Multi-element power supply safety fault grading evaluation method and system

Abstract

The invention discloses a multi-element power supply safety fault grading evaluation method and system, and belongs to the technical field of electrical fault detection, and the method comprises the steps: obtaining a power signal and a characteristic matrix equation of a power system, and obtaining a fault type of the power system through a pre-trained first neural network model; according to the fault type and the characteristic matrix equation of the power system, and in combination with a pre-trained second neural network model, obtaining an electricity utilization safety risk type corresponding to the fault type; obtaining an influence factor of the power utilization safety risk type, and inputting the influence factor into the multi-dimensional risk level evaluation model to obtain a safety risk level corresponding to the power utilization safety risk type; performing partition processing on the power system according to the safety risk level; accurate identification and early warning of safety faults of a power supply system are realized, so that the safety and reliability of the system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical fault detection, and in particular to a multi-element power supply safety fault grading evaluation method and system. BACKGROUND

[0002] Under the global consensus of low-carbon economy, efforts to build a green and low-carbon energy system has become a key measure to promote energy transformation and address climate change. This trend not only concerns environmental protection and sustainable development, but also is an important way to promote economic growth and optimize energy structure. However, the introduction of uncertain source and load elements in the new energy system will lead to an increasing number of faults and an increasing probability of faults, affecting the efficient and stable operation of the system. The importance of fault classification evaluation and monitoring and early warning technology for low-voltage power systems is increasingly prominent, and it has become an indispensable part of new energy technology development.

[0003] During the operation of the power system, overload, short circuit, overvoltage, undervoltage, surge, residual current, fault arc and cable temperature may all act as disaster-causing factors to trigger system faults, affecting the continuity and stability of system power supply. Therefore, establishing an effective low-voltage system fault classification and early warning mechanism, through real-time monitoring of the operation state of the power system, using advanced algorithms and data analysis methods, can identify potential fault risks in advance and issue warning signals, enabling operation and maintenance personnel to take timely measures, which is of great significance to ensure the safe and efficient operation of new energy systems, and can effectively improve the reliability of the system and reduce economic losses and social impacts caused by faults.

[0004] In recent years, with the progress of science and technology, computer technology is widely used in fault classification evaluation and monitoring and early warning technology research. Combined with the latest intelligent algorithm analysis, computer real-time information processing makes the research on alternating current and direct current power system more and more mature and perfect. At present, the main methods for monitoring and early warning of fault safety risk at home and abroad are electrical signal processing (such as wavelet transform and fractional Fourier transform), physical signal monitoring (such as heat, sound, light and electromagnetic radiation), algorithm grading evaluation (such as analytic hierarchy process, support vector machine and convolutional neural network). In the aspect of electrical signal processing (feature extraction), the wavelet feature extraction has the disadvantage of a large decline in diagnostic rate in special scenarios, poor adaptability in multiple working conditions, and even feature weakening in some scenarios, which brings challenges to the improvement of fault arc feature extraction and detection accuracy. In the aspect of physical signal monitoring, the single monitoring and fault analysis of system disaster factors are carried out at present, and the problems of coupling between multiple faults causing safety risk are not considered, and the problems of formation path, accurate identification and grading evaluation of multiple safety risks are not considered. With the development of safety science and technology, the quantitative evaluation technology of fault safety risk plays an important role in emergency management and disaster prevention and mitigation. In the aspect of algorithm grading evaluation, the existing time-frequency domain threshold method is easily disturbed by the field working condition in different systems, and is not suitable for fault positioning in a large range and open space. With the rapid development of deep learning, artificial intelligence and other technologies, it is an inevitable trend to monitor and early warn the electrical fire risk by using various characteristic parameters of electrical fire. However, the problems of low model accuracy caused by unbalanced fault data, low model robustness caused by the fact that the parameters of the constructed neural network do not have physical meaning, and the like limit the ability to improve the accuracy of fault monitoring and early warning to some extent. SUMMARY

[0005] In view of the problems in the prior art, the present application provides a multi-element power supply safety fault grading evaluation and early warning method and system, which realizes accurate identification and early warning of power supply system safety faults, so as to improve the safety and reliability of the system.

[0006] The present application is realized by the following technical solutions: In a second aspect, the present application provides a multi-element power supply safety fault grading evaluation method, comprising the following steps: Step 1, obtaining the power signal and characteristic matrix equation of the power system, and obtaining the fault type of the power system through the pre-trained first neural network model; Step 2, obtaining the power system fault type and characteristic matrix equation, and combining the pre-trained second neural network model to obtain the power consumption safety risk type corresponding to the fault type; Step 3, obtain the influence factors of the power utilization safety risk type, input the influence factor weights into the risk evaluation method to calculate the risk value, and then input the multi-dimensional risk level evaluation model to obtain the safety risk level corresponding to the power utilization safety risk type.

[0007] Preferably, the training method of the first neural network model in step 1 is as follows: Obtain the power signal of different fault types of the power system, construct a characteristic matrix equation according to the time-frequency domain characteristics of the power signal, and train the first neural network model using the characteristic matrix equation. The trained first neural network model outputs the fault type of the power system.

[0008] Preferably, the training method of the second neural network model in step 2 is as follows: Obtain the fault type and characteristic matrix equation of the power system, construct a data set according to the fault type and characteristic matrix equation, and train the second neural network model using the data set. The trained second neural network model outputs the power utilization safety risk type.

[0009] Preferably, the method for obtaining the influence factors of the power utilization safety risk type in step 3 includes: Obtaining the influence factors of the power utilization safety risk type; Constructing a multi-level judgment matrix according to the analytic hierarchy process and combining the influence factors; Scaling the relationship between each influence factor in each level judgment matrix according to the disaster-causing factor evaluation standard; Conducting consistency verification on the scaled judgment matrix according to the consistency index; When each level judgment matrix passes the consistency test, the weights of each influence factor are calculated according to the multi-level judgment matrix, and the risk value is calculated by inputting the weights of each influence factor into the risk evaluation method.

[0010] Preferably, the method for calculating the weights of each influence factor according to the multi-level judgment matrix includes: Normalizing the judgment matrix by column, then summing the row elements of the normalized matrix to obtain the unnormalized weights of each influence factor;

[0011] Then, the unnormalized weights are normalized to obtain the weights of each influence factor.

[0012]

[0013] wherein, is the unnormalized weight, is the weight of the influence factor, is the weight of the i-th influence factor, i is the weight of the j-th influence factor.j Column elements.

[0014] Preferably, obtaining the safety risk level corresponding to the type of electricity safety risk based on the multi-dimensional risk level assessment model includes: A multi-dimensional risk level assessment model is constructed based on the fault amplitude deviation, temperature change and energy release caused by the fault. The weights of the influencing factors are input into the multi-dimensional risk level assessment model to obtain the safety risk level.

[0015] Preferably, the method for calculating the security risk level is as follows:

[0016] Wherein, W1, W2, and W3 represent the proportions of voltage relative offset, current relative offset, and voltage and current harmonics in the fault amplitude offset, respectively; W4 and W5 represent the proportions of absolute temperature offset and relative temperature offset in the risk temperature change, respectively; and W6, W7, and W8 represent the proportions of energy amplitude offset, power amplitude offset, and fault duration in the energy release, respectively.

[0017] Preferably, in step 3, the power system is isolated according to the safety risk level; When the safety risk level is high, the fault in the power system is isolated. When the safety risk level is not high, the development trend of the safety risk is analyzed. If the risk is predicted to be severe, it is isolated, and the isolation effect is output. If the risk is predicted to be mild, isolation is not necessary, and the risk development trend is output directly.

[0018] Secondly, this application provides a graded assessment system for multi-source power supply safety faults, including: The fault classification module is used to obtain the power signals and feature matrix equations of the power system, and to obtain the fault types of the power system through a pre-trained first neural network model. The risk classification module is used to obtain the power safety risk type corresponding to the fault type based on the fault type and feature matrix equation of the power system, combined with the pre-trained second neural network model. The classification module is used to obtain the influencing factors of the electricity safety risk type, substitute the weights of each influencing factor into the risk assessment method to calculate the risk value, and then input it into the multi-dimensional risk level assessment model to obtain the safety risk level corresponding to the electricity safety risk type.

[0019] In a third aspect, the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the hierarchical assessment method for safety faults of multi-power supply when executing the computer program.

[0020] Compared with the prior art, the present application has the following beneficial technical effects: The hierarchical assessment method for safety faults of multi-power supply provided by the present application extracts the features of the fault signals, uses a deep neural network (DNN) model to identify the fault types, then uses the deep neural network (DNN) model to identify the power consumption safety risk types through the fault type and the feature matrix equation, and finally uses a multi-dimensional risk level assessment model to assess the risk through the power consumption safety risk influence factor, thereby realizing accurate identification and early warning of safety faults of the power supply system, and improving the safety and reliability of the system.

[0021] The present application also provides a hierarchical assessment system for safety faults of multi-power supply, an electronic device, and a computer storage medium, which have all the advantages of the hierarchical assessment method for safety faults of multi-power supply. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0023] Figure 1 Flowchart of the hierarchical assessment and early warning method for safety faults of multi-power supply of the present application; Figure 2 Training process of the intelligent model for multi-fault type identification of the present application; Figure 3 Test result comparison (left) and corresponding confusion matrix (right) of the intelligent model for multi-fault type identification of the present application; Figure 4 Test result comparison (left) and corresponding confusion matrix (right) of the intelligent model for multi-fault type identification of the present application; Figure 5 Flowchart of the PINN fault classification model of the present application; Figure 6 Feature extraction result of a typical fault (arc) of the present application; Figure 7 Confusion matrix of the result of the PINN model quantitative test arc risk assessment coefficient of the present application; Figure 8 A relationship diagram is constructed for mapping of system risks and disaster factors; Figure 9 The identification result of the system risk type identification intelligent model of the present application; Figure 10 A three-dimensional evaluation network structure diagram for quantitative grading evaluation of fire risk of the system of the present application; Figure 11 A multi-dimensional evaluation splitting diagram for each fire risk grade of the present application; Table 1 is a performance comparison of different models for arc risk prediction. DETAILED DESCRIPTION

[0024] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0025] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0026] Embodiment 1 A multi-element power supply safety fault grading evaluation method, comprising the following steps: Step 1, obtaining the power signal of the power system, and constructing a signal data matrix according to the power signal.

[0027] Data acquisition and preprocessing. The safety risk of the power utilization system can be caused by multiple faults, so multiple signals of the power utilization line need to be sampled. In this embodiment, a perfect sensor network is constructed by means of network technology to realize real-time transmission and sharing of power signals, and voltage signals, current signals, temperature signals and residual current signals are sampled. Since the collected signals are inevitably disturbed by noise and clutter, the collected data is further filtered and preprocessed to obtain a signal data matrix with higher signal-to-noise ratio.

[0028] Step 2, constructing a feature matrix equation according to the time-frequency domain features of the signal data matrix, and determining the fault type of the power system in combination with a first neural network model; Specifically, the time-frequency domain features of the signal data matrix obtained in step 1 are extracted to obtain a feature matrix equation, and then a neural network model is used for fault type judgment.

[0029] The time-frequency domain feature extraction method is preferably a fractional order wavelet transform.

[0030] As the basic link of power utilization safety risk analysis, the core task of fault classification is to build an efficient feature representation and classification model.

[0031] For fault category judgment task, the selected neural network should be good at solving classification and regression problems, and can select DNN (deep neural network), PINN (physical information neural network) and neural network based on decision tree (such as random forest, gradient boosting tree, etc.). For massive data (tens of thousands or even more samples), the complex structure of deep neural network (DNN) can give full play to its advantages. It can fit complex data distribution through a large number of parameters, and mine deep features and patterns from large-scale data. Therefore, the neural network model in this embodiment is preferably a deep neural network model.

[0032] Deep neural network (Deep Neural Networks, DNN) is a kind of multi-layer neural network, which connects multiple neurons together to form a deep structure. DNN can be used to process various types of data, including images, text, speech, etc. In deep neural network, the number of hidden layers and the number of neurons in each layer varies with specific tasks and data types. The training process of DNN usually adopts back propagation algorithm and gradient descent optimization method, which continuously adjusts the parameters of neural network to minimize prediction error and loss function. Deep neural network has multiple nonlinear mapping feature transformations, which can fit highly complex functions. Compared with shallow modeling method, deep modeling can represent actual complex nonlinear problems more meticulously and efficiently. Deep neural network has achieved remarkable results in many fields, such as image classification, speech recognition, natural language processing, etc. However, deep neural network also faces some challenges, such as overfitting, gradient vanishing and computing resources, etc.

[0033] The basic structure of DNN is a kind of multi-layer neural network, which extracts and transforms data features through multiple hidden layers to get the final output result. This structure can handle complex nonlinear problems and has strong expression ability and flexibility.

[0034] The input layer of DNN is responsible for receiving raw data and converting the data into a format that can be processed by the neural network. For different data types and tasks, the structure of input layer may be different.

[0035] The hidden layers of a DNN are the core components of a deep neural network and can contain multiple neurons. The number of hidden layers and the number of neurons in each layer can be determined based on the specific task and data type. In the hidden layers, each neuron receives information from the previous layer and passes it on to the next layer. This process is achieved through the adjustment of weights and biases. Each neuron applies an activation function to transform the input information into output information. Common activation functions include the sigmoid function, ReLU function, and others. Another important feature of hidden layers is that they can non-linearly map input data. Through the combination of multiple non-linear mappings, neural networks can better fit complex function relationships. This ability allows neural networks to handle various complex classification and regression problems.

[0036] The output layer of a DNN is the last layer in a deep neural network and is responsible for converting the features extracted by the hidden layers into specific outputs. The structure of the output layer depends on the specific task and data type.

[0037] DNN constructs training sets, validation sets, and test sets through multi-dimensional parameters, achieving high-precision classification of faults (accuracy improved by 30%), laying the foundation for data and algorithms; the dynamic risk assessment system is based on the DNN feature library, combined with Bayesian networks to establish a "fault-risk-strategy" mapping model, quantifying risk levels and forming a "detection-evaluation-response" closed loop, perfecting the risk control framework.

[0038] Physics-Informed Neural Networks (PINN) is a neural network that incorporates physical laws such as electromagnetism, thermodynamics, and mechanics as constraints in the loss function during neural network training. By approximating the solution function of the problem using a neural network and using automatic differentiation to calculate the derivative of the input variable, the physical equation is transformed into a constraint condition for the neural network output, guiding the network to learn a solution that conforms to the physical laws.

[0039] The components are as follows: Neural network architecture: commonly uses fully connected neural networks or other suitable structures, the input is generally the spatial and temporal coordinates of the problem and other variables, and the output is the solution of the problem, such as physical field values, and the structure design aims to effectively process and approximate the solution of the target physical problem.

[0040] Data points: data from various sources such as experimental measurements and numerical simulations play a key role in constraining the neural network output during the training process, prompting it to be as close to the true value as possible, ensuring the accuracy and reliability of the model.

[0041] Physical equation constraints: For different physical problems such as heat conduction, fluid mechanics, etc., the corresponding physical equations (such as heat conduction equation, Navier-Stokes equation, etc.) are converted into constraint conditions for the output of the neural network, which is one of the core features of PINN that distinguishes it from traditional neural networks, enhancing the physical reasonableness of the model.

[0042] PINN embeds electromagnetic, heat conduction, and other physical constraints, breaking through the limitations of pure data-driven, achieving fault energy and type joint inversion, and improving detection interpretability from the physical mechanism level; Feature injection positioning technology breaks through the system topology limit to realize accurate positioning through specific signal injection and high sensitivity detection, and the four algorithms are the core, from fault classification to risk assessment, from physical modeling to positioning optimization, gradually deepening, and building a complete technology system of "classification-evaluation-detection-disposal".

[0043] Step 2 aims to use deep learning models for fault detection on current data, using models including physical information neural network (PINN), deep neural network (DNN), and long short-term memory network (LSTM). The following is a comparative analysis of the three models: Accuracy is the proportion of correctly predicted samples (positive and negative classes) by the model. Recall is the proportion of correctly identified fault samples (positive class) by the model, which is particularly important in fault detection tasks, especially when fault samples are few or costly.

[0044] The data set comes from the same time series of fault arc current and voltage values, containing normal current samples and fault current samples. Initial experiments found that the number of normal samples was much larger than the number of fault samples, resulting in a low recall rate (Recall) in fault detection, which could not effectively identify faults. Therefore, the focus of the experiment is to improve the fault recall rate through data adjustment and model optimization, while trying to maintain the overall accuracy (Accuracy).

[0045] The initial data set is not balanced and is directly loaded and divided into training, test, and validation sets (the ratio is usually 6:2:2). We designed three models: PINN: Combines physical constraints (such as current rate of change and spectral characteristics) with classification tasks.

[0046] DNN: Standard fully connected neural network, only using classification loss.

[0047] LSTM: Suitable for time series data, input current segment, and output classification results.

[0048] The performance of PINN can still be further improved. We plan to introduce hyperparameter optimization methods such as grid search, Bayesian optimization, genetic algorithms or AutoML to systematically tune key parameters such as learning rate, hidden layer dimension, Dropout rate, etc. This process aims to further improve the overall accuracy rate while maintaining a high fault recall rate, ultimately creating an efficient and robust current fault detection system.

[0049] Table 1 Comparison of performance of different models. In terms of accuracy and fault recall rate, DNN (86.35% accuracy, 35.71% recall) and LSTM (86.43% accuracy, 20.30% recall) slightly outperform in overall classification performance, suitable for scenarios where accuracy is highly required and the cost of fault detection is low, and DNN has higher accuracy.

[0050]

[0051] Based on the above comparison, the neural network model preferred in this embodiment is a deep neural network model (DNN).

[0052] During the training of the DNN model, we constructed an objective function and used optimization algorithms to find the best method for model training. The design of the objective function aims to make the model have extremely high accuracy and extremely short response time after training. These two indicators are usually contradictory, and improving accuracy may require more complex models or more data, which increases the time. Therefore, we need to combine the two into a comprehensive index and perform weighted summation.

[0053] The main factors affecting the objective function (after the same quantization) include: model learning rate a during training, training iteration number b, data volume corresponding to each label in the data set used for training c, feature extraction type d of the original data, feature number e, and feature extraction method selection f.

[0054]

[0055] The optimization objectives are accuracy A and response time T; To avoid dimensional differences, the indicators need to be standardized (i.e. normalized):

[0056] The response time T needs to be divided by the reference time, i.e.

[0057] The linear weighting of the two optimization objectives is

[0058] where, , is a weight coefficient.

[0059] After constructing the objective function, we determine the optimization algorithm as the Bayesian optimization algorithm.

[0060] The fault type classification model based on the DNN algorithm has four hidden layers, each layer has ten nodes, the training set accounts for 70% of the data set, and the number of nodes in the output layer is set to four. Since the DNN model has nonlinear classification ability, the input value of the DNN model of the application directly uses the original signal data of each disaster factor. The existing sensor data under alternating current working conditions is constructed to form a data set suitable for inputting the DNN model, training is performed, the training period is set to 10000, and the accuracy of the training set during the training process is as follows Figure 2 is the progress in the training process, and it can be seen that after 10000 iterations, the model accuracy is close to 100%.

[0061] Step 3, according to the fault type and characteristic matrix equation of the power system, and combining the second neural network model, the power safety risk type corresponding to the fault type is obtained; Specifically, the fault type and characteristic matrix equation obtained in step 2 are introduced into the second neural network model, and the power safety risk type caused by the fault is judged through the second neural network model.

[0062] The characteristic matrix equation is the wavelet feature extracted from the power signal when the DNN classifies the fault type. It is a feature group for classifying safety risk types, including statistical parameters, frequency domain features, time domain features, and classification features.

[0063] Power safety risks include electric shock, fire, equipment damage, power failure, data loss, etc., which will cause different degrees of safety risks or economic losses.

[0064] The neural network model is preferably a deep neural network (DNN) model in the embodiment.

[0065] The DNN model training process is: the fault type and feature data matrix output by the first neural network model are introduced into the data set for DNN model training, the data categories and sample quantities are analyzed, and the training set, validation set and test set are divided in proportion. When setting the weight, the physical meaning should be considered, and the weight is adjusted from the energy angle combined with the fault internal classification, and the sample imbalance problem and the influence of different fault relationships on the weight are fully considered. Then create a DNN model, train the network and perform simulation test using the processed data, verify the model training effect according to the training result graph, and perform traction learning, considering the applicability of the alternating current model in the direct current scene.

[0066] Step 4, obtain the influence factors of the power utilization safety risk type, put the influence factor weights into the risk evaluation method to calculate the risk value, then input the risk value into the multi-dimensional risk level evaluation model to obtain the safety risk level corresponding to the power utilization safety risk type.

[0067] Risk level division, the risk is evaluated by the multi-dimensional risk level evaluation method. The multi-dimensional risk level evaluation method evaluates the risk level from three dimensions of fault amplitude offset degree, temperature variation degree and energy release amount, then makes a comprehensive consideration to form a three-dimensional matrix, and comprehensively determines the risk level according to the spatial position of the risk in the matrix. The entire cube will be divided into four levels, green represents low fire risk; yellow represents general fire risk; orange represents greater fire risk; red represents major fire risk.

[0068] In this embodiment, the dimensions are divided into the following three: (1) The degree of voltage and current deviation from the normal value caused by the fault. In an electrical system, voltage and current usually operate according to design standards and equipment ratings. Their stability is the basis for ensuring the safe and efficient operation of the system. The greater the deviation from the normal value, the more likely it is to cause the severity of various risks to rise.

[0069] (2) Temperature variation degree. Electrical equipment has a relatively stable normal temperature range during design and operation. This range is determined based on factors such as the material properties, heat dissipation capacity, internal structure, and working principle of the equipment. The greater the temperature deviation from the normal value, the higher the probability of equipment failure. A slight temperature deviation may only cause some minor problems with the equipment, such as a slight decrease in performance, but the equipment can still operate normally. However, as the temperature deviation increases, the probability of various failure mechanisms within the equipment (such as insulation breakdown, material deformation, uncontrolled chemical reactions, etc.) being triggered grows exponentially. For example, when a transformer's temperature is 20°C higher than normal, the speed of insulation material aging may be several times faster than at normal temperature, and the probability of short-circuit failure also increases significantly.

[0070] (3) Energy release amount. In an electrical system, there are various forms of energy, including electrical energy, thermal energy, electromagnetic energy, and mechanical energy. When an electrical risk occurs, these energies may be released in an uncontrolled manner, causing damage to electrical equipment, personnel, and the surrounding environment. When the energy released during an electrical risk is small, the damage to electrical equipment is usually local and minor, and the threat to personnel safety is relatively low; as the energy release amount increases, the degree of damage to electrical equipment rises sharply, and the threat to personnel safety also increases. High energy release may also cause serious damage to the surrounding environment, such as the occurrence of an electrical fire.

[0071] For example,Figure 10 The three-dimensional evaluation network structure diagram for quantitatively grading the fire risk of the system of the application. The three dimensions of fault amplitude offset degree, risk temperature change and energy release amount are A, B and C respectively. The influencing factors of fire risk under fault amplitude offset degree are voltage relative offset degree (R1), current relative offset degree (R2) and voltage and current harmonics (R3); under risk temperature change are absolute temperature offset degree (R4) and relative temperature offset degree (R5), and under energy release amount are energy amplitude offset degree (R6), power amplitude offset degree (R7) and fault duration (R8). According to the discrimination criteria and weights of each index, the calculation formula of fire risk evaluation is obtained:

[0072] Wherein, W1, W2 and W3 respectively represent the proportion of voltage relative offset degree, current relative offset degree and voltage and current harmonics in fault amplitude offset degree; W4 and W5 respectively represent the proportion of absolute temperature offset degree and relative temperature offset degree in risk temperature change; W6, W7 and W8 respectively represent the proportion of energy amplitude offset degree, power amplitude offset degree and fault duration in energy release amount.

[0073] For W i , the proportion is determined by the method of constructing a judgment matrix: The analytic hierarchy process (AHP) first identifies the influencing factors, then arranges the influencing factors into a hierarchical structure, establishes a hierarchical structure model to reduce the complexity of the problem, and then compares the influencing factors in the same layer with each other to construct the judgment matrix of each layer. The judgment matrix in the analytic hierarchy process is an n-order square matrix and is a positive reciprocal matrix. The form of the judgment matrix is:

[0074] In the formula, a ij is the element in the i-th row and the j-th column, i, j = 1, 2, …, n, the elements on the main diagonal are a ii = 1, the elements in the upper half of the triangle are reciprocal to the elements in the lower half of the triangle a ij × a ji = 1.

[0075] The elements in the judgment matrix are scaled according to the nine-level scale method, and the relationship between the influencing factors is scaled according to the evaluation standard of the disaster-causing factor.

[0076]

[0077] In fact, the evaluators cannot always make perfect judgments. In some cases, there may be inconsistencies, so a consistency index C Iconsistency index C I is defined as:

[0078] where λ m is the largest eigenvalue of the judgment matrix Y and n is the order of the judgment matrix Y.

[0079] In practical problems, pair-wise comparison is unable to obtain a perfect consistent judgment matrix, so it is the goal of the evaluator to obtain a judgment matrix close to a completely consistent matrix. The consistency index can determine how far a certain matrix is from a consistent matrix, C I function should have a unique local minimum value. When the consistency index exceeds the critical value, the previous judgment needs to be modified. If the judgment matrix Y is completely consistent, the average of other eigenvalues must be 0, so C I takes the value of 0. The random consistency index R I is introduced, which is the average of C I . The R I values of common matrix orders are shown in the table.

[0080]

[0081] The consistency ratio index C R is introduced, with a threshold value of 0.1. If C R > 0.1, i.e., the judgment matrix does not pass the consistency test, the judgment matrix needs to be reconstructed.

[0082]

[0083] When all the judgment matrices of the levels pass the consistency test, the weight of each factor is calculated. In the weight calculation process, the judgment matrix is normalized by column, and then the sum of the row elements of the normalized matrix is obtained, which is the unnormalized weight W i * of each factor. The calculation formula is:

[0084] The relative weight W i * of each influencing factor is obtained by normalizing W i , and the calculation formula is

[0085] The discrimination criteria of each index calculated above and the weight are substituted into the calculation formula of the first fire risk evaluation to calculate the fire risk value. According to the fire risk fire risk evaluation magic cube Figure 10 , the risk grade is obtained.

[0086] As Figure 10 , the risk evaluation magic cube diagram can more intuitively show the risk level.

[0087] Step 5, according to the safety risk level, the power system is isolated.

[0088] According to the risk level of the power safety risk, it is judged whether the risk is high risk, that is, major fire risk (red), if it is high risk, the fault is isolated in the first time, and further hidden danger tracing work is carried out, and it is determined that the risk is caused by which fault, and the equipment is quickly restored to normal operation; If the risk is judged to be low risk, that is, orange, yellow and green part, the risk is predicted. According to its future development trend, it is judged whether to isolate. If the risk is isolated, the isolation effect is output; if the risk is not isolated, the future development trend of the risk is output.

[0089] Figure 2 is the accuracy rate change of the training set in the DNN training process under alternating current fault. The disaster-causing risk rating model based on DNN algorithm built by the present application has four hidden layers, each layer has ten nodes, the training set accounts for 70% of the data set, and the number of nodes in the output layer is set to four. Due to the nonlinear classification ability of the DNN model, the input value of the DNN model of the present application directly uses the original signal data of each disaster-causing factor. The sensor data under the existing alternating current working condition is constructed to form a data set suitable for inputting the DNN model, and the training period is set to 10000. The accuracy rate of the training set during the training process is as follows Figure 2 .

[0090] Figure 3 , Figure 4 are the comparison between the prediction results and the actual results of the training set and the test set of the DNN model respectively. It can be seen that the accuracy rate of the training set after training is 99.4505%, and the prediction set accuracy rate is 94.9367%, which is relatively high. According to the corresponding confusion matrix, the accuracy rate of each type of fault type judgment can be more intuitively seen.

[0091] Figure 5 is a flow chart of a PINN fault classification model. The establishment process of the model is to first determine the architecture of PINN, including the input layer, the hidden layer and the output layer. The input layer receives the extracted feature vector, and the output layer corresponds to different fault categories (normal / fault). Then define the physical constraint condition, take the electromagnetic, thermal and mechanical equations as part of the model, and introduce the physical constraint through the loss function of PINN. These constraint conditions are integrated into the loss function of PINN.

[0092] Figure 6To extract features from the typical waveforms of the previous fault arc, the red curve is the result of feature extraction. It can be seen that the wavelet packet decomposition of the current signal has a significant effect on feature extraction. The amplitude changes significantly before and after the occurrence of the fault arc, and the system state can be clearly distinguished, and it is not affected by the "false arc interference" caused by load switching. Therefore, the current signal of the system is decomposed by wavelet packet to determine the system state of the fault arc. The wavelet packet decomposition is used as a feature layer scheme to construct a fault disaster risk warning model.

[0093] Figure 7 is the PINN model test result confusion matrix. The PINN model is trained using the Adam optimizer, with a learning rate of 0.001 and 400 iterations. The loss value is output every 100 rounds. After training is completed, we evaluate the performance of the model on the test set, and the accuracy of the model is 99.86%. This method performs well in the current fault detection task and can provide a high-accuracy fault detection model. Future work can further optimize the physical constraint loss function and combine more features for multi-task learning to improve the generalization ability and robustness of the model.

[0094] Figure 8 is the relationship diagram between power system risk and disaster-causing factors. During the operation of the power system, overload, short circuit, overvoltage, undervoltage, surge, residual current, fault arc, and cable temperature may all act as disaster-causing factors to trigger system failures, affecting the continuity and stability of system power supply. In this part, the risks of electric shock and fire caused by various disaster-causing factors are mainly studied. According to relevant national standards as risk judgment conditions, the relationship between system risks caused by various disaster-causing factors is described.

[0095] Figure 9 is the DNN system risk assessment result. The types of electrical safety risks include electric shock, fire, equipment damage, power outage, and data loss. This embodiment mainly studies the two types of electrical safety risks: electric shock and fire. Figure 9 The results of the DNN system in risk assessment are shown, including temperature changes, arc current fluctuations, and dynamic changes in safety risk types. The temperature at L1 point rises, so before L1 there is no risk, and the safety risk type is 1. The arc occurs at L2 point, so the temperature rises between L1 and L2, which corresponds to a safety risk type of 2. The fault arc ends at L3 point, so Figure 8 The system risk and disaster-causing factor relationship diagram shows that temperature rise causes fire risk, and arc causes electric shock and fire risk. Therefore, between L2 and L3, there is an over-temperature and an arc, which corresponds to a safety risk type of 4. Since the fault arc ends at L3, there is only an over-temperature after L3, which corresponds to a safety risk type of 2.

[0096] Figure 10 , Figure 11respectively are a fire risk evaluation network structure diagram and a fire risk evaluation magic cube diagram. A multi-dimensional risk level evaluation method is used in the risk evaluation link. Taking fire risk as an example, the multi-dimensional risk level evaluation method evaluates the risk level from three dimensions of fault amplitude offset degree, risk temperature change and energy release, and then comprehensively considers to form a three-dimensional matrix, and the risk level is comprehensively determined according to the spatial position of the risk in the matrix, and the fire risk evaluation magic cube diagram can more directly show the core idea of the multi-dimensional risk level evaluation method.

[0097] Table 1 is a comparison of different model performances. Under the comparison of accuracy and fault recall rate indicators, DNN (86.35% accuracy, 35.71% recall rate) and LSTM (86.43% accuracy, 20.30% recall rate) slightly outperform in overall classification performance, and are suitable for scenarios with high accuracy requirements and low fault detection cost, and the accuracy of DNN is higher than that of LSTM.

[0098]

[0099] Correspondingly, the application provides a multi-element power supply safety fault grading evaluation system, comprising: A fault classification module is configured to obtain power signals and a characteristic matrix equation of a power system, and obtain a fault type of the power system through a pre-trained first neural network model. A risk classification module is configured to obtain a fault type of the power system and a characteristic matrix equation, and obtain a power safety risk type corresponding to the fault type by combining a pre-trained second neural network model. A level classification module is configured to obtain an influence factor of the power safety risk type, calculate a risk value by substituting weights of the influence factor into a risk evaluation method, and then input a multi-dimensional risk level evaluation model to obtain a safety risk level corresponding to the power safety risk type.

[0100] It should be noted that in the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules can be combined or integrated into another device, or some features can be ignored or not executed. The modules described as separate components can be or can not be physically separated, and the components displayed as modules can be a physical unit or multiple physical units, that is, can be located in one place, or can be distributed to multiple different places. According to actual needs, part or all of the modules can be selected to achieve the purpose of the embodiments of the present application.

[0101] In addition, each module in various embodiments of the present application can be integrated in one processing unit, or each module can be physically present alone, or two or more modules can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0102] The electronic device provided in the embodiments of the present application includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the hierarchical evaluation method of the multi-power supply safety fault as described in any of the above embodiments when executing the computer program.

[0103] The electronic device provided in another embodiment of the present application can further include: an input port connected to the processor, configured to transmit the multi-modal data collected by an external collection device to the processor; a display unit connected to the processor, configured to display the processing result of the processor to the outside world; and a communication module connected to the processor, configured to realize the communication between the electronic device and the outside world. The display unit can be a display panel, a laser scanning display, etc. The communication mode adopted by the communication module includes but is not limited to mobile high-definition link technology (HML), universal serial bus (USB), high-definition multimedia interface (HDMI), wireless connection (including wireless fidelity technology (WiFi), Bluetooth communication technology, low-power Bluetooth communication technology, and IEEE 802.11s-based communication technology).

[0104] The computer readable storage medium provided in the embodiments of the present application stores a computer program, and the computer program is executed by the processor to implement the steps of the hierarchical evaluation method of the multi-power supply safety fault as described in any of the above embodiments.

[0105] The related parts of the hierarchical evaluation system of the multi-power supply safety fault, the electronic device and the computer readable storage medium provided in the embodiments of the present application are described in detail in the corresponding part of the hierarchical evaluation method of the multi-power supply safety fault provided in the embodiments of the present application, and will not be described here. In addition, the parts of the above technical solutions provided in the embodiments of the present application which are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail, so as not to be too redundant.

[0106] The above content only illustrates the technical idea of the present application, and cannot limit the protection scope of the present application. Any modification made according to the technical idea of the present application on the basis of the technical solution falls within the protection scope of the claims of the present application.

Claims

1. A method for hierarchical assessment of multiple power supply safety faults, characterized in that, The method comprises the following steps: Step 1, obtaining the power signal and the characteristic matrix equation of the power system, and obtaining the fault type of the power system through the pre-trained first neural network model; Step 2, obtaining the fault type corresponding to the power consumption safety risk type according to the fault type of the power system and the characteristic matrix equation, and combining the pre-trained second neural network model; Step 3, obtaining the influence factor of the power consumption safety risk type, substituting the weight of the influence factor into the risk evaluation method to calculate the risk value, and then inputting the multi-dimensional risk level evaluation model to obtain the safety risk level corresponding to the power consumption safety risk type.

2. The method of claim 1, wherein, The training method of the first neural network model in step 1 is as follows: Obtain the power signal of different fault types of the power system, construct the characteristic matrix equation according to the time-frequency domain characteristics of the power signal, train the first neural network model using the characteristic matrix equation, and the trained first neural network model outputs the fault type of the power system.

3. The method of claim 1, wherein, The training method of the second neural network model in step 2 is as follows: Obtain the fault type and the characteristic matrix equation of the power system, construct a data set according to the fault type and the characteristic matrix equation, train the second neural network model using the data set, and the trained second neural network model outputs the power consumption safety risk type.

4. The method of claim 1, wherein, Step 3 obtains the influence factor of the power consumption safety risk type, and substitutes the weight of the influence factor into the risk evaluation method to calculate the risk value, which comprises: Obtain the influence factor of the power consumption safety risk type; According to the analytic hierarchy process and combining the influence factor, a multi-level judgment matrix is constructed; According to the disaster-causing factor evaluation standard, the relationship between each influence factor in each layer judgment matrix is scaled; According to the consistency index, the scaled judgment matrix is verified for consistency; When each layer judgment matrix passes the consistency test, the weight of each influence factor is calculated according to the multi-level judgment matrix, and the weight of each influence factor is substituted into the risk evaluation method to calculate the risk value.

5. The method of claim 4, wherein, The weight of each influence factor is calculated according to the multi-level judgment matrix, which comprises: The judgment matrix is normalized by column, and then the sum of the row elements of the normalized matrix is obtained to obtain the non-normalized weight of each influence factor; Then, the non-normalized weight is normalized to obtain the weight of each influence factor; wherein, is the non-normalized weight, is the weight of the impact factor, is the first i row, first j column element.

6. The method of claim 5, wherein, According to the multi-dimensional risk level evaluation model, the safety risk level corresponding to the power consumption safety risk type is obtained, which comprises: According to the fault amplitude offset degree, the temperature change degree and the energy release amount caused by the fault, a multi-dimensional risk level evaluation model is constructed, the weight of the influence factor is input into the multi-dimensional risk level evaluation model, and the safety risk level is obtained.

7. The method of claim 6, wherein, The calculation method of the safety risk level is as follows: Wherein, W1, W2, W3 respectively represent the proportion of voltage phase offset degree, current phase offset degree and voltage and current harmonics in fault amplitude offset degree; W4, W5 respectively represent the proportion of absolute temperature offset degree and relative temperature offset degree in risk temperature change; W6, W7, W8 respectively represent the proportion of energy amplitude offset degree, power amplitude offset degree and fault duration in energy release amount.

8. The method of claim 1, wherein, In step 3, the power system is isolated according to the safety risk level. When the security risk level is high risk, the fault of the power system is isolated; When the security risk level is non-high risk, the development trend of the security risk is analyzed, if the risk prediction is serious, the isolation is performed, then the isolation effect is output, if the risk prediction is not serious, the risk development trend is directly output without isolation.

9. A hierarchical assessment system for multiple power supply safety faults, characterized by, It comprises: a fault classification module, configured to obtain power signals and characteristic matrix equations of the power system, and obtain a fault type of the power system through a pre-trained first neural network model; a risk classification module, configured to obtain a power safety risk type corresponding to the fault type according to the fault type and the characteristic matrix equations of the power system, and in combination with a pre-trained second neural network model; a level classification module, configured to obtain influence factors of the power safety risk type, calculate a risk value by substituting weights of the influence factors into a risk evaluation method, and then input a multi-dimensional risk level evaluation model to obtain a security risk level corresponding to the power safety risk type.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the multi-element power supply safety fault grading evaluation method according to any one of claims 1 to 8.

Images