operation status assessment method based on contrastive learning algorithm according to claim 1, characterized in that: Training the substation condition assessment comparative learning model also includes: The original data in the substation historical operation status dataset is preprocessed and divided into a training set and a test set; Perform feature extraction on the preprocessed training set to obtain the feature vector of each sample in the training set; Set a label for each sample pair, and calculate the contrast loss function based on the Euclidean distance between the feature vectors corresponding to the two samples in the sample pair and the label value; The test set is used to check whether the operation feature matching of the comparative learning model is correct. If so, the final trained substation status assessment comparative learning model is generated. Otherwise, a new operation status feature set is selected to retrain the model.

5. The substation operation status assessment method based on contrastive learning algorithm according to claim 4, characterized in that: The preprocessing includes: cleaning outliers and erroneous measurements, data scale standardization and data enhancement, wherein the data enhancement includes adding random noise, filter smoothing and time series slicing.

6. The substation operation status assessment method based on contrastive learning algorithm according to claim 5, characterized in that: Data enhancement also includes: performing sample balancing processing on the substation historical operation status data set to balance the number of each type of samples in the data set.

7. The substation operation status assessment method based on contrastive learning algorithm according to claim 1, characterized in that: Generate positive and negative sample pairs based on the following rules: The positive sample pairs are composed of data samples from the same device in normal operating state, and the negative samples are composed of data samples from the same device in normal state and fault state, or data samples between different devices.

8. The substation operation status assessment method based on contrastive learning algorithm according to claim 1, characterized in that: The contrast loss function is expressed as follows: L(x i ,x j ,y)=y·D(x i ,x j ) 2 +(1-y)·max(0,mD(x i ,x j )) 2 Where: D(x i ,x j ) represents the sample pair (x i , x j ) between ; m is the boundary threshold; y is the label.

9. The substation operation status assessment method based on contrastive learning algorithm according to claim 4, characterized in that: Using the test set to check whether the running feature matching of the contrast learning model is correct includes: The test samples are matched with the training samples to form new sample pairs, and the similarity between the new sample pairs and the known sample pairs is calculated. If all new sample pairs have known positive sample pairs or known negative sample pairs with a similarity less than a preset threshold, the match is considered correct, otherwise it is considered incorrect.

10. A substation operation status assessment system based on contrastive learning algorithm, running the substation operation status assessment method based on contrastive learning algorithm as described in any one of claims 1 to 9, characterized in that: include: The feature selection module is used to obtain the historical operating status data of the substation equipment, select the substation status evaluation indicators, and construct the substation historical operating status data set and operating status feature set; A state classification module is used to classify the operation state levels of samples in the substation historical operation state data set; The comparison model module is used to extract the feature vector of each sample in the substation historical operation status data set, generate positive sample pairs and negative sample pairs, calculate the comparison loss function, and generate a comparison learning model for substation status assessment; The result output module is used to obtain the real-time operating status data of the substation equipment and output the substation operating status level.

Images