A method and device for diagnosing abnormal power line losses in a distribution network

Abstract

The embodiments of this specification disclose a method and device for diagnosing abnormal power line losses in a distribution network, which relate to the technical field of data processing. Among them, the method includes: obtaining monitoring status features according to the monitoring and statistical data of the target line and the corresponding status information; sorting at least one sub-status feature in the monitoring status features according to the data acquisition time; respectively inputting each sub-status feature in the monitoring status features into the first feature extraction network and the second feature extraction network of the first target neural network in ascending time order and descending time order, and performing forward calculation and reverse calculation respectively to obtain a first data feature and a second data feature; obtaining the state change feature of the target line based on the first data feature and the second data feature; inputting the state change feature into the second target neural network to extract deep features to obtain at least one deep state feature; based on the at least one deep state feature, diagnosing and predicting abnormal power line losses of the target line.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and in particular, to a method and device for diagnosing abnormal line losses in a distribution network. Background Art

[0002] Due to the diverse and complex relationships of the factors affecting line losses, and the limited technical means of existing abnormal line loss diagnosis, most rely on the experience of management personnel for manual investigation. Therefore, it is difficult to accurately and effectively diagnose the real cause of abnormal line losses during abnormal line losses, seriously affecting the work efficiency of distribution network operation and maintenance.

[0003] Based on this, it is necessary to study a method and device for diagnosing abnormal line losses in a distribution network to improve the accuracy and effectiveness of abnormal line loss diagnosis, and further improve the work efficiency of distribution network operation and maintenance. Summary of the Invention

[0004] To solve the above problems, one aspect of the embodiments of this specification provides a method for diagnosing abnormal line losses in a distribution network, the method includes:

[0005] According to the monitoring and statistical data of the target line and the status information corresponding to the monitoring and statistical data, obtain the monitoring status characteristics of the target line, where the status information corresponding to the monitoring and statistical data is used to indicate the abnormal line loss status of the target line, the monitoring and statistical data is time series data generated based on the real-time monitoring data of the target line, the time series data includes the data acquisition time when the target line acquires real-time monitoring data each time, and the real-time monitoring data corresponding to each data acquisition time, and the real-time monitoring data includes power supply data, power sales data, meteorological data, and load data and electrical measurement data of each load point in the target line;

[0006] Sort at least one sub-status characteristic in the monitoring status characteristics in the order of the data acquisition time.

[0007] Input each sub-status characteristic in the monitoring status characteristics into the first feature extraction network of the first target neural network in the positive order corresponding to the data acquisition time, and perform positive order calculation on the monitoring status characteristics through the first feature extraction network to obtain a first data feature, where the first data feature is used to characterize the positive evolution law of the monitoring status characteristics;

[0008] Input each sub-status characteristic in the monitoring status characteristics into the second feature extraction network of the first target neural network in the reverse order corresponding to the data acquisition time, and perform reverse order calculation on the monitoring status characteristics through the second feature extraction network to obtain a second data feature, where the second data feature is used to characterize the reverse evolution law of the monitoring status characteristics;

[0009] Based on the first data feature and the second data feature, obtain the state change feature of the target line;

[0010] Input the state change feature into the second target neural network, and layer by layer extract the depth feature corresponding to the target line through the second target neural network to obtain at least one depth state feature of the target line;

[0011] Based on the at least one depth state feature of the target line, perform line loss anomaly diagnosis and prediction on the target line.

[0012] In some embodiments, the obtaining the monitoring state feature of the target line according to the monitoring statistical data of the target line and the state information corresponding to the monitoring statistical data includes:

[0013] Input at least one state information into a multi-layer perceptron;

[0014] Through the multi-layer perceptron, screen out the target line loss abnormal state in at least one line loss abnormal state corresponding to the at least one state information, and perform weight assignment on the target line loss abnormal state to obtain the features corresponding to the at least one line loss abnormal state;

[0015] Input the real-time monitoring data into a time series analysis model, and extract the time series feature corresponding to each item of data in the real-time monitoring data through the time series analysis model to obtain a monitoring data feature set;

[0016] Input the monitoring data feature set into a cross-processing network, and perform feature cross-processing on each time series feature in the monitoring data feature set through the cross-processing network to obtain the feature corresponding to the real-time monitoring data, where the cross-processing network includes a deep cross neural network;

[0017] Input the features corresponding to the at least one line loss abnormal state and the features corresponding to the real-time monitoring data into a depth feature extraction network, and extract the composite feature of the real-time monitoring data and the at least one line loss abnormal state through the depth feature extraction network to obtain the monitoring state feature of the target line.

[0018] In some embodiments, the inputting each sub-state feature in the monitoring state feature into the first feature extraction network of the first target neural network in the time positive order corresponding to the data acquisition time, and performing positive order calculation on the monitoring state feature through the first feature extraction network to obtain the first data feature includes: performing a positive order moving average calculation on the monitoring state feature with a first window size to obtain the first data feature;

[0019] Inputting each sub-state feature in the monitoring state features into the second feature extraction network of the first target neural network in reverse chronological order corresponding to the data acquisition time, and performing reverse chronological calculation on the monitoring state features through the second feature extraction network to obtain second data features, including: performing reverse moving average calculation on the monitoring state features with a second window size to obtain the second data features, where the second window size is different from the first window size.

[0020] In some embodiments, obtaining the state change features of the target line based on the first data features and the second data features includes:

[0021] Fusing the first data features and the second data features to obtain third data features;

[0022] Assigning weights to the third data features to obtain fourth data features, where the fourth data features are used to characterize the comprehensive evolution law of the monitoring state features;

[0023] Based on the third data features and the fourth data features, obtaining the state change features of the target line.

[0024] In some embodiments, assigning weights to the third data features to obtain fourth data features includes:

[0025] Processing the third data features through a first attention model to obtain at least one first weight parameter, where each first weight parameter is used to represent the importance measure of a piece of real-time monitoring data and the corresponding line loss abnormal state;

[0026] Performing normalization processing on the first weight parameter to obtain at least one second weight parameter;

[0027] Assigning weights to the third data features according to the second weight parameter to obtain the fourth data features.

[0028] In some embodiments, performing line loss abnormal diagnosis and prediction on the target line based on the at least one depth state feature of the target line includes:

[0029] Performing weight parameter learning based on a second attention model and the at least one depth state feature to obtain at least one third weight parameter, where each third weight parameter is used to represent the importance measure of the corresponding depth state feature;

[0030] Assigning weights to the at least one depth state feature according to the third weight parameter to obtain target depth state features;

[0031] Based on the target depth state feature, perform line loss anomaly diagnosis and prediction on the target line.

[0032] In some embodiments, each layer of the second target neural network outputs one of the depth state features, and the method further includes:

[0033] Generate the depth state feature output by the first processing layer according to the network layer feature of the first processing layer in the second target neural network and the depth state feature generated by the second processing layer, where the network layer feature of the first processing layer is used to represent the state of the state change feature in the first processing layer, and the second processing layer is the previous processing layer in the second target neural network located before the first processing layer.

[0034] In some embodiments, the network layer feature of the first processing layer is determined based on the state change feature and the network layer feature of the second processing layer.

[0035] In some embodiments, the performing line loss anomaly diagnosis and prediction on the target line based on the target depth state feature includes: inputting the target depth state feature into a trained line loss anomaly recognition model to obtain a prediction result corresponding to the target depth state feature, where the prediction result includes a predicted line loss anomaly type and its corresponding predicted anomaly reason;

[0036] The line loss anomaly recognition model is trained based on the following method:

[0037] Obtain a sample depth state feature processed based on sample monitoring statistical data, and a sample label corresponding to the sample depth state feature, where the sample label is used to characterize the line loss anomaly type and its anomaly reason corresponding to the sample monitoring statistical data;

[0038] Use the sample depth state feature as an input and the sample label corresponding to the sample depth state feature as an output to train an initial line loss anomaly recognition model until a preset condition is met, and obtain the trained line loss anomaly recognition model.

[0039] Another aspect of the embodiments of this specification further provides a device for diagnosing line loss anomalies in a distribution network, and the device includes:

[0040] A first acquisition module, configured to obtain a monitoring status feature of the target line according to the monitoring statistical data of the target line and the status information corresponding to the monitoring statistical data, where the status information corresponding to the monitoring statistical data is used to indicate the line loss abnormal status of the target line, the monitoring statistical data is time series data generated based on the real-time monitoring data of the target line, the time series data includes the data acquisition time when the real-time monitoring data of the target line is acquired each time, and the real-time monitoring data corresponding to each data acquisition time, and the real-time monitoring data includes power supply amount data, power sales amount data, meteorological data, and load data and electrical measurement data of each load point in the target line;

[0041] A sorting module, configured to sort at least one sub-status feature in the monitoring status feature in the order of the data acquisition time;

[0042] A first data feature calculation module, configured to input each sub-status feature in the monitoring status feature into a first feature extraction network of a first target neural network in the positive order corresponding to the data acquisition time, and perform a positive order calculation on the monitoring status feature through the first feature extraction network to obtain a first data feature, where the first data feature is used to characterize the positive evolution law of the monitoring status feature;

[0043] A second data feature calculation module, configured to input each sub-status feature in the monitoring status feature into a second feature extraction network of the first target neural network in the reverse order corresponding to the data acquisition time, and perform a reverse order calculation on the monitoring status feature through the second feature extraction network to obtain a second data feature, where the second data feature is used to characterize the reverse evolution law of the monitoring status feature;

[0044] A second acquisition module, configured to obtain a status change feature of the target line based on the first data feature and the second data feature;

[0045] A deep status feature extraction module, configured to input the status change feature into a second target neural network, and layer by layer extract the deep features corresponding to the target line through the second target neural network to obtain at least one deep status feature of the target line;

[0046] A line loss abnormal identification module, configured to perform line loss abnormal diagnosis and prediction on the target line based on the at least one deep status feature of the target line.

[0047] The beneficial effects that the method and device for diagnosing abnormal line loss of a distribution network provided by the embodiments of this specification may bring at least include:

[0048] (1) By processing the monitoring statistical data and the corresponding status information of the monitoring statistical data to obtain the monitoring status characteristics, compared with the characteristics obtained only based on the monitoring statistical data, it can more accurately reflect the abnormal line loss status reflected by the target line during the detection process, and thus can improve the accuracy of the line loss abnormal diagnosis result in the subsequent process;

[0049] (2) By respectively performing forward calculation and reverse calculation on the sub-status characteristics at each time series in the monitoring status characteristics through the first target neural network, the first data characteristic and the second data characteristic are obtained, which can respectively reflect the data change characteristics of the monitoring status characteristics in different evolution directions and different time scales, and thus can improve the accuracy and reliability of the data characteristics to a certain extent;

[0050] (3) By extracting the deep status characteristic corresponding to the target line through the second target neural network and performing line loss abnormal diagnosis and prediction on the target line based on the deep status characteristic, potential line loss problems can be discovered in time through the monitoring statistical data, so as to realize line loss abnormal prediction while realizing line loss abnormal diagnosis, providing strong technical support for the stable operation of the power grid.

[0051] Additional features will be described in part in the following description. For those skilled in the art, it will become obvious by referring to the following content and the drawings, or can be understood by generating or operating examples. The features of this specification can be realized and obtained by practicing or using various aspects of the methods, tools and combinations described in the following detailed examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] This specification will be further described in the form of exemplary embodiments, and these exemplary embodiments will be described in detail through the drawings. These embodiments are not restrictive. In these embodiments, the same numbers represent the same structures, where:

[0053] Figure 1 is an exemplary module diagram of a distribution network line loss abnormal diagnosis device shown in some embodiments of this specification;

[0054] Figure 2 is an exemplary flowchart of a distribution network line loss abnormal diagnosis method shown in some embodiments of this specification;

[0055] Figure 3 is an exemplary sub-step flowchart of a distribution network line loss abnormal diagnosis method shown in some embodiments of this specification;

[0056] Figure 4 is an exemplary structural schematic diagram of a computer processing device provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] To more clearly illustrate the technical solutions of the embodiments of this specification, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some examples or embodiments of this specification. For those of ordinary skill in the art, without creative efforts, this specification can also be applied to other similar scenarios based on these drawings. Unless obvious from the language context or otherwise stated, the same reference numerals in the drawings represent the same structure or operation.

[0058] It should be understood that the "system", "device", "unit" and / or "module" used in this specification are a way to distinguish different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the said words can be replaced by other expressions.

[0059] As shown in this specification and the claims, unless the context clearly indicates an exceptional situation, words such as "a", "an", "one" and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0060] Flowcharts are used in this specification to illustrate the operations performed by the system according to the embodiments of this specification. It should be understood that the previous or subsequent operations do not necessarily need to be executed precisely in sequence. On the contrary, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps of operations can be removed from these processes.

[0061] A distribution network refers to a power grid that receives electric energy from a transmission network or a regional power plant and distributes it locally through distribution facilities or step by step according to voltage levels to various users. A distribution network usually consists of distribution equipment such as overhead lines, poles, cables, distribution transformers, switchgear, reactive power compensation capacitors and ancillary facilities, and its main function is to distribute electric energy.

[0062] Distribution network line loss refers to the energy loss caused by various reasons during the process of transmitting electric energy from a power plant to users. The main reasons for its generation include technical line loss, management line loss and other line losses. Among them, technical line loss is the energy loss caused by the physical characteristics of power equipment and lines, including resistance loss, transformer loss, capacitor loss, insulation loss, etc.; management line loss is the energy loss caused by poor management or human factors, including meter reading errors, measurement errors, electricity theft, data processing errors, etc.; other line losses refer to the energy loss caused by unforeseen or special reasons, such as fault loss, energy loss caused by environmental factors, etc.

[0063] For technical line losses, it can be understood as the energy loss caused by the heat dissipated during the operation of the distribution network. This belongs to the active power consumed by conductance and resistance in the power grid. Once the technical line losses in the distribution network are too high, the speed of line aging and damage will accelerate, thus having a greater impact on the safe operation of the power grid. Therefore, in actual work, it is necessary to fully attach importance to the problem of distribution network line losses, and take effective measures to solve them according to the causes of abnormal distribution network line losses, so as to comprehensively improve the level of line loss management.

[0064] However, due to the diverse and complex relationships of the factors affecting line losses, and the limited existing technical means for diagnosing abnormal line losses, most rely on the experience of management personnel for manual investigation. Therefore, it is difficult to accurately and effectively diagnose the real cause of abnormal line losses when line losses are abnormal, not only the operation and maintenance tasks are heavy, but also the efficiency is low.

[0065] Based on the above technical problems, this specification provides a method and device for diagnosing abnormal distribution network line losses based on artificial intelligence technology to improve the accuracy and effectiveness of diagnosing abnormal line losses. The following will describe in detail the method and device for diagnosing abnormal distribution network line losses provided in the embodiments of this specification with reference to the accompanying drawings.

[0066] Figure 1 It is an exemplary module schematic diagram of a device for diagnosing abnormal distribution network line losses provided in an embodiment of this application. Refer to Figure 1 As shown in the figure, the device 100 for diagnosing abnormal distribution network line losses may include a first acquisition module 110, a sorting module 120, a first data feature calculation module 130, a second data feature calculation module 140, a second acquisition module 150, a deep state feature extraction module 160, and a line loss abnormality identification module 170.

[0067] The first acquisition module 110 may be configured to obtain the monitoring state features of the target line according to the monitoring statistical data of the target line and the state information corresponding to the monitoring statistical data, where the state information corresponding to the monitoring statistical data is used to indicate the line loss abnormal state of the target line, the monitoring statistical data is time-series data generated based on the real-time monitoring data of the target line, the time-series data includes the data acquisition time when the target line acquires real-time monitoring data each time, and the real-time monitoring data corresponding to each data acquisition time, and the real-time monitoring data includes power supply data, power sales data, meteorological data, and load data and electrical measurement data of each load point in the target line.

[0068] The sorting module 120 may be configured to sort at least one sub-state feature in the monitoring state features in the order of the data acquisition time.

[0069] The first data feature calculation module 130 can be used to input each sub-state feature in the monitoring state feature into the first feature extraction network of the first target neural network in the time forward order corresponding to the data acquisition time, and perform forward calculation on the monitoring state feature through the first feature extraction network to obtain a first data feature, where the first data feature is used to characterize the forward evolution law of the monitoring state feature.

[0070] The second data feature calculation module 140 can be used to input each sub-state feature in the monitoring state feature into the second feature extraction network of the first target neural network in the time reverse order corresponding to the data acquisition time, and perform reverse calculation on the monitoring state feature through the second feature extraction network to obtain a second data feature, where the second data feature is used to characterize the reverse evolution law of the monitoring state feature.

[0071] The second acquisition module 150 can be used to obtain the state change feature of the target line based on the first data feature and the second data feature.

[0072] The deep state feature extraction module 160 can be used to input the state change feature into the second target neural network, and layer by layer extract the deep feature corresponding to the target line through the second target neural network to obtain at least one deep state feature of the target line.

[0073] The line loss anomaly identification module 170 can be used to perform line loss anomaly diagnosis and prediction on the target line based on the at least one deep state feature of the target line.

[0074] For more details about the above-mentioned various modules, reference can be made to other locations in this specification (such as Figures 2 - 3 its relevant descriptions), and no detailed discussion will be made here.

[0075] It should be understood that Figure 1The power distribution network line loss anomaly diagnosis device 100 and its modules shown can be implemented in various ways. For example, in some embodiments, the device and its modules can be implemented through hardware, software, or a combination of software and hardware. Among them, the hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art can understand that the above methods and systems can be implemented using computer-executable instructions and / or included in processor control code. For example, such code is provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device and its modules in this specification can be implemented not only by a hardware circuit of a programmable hardware device such as a very large scale integrated circuit or gate array, a semiconductor such as a logic chip or transistor, or a field programmable gate array or programmable logic device, but also by software executed by various types of processors, or by a combination of the above hardware circuit and software (for example, firmware).

[0076] It should be noted that the above description of the power distribution network line loss anomaly diagnosis device 100 is provided for illustrative purposes only and is not intended to limit the scope of this specification. It can be understood that for those skilled in the art, various modules can be arbitrarily combined or a subsystem can be formed and connected to other modules according to the description of this specification without departing from this principle. For example, Figure 1 the first acquisition module 110, sorting module 120, first data feature calculation module 130, second data feature calculation module 140, second acquisition module 150, depth state feature extraction module 160, and line loss anomaly recognition module 170 described in Figure 1 can be different modules in a system, or a module can implement the functions of two or more of the above modules. For another example, the power distribution network line loss anomaly diagnosis device 100 can further include a preprocessing module (

[0077] Figure 2 not shown in Figure 2 ), and this preprocessing module can be used to perform preprocessing such as filtering and denoising on the aforementioned monitoring and statistical data. Such variations are all within the protection scope of this specification.

[0078] Step S210: Obtain the monitoring status features of the target line according to the monitoring statistical data of the target line and the status information corresponding to the monitoring statistical data. In some embodiments, step S210 may be executed by the aforementioned first obtaining module 110.

[0079] In the embodiments of the present application, the target line may refer to any power transmission line and distribution line that needs to perform line loss anomaly diagnosis or status monitoring. The monitoring statistical data of the target line may include a plurality of historical real-time monitoring data obtained by monitoring the target line. Among them, each time of monitoring obtains monitoring statistical data corresponding to a data acquisition time. In the embodiments of the present application, the monitoring statistical data may be time series data generated based on a plurality of real-time monitoring data of the target line. The time series data includes the data acquisition time when the target line obtains real-time monitoring data each time, and the real-time monitoring data corresponding to each data acquisition time. Specifically, in the embodiments of the present application, the real-time monitoring data may include power supply data, power sales data, meteorological data, and load data and electrical measurement data of each load point in the target line. Among them, the power supply data and power sales data can be obtained through the metering system and billing system of the power company (these data can be collected through smart meters, Supervisory Control and Data Acquisition (SCADA) systems, or Energy Management Systems (EMS)); meteorological data can be measured by various sensors. Exemplary meteorological data may include parameters such as weather patterns, temperature, humidity, wind speed, and wind direction; the load data and electrical measurement data of each load point in the target line can be obtained through one or more of smart meters (smart meters can not only record power supply and power sales data, but also measure electrical parameters such as voltage, current, and power factor of the load point), distribution automation systems (distribution automation systems can monitor and control the operation status of the distribution network in real time through various sensors and controllers installed in the distribution network, and collect load data and electrical measurement data), and substation automation systems (substation automation systems can collect the operation status and electrical parameters of each device in the substation, including load data and electrical measurement data such as voltage, current, and power factor).

[0080] It can be understood that the power supply data and the power sold data are the key data for calculating the line loss. Meteorological factors such as temperature, humidity, and wind speed will affect the operating status of power equipment and power demand, thereby affecting the effectiveness and accuracy of line loss calculation. Similarly, the load data can reflect the power demand situation at each load point. By analyzing the load data, the change trend of power demand and the peak load can be understood, which helps to identify the time periods and regions with high line loss, and further evaluate the operating efficiency of power equipment and the line loss situation. The electrical measurement data can directly reflect the operating status of power equipment and power quality, helping to judge whether there are faults or abnormal operations in power equipment, so as to find out the potential causes of line loss. In the embodiment of the present application, by combining the above data to perform line loss abnormal diagnosis on the distribution network, the accuracy and reliability of the line loss abnormal diagnosis result can be improved to a certain extent, and the line loss abnormal type and its abnormal cause corresponding to the target line can be diagnosed more accurately.

[0081] In the embodiment of the present application, the status information corresponding to the foregoing monitoring and statistical data can be used to indicate whether there is a line loss abnormal status in the target line. In other words, in the embodiment of the present application, the status flag can be used to indicate whether there is a corresponding line loss abnormal status in the target line. Exemplary line loss abnormal statuses can include but are not limited to line aging (which can be judged according to the service life and lifespan of the line or equipment), abnormal insulation performance, poor contact, overload operation, equipment failure, voltage abnormality, current imbalance, exceeding the power generation rule, bad weather, etc. In the embodiment of the present application, the status flag can include a first flag and a second flag. Among them, the first flag can be used to indicate that the target line has a corresponding abnormal status, and the second flag can be used to indicate that the target line does not have a corresponding abnormal status. For example, in some embodiments, the status information can be expressed in the form of "01……001", where 1 represents the first flag and 0 represents the second flag. When any position in the status information (i.e., the digit position of the flag in the status information) has the first flag, it indicates that the target line has the abnormal status corresponding to this position. When any position in the status information has the second flag, it indicates that the target line does not have the abnormal status corresponding to this position. It should be noted that in the embodiment of the present application, other characters can also be used to represent the above first flag and second flag. Only as an example, in some embodiments, A can also be used to represent the foregoing first flag, and B can be used to represent the foregoing second flag. In some embodiments of the present application, the status information can be obtained by processing according to the corresponding judgment rules based on expert experience.

[0082] After obtaining the real-time monitoring data for the target line each time, the monitoring statistical data corresponding to the current data acquisition time can be obtained. Based on the monitoring statistical data obtained each time, the status information of the target line can be determined. Therefore, in the embodiments of the present application, each status information can correspond to a data acquisition time, and the monitoring statistical data of the target line includes at least one data acquisition time, so the monitoring statistical data of the target line corresponds to at least one status information. Only as an example, in the embodiments of the present application, the status information corresponding to each data acquisition time in the monitoring statistical data can be represented as a status code, which can be composed of 0 and 1. Each digit in the status code can correspond to an abnormal situation. When the data in a certain digit is 0, it means that the target line does not have the abnormal situation corresponding to this digit. When the data in any digit is 1, it means that the target line has the abnormal situation corresponding to this digit.

[0083] For the convenience of subsequent calculations, in some embodiments, the monitoring statistical data of the target line and the status information corresponding to the monitoring statistical data can be preprocessed so that the monitoring statistical data and the status information corresponding to the monitoring statistical data conform to the corresponding formats, and then the status features of the preprocessed data can be extracted to obtain the monitoring status features of the target line. Exemplarily, referring to Figure 3 , in some possible implementation manners, step S210 may include the following sub-steps:

[0084] Sub-step S2101: Input at least one status information into a multi-layer perceptron.

[0085] Sub-step S2102: Screen out the target line loss abnormal status among the at least one line loss abnormal status corresponding to the at least one status information through the multi-layer perceptron, and assign weights to the target line loss abnormal status to obtain the features corresponding to the at least one line loss abnormal status.

[0086] In the embodiments of the present application, since the status information of the target line includes multiple preset statuses (each preset status corresponds to a line loss abnormal situation), the target line has some or all of the multiple preset statuses. In some embodiments, at least one status information corresponding to at least one data acquisition time can be input into a multi-layer perceptron for feature screening, and the target line loss abnormal status can be determined based on the abnormal status (i.e., the preset status corresponding to the first identifier in the foregoing status information) of the target line at each data acquisition time.

[0087] In some embodiments, a multi-dimensional one-hot vector may be used to represent the status information corresponding to the monitoring statistical data, so as to preprocess the status information corresponding to the monitoring statistical data. The multi-dimensional one-hot vector may be composed of 0s and 1s, where 0 represents that the target line does not have the abnormal condition corresponding to the corresponding digit, and 1 represents that the target line has the abnormal condition corresponding to the digit.

[0088] In some embodiments, the multi-dimensional one-hot vector may be input into a multi-layer perceptron. The multi-layer perceptron network of the multi-layer perceptron may screen out the target line loss abnormal status of the target line through an encoding matrix, and perform weight assignment on the screened target line loss abnormal status to obtain the characteristics of the status information corresponding to the monitoring statistical data. It should be noted that in the embodiments of the present application, by performing weight assignment on the screened target line loss abnormal status, the characteristics of the status information corresponding to the multi-dimensional one-hot vector in the processed result can be concentrated.

[0089] Specifically, in the embodiments of the present application, each network node in the multi-layer perceptron network may calculate the data in the multi-dimensional one-hot vector through the following formula:

[0090]

[0091] where is the feature vector of the status information corresponding to the j-th data acquisition time, and this feature vector represents the feature of the status information corresponding to the j-th data acquisition time, where j is an integer greater than or equal to 1; represents a processing function; is a feature encoding matrix, which may be a matrix obtained by pre-training, or a matrix trained by the multi-layer perceptron network during the process of calculating the features of the status information; is the multi-dimensional one-hot vector of the status information corresponding to the j-th data acquisition time; represents a bias term.

[0092] Exemplarily, when j represents the total number of times of obtaining real-time monitoring data for the target line, after calculating all the status information corresponding to the monitoring statistical data through the above multi-layer perceptron network, the feature G = [G1, G2,..., G j of the status information corresponding to the monitoring statistical data can be obtained. It should be noted that in the embodiments of the present application, in addition to calculating the features of the foregoing status information through the multi-layer perceptron network, the features of the status information may also be calculated through other neural networks. For example, in some embodiments, a fully connected neural network may also be used.

[0093] It can be understood that since in the embodiments of the present application, a status information is used to indicate at least one status of the target line at a data acquisition time, then when the status indicated by the status information is more, the dimension of the multi-dimensional one-hot vector may be higher, which will increase the complexity of subsequent calculations. Based on this, in the embodiments of the present application, by screening out the target line loss abnormal status from the multiple status information corresponding to multiple data acquisition times, the feature dimension of the status information can be reduced compared to the dimension of the original multi-dimensional one-hot vector, so as to achieve data dimensionality reduction for subsequent calculations.

[0094] Sub-step S2103: Input the real-time monitoring data into the time series analysis model, and extract the time series features corresponding to each item of data in the real-time monitoring data through the time series analysis model to obtain a monitoring data feature set.

[0095] In the embodiments of the present application, the time series analysis model can be a recurrent neural network model based on deep learning, such as a long short-term memory network (LSTM) or a gated recurrent unit (GRU). These models can capture the time dependence in the data sequence, so as to effectively identify and predict the patterns and trends in the time series data. By training this model, it can be ensured that it has a high recognition accuracy for the time series features of specific types of monitoring data. In addition, the time series analysis model can also optimize its performance by continuously learning new monitoring data to adapt to the changes in the monitoring environment.

[0096] In some embodiments, each item of data in the foregoing real-time monitoring data can be processed separately to obtain the features corresponding to each item of data. Only as an example, in the embodiments of the present application, the time series features corresponding to each item of data may include but are not limited to features such as the data distribution, entropy, and scaling attributes of this type of data, and these features can represent the internal correlation of this type of data.

[0097] Exemplarily, when the monitoring statistical data obtained at the jth data acquisition time is input into the time series analysis model, the time series analysis model can extract the features of each type of data in the monitoring statistical data at the jth time based on a preset feature extraction rule. . Among them, represents the time series feature of the kth data type (i.e., the kth item of data) in the monitoring statistical data obtained at the jth data acquisition time, z is an integer greater than 1, and k is the number of the data type.

[0098] After the time series analysis model processes the monitoring statistical data corresponding to all data acquisition times, the time series analysis model can store the features of the monitoring statistical data of each data acquisition time extracted in a monitoring data feature set. , finally, the timing analysis model can output the monitoring data feature set.

[0099] Since the features in the aforementioned monitoring data feature set can only represent the internal correlations of this various types of data and cannot reflect the features of all monitoring statistical data, therefore, in the embodiments of the present application, it is also necessary to obtain the features of the monitoring statistical data by performing the following steps.

[0100] Sub-step S2104: Input the monitoring data feature set into the cross-processing network, and perform feature cross-processing on each timing feature in the monitoring data feature set through the cross-processing network to obtain the features corresponding to the real-time monitoring data, where the cross-processing network includes a deep cross neural network.

[0101] In the embodiments of the present application, the cross-processing network can be understood as a deep learning model, which extracts and learns features from the input data through the non-linear transformation of multiple layers of neurons. During the feature cross-processing, the network can automatically identify and combine the correlations between different features, thereby enhancing the model's ability to capture the internal laws of the data. In this way, the cross-processing network can generate richer and more abstract feature representations, providing more accurate data support for subsequent line loss anomaly diagnosis and prediction tasks.

[0102] In some embodiments, the cross-processing network can include a deep cross neural network, which is composed of a cross network and a deep network. By respectively inputting the aforementioned monitoring data feature set into the cross network and the deep network, multi-layer abstraction and feature extraction of the data can be realized. Among them, the cross network part can capture the complex relationships between the input features through multi-layer cross operations, while the deep network can learn the deep representations of the data through its deep structure. Specifically, through the cross network, multiple timing features in the monitoring data feature set can be crossed to output the crossed features, and through the deep network, the deep features of all features in the monitoring data feature set can be extracted. Finally, by combining the crossed features with the deep features extracted by the deep network, the features of the monitoring statistical data are obtained.

[0103] Sub-step S2105: Input the features corresponding to the at least one line loss anomaly state and the features corresponding to the real-time monitoring data into the deep feature extraction network, and extract the composite features of the real-time monitoring data and the at least one line loss anomaly state through the deep feature extraction network to obtain the monitoring state features of the target line.

[0104] In some embodiments, the features of the status information and the features of the monitoring statistical data may be first fused to obtain a fused feature, and then the fused feature is input into the deep feature extraction network, so as to extract the composite features of the real-time monitoring data and the at least one abnormal line loss status, and obtain the monitoring status features of the target line.

[0105] Exemplarily, in some embodiments, the features of the status information corresponding to the j-th data acquisition time and the features of the monitoring statistical data acquired at the j-th data acquisition time ( = ) can be fused into , then = concat , . Then, is input into the deep feature extraction network, and each node in the deep feature extraction network can process the input according to the following formula, so as to extract the composite features of the real-time monitoring data and the at least one abnormal line loss status, and obtain the monitoring status features of the target line.

[0106]

[0107] Wherein, represents the monitoring status features of the target line at the j-th data acquisition time; is the first weight parameter matrix, is the first bias term. It should be noted that, since each weight parameter in represents the importance measure corresponding to each element in , through the weight assignment for each element in can be realized, and then the elements in can be integrated. The ReLU function can better mine the relevant features between data. Therefore, in the embodiments of the present application, by using the ReLU function to process to obtain the monitoring status features , the composite features of the real-time monitoring data and the at least one abnormal line loss status can be better expressed.

[0108] Continuing to refer to Figure 2 , after step S210, the distribution network line loss abnormal diagnosis method 200 further includes:

[0109] Step S220: Sort at least one sub - status feature among the monitoring status features in the order of the data acquisition time. In some embodiments, step S220 may be executed by the aforementioned sorting module 120.

[0110] In the embodiments of the present application, by fusing the features of the status information corresponding to each data acquisition time and the features of the monitoring statistical data, and then through the aforementioned deep feature extraction network, the monitoring status features at each data acquisition time can be extracted. For ease of description, in this specification, the monitoring status features at each data acquisition time are referred to as sub - status features. Therefore, the monitoring status features finally output by the deep feature extraction network may include at least one sub - status feature.

[0111] In the embodiments of the present application, at least one sub - status feature among the monitoring status features can be sorted in the order of the data acquisition time, and the at least one sub - status feature can be stored in the aforementioned monitoring status features in this order to obtain the monitoring status feature W = [W1, W2,..., W j , where W j represents the j - th sub - status feature, that is, the monitoring status feature corresponding to the j - th data acquisition time.

[0112] In the embodiments of the present application, since the features of the monitoring statistical data and the features of the status information include the features of various types of data at each time series, the monitoring status features have multiple dimensions and multiple modalities, and the processing process of the monitoring status features can be regarded as a multi - modal feature extraction process.

[0113] Since in the embodiments of the present application, the aforementioned monitoring status features are processed based on the monitoring statistical data and the status information corresponding to the monitoring statistical data, compared with the features obtained only based on the monitoring statistical data, they can more accurately reflect the abnormal line loss state reflected by the target line during the detection process. And the monitoring statistical data is the actually monitored data, which can be used as an objective basis, making the obtained monitoring status features interpretable. The status information corresponding to the monitoring statistical data is the result obtained based on expert experience judgment. Therefore, obtaining the monitoring status features based on the status information and the monitoring statistical data, and performing subsequent calculations based on the monitoring status features to achieve the diagnosis of abnormal line loss in the distribution network can improve the accuracy of the diagnosis results to a certain extent.

[0114] Step S230: Input each sub - state feature in the monitoring status feature into the first feature extraction network of the first target neural network in the time positive order corresponding to the data acquisition time, and perform positive - order calculation on the monitoring status feature through the first feature extraction network to obtain the first data feature. In some embodiments, step S230 may be executed by the aforementioned first data feature calculation module 130.

[0115] Step S240: Input each sub - state feature in the monitoring status feature into the second feature extraction network of the first target neural network in the time reverse order corresponding to the data acquisition time, and perform reverse - order calculation on the monitoring status feature through the second feature extraction network to obtain the second data feature. In some embodiments, step S240 may be executed by the aforementioned second data feature calculation module 140.

[0116] In the embodiments of the present application, the first data feature may be used to characterize the forward evolution law of the monitoring status feature, and the second data feature is used to characterize the reverse evolution law of the monitoring status feature.

[0117] Specifically, in the embodiments of the present application, the monitoring status feature obtained by processing the foregoing process may be input into the first target neural network. Through the first target neural network, positive - order calculation and reverse - order calculation are respectively performed on the sub - state features at each time sequence in the monitoring status feature to obtain the first data feature and the second data feature. The first data feature is used to represent the forward evolution law of the monitoring status feature, and the second data feature is used to represent the reverse evolution law of the monitoring status feature. In the embodiments of the present application, the first target neural network may be a bidirectional RNN (Bidirectional Recurrent Neural Networks, BiRNN) with an attention mechanism. The bidirectional RNN may be composed of a first feature extraction network and a second feature extraction network. Among them, the first feature extraction network is used to obtain the aforementioned first data feature, and the second feature extraction network is used to obtain the aforementioned second data feature.

[0118] In the embodiments of the present application, the monitoring status feature may be input into the first feature extraction network of the first target neural network in the time positive order corresponding to the data acquisition time, and positive - order calculation is performed on the monitoring status feature through the first feature extraction network to obtain the first data feature. Similarly, the monitoring status feature may also be input into the second feature extraction network of the first target neural network in the time reverse order corresponding to the data acquisition time, and reverse - order calculation is performed on the monitoring status feature through the first feature extraction network to obtain the second data feature.

[0119] For example, in some embodiments of the present application, the monitoring status feature may be calculated by a forward moving average with a first window size to obtain the first data feature, and the monitoring status feature may be calculated by a reverse moving average with a second window size to obtain the second data feature, where the second window size is different from the first window size. For example, the first window size may be 5 data acquisition cycles, and the second window size may be 10 data acquisition cycles. In this way, two different data features can be obtained, which respectively reflect the change characteristics of the monitoring status feature on different time scales and different evolution directions, thereby improving the accuracy and reliability of the data feature. The specific implementation of the moving average calculation can be regarded as the prior art and will not be described in detail in this specification.

[0120] In some embodiments, other data processing techniques such as filtering and normalization processing may also be combined to further optimize the foregoing first data feature and second data feature. In addition, in some embodiments, by comparing the first data feature and the second data feature, the mutation points of the monitoring status feature can be identified, so as to timely discover potential problems or anomalies. It should be noted that in the embodiments of the present application, by using this dual-window moving average calculation method, not only can the long-term trend of the monitoring status be captured, but also the short-term fluctuations can be sensitively reflected, providing strong technical support for the real-time monitoring and abnormal diagnosis of the distribution network line loss.

[0121] It can be understood that since at least one sub-status feature in the monitoring status feature is sorted in the order of the data acquisition time in the embodiments of the present application, therefore, each sub-status feature in the monitoring status feature can be input into the first feature extraction network in the forward time sequence manner to obtain the first data feature. Similarly, each sub-status feature in the monitoring status feature can be input into the second feature extraction network in the reverse time sequence manner to obtain the second data feature.

[0122] Specifically, in some embodiments of the present application, each sub-status feature in the monitoring status feature W = [W1, W2,..., W j can be sequentially input into the nodes of the input layer of the first feature extraction network in the order from front to back. In other words, W1 is input into the first node of the input layer of the first feature extraction network, W2 is input into the second node of the input layer of the first feature extraction network, and so on... When the monitoring status feature is input into the first feature extraction network, the first feature extraction network can calculate each sub-status feature in the monitoring status feature based on the preset calculation rules in the first feature extraction network, and finally output the first data feature Y Z ={ ,……, , where z represents the forward order, and m represents the number of features included in the first data feature (in some embodiments, this number depends on the value of the first window size).

[0123] Similarly, in some embodiments, in the order from back to front, each sub-state feature in the monitoring status feature W = [W1, W2,..., W j can be sequentially input into the nodes of the input layer within the second feature extraction network. For example, W j is input into the first node of the input layer of the second feature extraction network, and W j-1 is input into the second node of the input layer of the second feature extraction network, and so on... After the monitoring status feature is input into the second feature extraction network, the second feature extraction network can calculate each sub-state feature within the monitoring status feature based on the preset calculation rules within the second feature extraction network, and finally output the second data feature Y F ={ , ……, }, where F represents the reverse order, and n represents the number of features included in the second data feature (in some embodiments, this number depends on the value of the second window size).

[0124] Step S250, based on the first data feature and the second data feature, obtain the state change feature of the target line. In some embodiments, step S250 can be executed by the aforementioned second acquisition module 150.

[0125] In the embodiments of the present application, in order to obtain a more accurate state change feature of the target line, based on the first data feature and the second data feature, obtain the state change feature of the target line. Specifically, it is possible to first obtain a data feature representing the comprehensive evolution law of the monitoring status feature based on the first data feature and the second data feature, and then obtain the state change feature of the target line according to this data feature.

[0126] Exemplarily, in some embodiments, the first data feature and the second data feature can be fused to obtain a third data feature Y = {Y1, Y2, ……, Y l}, where l represents the number of features included in the third data feature (in some embodiments, l = m + n).

[0127] Furthermore, weight assignment can be performed on the third data feature to obtain a fourth data feature. In the embodiments of the present application, this fourth data feature can be used to represent the comprehensive evolution law of the monitoring status feature.

[0128] Specifically, in the embodiments of the present application, the first attention model in the first target neural network can be used to assign weights to the third data feature. For example, in a possible implementation manner, weight parameter learning can be performed based on the first attention model and the third data feature to obtain at least one first weight parameter, and the first weight parameter is used to indicate the importance of the monitoring statistical data and the status information corresponding to the monitoring statistical data. In other words, each of the first weight parameters is used to represent the importance measure of a piece of real-time monitoring data and the corresponding line loss abnormal state. Among them, the first attention model is any attention model in the first target neural network. In the embodiments of the present application, weight parameter learning can be performed based on the weight parameter learning strategy in the first attention model. In some embodiments of the present application, the weight parameter learning strategy can be an attention weight parameter learning strategy based on position, and the learning strategy can be expressed as:

[0129]

[0130] Wherein, is the first weight parameter corresponding to the jth data acquisition time; B τ represents the second weight parameter vector; represents the jth feature element in the third data feature; b τ represents the second bias term.

[0131] It should be noted that in the embodiments of the present application, the weight parameter learning strategy in the first attention model can also be other attention weight parameter learning strategies, and the embodiments of the present application do not make specific limitations on the weight parameter learning strategy in the first attention model.

[0132] Furthermore, in the embodiments of the present application, the at least one first weight parameter can be normalized to obtain at least one second weight parameter.

[0133] Since it is considered that the first weight parameter calculated in the above process is a value obtained through mathematical calculation, and it may be too large or too small. Therefore, in the embodiments of the present application, for the convenience of calculation, the at least one first weight parameter can be normalized so that the sizes of the obtained second weight parameters are appropriate. In the embodiments of the present application, the normalization processing can include normalization. It can be understood that in the embodiments of the present application, since each second weight parameter is only the result of normalizing a first weight parameter, the second weight parameter and the first weight parameter have the same function, and both are used to indicate the importance measure of a monitoring statistical data and the status information corresponding to the monitoring statistical data. Based on this, in some embodiments, the third data feature can also be weighted based on the at least one second weight parameter to obtain the fourth data feature.

[0134] Specifically, the at least one second weight parameter , , ……, , ,] can be substituted into the formula , and then the output of the formula is used as the fourth data feature to achieve weight allocation for the third data feature. Wherein, c is the fourth data feature, is the second weight parameter corresponding to the j-th data acquisition time, is the j-th feature element in the third data feature, and l is the number of features included in the third data feature.

[0135] It can be understood that since a second weight parameter is used to represent the importance measure of a monitoring statistical data and the status information corresponding to the monitoring statistical data, by allocating weights to the third data feature through at least one second weight parameter, the third data feature can have a more accurate feature expression. Therefore, in the embodiments of the present application, by performing feature fusion on the first data feature and the second data feature to obtain a third data feature, and then performing weight allocation on the third data feature to obtain a fourth data feature, the evolution law of the foregoing monitoring status feature can be more accurately reflected.

[0136] Since weight allocation is performed on the first data feature and the second data feature, the first data feature and the second data feature can be comprehensively represented by the fourth data feature. Therefore, the fourth data feature can reflect both the positive evolution law represented by the first data feature and the reverse evolution law represented by the second data feature. In other words, the fourth data feature can represent the comprehensive evolution law of the foregoing monitoring status feature.

[0137] In some embodiments of the present application, the state change feature can be obtained based on the third data feature and the fourth data feature. Specifically, since the correlation between adjacent monitoring statistical data is the highest, and the correlation between adjacent status information is the highest, therefore, in order to further predict the status feature at the next detection of the target line, the data feature corresponding to the last data acquisition time in the third data feature and the fourth data feature can be substituted into the formula = , and the output is used as the state change feature. Wherein, is the data feature corresponding to the last data acquisition time of the target line in the third data feature, is the vector after fusion of d and c, B is the third weight parameter matrix,

[0138] Since the fourth data feature represents the comprehensive evolution law of the monitoring state feature, the first data feature can represent the positive evolution law of the monitoring state feature, and the second data feature can represent the reverse evolution law of the monitoring state feature. Therefore, by assigning weights to these three data features, the obtained result can more accurately represent the state change feature of the target line.

[0139] Step S260: Input the state change feature into the second target neural network, and layer by layer extract the depth feature corresponding to the target line through the second target neural network to obtain at least one depth state feature of the target line. In some embodiments, step S260 may be executed by the aforementioned depth state feature extraction module 160.

[0140] In the embodiments of the present application, the second target neural network may be a hierarchical and multi-label classification network, such as HMCN (Hierarchical Multi-label Classification Networks). Since the state change feature obtained in the foregoing process cannot reflect the depth feature of the target line, in some embodiments of the present application, the depth feature corresponding to the target line may be extracted through the second target neural network.

[0141] Specifically, in some embodiments of the present application, the second target neural network may layer by layer extract the depth feature of the target line, so that the finally extracted detailed feature can meet the requirements for state information prediction. Each processing layer of the second target neural network may output a depth state feature. When the state change feature is input into the second target neural network, it can be input from the input layer all the way to the output layer of the second target neural network. The second target neural network may perform layer by layer calculation on the state change feature. When calculating, the first processing layer of the second target neural network may, based on the output data of the second processing layer, calculate the network layer feature of the first processing layer and calculate the depth state feature of the target line at the first processing layer. Herein, the first processing layer may refer to any layer of the second target neural network, and the second processing layer is the previous processing layer located before the first processing layer in the second target neural network. The network layer feature is used to represent the state of the state change feature in the network layer of the second target neural network, and the network layer feature of the first processing layer is determined by the state change feature and the network layer feature of the second processing layer.

[0142] After the second processing layer of the second target neural network generates the network layer features of the second processing layer and the depth state features of the target line in the second processing layer, the second processing layer can output the network layer features of the second processing layer and the state change features (i.e., the output data of the second processing layer) to the first processing layer, so that the first processing layer can receive the network layer features of the second processing layer and the state change features. In this way, with the state change features output by each network layer of the second target neural network, the state change features can be input to each network layer of the second target neural network.

[0143] Since the network layer features of the first processing layer are determined by the state change features and the network layer features of the second processing layer, when the first processing layer receives the network layer features of the second processing layer and the state change features, the first processing layer can calculate the network layer features of the first processing layer based on the network layer features of the second processing layer and the state change features. Specifically, in some embodiments, the network layer features of the i-th layer in the second target neural network can be expressed as: , where G represents global, is the fourth weight parameter matrix, is the network layer features of the (i - 1)-th processing layer (the i-th layer can be regarded as the first processing layer), represents the state change features obtained by processing the foregoing process, represents the fourth bias term.

[0144] In the embodiments of the present application, the nodes on the first processing layer can obtain the depth state features of the target line based on the network layer features of the first processing layer and the state change features. Only as an example, in some embodiments, the depth state features of the target line in the i-th layer can be expressed as: . Where L represents the network layer, is the fifth weight parameter matrix, is the fifth bias term.

[0145] It can be understood that in the embodiments of the present application, since each layer of the second target neural network is calculated based on the network layer features and state change features of the previous layer, the depth state features of the target line on each layer of the second target neural network are affected by the depth state features of the previous layer. And since the hierarchical expression of each layer is determined by the network layer features of this layer, the depth state features generated by any network layer in the second target neural network can be used as the parent of the depth state features generated by the next network layer. Based on this, the second target neural network can extract the depth features of the target line layer by layer.

[0146] Step S270: Based on the at least one depth state feature of the target line, perform line loss anomaly diagnosis and prediction on the target line. In some embodiments, step S270 may be executed by the aforementioned line loss anomaly recognition module 170.

[0147] As can be seen from the above, in the embodiments of the present application, the depth state feature can reflect the monitoring and statistical data of the target line at different levels of detail. Since it is considered that there may be many details involved in the foregoing process, in some embodiments of the present application, these different levels of detail can be centrally processed to obtain a more detailed depth state feature, and then based on this more detailed depth state feature, perform line loss anomaly diagnosis and prediction.

[0148] In some embodiments, weight assignment may be performed on at least one depth state feature of the target line to obtain a target depth state feature. The target depth state feature is also a more detailed depth state feature. Specifically, in some embodiments, the depth state feature of the target line may be weighted by a second attention model in the second target neural network. Specifically, weight parameter learning may be performed based on the second attention model and the at least one depth state feature to obtain at least one third weight parameter, where the third weight parameter is used to represent the importance measure of the corresponding depth state feature.

[0149] In the embodiments of the present application, the second attention model may be any attention model in the second target neural network. In some embodiments, weight parameter learning may be performed based on the second attention model and the at least one depth state feature. Specifically, weight parameters may be learned based on the weight parameter learning strategy in the second attention model, and the weight parameter learning strategy in the second attention model may be expressed as:

[0150]

[0151]

[0152] Among them, represents the i-th third weight parameter, represents the exponential function with base e, and represent the reference weight parameters corresponding to the i-th and j-th data acquisition times respectively, represents the sixth weight parameter matrix, is the sixth bias term, represents the hyperbolic tangent function.

[0153] It should be noted that other learning strategies can also be adopted for the weight parameter learning strategy in the second attention model, and the embodiments of the present application do not specifically limit the weight parameter learning strategy in the second attention model. It should also be noted that in the embodiments of the present application, the weight parameter matrix and the bias term involved in the foregoing content can both be obtained through training, and during the training process, the gradient descent method or other optimization algorithms can be used to continuously adjust these parameters to minimize the model error.

[0154] Further, based on the at least one third weight parameter, weight distribution can be performed on at least one depth state feature of the target line to obtain the target depth state feature. Specifically, in the embodiments of the present application, at least one third weight parameter and at least one depth state feature of the target line can be substituted into the formula: , and then its output is used as the target depth state feature to achieve weight distribution for at least one depth state feature of the target line. Wherein, is the depth state feature corresponding to the monitoring statistical data corresponding to the current time sequence, is the number of layers of the second target neural network.

[0155] It can be understood that since in the embodiments of the present application, one third weight parameter represents the importance measure of a depth state feature, and weight distribution is performed on at least one depth state feature through the at least one third weight parameter to obtain the target depth state feature, the obtained target depth state feature can be made more detailed.

[0156] After obtaining the target depth state feature through the above steps, line loss anomaly diagnosis and prediction can be performed on the target line based on the target depth state feature. Specifically, in some embodiments, the target depth state feature can be input into a trained line loss anomaly recognition model to obtain a prediction result corresponding to the target depth state feature, where the prediction result includes the predicted line loss anomaly type and its corresponding predicted anomaly cause.

[0157] In the embodiments of the present application, the line loss anomaly recognition model can be trained based on the following method:

[0158] First, obtain sample depth state features processed based on sample monitoring statistical data, and sample labels corresponding to the sample depth state features, where the sample labels are used to characterize the line loss anomaly type and its anomaly cause corresponding to the sample monitoring statistical data, and the sample monitoring statistical data can be obtained through experiments. For example, in some embodiments, line loss anomaly situations can be artificially created or selected, and then the corresponding monitoring statistical data is collected as the sample monitoring statistical data. It should be noted that in the embodiments of the present application, the sample monitoring statistical data is time sequence data collected within a certain period of time.

[0159] Then, the above-mentioned method can be used to process the sample monitoring statistical data to obtain the sample depth state features corresponding to the sample monitoring statistical data. The specific implementation details of this process can refer to the above text and will not be elaborated here. At the same time, according to the true line loss anomaly situation and its anomaly cause corresponding to each sample monitoring statistical data, a corresponding sample label can be determined for each sample depth state feature, and this sample label can reflect the line loss anomaly type and its anomaly cause corresponding to each sample monitoring statistical data.

[0160] Finally, the sample depth state features can be used as the input, and the sample labels corresponding to the sample depth state features can be used as the output to train the initial line loss anomaly recognition model until the preset conditions are met, and the trained line loss anomaly recognition model is obtained. In the embodiments of the present application, the initial line loss anomaly recognition model can include one or more neural network layers, such as a convolutional neural network (CNN) layer, a recurrent neural network (RNN) layer, or a fully connected layer. Among them, the convolutional neural network layer can be used to extract features, the recurrent neural network layer can be used to process time series data, and the fully connected layer is used to integrate the relationships between different features. Through such a multi-layer network structure, the line loss anomaly recognition model can learn the complex patterns and correlations in the sample depth state features corresponding to the sample monitoring statistical data.

[0161] During the training process, the backpropagation algorithm can be used to adjust the weights of the model to minimize the difference between the prediction result and the sample label. In the embodiments of the present application, this difference can be measured by a loss function (such as mean squared error (MSE)). Through continuous iterative optimization, the model can learn the complex patterns and relationships in the sample data, thereby improving the recognition accuracy of line loss anomalies. In addition, in some embodiments of the present application, to prevent the model from overfitting, regularization techniques such as L1 or L2 regularization can be adopted, or the dropout method can be used to randomly discard some neurons to enhance the generalization ability of the model.

[0162] When the foregoing loss function reaches the preset threshold, the training can be terminated to obtain the trained line loss anomaly recognition model. After the training is completed, the model can be tested, and the performance of the model can be evaluated by comparing the prediction result on the test set with the actual label. If the line loss anomaly recognition model obtained by training through the above steps performs well on the test set, the trained line loss anomaly recognition model can be used for actual line loss anomaly detection tasks. More details about model training can be regarded as prior art and will not be elaborated in this specification.

[0163] It can be understood that in the embodiments of the present application, after obtaining the above-trained line loss anomaly recognition model, the trained line loss anomaly recognition model can be used to diagnose the real-time monitoring data of any target line, so as to identify or predict the corresponding line loss anomaly type and give the relevant anomaly reasons, thereby realizing the real-time monitoring and management of the power grid operation state. In this way, the power company can timely discover and handle potential line loss problems, reduce economic losses, and improve the operation efficiency and reliability of the power grid. In addition, in some embodiments of the present application, the line loss anomaly recognition model can also be continuously optimized and adjusted according to the comparative analysis of historical data and real-time data to adapt to the changes in the power grid operation environment and ensure the accuracy and timeliness of the diagnosis results.

[0164] Figure 4 is an exemplary structural schematic diagram of a computer processing device provided by an embodiment of the present application. Refer to Figure 4 , wherein the computer processing device 400 may include a memory and a processor. The memory and the processor are directly or indirectly electrically connected to achieve data transmission or interaction. For example, they can be electrically connected through one or more communication buses or signal lines. The memory may store at least one software function module (computer program) that can exist in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, thereby implementing the distribution network line loss anomaly diagnosis method provided by the embodiments of the present application.

[0165] Optionally, in the embodiments of the present application, the memory may be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc. The processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0166] It should be noted that Figure 4 the structure shown is only illustrative, and the computer processing device may also include more or fewer components than those shown in Figure 4 it, or have a configuration different from that shown in Figure 4 it. For example, it may include a communication unit for information interaction with other devices. Among them, in an alternative example, the computer processing device may be a server with data processing capabilities. In some embodiments, the aforementioned distribution network line loss anomaly diagnosis device 100 may be a part of the computer processing device 400.

[0167] In summary, the beneficial effects that the embodiments of this specification may bring include but are not limited to: (1) In the distribution network line loss anomaly diagnosis method and device provided in some embodiments of this specification, by processing the monitoring statistical data and the corresponding status information of the monitoring statistical data to obtain the monitoring status features, compared with the features obtained only based on the monitoring statistical data, it can more accurately reflect the line loss anomaly status reflected by the target line during the detection process, and thus can improve the accuracy of the line loss anomaly diagnosis result in the subsequent process; (2) In the distribution network line loss anomaly diagnosis method and device provided in some embodiments of this specification, by respectively performing forward calculation and reverse calculation on the sub-status features at each time sequence in the monitoring status features through the first target neural network, the first data feature and the second data feature are obtained, which can respectively reflect the data change features of the monitoring status features in different evolution directions and different time scales, and thus can improve the accuracy and reliability of the data features to a certain extent; (3) In the distribution network line loss anomaly diagnosis method and device provided in some embodiments of this specification, by extracting the depth status features corresponding to the target line through the second target neural network and performing line loss anomaly diagnosis and prediction on the target line based on the depth status features, potential line loss problems can be detected in a timely manner through the monitoring statistical data, so as to realize line loss anomaly prediction while realizing line loss anomaly diagnosis, providing strong technical support for the stable operation of the power grid.

[0168] It should be noted that the beneficial effects that different embodiments may produce are different. In different embodiments, the beneficial effects that may be produced may be a combination of any one or more of the above, or any other beneficial effects that may be obtained.

[0169] The basic concepts have been described above. Obviously, for those skilled in the art, the above detailed disclosure is only an example and does not constitute a limitation to this specification. Although not explicitly stated here, those skilled in the art may make various modifications, improvements and corrections to this specification. Such modifications, improvements and corrections are proposed in this specification, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of this specification.

[0170] In the meantime, this specification uses specific terms to describe the embodiments of this specification. For example, "an embodiment", "one embodiment", and / or "some embodiments" mean a certain feature, structure, or characteristic related to at least one embodiment of this specification. Therefore, it should be emphasized and noted that the "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more at different positions in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this specification can be combined appropriately.

[0171] In addition, those skilled in the art can understand that various aspects of this specification can be described and illustrated by several patentable types or situations, including any new and useful process, machine, product, or composition of matter, or any new and useful improvement thereof. Accordingly, various aspects of this specification can be executed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above-mentioned hardware or software can all be referred to as "data block", "module", "engine", "unit", "component", or "system". In addition, various aspects of this specification may be embodied as a computer product located in one or more computer-readable media, and this product includes computer-readable program code.

[0172] A computer storage medium may contain a propagated data signal that contains computer program code, such as on a baseband or as part of a carrier wave. This propagated signal may have various forms of representation, including electromagnetic form, optical form, etc., or a suitable combination of forms. A computer storage medium can be any computer-readable medium other than a computer-readable storage medium, and this medium can be connected to an instruction execution system, apparatus, or device to implement communication, propagation, or transmission for use of a program. The program code located on the computer storage medium can be propagated through any suitable medium, including radio, cable, fiber optic cable, RF, or similar media, or any combination of the above media.

[0173] The computer program codes required for the operations of various parts of this specification can be written in any one or more programming languages, including object-oriented programming languages such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages such as C language, Visual Basic, Fortran2003, Perl, COBOL2002, PHP, ABAP, dynamic programming languages such as Python, Ruby, and Groovy, or other programming languages. The program codes can run entirely on the user's computer, or run as an independent software package on the user's computer, or partially run on the user's computer and partially on a remote computer, or run entirely on a remote computer or processing device. In the latter case, the remote computer can be connected to the user's computer through any network form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., through the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0174] In addition, unless clearly stated in the claims, the order of the processing elements and sequences, the use of numerical and alphabetical characters, or the use of other names in this specification are not used to limit the order of the processes and methods of this specification. Although some currently considered useful embodiments of the invention are discussed through various examples in the above disclosure, it should be understood that such details only serve the purpose of illustration, and the appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that conform to the essence and scope of the embodiments of this specification. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only through software solutions, such as installing the described system on existing processing devices or mobile devices.

[0175] Similarly, it should be noted that, in order to simplify the expression of the disclosure of this specification and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of this specification, sometimes multiple features are merged into one embodiment, drawing, or description thereof. However, this disclosure method does not mean that the features required by the object of this specification are more than those mentioned in the claims. In fact, the features of the embodiments are fewer than all the features of the individual embodiments disclosed above.

[0176] In some embodiments, numbers are used to describe components and the quantity of attributes. It should be understood that such numbers used in the description of embodiments are, in some examples, modified by the modifiers "about", "approximately" or "substantially". Unless otherwise specified, "about", "approximately" or "substantially" indicate that the said numbers allow a variation of ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may vary according to the characteristics required by individual embodiments. In some embodiments, the numerical parameters should consider the specified significant digits and adopt the method of retaining the general number of digits. Although the numerical ranges and parameters used in some embodiments of this specification to confirm the breadth of their scope are approximate values, in specific embodiments, such numerical settings are as precise as possible within the feasible range.

[0177] For each patent, patent application, patent application publication and other materials cited in this specification, such as articles, books, specifications, publications, documents, etc., their entire contents are hereby incorporated into this specification by reference. Except for the application history documents that are inconsistent with or conflict with the content of this specification, and also except for the documents that limit the broadest scope of the claims of this specification (currently or subsequently attached to this specification). It should be noted that if there are inconsistencies or conflicts between the descriptions, definitions, and / or uses of terms in the supplementary materials of this specification and the content described in this specification, the descriptions, definitions, and / or uses of terms in this specification shall prevail.

[0178] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be considered to be consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A method for diagnosing abnormal power losses in a distribution network, characterized in that, Including: Obtain the monitoring status characteristics of the target line according to the monitoring statistical data of the target line and the status information corresponding to the monitoring statistical data, where the status information corresponding to the monitoring statistical data is used to indicate the abnormal line loss status of the target line, the monitoring statistical data is time-series data generated based on the real-time monitoring data of the target line, the time-series data includes the data acquisition time when the target line acquires real-time monitoring data each time, and the real-time monitoring data corresponding to each data acquisition time, and the real-time monitoring data includes power supply data, power sales data, meteorological data, and load data and electrical measurement data of each load point in the target line; Sort at least one sub-status characteristic in the monitoring status characteristics in the order of the data acquisition time; Input each sub-status characteristic in the monitoring status characteristics into the first feature extraction network of the first target neural network in the forward order corresponding to the data acquisition time, and perform forward calculation on the monitoring status characteristics through the first feature extraction network to obtain the first data feature, where the first data feature is used to characterize the forward evolution law of the monitoring status characteristics; Input each sub-status characteristic in the monitoring status characteristics into the second feature extraction network of the first target neural network in the reverse order corresponding to the data acquisition time, and perform reverse calculation on the monitoring status characteristics through the second feature extraction network to obtain the second data feature, where the second data feature is used to characterize the reverse evolution law of the monitoring status characteristics; Based on the first data feature and the second data feature, obtain the status change characteristics of the target line; Input the status change characteristics into the second target neural network, and layer by layer extract the depth characteristics corresponding to the target line through the second target neural network to obtain at least one depth status characteristic of the target line; Based on the at least one depth status characteristic of the target line, perform line loss abnormal diagnosis and prediction on the target line; Wherein, each layer of the second target neural network outputs one depth status characteristic, and the inputting the status change characteristics into the second target neural network and layer by layer extracting the depth characteristics corresponding to the target line through the second target neural network to obtain at least one depth status characteristic of the target line includes: generating the depth status characteristic output by the first processing layer according to the network layer characteristic of the first processing layer in the second target neural network and the depth status characteristic generated by the second processing layer, where the network layer characteristic of the first processing layer is used to represent the status of the status change characteristics in the first processing layer, and the second processing layer is the previous processing layer in the second target neural network located before the first processing layer.

2. The method according to claim 1, wherein The obtaining the monitoring status characteristics of the target line according to the monitoring statistical data of the target line and the status information corresponding to the monitoring statistical data includes: Input at least one status information into a multi-layer perceptron; Filter out the target line loss abnormal state among the at least one line loss abnormal state corresponding to the at least one state information through the multi-layer perceptron, and perform weight assignment on the target line loss abnormal state to obtain the features corresponding to the at least one line loss abnormal state; Input the real-time monitoring data into the time series analysis model, and extract the time series features corresponding to each item of data in the real-time monitoring data through the time series analysis model to obtain a monitoring data feature set; Input the monitoring data feature set into the cross-processing network, and perform feature cross-processing on each time series feature in the monitoring data feature set through the cross-processing network to obtain the features corresponding to the real-time monitoring data, where the cross-processing network includes a deep cross neural network; Input the features corresponding to the at least one line loss abnormal state and the features corresponding to the real-time monitoring data into the deep feature extraction network, and extract the composite features of the real-time monitoring data and the at least one line loss abnormal state through the deep feature extraction network to obtain the monitoring state features of the target line.

3. The method according to claim 1, characterized in that, Input each sub-state feature in the monitoring state features into the first feature extraction network of the first target neural network in the positive order corresponding to the data acquisition time, and perform positive order calculation on the monitoring state features through the first feature extraction network to obtain the first data feature, including: performing a positive order moving average calculation on the monitoring state features with a first window size to obtain the first data feature; Input each sub-state feature in the monitoring state features into the second feature extraction network of the first target neural network in the reverse order corresponding to the data acquisition time, and perform reverse order calculation on the monitoring state features through the second feature extraction network to obtain the second data feature, including: performing a reverse order moving average calculation on the monitoring state features with a second window size to obtain the second data feature, where the second window size is different from the first window size.

4. The method according to claim 1, wherein Obtain the state change features of the target line based on the first data feature and the second data feature, including: Fuse the first data feature and the second data feature to obtain a third data feature; Perform weight assignment on the third data feature to obtain a fourth data feature, where the fourth data feature is used to characterize the comprehensive evolution law of the monitoring state features; Obtain the state change features of the target line based on the third data feature and the fourth data feature.

5. The method according to claim 4, characterized in that The performing weight assignment on the third data feature to obtain a fourth data feature includes: Process the third data feature through a first attention model to obtain at least one first weight parameter, where each first weight parameter is used to represent the importance measure of a piece of real-time monitoring data and the corresponding line loss abnormal state; Perform normalization processing on the first weight parameter to obtain at least one second weight parameter; Perform weight assignment on the third data feature according to the second weight parameter to obtain the fourth data feature.

6. The method according to claim 1, characterized in that, Performing line loss anomaly diagnosis and prediction on the target line based on the at least one depth state feature of the target line includes: Performing weight parameter learning based on a second attention model and the at least one depth state feature to obtain at least one third weight parameter, where each third weight parameter is used to represent the importance measure of the corresponding depth state feature; Performing weight assignment on the at least one depth state feature according to the third weight parameter to obtain a target depth state feature; Performing line loss anomaly diagnosis and prediction on the target line based on the target depth state feature.

7. The method according to claim 1, characterized in that, The network layer feature of the first processing layer is determined based on the state change feature and the network layer feature of the second processing layer.

8. The method according to claim 6, wherein Performing line loss anomaly diagnosis and prediction on the target line based on the target depth state feature includes: inputting the target depth state feature into a trained line loss anomaly recognition model to obtain a prediction result corresponding to the target depth state feature, where the prediction result includes a predicted line loss anomaly type and its corresponding predicted anomaly cause; The line loss anomaly recognition model is trained based on the following method: Obtaining sample depth state features processed based on sample monitoring statistical data and sample labels corresponding to the sample depth state features, where the sample labels are used to characterize the line loss anomaly type and its anomaly cause corresponding to the sample monitoring statistical data; Using the sample depth state features as inputs and the sample labels corresponding to the sample depth state features as outputs to train an initial line loss anomaly recognition model until a preset condition is met, obtaining the trained line loss anomaly recognition model.

9. A device for diagnosing abnormal power losses in a distribution network, characterized in that, Including: A first acquisition module, configured to obtain the monitoring state feature of the target line according to the monitoring statistical data of the target line and the state information corresponding to the monitoring statistical data, where the state information corresponding to the monitoring statistical data is used to indicate the line loss anomaly state of the target line, the monitoring statistical data is time-series data generated based on the real-time monitoring data of the target line, the time-series data includes the data acquisition time when the target line obtains real-time monitoring data each time, and the real-time monitoring data corresponding to each data acquisition time, and the real-time monitoring data includes power supply amount data, power sales amount data, meteorological data, and load data and electrical measurement data of each load point in the target line; A sorting module, configured to sort at least one sub-state feature in the monitoring state feature in the order of the data acquisition time; A first data feature calculation module, configured to input each sub-state feature in the monitoring state feature into a first feature extraction network of a first target neural network in the forward order corresponding to the data acquisition time, and perform forward calculation on the monitoring state feature through the first feature extraction network to obtain a first data feature, where the first data feature is used to characterize the forward evolution law of the monitoring state feature; The second data feature calculation module is configured to input each sub-state feature in the monitoring state feature into the second feature extraction network of the first target neural network in reverse time order corresponding to the data acquisition time, and perform reverse calculation on the monitoring state feature through the second feature extraction network to obtain a second data feature, where the second data feature is used to characterize the reverse evolution law of the monitoring state feature; The second acquisition module is configured to acquire the state change feature of the target line based on the first data feature and the second data feature; The deep state feature extraction module is configured to input the state change feature into a second target neural network, and layer by layer extract the deep feature corresponding to the target line through the second target neural network to obtain at least one deep state feature of the target line; The line loss anomaly identification module is configured to perform line loss anomaly diagnosis and prediction on the target line based on the at least one deep state feature of the target line; Wherein, each layer of the second target neural network outputs one of the deep state features, and the deep state feature extraction module is specifically configured to: generate the deep state feature output by the first processing layer according to the network layer feature of the first processing layer in the second target neural network and the deep state feature generated by the second processing layer, where the network layer feature of the first processing layer is used to represent the state of the state change feature in the first processing layer, and the second processing layer is the previous processing layer in the second target neural network located before the first processing layer.

Images