A power distribution network cable joint fault monitoring method and system based on voiceprint analysis

Abstract

The application discloses a power distribution network cable joint fault monitoring method and system based on voiceprint analysis, and relates to the technical field of power distribution networks.The method comprises the following steps: performing monitoring urgency evaluation of partial discharge fault monitoring based on predicted material state information and predicted scene state information of a cable joint area in a preset time zone, and setting an adaptive identification time length; taking maximum fault identification accuracy and approximation of the adaptive identification time length as scheme optimization targets, and outputting an optimal fault identification scheme; periodically acquiring original acoustic signals of the cable joint area, and performing signal processing and fault identification on the original acoustic signals according to the optimal fault identification scheme, and outputting a fault monitoring result.The technical problem that the existing cable joint fault identification scheme lacks a dynamic adjustment mechanism, cannot be dynamically adjusted according to real-time equipment states and operating environments, results in low detection efficiency, high missed detection rate and inability to realize real-time monitoring is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution networks, in particular to a power distribution network cable joint fault monitoring method and system based on voiceprint analysis. BACKGROUND

[0002] With the continuous expansion of the power distribution network scale and the increasing application of cables, as the key connection point of power transmission, cable joints are long-term subjected to high voltage, large current and complex environmental influences, and are prone to partial discharge due to insulation aging, poor contact or mechanical damage, which further leads to overheating, breakdown and even fire accidents. The operation state of the cable joint is directly related to the safety and stability of the entire power distribution network. However, in the complex and changeable operating environment, the cable joint is prone to partial discharge due to material aging, environmental factors and overload operation. The traditional cable joint fault detection method relies on periodic inspection and manual judgment, resulting in low detection efficiency, high missed detection rate and inability to monitor in real time. SUMMARY

[0003] The embodiments of the present application provide a power distribution network cable joint fault monitoring method and system based on voiceprint analysis, which solves the technical problem that the existing cable joint fault identification scheme lacks a dynamic adjustment mechanism and cannot be dynamically adjusted according to the real-time state of the equipment and the operating environment, resulting in low detection efficiency, high missed detection rate and inability to monitor in real time.

[0004] The technical solution of the present application to solve the above technical problems is as follows:

[0005] In a first aspect, the present application provides a power distribution network cable joint fault monitoring method based on voiceprint analysis, which comprises:

[0006] Performing a monitoring urgency evaluation based on the predicted material state information and the predicted scene state information of the cable joint area in a preset time zone for partial discharge fault monitoring, and setting an adaptive identification time length according to the urgency evaluation result;

[0007] Taking the maximum fault identification accuracy and approximating the adaptive identification time length as the optimization target, iteratively optimizing and searching the acoustic signal processing strategy and the fault identification strategy in the preset time zone, and outputting an optimal fault identification scheme;

[0008] In the preset time zone, continuously sampling through an acoustic sensor, periodically acquiring the original acoustic signal of the cable joint area, and performing signal processing and fault identification on the original acoustic signal according to the optimal fault identification scheme, and outputting a fault monitoring result.

[0009] In a second aspect, the present application provides a power distribution network cable joint fault monitoring system based on voiceprint analysis, which comprises:

[0010] The recognition duration configuration module is configured to set an adaptive recognition duration based on the monitoring urgency evaluation of the partial discharge fault monitoring based on the predicted material state information and the predicted scene state information of the cable joint area in the preset time zone;

[0011] The recognition scheme output module is configured to maximize the fault recognition accuracy and approximate the adaptive recognition duration as the scheme optimization target, iteratively optimize and search the acoustic signal processing strategy and the fault recognition strategy in the preset time zone, and output an optimal fault recognition scheme.

[0012] The detection result output module is configured to continuously sample through the acoustic sensor in the preset time zone, periodically acquire the original acoustic signal of the cable joint area, and perform signal processing and fault recognition on the original acoustic signal according to the optimal fault recognition scheme, and output a fault monitoring result.

[0013] The present application provides one or more technical solutions, at least with the following technical effects or advantages:

[0014] The power distribution network cable joint fault monitoring method and system based on acoustic analysis provided by the embodiment of the present application first dynamically evaluates the urgency of the partial discharge fault monitoring based on the predicted material state information and the predicted scene state information of the cable joint area, and sets an adaptive recognition duration accordingly, thereby avoiding the problems of untimely detection or resource waste caused by a fixed detection period. Secondly, the acoustic signal processing strategy and the fault recognition strategy are iteratively optimized and searched with the goal of maximizing the fault recognition accuracy and approximating the adaptive recognition duration, which can automatically generate an optimal fault recognition scheme, thereby improving the accuracy and efficiency of fault recognition. Finally, the acoustic sensor is continuously sampled in the preset time zone, and the original acoustic signal is processed and recognized according to the optimal fault recognition scheme, thereby realizing real-time monitoring of the cable joint fault and providing protection for the safe and stable operation of the power distribution network.

[0015] Through the above technical solutions, the present application realizes accurate prediction and real-time monitoring of the partial discharge fault based on the acoustic analysis technology and in combination with the material state and environmental scene information of the cable joint area. The mechanism of dynamically evaluating the monitoring urgency and setting an adaptive recognition duration makes the detection process more flexible and efficient, and can timely adjust the detection strategy according to the actual situation, thereby avoiding unnecessary resource waste. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 is a flowchart of a power distribution network cable joint fault monitoring method based on voiceprint analysis provided by an embodiment of the present application.

[0018] Figure 2 is a structural diagram of a power distribution network cable joint fault monitoring system based on voiceprint analysis provided by an embodiment of the present application.

[0019] In the drawings, the components represented by the respective reference numerals are described as follows:

[0020] The identification duration configuration module 11, the identification scheme output module 12, and the detection result output module 13. DETAILED DESCRIPTION

[0021] The embodiments of the present application provide a power distribution network cable joint fault monitoring method and system based on voiceprint analysis, which are used to solve the technical problem that the existing cable joint fault identification scheme lacks a dynamic adjustment mechanism and cannot be dynamically adjusted according to the real-time state of the equipment and the operating environment, resulting in low detection efficiency, high missed detection rate, and inability to monitor in real time.

[0022] 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 part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative labor fall within the scope of protection of the present application.

[0023] In the description of the present application, the terms "first" and "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0024] In the description of the present application, the term "for example" is used to mean "serving as an example, instance, or illustration." Any embodiment described as "for example" in the present application is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the present application. In the following description, for the purposes of explanation, numerous details are set forth in order to provide a thorough understanding. It should be apparent to those skilled in the art that the present application can be practiced without the use of these specific details. In other instances, well known structures and processes are not elaborated upon in order to avoid unnecessary detail, which might obscure the description of the present application. Thus, the present application is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0025] Embodiment one, as shown in the present application, provides a power distribution network cable joint fault monitoring method based on voiceprint analysis, comprising: Figure 1

[0026] S10: monitoring emergency evaluation of partial discharge fault monitoring based on the predicted material state information and the predicted scene state information of the cable joint area in the preset time zone, and setting an adaptive identification duration according to the emergency evaluation result;

[0027] In the present application, first, the monitoring emergency evaluation is performed based on the predicted material state information and the predicted scene state information of the cable joint area, the predicted material state information covers the insulation performance degradation degree and mechanical stress state of the cable joint material, and is collected and transmitted to the data processing center in real time by deploying various sensors in the cable joint area; the predicted scene state information includes external factors such as environmental temperature, humidity, and electromagnetic interference level, which affect the occurrence and development of partial discharge.

[0028] By comprehensively analyzing the predicted material state information and the predicted scene state information, the fuzzy comprehensive evaluation method or the neural network model is used to quantitatively evaluate the emergency degree of partial discharge fault monitoring.

[0029] Specifically, each item of state information is taken as an input variable, and after model processing, an emergency evaluation value between 0 and 1 is output, where 0 represents no emergency monitoring requirement, and 1 represents an urgent need for fault monitoring. According to the emergency evaluation value, the adaptive identification duration is set, when the emergency evaluation value is high, the identification duration is shortened to improve the monitoring response speed; when the emergency evaluation value is low, the identification duration is appropriately extended to save monitoring resources.

[0030] Among them, the monitoring emergency evaluation of partial discharge fault monitoring based on the predicted material state information and the predicted scene state information of the cable joint area in the preset time zone comprises:

[0031] ​Based on the inspection data of the cable joint area, a material aging index sequence in a historical time range is obtained, and a material aging index at the last inspection is extracted as a first material aging index;

[0032] A material aging trend coefficient is obtained according to the material aging index sequence analysis, and a predicted material aging index mean in a preset time zone is predicted according to the material aging trend coefficient and the first material aging index, as predicted material state information;

[0033] A predicted current load sequence, a predicted temperature sequence and a predicted humidity sequence of the cable joint area in the preset time zone are obtained;

[0034] The predicted current load sequence, the predicted temperature sequence and the predicted humidity sequence are calculated and feature-extracted respectively to obtain a predicted current maximum load, a predicted current load mean, a predicted current load fluctuation coefficient, a predicted maximum temperature, a predicted temperature mean, a predicted temperature fluctuation coefficient, a predicted maximum humidity, a predicted humidity mean and a predicted humidity fluctuation coefficient, as predicted scene state information;

[0035] A monitoring urgency evaluation plug-in of the cable joint area is constructed based on a deep learning model, a monitoring urgency evaluation of partial discharge fault monitoring is performed according to the predicted material state information and the predicted scene state information, and a predicted monitoring urgency is output as an urgency evaluation result.

[0036] In the embodiments of the present application, first, based on the inspection data of the cable joint area, a material aging index sequence in a historical time range is obtained, and the material aging index sequence reflects the aging degree change of the cable joint material in the past period of time. The material aging index at the last inspection is extracted as a first material aging index, and the first material aging index represents the latest aging state of the cable joint material.

[0037] Specifically, the evaluation of the material aging index is a quantitative process based on multi-source information fusion, which is realized by comprehensively analyzing various direct and indirect indicators reflecting the insulation condition of the cable joint. The indicators include historical partial discharge activity intensity such as average discharge quantity and cumulative energy, electrical test data such as change trend of insulation resistance and dielectric loss angle, operating temperature history estimated based on thermal aging model of load and environment, and appearance inspection records such as tracking and cracking. By normalizing and weighting the heterogeneous data, an aging evaluation model is constructed.

[0038] Further, the model finally outputs an index value standardized in the range of [0, 1], where 0 represents "no aging" and 1 represents "end of life". Through the quantitative index, the system can overcome the limitations of real-time monitoring, intelligently diagnose and predictively maintain the health status of the joint throughout its life cycle, and provide decision basis for adaptive adjustment of monitoring strategy.

[0039] Subsequently, according to the sequence of material aging indexes, a material aging trend coefficient is obtained by a statistical method or a machine learning algorithm. The material aging trend coefficient reflects the speed of material aging and the trend of the change in the aging direction. Using the material aging trend coefficient and the first material aging index, a predicted material aging index mean at multiple time points in a preset time zone is predicted as the predicted material state information. The preset time zone is relatively large, such as 15 days or a month.

[0040] At the same time, a predicted current load sequence, a predicted temperature sequence, and a predicted humidity sequence of the cable joint area in the preset time zone are obtained, which respectively reflect the current load that the cable joint may bear in the future, the change in the environmental temperature, and the change in the humidity.

[0041] Exemplarily, a machine learning model is used to predict the current load sequence, and historical load, weather forecast, date type, and the like are used as features to train a supervised learning model. A long short-term memory network is used to train the model, which is good at processing time series and learning long-term dependencies, and outputs a predicted current load sequence for 24 hours or more in the future.

[0042] Further, the predicted current load sequence, the predicted temperature sequence, and the predicted humidity sequence are respectively calculated and feature extracted to obtain parameters, including a predicted maximum current load, a predicted current load mean, a predicted current load fluctuation coefficient, a predicted maximum temperature, a predicted temperature mean, a predicted temperature fluctuation coefficient, a predicted maximum humidity, a predicted humidity mean, and a predicted humidity fluctuation coefficient. The above parameters collectively constitute the predicted scene state information, which provides comprehensive environmental scene data support for monitoring urgency evaluation.

[0043] Finally, a monitoring urgency evaluation plug-in of the cable joint area is constructed based on a deep learning model. The plug-in can receive the predicted material state information and the predicted scene state information as inputs, and perform monitoring urgency evaluation of partial discharge fault monitoring through the powerful processing capability of the deep learning model. After the model processing, a predicted monitoring urgency is output as the urgency evaluation result, which provides a basis for subsequent adaptive identification time length setting.

[0044] Specifically, the monitoring urgency evaluation plug-in is constructed and trained based on a deep learning model, and the specific steps are as follows:

[0045] First, data preparation. The input nodes of the monitoring urgency evaluation plug-in are the predicted material state information and the predicted scene state information, which are calculated and extracted based on the predicted current load sequence, the predicted temperature sequence, and the predicted humidity sequence.

[0046] Secondly, model construction, the input layer is the node number equal to the dimension of the input characteristics, such as predicting the average of the material aging index, predicting the maximum load of the current, predicting the average of the current load, predicting the fluctuation coefficient of the current load, predicting the maximum temperature, predicting the average of the temperature, predicting the fluctuation coefficient of the temperature, predicting the maximum humidity, predicting the average of the humidity and predicting the fluctuation coefficient of the humidity, a total of 10 characteristics, so the input layer contains 10 nodes; 1-3 layers of hidden layer are set, the number of nodes of each layer is adjusted through experiment, such as 64, 32, etc., and the activation function is selected as ReLU; the output layer generally does not use the activation function, such as 2 nodes of output time consumption, directly outputting continuous values.

[0047] Thirdly, model training, the predicted monitoring urgency is taken as the output, and the urgency evaluation result is obtained. An Adam optimizer and a mean square error loss function are used to construct a training framework, the batch size is set to 32, the total training rounds are set to 50, and an early stopping mechanism (patience=5) is introduced, when the loss of the verification set does not appear for 5 consecutive rounds, the training process is automatically terminated, and the trained monitoring urgency evaluation plug-in is obtained, so that the model overfitting is effectively avoided, and the model reaches the convergence state.

[0048] Further, the adaptive identification time length is set according to the urgency evaluation result, including:

[0049] The ratio of the preset standard monitoring urgency and the predicted monitoring urgency is set as the time limit compensation coefficient;

[0050] After the preset identification time length is compensated according to the time limit compensation coefficient, the adaptive identification time length is obtained.

[0051] In the embodiment of the application, firstly, the ratio of the preset standard monitoring urgency and the predicted monitoring urgency is set as the time limit compensation coefficient. For example, the preset standard monitoring urgency can be set to 0.5. The time limit compensation coefficient reflects the relative relationship between the predicted monitoring urgency and the preset standard monitoring urgency. When the predicted monitoring urgency is higher than the preset standard monitoring urgency, the time limit compensation coefficient is greater than 1, which means that the identification time length needs to be shortened to improve the timeliness of monitoring; when the predicted monitoring urgency is lower than the preset standard monitoring urgency, the time limit compensation coefficient is less than 1, at this time, the identification time length is lengthened to save monitoring resources.

[0052] Then, the preset identification time length is compensated according to the time limit compensation coefficient. The preset identification time length can be pre-set according to system performance, monitoring accuracy requirements and other factors, such as 1 hour. The specific calculation method is: adaptive identification time length = preset identification time length x time limit compensation coefficient. Through calculation, the identification time length is dynamically adjusted according to different monitoring urgency conditions, so that the monitoring process is more flexible and efficient, thereby better meeting the actual needs of the power distribution network cable joint fault monitoring.

[0053] For example, when the predicted monitoring urgency is 0.6 and the preset standard monitoring urgency is 0.5, the time limit compensation coefficient is 0.6 ÷ 0.5 = 1.2. If the preset identification time is 1 hour, then the adapted identification time = 1 × 1.2 = 1.2 hours, that is, the identification time needs to be appropriately shortened to speed up the monitoring and detect the partial discharge fault that may exist in the cable joint more promptly.

[0054] When the predicted monitoring urgency is 0.3 and the preset standard monitoring urgency is still 0.5, the time limit compensation coefficient is 0.3 ÷ 0.5 = 0.6. At this time, the adapted identification time is 1 × 0.6 = 0.6 hours. That is, when the monitoring urgency is low, the identification time is appropriately extended to avoid unnecessary frequent monitoring and save monitoring resources.

[0055] S20: With the goal of maximizing fault identification accuracy and approximating the adaptive identification time, iteratively optimize and search the acoustic signal processing strategy and fault identification strategy within the preset time zone, and output the optimal fault identification scheme.

[0056] In this embodiment, the optimization objective is to maximize fault identification accuracy and approximate the appropriate identification time. The acoustic signal processing strategy and fault identification strategy within a preset time zone are iteratively optimized and searched. Specifically, this process is achieved by constructing an optimization framework that includes signal preprocessing, feature extraction, and pattern recognition.

[0057] First, at the acoustic signal processing strategy level, the center frequency and bandwidth parameters of the bandpass filter are dynamically adjusted to suppress background noise interference, taking into account the characteristics of ultrasonic signals generated by partial discharge in cable joints. Simultaneously, the number of layers and basis function selection in wavelet packet decomposition are optimized to enhance the separability of fault features. Second, at the fault identification strategy level, a genetic algorithm is used to globally optimize the kernel function type, penalty coefficient, and kernel parameters of the support vector machine, or a particle swarm optimization algorithm is used to optimize the kernel size, stride, and pooling method of the convolutional neural network, thereby improving the model's ability to identify weak discharge signals.

[0058] During the iterative optimization process, fault identification accuracy and the deviation between the actual identification time and the adapted identification time are used as dual objective functions. A Pareto optimal solution set is generated through a multi-objective evolutionary algorithm. After each iteration, the strategy parameters are dynamically adjusted based on the objective function value until convergence conditions, such as the maximum number of iterations or the objective function improvement threshold, are met. The final output optimal fault identification scheme includes the optimized signal processing parameter set and fault identification model parameter set, ensuring the highest possible fault detection rate within the adapted identification time.

[0059] Specifically, step S20 in the method includes:

[0060] obtain a plurality of acoustic index selection schemes based on the plurality of acoustic indexes through permutation and combination;

[0061] construct a plurality of fault identification channels based on the plurality of acoustic index selection schemes, wherein each fault identification channel includes Q fault identification branches;

[0062] perform iterative optimization search on the acoustic signal processing strategy and the fault identification strategy in the preset time zone based on the plurality of acoustic index selection schemes and the plurality of fault identification channels, with the scheme optimization target being maximizing fault identification accuracy and approximating the adaptive identification time length, and output an optimal fault identification scheme.

[0063] In the embodiments of the present application, firstly, a plurality of acoustic indexes after acoustic signal processing are obtained. The acoustic indexes can reflect the characteristic information of the partial discharge of the cable joint from different angles, including time domain indexes, frequency domain indexes, time-frequency domain indexes, and nonlinear dynamic indexes.

[0064] The time domain indexes are direct measurement results of the acoustic signal in the time domain, such as the amplitude, duration, and rise time of the signal, which can intuitively show the transient characteristics of the partial discharge. The frequency domain indexes are characteristics obtained by converting the time domain signal to the frequency domain through Fourier transform or the like, such as the main frequency component and the frequency band energy distribution, which are helpful for identifying the frequency characteristics of the discharge signal. The time-frequency domain indexes combine the analysis methods of the time domain and the frequency domain, such as the time-frequency spectrum obtained by short-time Fourier transform or wavelet transform, which can simultaneously capture the time-varying characteristics and frequency distribution of the signal. The nonlinear dynamic indexes reveal the nonlinear dynamic process behind the partial discharge by analyzing the chaotic characteristics and fractal dimension of the acoustic signal from the perspective of complex systems.

[0065] The plurality of acoustic index selection schemes are obtained through permutation and combination based on the plurality of acoustic indexes. Different acoustic index selection schemes cover various possible index combinations, providing a basis for subsequent construction of diversified fault identification channels.

[0066] Secondly, a plurality of fault identification channels are constructed based on the plurality of acoustic index selection schemes, and each fault identification channel includes Q fault identification branches. Each fault identification branch uses different fault identification algorithms or models, such as support vector machines and neural networks, to perform fault identification analysis on the acoustic signal from different angles. The plurality of fault identification channels work in parallel, taking advantage of different acoustic index combinations and identification algorithms to improve the accuracy and reliability of fault identification.

[0067] Then, based on multiple acoustic index selection schemes and multiple fault identification channels, the scheme optimization target is to maximize the fault identification accuracy and approximate the adaptive identification time length, and the acoustic signal processing strategy and the fault identification strategy in the preset time zone are iteratively optimized and searched. In the iteration process, the parameters in the acoustic signal processing strategy, such as the parameters of the filter and the parameters of the wavelet transform, and the model parameters in the fault identification strategy are adjusted to optimize the fault identification effect. By comparing the fault identification accuracy and the deviation of the actual identification time length and the adaptive identification time length under different schemes, the optimal scheme is gradually selected. Finally, the optimal fault identification scheme is output, which includes the acoustic signal processing strategy and the fault identification strategy most suitable for the current situation, and can achieve the highest possible fault detection rate within the adaptive identification time length, meeting the actual needs of the power distribution network cable joint fault monitoring.

[0068] Among them, a plurality of fault identification channels are constructed based on the plurality of acoustic index selection schemes, comprising:

[0069] A first acoustic index selection scheme is randomly selected from the plurality of acoustic index selection schemes;

[0070] With the first acoustic index selection scheme as a constraint, a sample first acoustic index set is collected, and the historical fault event proportion of partial discharge fault corresponding to different sample first acoustic indexes is obtained as a sample first fault probability to obtain a sample first fault probability set;

[0071] The sample first acoustic index set and the sample first fault probability set are used as training data and are equally divided into Q parts, and Q times are selected with replacement to obtain a first training set, and Q times are iteratively selected to obtain Q training sets, wherein Q is an integer greater than or equal to 20 and less than or equal to 60;

[0072] The Q training sets are used to supervise the training of the deep learning model respectively until convergence, and Q first fault identification branches are obtained.

[0073] The Q first fault identification branches are integrated and fused according to the mean fusion strategy to obtain a first fault identification channel, which is added to the plurality of fault identification channels.

[0074] In the embodiments of the present application, first, a first acoustic index selection scheme is randomly selected from a plurality of acoustic index selection schemes, and the scheme includes a plurality of specific acoustic index combinations. The first acoustic index selection scheme is used as a constraint condition to collect a corresponding sample first acoustic index set. The sample first acoustic index can reflect the acoustic characteristics of the cable joint under different states. At the same time, the historical fault event proportion of the partial discharge fault corresponding to different sample first acoustic indexes is obtained as a sample first fault probability, and a sample first fault probability set is obtained. The sample first fault probability set reflects the possibility of the occurrence of the partial discharge fault under the specific acoustic index combination.

[0075] Then, the sample first acoustic index set and the sample first fault probability set are used as training data, and the collected data is equally divided into Q parts, Q is an integer greater than or equal to 20 and less than or equal to 60. Then, the first training set is obtained Q times with replacement, and the Q times are iteratively selected, and finally Q training sets are obtained.

[0076] After that, the Q training sets are used to supervise the training of the deep learning model. In the training process, the model parameters are constantly adjusted so that the model can better fit the training data. Until the model converges, that is, the performance index of the model on the training set no longer improves significantly, at this time Q first fault identification branches are obtained. Each first fault identification branch has fault identification capability and can analyze and judge the acoustic signal from a specific angle.

[0077] Finally, the Q first fault identification branches are integrated and fused according to the mean fusion strategy. The mean fusion strategy is to average the output results of each fault identification branch to obtain a comprehensive fault identification result. Through the integrated fusion mode, the advantages of each fault identification branch are utilized to improve the accuracy and reliability of fault identification. The first fault identification channel obtained after fusion is added to the plurality of fault identification channels to provide a reference for subsequent fault identification.

[0078] Further, based on the plurality of acoustic index selection schemes and the plurality of fault identification channels, the scheme optimization target is to maximize the fault identification accuracy and approximate the adaptive identification duration, and the acoustic signal processing strategy and the fault identification strategy in the preset time zone are iteratively optimized and searched to output an optimal fault identification scheme, including:

[0079] A first acoustic index selection scheme is randomly selected from the plurality of acoustic index selection schemes as a first acoustic signal processing strategy;

[0080] A first fault identification channel corresponding to the first acoustic index selection scheme is obtained, and any number of fault identification branches in the first fault identification channel are randomly selected as a first fault identification strategy;

[0081] The first acoustic signal processing strategy and the first fault identification strategy are combined as a first fault identification scheme, and scheme enumeration is performed based on the multiple acoustic index selection schemes and the multiple fault identification channels to generate a plurality of fault identification schemes.

[0082] The scheme optimization target is to maximize the fault identification accuracy and approximate the adaptive identification time length, and iterative optimization search is performed based on the plurality of fault identification schemes to output an optimal fault identification scheme.

[0083] In the embodiments of the present application, first, a first acoustic index selection scheme is randomly selected from the multiple acoustic index selection schemes as a first acoustic signal processing strategy. This strategy determines the acoustic index combination on which subsequent signal processing is based, laying the foundation for capturing the partial discharge characteristics of the cable joint.

[0084] Secondly, a first fault identification channel corresponding to the first acoustic index selection scheme is obtained, and any number of fault identification branches are randomly selected in the channel to form a first fault identification strategy. Different numbers of fault identification branch combinations can analyze fault features from multiple dimensions and enhance the comprehensiveness of fault identification.

[0085] Then, the first acoustic signal processing strategy and the first fault identification strategy are combined as a first fault identification scheme. On this basis, scheme enumeration is performed based on the multiple acoustic index selection schemes and the multiple fault identification channels, and a plurality of fault identification schemes are generated by comprehensively traversing various possible strategy combinations, covering acoustic signal processing and fault identification strategy combinations, providing options for selecting the optimal scheme.

[0086] Finally, the scheme optimization target is to maximize the fault identification accuracy and approximate the adaptive identification time length, and iterative optimization search is performed based on the plurality of fault identification schemes. In the iteration process, the fault identification accuracy of each scheme and the deviation of the actual identification time length from the adaptive identification time length are continuously evaluated. According to the evaluation results, the parameters in the acoustic signal processing strategy and the fault identification strategy, such as filter parameters and model parameters, are dynamically adjusted. After multiple iterations, the optimal fault identification scheme that meets the target is gradually selected, which realizes the highest possible fault detection rate within the adaptive identification time length and meets the actual needs of the power distribution network cable joint fault monitoring.

[0087] Specifically, the scheme optimization target is to maximize the fault identification accuracy and approximate the adaptive identification time length, and iterative optimization search is performed based on the plurality of fault identification schemes to output an optimal fault identification scheme, including:

[0088] According to the plurality of fault identification schemes, fault identification time length analysis and fault identification accuracy prediction are performed to obtain a plurality of predicted identification time lengths and a plurality of predicted identification accuracies;

[0089] A fitness evaluation function of the scheme is constructed by taking maximizing the fault identification accuracy and approximating the identification time length as the optimization target of the scheme.

[0090] A maximum fitness of the scheme is selected as the optimal fault identification scheme by using the fitness evaluation function of the scheme, according to the several predicted identification time lengths and the several predicted identification accuracies.

[0091] In the embodiments of the present application, firstly, the several generated fault identification schemes are respectively subjected to fault identification time length analysis and fault identification accuracy prediction. The purpose of the fault identification time length analysis is to estimate the time required for each scheme to complete fault identification in actual application. The fault identification accuracy prediction is to predict the fault identification accuracy that can be achieved by each scheme through historical data simulation or theoretical analysis, which depends on the selection of acoustic indicators, the performance of fault identification algorithm and other factors. Through the above steps, the several predicted identification time lengths and the several predicted identification accuracies are obtained.

[0092] Then, the fitness evaluation function of the scheme is constructed by taking maximizing the fault identification accuracy and approximating the identification time length as the optimization target of the scheme. The function is a multi-objective optimization function, taking the fault identification accuracy and the deviation of the identification time length from the adaptive identification time length as the key evaluation indexes, and then assigning the weights. For example, the weight of the fault identification accuracy is set to be high, so as to ensure that the finally selected scheme has a high fault detection rate; at the same time, the deviation of the identification time length from the adaptive identification time length is considered, so as to avoid that the scheme affects the monitoring efficiency in actual application due to too long time consumption. Through the above way, the fitness evaluation function which can comprehensively reflect the advantages and disadvantages of the scheme is constructed.

[0093] Finally, the fitness evaluation function of the scheme is constructed, and each scheme is evaluated according to the several predicted identification time lengths and the several predicted identification accuracies obtained before, so as to obtain the several fitnesses of the schemes. The fitness of the scheme is a comprehensive index, which reflects the comprehensive performance of the scheme in terms of fault identification accuracy and identification time length. In the evaluation process, the predicted identification time length and the predicted identification accuracy of each scheme are substituted into the fitness evaluation function, and the corresponding fitness of the scheme is calculated. Then, the fault identification scheme corresponding to the maximum fitness of the scheme is selected as the optimal fault identification scheme. The optimal scheme can achieve the highest possible fault detection rate within the adaptive identification time length, and meets the actual needs of the cable joint fault monitoring of the distribution network.

[0094] Specifically, assuming that there are three fault identification schemes A, B and C, firstly, the three schemes are respectively subjected to fault identification time length analysis and fault identification accuracy prediction. For scheme A, the identification time length is t1 and the identification accuracy is p1; for scheme B, the identification time length is t2 and the identification accuracy is p2; for scheme C, the identification time length is t3 and the identification accuracy is p3.

[0095] Then the fitness evaluation function of the scheme is constructed, assuming that the weight of fault identification accuracy is set to 0.7, the weight of the deviation of identification time length from the adaptive identification time length is set to 0.3, and the adaptive identification time length is set to T. The scheme fitness evaluation function F can be expressed as: F = 0.7 x predicted identification accuracy - 0.3 x (|predicted identification time length - T| / T). The time length deviation is normalized to make it comparable to the accuracy in the same order of magnitude.

[0096] The fitness F1 of scheme A is 0.7 x p1 - 0.3 x (|t1 - T| / T); the scheme fitness F2 of scheme B is 0.7 x p2 - 0.3 x (|t2 - T| / T); and the scheme fitness F3 of scheme C is 0.7 x p3 - 0.3 x (|t3 - T| / T).

[0097] Then the values of F1, F2 and F3 are calculated respectively. Assuming that F1 = 0.65, F2 = 0.72 and F3 = 0.68 are obtained after calculation. Comparing the sizes of F1, F2 and F3, because 0.72 > 0.68 > 0.65, i.e. F2 is the largest, scheme B is selected as the optimal fault identification scheme.

[0098] The fault identification time length analysis and the fault identification accuracy prediction are performed according to the plurality of fault identification schemes to obtain a plurality of predicted identification time lengths and a plurality of predicted identification accuracies, including:

[0099] A first fault identification scheme is randomly selected from the plurality of fault identification schemes, and a plurality of simulations of acoustic signal processing and fault identification based on the first fault identification scheme are performed. The mean of the simulation time lengths is taken as the first predicted identification time length, which is added to the plurality of predicted identification time lengths.

[0100] A scheme identification accuracy predictor is constructed based on deep learning, and the identification accuracy predictor is used to predict the identification accuracy according to the plurality of fault identification schemes, and output a plurality of predicted identification accuracies.

[0101] In the embodiments of the present application, first, a first fault identification scheme is randomly selected from the plurality of fault identification schemes, which contains a specific acoustic signal processing strategy and a fault identification strategy. A plurality of simulations of acoustic signal processing and fault identification based on the first fault identification scheme are performed. The simulation process needs to restore the actual monitoring environment as much as possible to obtain more accurate simulation time lengths. The mean of the simulation time lengths is taken as the first predicted identification time length, which reflects the approximate time required for the scheme to complete fault identification in actual application, and is added to the plurality of predicted identification time lengths.

[0102] Secondly, a scheme recognition accuracy predictor is constructed based on deep learning. The deep learning model can learn the relationship between fault recognition accuracy and various factors from a large amount of historical data and scheme information. Using the constructed recognition accuracy predictor, the recognition accuracy of each scheme is predicted. In the prediction process, the relevant parameters of each scheme, such as acoustic index selection and fault recognition algorithm type, are input into the predictor, and the corresponding predicted recognition accuracy is output after model calculation, obtaining several predicted recognition accuracies.

[0103] For example, a scheme recognition accuracy predictor is constructed and trained based on a deep learning model, and the specific steps are as follows:

[0104] First, data preparation. The input nodes of the monitoring urgency evaluation plug-in are fault recognition schemes. A first fault recognition scheme is randomly selected from the fault recognition schemes, and multiple simulations of acoustic signal processing and fault recognition are performed to obtain several predicted recognition durations.

[0105] Secondly, model construction. The input layer has a node number equal to the dimension of the input features, such as 2 features of fault recognition schemes and predicted recognition durations, so the input layer contains 2 nodes. The number of nodes in each hidden layer is adjusted through experiments, such as 64, 32, etc., and the activation function is ReLU. The output layer generally does not use an activation function, such as 2 nodes of output time consumption, which directly outputs continuous values.

[0106] Thirdly, model training. The predicted recognition accuracy is used as the output, and the urgency evaluation result is used as the input. The Adam optimizer and mean square error loss function are used to construct the training framework, the batch size is set to 32, the total training rounds are set to 50, and the early stopping mechanism (patience=5) is introduced. When the validation set loss does not decrease for 5 consecutive rounds, the training process is automatically terminated, and the trained scheme recognition accuracy predictor is obtained. This ensures that the model reaches a convergent state while effectively avoiding overfitting.

[0107] Through the above method, data support is provided for subsequent scheme fitness evaluation and optimal scheme selection, which helps to select the scheme with the best performance in fault recognition accuracy and recognition duration, and meets the actual needs of power distribution network cable joint fault monitoring.

[0108] Further, a scheme fitness evaluation function is constructed with the maximum fault recognition accuracy and the approximate adaptive recognition duration as the optimization target, including:

[0109] The recognition duration difference is obtained by subtracting the adaptive recognition duration from the preset recognition duration. If the recognition duration difference is less than or equal to 0, the scheme fitness evaluation function is constructed according to the initial recognition accuracy weight and the initial recognition efficiency weight, where the initial recognition accuracy weight and the initial recognition efficiency weight are both 0.5.

[0110] If the identified time length difference is greater than 0, an efficiency weight compensation coefficient is set according to the identified time length difference, the initial identification efficiency weight is compensated to obtain an adapted identification efficiency weight, wherein the efficiency weight compensation coefficient is greater than 1 and positively correlated with the identified time length difference, and the adapted identification efficiency weight is not greater than 0.75.

[0111] An adapted identification precision weight is obtained by subtracting the adapted identification efficiency weight from 1, and a scheme adaptability evaluation function is constructed according to the adapted identification precision weight and the adapted identification efficiency weight.

[0112] In the embodiment of the application, first, a difference between the preset identification time length and the adapted identification time length, i.e., an identification time length difference, is calculated. If the identification time length difference is less than or equal to 0, it means that the preset identification time length does not exceed the adapted identification time length. In this case, the initial identification precision weight and the initial identification efficiency weight are both set to 0.5 to construct a basic scheme adaptability evaluation function, so as to ensure that equal attention is paid to identification precision and efficiency under the premise that the identification time length meets the requirements.

[0113] If the identification time length difference is greater than 0, it means that the preset identification time length exceeds the adapted identification time length. In this case, an efficiency weight compensation coefficient is set according to the size of the identification time length difference. The coefficient is greater than 1 and positively correlated with the identification time length difference, i.e., the greater the identification time length difference, the greater the efficiency weight compensation coefficient. However, the upper limit of the adapted identification efficiency weight is set to 0.75 to prevent the efficiency weight from being too high and ignoring the identification precision. The initial identification efficiency weight is adjusted by the efficiency weight compensation coefficient to obtain the adapted identification efficiency weight. Meanwhile, the adapted identification precision weight is obtained by subtracting the adapted identification efficiency weight from 1.

[0114] Exemplarily, assuming that the adapted identification time length is T and the preset identification time length is t, when t-T>0, the identification time length difference is Δt=t-T. If Δt is in a certain interval, such as 0<Δt≤a, a is a preset time length difference threshold, the efficiency weight compensation coefficient is set to k1, k1>1, the adapted identification efficiency weight is 0.5×k1, and the adapted identification precision weight is 1-0.5×k1. If Δt is in another interval, such as a<Δt≤b (b>a), the efficiency weight compensation coefficient is set to k2, k2>k1, the adapted identification efficiency weight is 0.5×k2, and the adapted identification precision weight is 1-0.5×k2. The same applies to other intervals.

[0115] Finally, a scheme adaptability evaluation function that is more in line with actual requirements is constructed according to the adjusted adapted identification precision weight and the adapted identification efficiency weight. This function can dynamically reflect the comprehensive performance of a scheme in terms of identification precision and efficiency under different identification time lengths, and provide strong support for screening the optimal fault identification scheme.

[0116] S30: In the preset time zone, the acoustic sensor is continuously sampled, the original acoustic signal of the cable joint area is periodically acquired, the original acoustic signal is processed and fault is identified according to the optimal fault identification scheme, and a fault monitoring result is output.

[0117] In the embodiment of the application, in the preset time zone, the acoustic sensor pre-installed in the cable joint area is continuously sampled. The acoustic sensor has high sensitivity and wide frequency response characteristics, and can capture weak acoustic signals generated by the cable joint during operation. The sampling process is periodically performed according to the set time interval, ensuring comprehensive and continuous acquisition of the original acoustic signal of the cable joint area.

[0118] Secondly, after the original acoustic signal is acquired, the original acoustic signal is immediately processed and fault is identified according to the optimal fault identification scheme screened out before. The signal processing link includes filtering, denoising, feature extraction and other operations on the original acoustic signal, so as to eliminate interference signals and highlight the feature information related to the cable joint fault. The fault identification link uses the processed acoustic signal, combines the fault identification algorithm and threshold set in the optimal fault identification scheme, and judges whether the cable joint has a fault and the type and degree of the fault.

[0119] Thirdly, after signal processing and fault identification, the fault monitoring result, i.e. the fault probability of the partial discharge signal, is output. The fault monitoring result includes information such as whether a fault occurs, the fault position, the fault type and the fault severity. At the same time, the fault monitoring result is transmitted to the terminal equipment of the monitoring center in real time, so that corresponding maintenance and processing measures can be taken in time, and the safe and stable operation of the power distribution network cable joint is ensured.

[0120] In summary, compared with the prior art, the application realizes real-time monitoring and early warning of the cable joint fault, effectively improving the reliability and safety of the power distribution network operation.

[0121] In summary, the embodiment of the application has at least the following technical effects:

[0122] The embodiment of the application provides a power distribution network cable joint fault monitoring method and system based on voiceprint analysis. First, based on the predicted material state information and the predicted scene state information of the cable joint area, the emergency degree of partial discharge fault monitoring is dynamically evaluated, and an adaptive identification time length is set, thereby avoiding the problems of untimely detection or resource waste caused by a fixed detection period. Second, by maximizing the fault identification accuracy and approaching the adaptive identification time length as the target, the acoustic signal processing strategy and the fault identification strategy are iteratively optimized and searched, and the optimal fault identification scheme is automatically generated, thereby improving the accuracy and efficiency of fault identification. Finally, in the preset time zone, the acoustic sensor is used for continuous sampling, and the original acoustic signal is processed and identified according to the optimal fault identification scheme, thereby realizing real-time monitoring of the cable joint fault and providing protection for the safe and stable operation of the power distribution network. Through the above technical scheme, the application is based on the voiceprint analysis technology, and the material state and environmental scene information of the cable joint area are combined, thereby realizing accurate prediction and real-time monitoring of the partial discharge fault. The mechanism of dynamically evaluating the monitoring emergency degree and setting the adaptive identification time length makes the detection process more flexible and efficient, and the detection strategy can be adjusted in a timely manner according to the actual situation, thereby avoiding unnecessary resource waste.

[0123] Embodiment two, as shown in Figure 2 the same inventive concept as the power distribution network cable joint fault monitoring method based on voiceprint analysis provided in embodiment one, the embodiment of the application further provides a power distribution network cable joint fault monitoring system based on voiceprint analysis, comprising:

[0124] The identification time length configuration module 11 is used for evaluating the monitoring emergency degree of the partial discharge fault monitoring based on the predicted material state information and the predicted scene state information of the cable joint area in the preset time zone, and setting an adaptive identification time length according to the emergency degree evaluation result.

[0125] The identification scheme output module 12 is used for iteratively optimizing and searching the acoustic signal processing strategy and the fault identification strategy in the preset time zone, taking maximizing the fault identification accuracy and approaching the adaptive identification time length as the scheme optimization target, and outputting an optimal fault identification scheme.

[0126] The detection result output module 13 is used for continuously sampling through the acoustic sensor in the preset time zone, regularly acquiring the original acoustic signal of the cable joint area, and performing signal processing and fault identification on the original acoustic signal according to the optimal fault identification scheme, and outputting a fault monitoring result.

[0127] Further, in an application embodiment, the monitoring emergency degree evaluation of the partial discharge fault monitoring based on the predicted material state information and the predicted scene state information of the cable joint area in the preset time zone comprises:

[0128] Based on the inspection data of the cable joint area, a material aging index sequence in a historical time range is obtained, and a material aging index at the last inspection is extracted as a first material aging index;

[0129] A material aging trend coefficient is obtained according to the material aging index sequence analysis, and a predicted material aging index mean in a preset time zone is predicted according to the material aging trend coefficient and the first material aging index, as the predicted material state information;

[0130] A predicted current load sequence, a predicted temperature sequence and a predicted humidity sequence of the cable joint area in the preset time zone are obtained;

[0131] The predicted current load sequence, the predicted temperature sequence and the predicted humidity sequence are calculated and feature extracted respectively, to obtain a predicted current maximum load, a predicted current load mean, a predicted current load fluctuation coefficient, a predicted maximum temperature, a predicted temperature mean, a predicted temperature fluctuation coefficient, a predicted maximum humidity, a predicted humidity mean and a predicted humidity fluctuation coefficient, as the predicted scene state information;

[0132] A monitoring urgency evaluation plug-in of the cable joint area is constructed based on a deep learning model, a monitoring urgency evaluation of partial discharge fault monitoring is performed according to the predicted material state information and the predicted scene state information, and a predicted monitoring urgency is output as the urgency evaluation result.

[0133] Further, an adaptive identification duration is set according to the urgency evaluation result, including:

[0134] A ratio of a preset standard monitoring urgency to the predicted monitoring urgency is set as a time limit compensation coefficient;

[0135] After the preset identification duration is compensated according to the time limit compensation coefficient, the adaptive identification duration is obtained.

[0136] In one embodiment, the identification scheme output module 12 is specifically configured to:

[0137] Obtain a plurality of acoustic indicators processed from the acoustic signal, perform permutation and combination based on the plurality of acoustic indicators, and obtain a plurality of acoustic indicator selection schemes;

[0138] Based on the plurality of acoustic indicator selection schemes, a plurality of fault identification channels are constructed, wherein each fault identification channel includes Q fault identification branches;

[0139] Based on the plurality of acoustic indicator selection schemes and the plurality of fault identification channels, maximizing fault identification accuracy and approximating the adaptive identification duration are taken as scheme optimization targets, the acoustic signal processing strategy and the fault identification strategy in the preset time zone are iteratively optimized and searched, and an optimal fault identification scheme is output.

[0140] Further, in one application embodiment, based on the plurality of acoustic index selection schemes, a plurality of fault identification channels are constructed respectively, including:

[0141] A first acoustic index selection scheme is randomly selected from the plurality of acoustic index selection schemes;

[0142] With the first acoustic index selection scheme as a constraint, a sample first acoustic index set is collected, and the historical fault event proportion of partial discharge fault corresponding to different sample first acoustic indexes is obtained as a sample first fault probability, to obtain a sample first fault probability set;

[0143] The sample first acoustic index set and the sample first fault probability set are used as training data and equally divided into Q parts, and Q times are selected with replacement to obtain a first training set, and Q times are iteratively selected to obtain Q training sets, wherein Q is an integer greater than or equal to 20 and less than or equal to 60;

[0144] The Q training sets are used to supervise the training of a deep learning model respectively until convergence, to obtain Q first fault identification branches;

[0145] The Q first fault identification branches are integrated and fused according to a mean fusion strategy to obtain a first fault identification channel, which is added to the plurality of fault identification channels.

[0146] Further, in one embodiment, based on the plurality of acoustic index selection schemes and the plurality of fault identification channels, the acoustic signal processing strategy and the fault identification strategy in the preset time zone are iteratively optimized and searched to maximize the fault identification accuracy and approximate the adaptive identification duration, and an optimal fault identification scheme is output, including:

[0147] A first acoustic index selection scheme is randomly selected from the plurality of acoustic index selection schemes as a first acoustic signal processing strategy;

[0148] A first fault identification channel corresponding to the first acoustic index selection scheme is obtained, and any number of fault identification branches in the first fault identification channel are randomly selected as a first fault identification strategy;

[0149] The first acoustic signal processing strategy and the first fault identification strategy are taken as a first fault identification scheme, and scheme enumeration is performed based on the plurality of acoustic index selection schemes and the plurality of fault identification channels to generate a plurality of fault identification schemes;

[0150] With the maximum fault identification accuracy and the approximate adaptive identification duration as the scheme optimization target, the plurality of fault identification schemes are iteratively optimized and searched to output an optimal fault identification scheme.

[0151] Further, in one application embodiment, the fault identification duration analysis and the fault identification accuracy prediction are performed according to the plurality of fault identification schemes respectively, to obtain a plurality of predicted identification durations and a plurality of predicted identification accuracies, including:

[0152] A first fault identification scheme is randomly selected from the plurality of fault identification schemes, and a plurality of simulations of acoustic signal processing and fault identification based on the first fault identification scheme are performed, and the average of the simulation durations is taken as a first predicted identification duration, which is added to the plurality of predicted identification durations.

[0153] A scheme identification accuracy predictor is constructed based on deep learning, and the identification accuracy predictor is used to perform identification accuracy prediction according to the plurality of fault identification schemes respectively, and output a plurality of predicted identification accuracies.

[0154] Further, in one application embodiment, a scheme fitness evaluation function is constructed with the optimization target of maximizing the fault identification accuracy and approximating the adaptive identification duration, including:

[0155] The identification duration difference is obtained by subtracting the adaptive identification duration from the preset identification duration, and if the identification duration difference is less than or equal to 0, the scheme fitness evaluation function is constructed according to the initial identification accuracy weight and the initial identification efficiency weight, wherein the initial identification accuracy weight and the initial identification efficiency weight are both 0.5.

[0156] If the identification duration difference is greater than 0, the initial identification efficiency weight is compensated to obtain an adaptive identification efficiency weight according to the efficiency weight compensation coefficient set according to the identification duration difference, wherein the efficiency weight compensation coefficient is greater than 1 and positively correlated with the identification duration difference, and the adaptive identification efficiency weight is not greater than 0.75.

[0157] The adaptive identification accuracy weight is obtained by subtracting the adaptive identification efficiency weight from 1, and the scheme fitness evaluation function is constructed according to the adaptive identification accuracy weight and the adaptive identification efficiency weight.

[0158] It should be noted that the above sequence of the embodiments of the application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above describes a specific embodiment of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or may be advantageous.

[0159] The above only describes the preferred embodiments of the application, and does not limit the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

[0160] The specification and drawings are to be regarded in all respects as illustrative and

[0161] modifications, combinations or equivalents. It is therefore evident that various

[0162] modifications and changes can be made by those skilled in the art without departing

[0163] from the scope of the application. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense, the only purpose being to

Claims

1. A voiceprint analysis based power distribution network cable joint fault monitoring method, characterized in that, The method comprises: based on the predicted material state information and the predicted scene state information of the cable joint area in the preset time zone, monitoring emergency degree evaluation of partial discharge fault monitoring, and setting an adaptive identification time length according to the emergency degree evaluation result; maximizing fault identification accuracy and approximating the adaptive identification time length as the scheme optimization target, iteratively optimizing and searching the acoustic signal processing strategy and fault identification strategy in the preset time zone, and outputting an optimal fault identification scheme; in the preset time zone, continuously sampling through an acoustic sensor, regularly acquiring original acoustic signals of the cable joint area, and performing signal processing and fault identification on the original acoustic signals according to the optimal fault identification scheme, and outputting a fault monitoring result; wherein, based on the predicted material state information and the predicted scene state information of the cable joint area in the preset time zone, the monitoring emergency degree evaluation of the partial discharge fault monitoring comprises: based on the inspection data of the cable joint area, obtaining a material aging index sequence in a historical time range, and extracting the material aging index at the last inspection as a first material aging index; obtaining a material aging trend coefficient according to the material aging index sequence, and predicting a predicted material aging index mean in the preset time zone as the predicted material state information according to the material aging trend coefficient and the first material aging index; obtaining a predicted current load sequence, a predicted temperature sequence and a predicted humidity sequence of the cable joint area in the preset time zone; respectively calculating and extracting features of the predicted current load sequence, the predicted temperature sequence and the predicted humidity sequence, obtaining a predicted current maximum load, a predicted current load mean, a predicted current load fluctuation coefficient, a predicted maximum temperature, a predicted temperature mean, a predicted temperature fluctuation coefficient, a predicted maximum humidity, a predicted humidity mean and a predicted humidity fluctuation coefficient as the predicted scene state information; based on a deep learning model, constructing a monitoring emergency degree evaluation plug-in of the cable joint area, performing monitoring emergency degree evaluation of the partial discharge fault monitoring according to the predicted material state information and the predicted scene state information, and outputting a predicted monitoring emergency degree as the emergency degree evaluation result; wherein, maximizing fault identification accuracy and approximating the adaptive identification time length as the scheme optimization target, iteratively optimizing and searching the acoustic signal processing strategy and fault identification strategy in the preset time zone, and outputting an optimal fault identification scheme, comprising: obtaining a plurality of acoustic indexes after acoustic signal processing, arranging and combining based on the plurality of acoustic indexes to obtain a plurality of acoustic index selection schemes; based on the plurality of acoustic index selection schemes, constructing a plurality of fault identification channels, wherein each fault identification channel comprises Q fault identification branches; based on the plurality of acoustic index selection schemes and the plurality of fault identification channels, maximizing fault identification accuracy and approximating the adaptive identification time length as the scheme optimization target, iteratively optimizing and searching the acoustic signal processing strategy and fault identification strategy in the preset time zone, and outputting an optimal fault identification scheme.

2. A power distribution network cable joint fault monitoring method based on voiceprint analysis according to claim 1, characterized in that, setting an adaptive identification time length according to the emergency degree evaluation result, comprising: A ratio of the preset standard monitoring emergency degree and the predicted monitoring emergency degree is set as a time limit compensation coefficient; After compensating a preset identification time length according to the time limit compensation coefficient, an adaptive identification time length is obtained.

3. The power distribution network cable joint fault monitoring method based on voiceprint analysis according to claim 1, characterized in that, Based on the multiple acoustic index selection schemes, multiple fault identification channels are constructed, including: A first acoustic index selection scheme is randomly selected from the multiple acoustic index selection schemes; A first acoustic index set of a sample is collected and a first fault probability corresponding to a historical fault event proportion of partial discharge fault of different sample first acoustic indexes is obtained as a sample first fault probability, to obtain a sample first fault probability set, with the first acoustic index selection scheme as a constraint; The sample first acoustic index set and the sample first fault probability set are used as training data and are equally divided into Q parts, and Q times are selected with replacement to obtain a first training set, and Q times are iteratively selected to obtain Q training sets, wherein Q is an integer greater than or equal to 20 and less than or equal to 60; The Q training sets are used to supervise training of a deep learning model, respectively, until convergence, to obtain Q first fault identification branches; The Q first fault identification branches are integrated and fused according to a mean fusion strategy to obtain a first fault identification channel, which is added to the multiple fault identification channels.

4. The power distribution network cable joint fault monitoring method based on voiceprint analysis according to claim 1, characterized in that, Based on the multiple acoustic index selection schemes and the multiple fault identification channels, a maximum fault identification accuracy and an approximation of the adaptive identification time length are used as scheme optimization targets to iteratively optimize and search acoustic signal processing strategies and fault identification strategies in the preset time zone, and an optimal fault identification scheme is output, including: A first acoustic index selection scheme is randomly selected from the multiple acoustic index selection schemes as a first acoustic signal processing strategy; A first fault identification channel corresponding to the first acoustic index selection scheme is obtained, and any number of fault identification branches in the first fault identification channel are randomly selected as a first fault identification strategy; The first acoustic signal processing strategy and the first fault identification strategy are used as a first fault identification scheme, and scheme enumeration is performed based on the multiple acoustic index selection schemes and the multiple fault identification channels to generate a plurality of fault identification schemes; A maximum fault identification accuracy and an approximation of the adaptive identification time length are used as scheme optimization targets to iteratively optimize and search the plurality of fault identification schemes, and an optimal fault identification scheme is output.

5. A power distribution network cable joint fault monitoring method based on voiceprint analysis according to claim 4, characterized in that, A maximum fault identification accuracy and an approximation of the adaptive identification time length are used as scheme optimization targets to iteratively optimize and search the plurality of fault identification schemes, and an optimal fault identification scheme is output, including: Fault identification time length analysis and fault identification accuracy prediction are performed according to the plurality of fault identification schemes to obtain a plurality of predicted identification time lengths and a plurality of predicted identification accuracies; A scheme fitness evaluation function is constructed with a maximum fault identification accuracy and an approximation of the adaptive identification time length as scheme optimization targets; A plurality of scheme fitnesses are evaluated by using the scheme fitness evaluation function according to the plurality of predicted identification time lengths and the plurality of predicted identification accuracies, and a fault identification scheme corresponding to a maximum scheme fitness is selected as an optimal fault identification scheme.

6. The power distribution network cable joint fault monitoring method based on voiceprint analysis according to claim 5, characterized in that, The fault identification duration analysis and the fault identification precision prediction are performed according to the plurality of fault identification schemes, to obtain a plurality of predicted identification durations and a plurality of predicted identification precisions, including: A first fault identification scheme is randomly selected from the plurality of fault identification schemes, acoustic signal processing and multiple simulations of fault identification are performed based on the first fault identification scheme, an average of the multiple simulation durations is taken as a first predicted identification duration, and the first predicted identification duration is added to the plurality of predicted identification durations; An identification precision predictor is constructed based on a deep learning scheme, the identification precision predictor is used to perform identification precision prediction according to the plurality of fault identification schemes, and a plurality of predicted identification precisions are output.

7. The power distribution network cable joint fault monitoring method based on voiceprint analysis according to claim 5, characterized in that, A scheme fitness evaluation function is constructed with the scheme optimization target of maximizing the fault identification precision and approximating the adaptive identification duration, including: An identification duration difference is obtained by subtracting the adaptive identification duration from a preset identification duration, if the identification duration difference is less than or equal to 0, a scheme fitness evaluation function is constructed according to an initial identification precision weight and an initial identification efficiency weight, wherein the initial identification precision weight and the initial identification efficiency weight are both 0.5; If the identification duration difference is greater than 0, an efficiency weight compensation coefficient is set according to the identification duration difference, the initial identification efficiency weight is compensated to obtain an adaptive identification efficiency weight, wherein the efficiency weight compensation coefficient is greater than 1 and positively correlated with the identification duration difference, and the adaptive identification efficiency weight is not greater than 0.75; An adaptive identification precision weight is obtained by subtracting the adaptive identification efficiency weight from 1, and a scheme fitness evaluation function is constructed according to the adaptive identification precision weight and the adaptive identification efficiency weight.

8. A voiceprint analysis based power distribution network cable joint fault monitoring system characterized in that, A power distribution network cable joint fault monitoring method based on voiceprint analysis for performing any one of claims 1-7, comprising: An identification duration configuration module configured to perform a local discharge fault monitoring emergency evaluation based on predicted material state information and predicted scene state information of a cable joint area in a preset time zone, and set an adaptive identification duration according to the emergency evaluation result; An identification scheme output module configured to maximize the fault identification precision and approximate the adaptive identification duration as a scheme optimization target, iteratively optimize and search the acoustic signal processing strategy and the fault identification strategy in the preset time zone, and output an optimal fault identification scheme; A detection result output module configured to continuously sample through an acoustic sensor in the preset time zone, periodically acquire original acoustic signals of the cable joint area, perform signal processing and fault identification on the original acoustic signals according to the optimal fault identification scheme, and output a fault monitoring result.

Images