An adaptive probabilistic neural network method for high-voltage cable condition assessment based on multi-source data

Abstract

The present invention provides an adaptive probabilistic neural network high-voltage cable state assessment method based on multi-source data, the method comprising: selecting cable state assessment indicators, establishing an assessment indicator set and state level division, building a sensor network layout and data collection mechanism for multi-source data, collecting multi-source data related to high-voltage cable state assessment; performing data preprocessing on the collected high-voltage cable multi-source data; constructing an APNN adaptive probabilistic neural network, dividing the preprocessed sample data as a training set and a test set, using historical data and known cable fault cases to train the neural network, and training the APNN network; using the trained APNN network to assess the high-voltage cable state, and outputting the assessment value of the relevant state classification. The cable-related data of the present invention is easy to obtain, has strong feasibility, is used for intelligent assessment of the high-voltage cable state, and shows great potential in improving the accuracy of cable monitoring and reducing operation and maintenance costs.

Description

Technical Field

[0001] The invention relates to the technical field of high-voltage cable maintenance, and in particular to a method for evaluating the state of a high-voltage cable using an adaptive probabilistic neural network based on multi-source data. Background Art

[0002] As a key component of the power system, the reliability of high-voltage cables directly affects the stability and safety of the entire power network. These cables are usually buried underground or underwater, making their detection and maintenance particularly complex and difficult. Traditional cable status assessment methods, such as electrical testing, visual inspection, and thermal imaging technology, are usually intermittent and cannot provide real-time data. These methods often rely on the experience and judgment of professionals, may be affected by subjective factors, and have limited effectiveness in early fault prediction.

[0003] With the development of the Internet of Things and big data technologies, we can now collect data about high-voltage cables in real time from multiple sources, including temperature, current, humidity, and environmental factors. Comprehensive analysis of multi-source data can provide a more comprehensive and accurate view of the cable status, thus facilitating the transition from preventive maintenance to predictive maintenance. In this context, probabilistic neural networks show great potential in dealing with uncertainty and nonlinear problems, and are particularly suitable for the evaluation of complex system status. Such networks can learn the pattern of cable faults based on historical and real-time data, and show higher flexibility and accuracy in data-driven decision support systems compared to traditional algorithms.

[0004] Considering the variability of cable operating environment and status, traditional algorithms are inadequate in processing dynamic changes, abnormal data and non-stationary data. An adaptive evaluation method is particularly important. This method can automatically adjust the evaluation model according to real-time data to adapt to new data and conditions. The adaptive method can adjust the evaluation parameters in real time, reduce false positives and false negatives, and improve the reliability of cable fault prediction. In recent years, artificial intelligence and machine learning have been increasingly used in the power industry. Studies have shown that combining multi-source data and advanced algorithms can significantly improve the accuracy of power equipment status assessment.

[0005] In summary, the evaluation of high-voltage cable status by adaptive probabilistic neural network (APNN) based on multi-source data is not only an innovation, but also an important supplement and improvement to the existing technology. This method has great practical value and broad application prospects in improving the accuracy and efficiency of cable monitoring. Therefore, it is necessary to propose an adaptive probabilistic neural network high-voltage cable status evaluation method based on multi-source data. Summary of the invention

[0006] In view of this, it is necessary to provide an adaptive probabilistic neural network high-voltage cable status assessment method based on multi-source data, which improves the sensitivity of early diagnosis of cable faults and can maintain a high level of prediction accuracy under complex and changeable operating environments, solving the existing traditional high-voltage cable status assessment algorithms. The network has insufficient learning and adaptability when facing changing data characteristics, and the accuracy and efficiency of the assessment need to be further improved.

[0007] The present invention provides a method for evaluating the state of a high-voltage cable based on an adaptive probabilistic neural network of multi-source data, comprising the following steps:

[0008] Step S1: Select cable status evaluation indicators, establish evaluation indicator sets and status level division, build a sensor network layout and data collection mechanism for multi-source data, and collect multi-source data related to high-voltage cable status evaluation;

[0009] Step S2: preprocessing the collected high-voltage cable multi-source data, performing outlier detection and missing value processing, standardizing the relevant data, and obtaining preprocessed sample data;

[0010] Step S3: construct an APNN adaptive probabilistic neural network, divide the preprocessed sample data into training sets and test sets, use historical data and known cable fault cases to train the neural network, and during the training process, the APNN automatically updates its parameters according to the real-time data stream to adapt to new data patterns and changing trends;

[0011] Step S4: Use the trained APNN network to evaluate the status of the high-voltage cable and output the evaluation value of the relevant status classification.

[0012] Furthermore, the evaluation index set established in step S1 is divided into static basic indicators and dynamic variable indicators, wherein the static basic indicators include basic equipment information: voltage level, operating years, historical failure rate, and load type; the dynamic indicators include operation and inspection status information, operation status information, and tunnel environment information, wherein the operation and inspection status information includes insulation resistance, eccentricity, partial discharge, and dielectric loss; the operation status information includes cable temperature, temperature rise rate, cable load, circulating current, and mechanical stress; the tunnel environment information includes ambient humidity, ambient temperature, oxygen concentration, methane concentration, carbon monoxide concentration, hydrogen sulfide concentration, chlorine concentration, and water level information; the established status levels are divided, and specifically the evaluation levels are divided into five levels: healthy, attention, abnormal, fault, and serious fault.

[0013] Furthermore, the collection of multi-source data related to the high-voltage cable status assessment in step S1 specifically includes: deploying various types of sensors at key nodes of the cable, including thermocouples, voltage and current sensors, insulation monitoring equipment, and environmental monitoring equipment, implementing a high-frequency sampling strategy to capture instant changes in the cable status, and using data fusion technology to ensure the consistency of data collected from different sources in time and space.

[0014] Furthermore, in step S2, data preprocessing is performed on the collected high-voltage cable multi-source data, specifically including:

[0015] S21. Apply interpolation methods to estimate missing values ​​to complete the processing of missing data, and use the interquartile range method to detect outliers;

[0016] S22. Use Z-score standardization to standardize the data to eliminate the differences between the attributes of the influencing factors:

[0017]

[0018] Wherein, μ is the mean value, σ is the standard deviation, and the influencing factors include temperature and load.

[0019] Furthermore, the step S3 specifically includes:

[0020] S31. Construct an APNN adaptive probabilistic neural network suitable for high-voltage cable status assessment;

[0021] S32. The preprocessed sample data is divided into training set and test set respectively to ensure sufficient training samples for learning the APNN model. The performance of the model is evaluated using an independent test set. The network is trained to minimize the prediction error through back propagation algorithm and gradient descent. The learning rate in the gradient descent process is dynamically adjusted to improve the convergence speed and avoid falling into local minima. The cross-validation method is introduced to train and evaluate the APNN model multiple times to verify its generalization ability on different data subsets and ensure the adaptability of the model to diverse data.

[0022] S33. Real-time monitoring of network input and high-frequency sampling of cable status input data ensure that the model can perceive changes in cable status in a timely manner. When significant changes in dynamic indicators are detected, the adaptive adjustment mechanism is triggered, and the network automatically adjusts its parameters based on real-time data to adapt to new data distribution and characteristics, realizing online learning of the network to continuously adapt to dynamic changes in cable status and adapt to the evolution of cable status without interrupting evaluation and status prediction.

[0023] Furthermore, the network architecture of the APNN adaptive probabilistic neural network consists of an input layer, a pattern layer, a summary layer, and an output layer. The adaptability of the APNN is achieved by dynamically adjusting the parameters in the pattern layer. When the network receives new data, it will update these parameters to better reflect the characteristics of the current data.

[0024] Furthermore, the parameters in the pattern layer include the mean vector m of the i-th training sample i and standard deviation σ i .

[0025] The beneficial effects brought by adopting the above technical solution are:

[0026] Compared with the prior art, the present invention selects reasonable and comprehensive cable status evaluation indicators, constructs an evaluation indicator set and status level division, builds a sensor network layout and data collection mechanism for multi-source data, collects multi-source data related to high-voltage cable status evaluation, and constructs a comprehensive and scientific high-voltage cable status evaluation indicator set, so that the cable status evaluation is more comprehensive, can accurately reflect the actual operating conditions of the cable, and provides a richer and more comprehensive information basis for subsequent status evaluation, which is helpful to timely discover potential problems and improve the reliability of the cable system; the present invention uses an adaptive probabilistic neural network to perform intelligent evaluation of the status of the high-voltage cable, combining the advantages of probability theory and neural networks, Compared with the existing technology, it can efficiently process and analyze a large amount of cable-related data, so that the network can learn and adapt to changing data characteristics more quickly and accurately when facing changing data characteristics, thereby improving the accuracy and efficiency of the evaluation; in practical applications, the present invention introduces an adaptive mechanism, so that the network can receive new monitoring data in real time and automatically adjust its structure and parameters to adapt to new data characteristics. Compared with traditional cable status evaluation methods, the addition of an adaptive mechanism not only improves the sensitivity of early diagnosis of cable faults, but also can maintain a high level of prediction accuracy in complex and changeable operating environments, providing unprecedented flexibility and efficiency for intelligent monitoring and management of cable systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram of the process of cable joint temperature prediction provided by the present invention;

[0028] Figure 2 A schematic diagram of the process of step S1 provided by the present invention;

[0029] Figure 3 A schematic diagram of the process of step S2 provided by the present invention;

[0030] Figure 4 A schematic diagram of the process of step S3 provided by the present invention;

[0031] Figure 5 A schematic diagram of an APNN adaptive probabilistic neural network provided by the present invention; DETAILED DESCRIPTION

[0032] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0033] Example 1

[0034] The embodiment of the present invention provides a method for evaluating the state of a high-voltage cable based on an adaptive probabilistic neural network of multi-source data. Figure 1 Come and see, Figure 1 The present invention provides a schematic diagram of the process flow of cable joint temperature prediction, wherein the method includes the following contents:

[0035] S1. Select reasonable and comprehensive cable status evaluation indicators, construct evaluation indicator sets and status level division, build multi-source data sensor network layout and data collection mechanism, collect multi-source data related to high-voltage cable status evaluation, and combine Figure 2 Come and see, Figure 2 This is a flow chart of step S1 provided by the present invention, wherein the specific steps in step S1 include the following steps S11 to S13.

[0036] S11. The evaluation index set is divided into static basic indicators and dynamic variable indicators. The specific state information categories and state parameters are shown in Table 1, which statistically includes 21 state quantities.

[0037] Table 1

[0038]

[0039] S12. Classify the status of high-voltage cables into five levels: healthy, caution, abnormal, fault, and severe fault;

[0040] S13. Deploy various types of sensors at key cable nodes, such as thermocouples, voltage and current sensors, insulation monitoring equipment, and environmental monitoring equipment (such as humidity and temperature sensors). As a specific embodiment, collect 200 sets of high-voltage cable related historical data containing 21 state parameters in S11 to construct a data set. Implement a high-frequency sampling strategy to capture instant changes in cable status, ensure the timeliness and accuracy of data, and use data fusion technology to ensure the consistency of data collected from different sources in time and space.

[0041] S2. Preprocess the collected high-voltage cable multi-source data, perform outlier detection and missing value processing, standardize the relevant data, and combine Figure 3Look, Figure 3 FIG. Figure 3 is a schematic flow chart of step S2 provided by the present invention. The specific steps in step S2 include the following steps S21-S22:

[0042] S21. Apply an interpolation method to estimate missing values to complete the processing of missing data, and use the Interquartile Range (IQR) method to detect outliers:

[0043] Calculate the first quartile Q1 and the third quartile Q3, calculate the interquartile range IQR = Q3 - Q1, and outlier determination: x < Q1 - 1.5×IQR or x > Q3 + 1.5×IQR, that is, the values in the data that are less than the lower limit or greater than the upper limit are regarded as outliers.

[0044] S22. The prediction model eliminates the differences between the influencing factor attributes such as temperature and load, and uses Z-score standardization to standardize the data:

[0045]

[0046] where μ is the mean value and σ is the standard deviation.

[0047] S3. Construct an APNN adaptive probabilistic neural network, divide the preprocessed sample data into a training set and a test set, and use historical data and known cable fault cases to train the neural network. During the training process, the APNN automatically updates its parameters according to the real-time data stream to adapt to new data patterns and changing trends;

[0048] Preferably, in combination with Figure 4 Look, Figure 4 FIG. Figure 4 is a schematic flow chart of step S3 provided by the present invention. The specific steps in step S3 include the following steps S31-S33:

[0049] S31. Construct an APNN adaptive probabilistic neural network suitable for high-voltage cable condition assessment, in combination with Figure 5 Look, Figure 5 FIG. Figure 5 is a schematic diagram of the APNN adaptive probabilistic neural network provided by the present invention. The specific network architecture can be expressed as follows:

[0050] 1. Input layer: Receive the input vector x, where x is a feature vector representing different parameters of the system to be evaluated, such as the temperature and current of a high-voltage cable.

[0051] 2. Pattern layer: Each neuron represents a training sample and describes the data through a probability distribution function (usually a Gaussian distribution). For the i-th pattern neuron, its output y i is a function of the input x, usually represented by a Gaussian function:

[0052]

[0053] Where m i is the mean vector of the i-th training sample, σ i is the standard deviation.

[0054] 3. Summary layer: The output of the pattern layer is classified and summarized. For each category Cj, its output Sj is the output y of all pattern neurons belonging to this category. i The sum of:

[0055]

[0056] 4. Output layer: Provides the final judgment or classification result based on the output of the summary layer. If there are K categories, the output of the output layer is a K-dimensional vector, where each element represents the probability that the input x belongs to the corresponding category.

[0057] 5. Adaptive mechanism: The adaptability of APNN is achieved by dynamically adjusting the parameters in the pattern layer (such as m i and σ i ). When the network receives new data, it updates these parameters to better reflect the characteristics of the current data.

[0058] As a specific embodiment, in the present invention, for the evaluation of the high-voltage cable status, the constructed APNN network structure is 21×200×5×1, corresponding to the number of neurons in the input layer, hidden layer, summation layer and output layer respectively. The APNN performance is better when the smoothing factor σ<0.10. The initialization smoothing factor σ in the APNN model of the present invention is 0.05. For the convenience of network calculation, the output state values ​​1, 2, 3, 4, and 5 are set to represent five different states: healthy, attention, abnormal, faulty, and serious.

[0059] S32. The pre-processed sample data is divided into a training set and a test set. As a specific embodiment, 80% of the sample data is randomly selected as a training sample set, and the remaining 20% ​​is used as a test set. During the network training process, the network is trained by back propagation algorithm and gradient descent techniques to minimize the prediction error. At the same time, the k-fold cross-validation method is used to evaluate the generalization ability of the model. The data set is divided into k subsets, and the model is trained k times. Each time, one of the subsets is used as a test set, and the rest are used as training sets. The above steps are repeated so that each subset has the opportunity to become a test set to ensure that the model performs stably on unseen data.

[0060] S33. Monitor network input in real time. When a significant change in data pattern is detected, the adaptive adjustment mechanism is triggered. The network automatically adjusts its parameters based on real-time data to adapt to the new data distribution and characteristics. The network can learn online to continuously adapt to the dynamic changes in cable status.

[0061] As a specific embodiment, during the training process of the model, the difference between the current training error and the previous training error is calculated and compared. If the former is smaller than the latter, the smoothing factor is increased; if the former is larger than the latter and exceeds 1.03 times of the previous error, the training of the next sample is started; otherwise, the smoothing factor is reduced.

[0062] S4. Use the trained APNN network to evaluate the status of the high-voltage cable and output the evaluation value of the relevant status classification, that is, the high-voltage cable status is evaluated as healthy, caution, abnormal, faulty, and severe corresponding levels.

[0063] In order to evaluate the adaptive probabilistic neural network high-voltage cable state assessment method based on multi-source data, as a specific embodiment, the effect of the model proposed in the present invention on the test data set is statistically analyzed, and the prediction accuracy of the high-voltage cable state assessment result is shown in Table 2. It can be seen from Table 2 that the average accuracy of the high-voltage cable state assessment of the APNN model constructed by the present invention reaches 90.16%. During the training process, its network parameters are adjusted in time according to the result error feedback to make the state assessment accurate. It combines advanced data processing technology and adaptive machine learning algorithms, can accurately assess the cable state and guide maintenance decisions in time, thereby improving the reliability and safety of the power system, and shows great potential in improving the accuracy of cable monitoring and reducing operation and maintenance costs.

[0064] Table 2

[0065]

[0066]

[0067] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. An adaptive probabilistic neural network high-voltage cable status assessment method based on multi-source data, characterized in that: include: Step S1: Select cable status evaluation indicators, establish evaluation indicator sets and status level division, build a sensor network layout and data collection mechanism for multi-source data, and collect multi-source data related to high-voltage cable status evaluation; Step S2: preprocessing the collected high-voltage cable multi-source data, performing outlier detection and missing value processing, standardizing the relevant data, and obtaining preprocessed sample data; Step S3: construct an APNN adaptive probabilistic neural network, divide the preprocessed sample data into training sets and test sets, use historical data and known cable fault cases to train the neural network, and during the training process, the APNN automatically updates its parameters according to the real-time data stream to adapt to new data patterns and changing trends; Step S4: using the trained APNN network to evaluate the state of the high-voltage cable and output the evaluation value of the relevant state classification; The step S3 specifically includes: S31. Construct an APNN adaptive probabilistic neural network suitable for high-voltage cable status assessment; S32. The preprocessed sample data is divided into training set and test set respectively to ensure sufficient training samples for learning the APNN model. The performance of the model is evaluated using an independent test set. The network is trained to minimize the prediction error through back propagation algorithm and gradient descent. The learning rate in the gradient descent process is dynamically adjusted to improve the convergence speed and avoid falling into local minima. The cross-validation method is introduced to train and evaluate the APNN model multiple times to verify its generalization ability on different data subsets and ensure the adaptability of the model to diverse data. S33. Real-time monitoring of network input, high-frequency sampling of cable status input data, to ensure that the model can perceive changes in cable status in a timely manner. When a significant change in dynamic indicators is detected, the adaptive adjustment mechanism is triggered, and the network automatically adjusts its parameters according to real-time data to adapt to the new data distribution and characteristics, so as to achieve online learning of the network to continuously adapt to the dynamic changes of cable status and adapt to the evolution of cable status without interrupting evaluation and status prediction; The evaluation index set established in step S1 is divided into static basic indicators and dynamic variable indicators, wherein the static basic indicators include basic equipment information: voltage level, operating years, historical failure rate, load type; dynamic indicators include operation and inspection status information, operation status information, and tunnel environment information, wherein the operation and inspection status information includes insulation resistance, eccentricity, partial discharge, and dielectric loss; the operation status information includes cable temperature, temperature rise rate, cable load, circulating current, and mechanical stress; the tunnel environment information includes ambient humidity, ambient temperature, oxygen concentration, methane concentration, carbon monoxide concentration, hydrogen sulfide concentration, chlorine concentration, and water level information; the established status level is divided, and the evaluation level is specifically divided into five levels: healthy, attention, abnormal, fault, and serious fault; The step S1 of collecting multi-source data related to the high-voltage cable status assessment specifically includes: deploying various types of sensors at key cable nodes, including thermocouples, voltage and current sensors, insulation monitoring equipment, and environmental monitoring equipment, implementing a high-frequency sampling strategy to capture instant changes in the cable status, and using data fusion technology to ensure the consistency of data collected from different sources in time and space; The step S2 performs data preprocessing on the collected high-voltage cable multi-source data, specifically including: S21. Apply interpolation methods to estimate missing values ​​to complete the processing of missing data, and use the interquartile range method to detect outliers; S22. Use Z-score standardization to standardize the data to eliminate the differences between the attributes of the influencing factors: Wherein, μ is the mean value, σ is the standard deviation, and the influencing factors include temperature and load; The network architecture of the APNN adaptive probabilistic neural network consists of an input layer, a pattern layer, a summary layer, and an output layer. The adaptability of the APNN is achieved by dynamically adjusting the parameters in the pattern layer. The network updates these parameters when receiving new data to better reflect the characteristics of the current data. The parameters in the pattern layer include the mean vector m of the i-th training sample i and standard deviation σ i .

Images