Distribution network equipment operation state evaluation method and system based on data analysis

Abstract

The application provides a data analysis-based distribution network equipment operation state evaluation method and system, and relates to the technical field of data analysis.In the application, firstly, a plurality of distribution network equipment state data are acquired;secondly, for each distribution network equipment state data, a time domain state feature of the distribution network equipment state data is extracted, and the distribution network equipment state data is converted into equipment state frequency domain data;then, based on the semantic information possessed by the time domain state feature, semantic mining is performed on the semantic information possessed by the equipment state frequency domain data to form a first state semantic vector; further, the first state semantic vector corresponding to each distribution network equipment state data is fused to form a second state semantic vector; finally, state evaluation is performed based on the second state semantic vector to obtain an operation state evaluation result.Based on the above, the problem that the reliability of the operation state evaluation of the distribution network equipment is relatively low in the prior art can be improved.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, and more specifically, to a method and system for evaluating the operating status of distribution network equipment based on data analysis. Background Technology

[0002] Distribution network equipment is a core component of the power system, including key equipment such as transmission lines, substations, and distribution transformers. The normal operation of this equipment is fundamental to ensuring the safe, stable, and economical operation of the power supply. However, the operating status of distribution network equipment is complex and diverse, easily affected by multiple factors such as the external environment, internal operating conditions, and interdependencies between equipment. Therefore, timely and accurate assessment of the operating status of distribution network equipment is a crucial guarantee for the safe operation of the power system.

[0003] However, existing technologies typically rely on single operating parameters, such as voltage, current, and frequency. These parameters only reflect a localized aspect of equipment operation and cannot comprehensively reflect the overall operating status of the distribution network system. Furthermore, existing technologies generally compare these operating parameters with corresponding thresholds, failing to fully utilize the potential semantic information inherent in these parameters. Therefore, existing technologies suffer from relatively low reliability in assessing the operating status of distribution network equipment. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a data analysis-based method and system for evaluating the operating status of distribution network equipment, so as to improve the problem of relatively low reliability of the existing technology for evaluating the operating status of distribution network equipment.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] A data analysis-based method for assessing the operational status of distribution network equipment includes:

[0007] The status data of the target distribution network device in multiple dimensions are obtained to obtain multiple distribution network device status data. The multiple dimensions include at least two of vibration, temperature, sound, voltage and current. Each distribution network device status data includes status data at multiple time points.

[0008] For each of the distribution network device status data, the time-domain status features of the distribution network device status data are extracted, and the distribution network device status data is converted into device status frequency-domain data;

[0009] based on semantic information possessed by the time domain state feature, performing semantic mining on semantic information possessed by the device state frequency domain data to form a first state semantic vector, wherein the first state semantic vector is used to reflect semantic information of the target network configuration device in the time domain and the frequency domain in the dimension;

[0010] fusing the first state semantic vector corresponding to each of the network configuration device state data to form a second state semantic vector, wherein the second state semantic vector is used to reflect global semantic information possessed by the target network configuration device in the plurality of dimensions;

[0011] based on the second state semantic vector, performing state evaluation to obtain a running state evaluation result, wherein the running state evaluation result is used to represent whether the target network configuration device has an abnormality.

[0012] In a preferred selection of the present application, in the network configuration device running state evaluation method based on data analysis, the step of extracting, for each of the network configuration device state data, a time domain state feature of the network configuration device state data and converting the network configuration device state data into device state frequency domain data comprises:

[0013] for each of the network configuration device state data, determining a maximum value, a minimum value, a mean value and a standard deviation of state data at a plurality of time points included in the network configuration device state data, and based on the mean value and the standard deviation, determining a standardized value of a fourth central moment around the mean value of the network configuration device state data to obtain a kurtosis value corresponding to the network configuration device state data for measuring the steepness of the tail, and determining the maximum value, the minimum value, the mean value, the standard deviation and the kurtosis value as the time domain state feature of the network configuration device state data;

[0014] performing Fourier transform on the network configuration device state data to obtain device state frequency domain data.

[0015] In a preferred selection of the present application, in the network configuration device running state evaluation method based on data analysis, the step of forming, based on semantic information possessed by the time domain state feature, a first state semantic vector by performing semantic mining on semantic information possessed by the device state frequency domain data comprises:

[0016] performing vector space mapping on the time domain state feature to form a time domain state mapping vector;

[0017] performing convolution processing on the device state frequency domain data to form a frequency domain state convolution vector;

[0018] perform first deep semantic mining on the frequency domain state convolution vector based on the local mapping vector corresponding to the steepness value in the time domain state mapping vector corresponding to the power distribution device state data, to form a first deep semantic vector;

[0019] perform second deep semantic mining on the frequency domain state convolution vector based on other local mapping vectors except the local mapping vector corresponding to the steepness value in the time domain state mapping vector, to form a second deep semantic vector;

[0020] perform convolution, pooling and activation processing on the spliced vector of the first deep semantic vector and the second deep semantic vector, to form a first state semantic vector.

[0021] In a preferred selection of the present application, in the power distribution device operation state evaluation method based on data analysis, the step of performing first deep semantic mining on the frequency domain state convolution vector based on the local mapping vector corresponding to the steepness value in the time domain state mapping vector corresponding to the power distribution device state data, to form a first deep semantic vector, includes:

[0022] extract the local mapping vector corresponding to the steepness value for measuring the steepness of the tail part from the time domain state mapping vector corresponding to the power distribution device state data, to obtain a first local mapping vector;

[0023] perform multiple local hidden processing on the frequency domain state convolution vector to form multiple frequency domain state hidden vectors, wherein the hidden regions of each two local hidden processing do not at least completely coincide;

[0024] perform attention mining on each of the frequency domain state hidden vectors based on the first local mapping vector, to form a first state attention vector;

[0025] fuse each of the first state attention vectors to form a first deep semantic vector.

[0026] In a preferred selection of the present application, in the power distribution device operation state evaluation method based on data analysis, the step of performing second deep semantic mining on the frequency domain state convolution vector based on other local mapping vectors except the local mapping vector corresponding to the steepness value in the time domain state mapping vector, to form a second deep semantic vector, includes:

[0027] extract other local mapping vectors except the local mapping vector corresponding to the steepness value for measuring the steepness of the tail part from the time domain state mapping vector corresponding to the power distribution device state data, to obtain a second local mapping vector;

[0028] The frequency domain state convolution vector is subjected to multiple local hidden processing to form multiple frequency domain state hidden vectors, wherein hidden regions between every two local hidden processing are at least not completely overlapped;

[0029] Based on the second local mapping vector, each of the frequency domain state hidden vectors is subjected to attention mining to form each second state attention vector;

[0030] Each of the second state attention vectors is fused to form a second deep semantic vector.

[0031] In a preferred selection of the present application, in the above-mentioned data analysis-based distribution network equipment operating state evaluation method, the step of fusing each of the first state semantic vectors corresponding to the distribution network equipment state data to form a second state semantic vector comprises:

[0032] The two first state semantic vectors corresponding to the vibration and sound dimensions of the distribution network equipment state data are subjected to multiple hierarchical focused mining to form a first state focused vector;

[0033] The two first state semantic vectors corresponding to the voltage and current dimensions of the distribution network equipment state data are subjected to multiple hierarchical focused mining to form a second state focused vector;

[0034] Based on the first state semantic vector corresponding to the temperature dimension of the distribution network equipment state data, the first state focused vector and the second state focused vector are subjected to attention mining to form a first state mining vector and a second state mining vector, respectively;

[0035] The first state mining vector and the second state mining vector are subjected to focused mining to form a second state semantic vector.

[0036] In a preferred selection of the present application, in the above-mentioned data analysis-based distribution network equipment operating state evaluation method, the step of fusing each of the first state semantic vectors corresponding to the distribution network equipment state data to form a second state semantic vector comprises:

[0037] In the first hierarchical focused mining, the first state semantic vector corresponding to the vibration dimension of the distribution network equipment state data is mapped to a focused weight distribution of the first state semantic vector corresponding to the sound dimension of the distribution network equipment state data, and the first state semantic vector and the focused weight distribution are subjected to bitwise multiplication operation to form a first hierarchical focused mining vector;

[0038] In the second level of focused mining, the first level of focused mining vector is mapped to the focused weight distribution of the first state semantic vector corresponding to the vibration dimension of the distribution network equipment state data, and the first state semantic vector and the focused weight distribution are multiplied by bit to form the second level of focused mining vector;

[0039] Based on the second level of focused mining vector, a first state focused vector is formed.

[0040] In the preferred selection of the present application, in the above-mentioned distribution network equipment running state evaluation method based on data analysis, the step of performing multiple levels of focused mining on the two first state semantic vectors corresponding to the voltage and current dimensions of the distribution network equipment state data to form a second state focused vector comprises:

[0041] In the first level of focused mining, the first state semantic vector corresponding to the voltage dimension of the distribution network equipment state data is mapped to the focused weight distribution of the first state semantic vector corresponding to the current dimension of the distribution network equipment state data, and the first state semantic vector and the focused weight distribution are multiplied by bit to form the first level of focused mining vector;

[0042] In the second level of focused mining, the first level of focused mining vector is mapped to the focused weight distribution of the first state semantic vector corresponding to the voltage dimension of the distribution network equipment state data, and the first state semantic vector and the focused weight distribution are multiplied by bit to form the second level of focused mining vector;

[0043] Based on the second level of focused mining vector, a second state focused vector is formed.

[0044] In the preferred selection of the present application, in the above-mentioned distribution network equipment running state evaluation method based on data analysis, the step of performing multiple levels of focused mining on the two first state semantic vectors corresponding to the voltage and current dimensions of the distribution network equipment state data to form a second state focused vector comprises:

[0045] The second state semantic vector is fully connected to form a fully connected vector, wherein the size of the fully connected vector is 1*2;

[0046] The fully connected vector is linearly mapped to form a target probability distribution, wherein the size of the target probability distribution is 1*2, and the two probability values included therein are respectively used to represent the probability of existence of an abnormality and the probability of non-existence of an abnormality of the target distribution network equipment;

[0047] Based on the size relationship of the two probability values in the target probability distribution, a running state evaluation result is determined.

[0048] On the basis, the application further provides a data analysis-based power distribution equipment operation state evaluation method and system, comprising a memory and a processor connected with the memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to realize the data analysis-based power distribution equipment operation state evaluation method.

[0049] The data analysis-based power distribution equipment operation state evaluation method and system provided by the application firstly acquires a plurality of power distribution equipment state data; secondly, for each power distribution equipment state data, extracts the time domain state feature of the power distribution equipment state data, and converts the power distribution equipment state data into equipment state frequency domain data; then, based on the semantic information possessed by the time domain state feature, performs semantic mining on the semantic information possessed by the equipment state frequency domain data to form a first state semantic vector; further, fuses the first state semantic vector corresponding to each power distribution equipment state data to form a second state semantic vector; and finally, performs state evaluation based on the second state semantic vector to obtain an operation state evaluation result. Based on the above, on the one hand, since semantic mining is performed, the potential semantic information in the power distribution equipment state data can be utilized, and the basis for state evaluation is more sufficient, so as to guarantee the reliability of state evaluation. On the other hand, since the semantic information of the time domain and the frequency domain is associated and mined within the dimension, the semantic representation accuracy of the mined semantic vector can be higher, and the semantic vectors of at least two dimensions are fused, so that the basis for state evaluation can be more sufficient and reliable. Therefore, the technical solution of the application can improve the problem of relatively low reliability of power distribution equipment operation state evaluation in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the following preferred embodiments are specifically described below with reference to the accompanying drawings.

[0051] Figure 1 The structure block diagram of the data analysis-based power distribution equipment operation state evaluation system provided by the embodiments of the application.

[0052] Figure 2 The first schematic diagram of the data analysis-based power distribution equipment operation state evaluation method provided by the embodiments of the application.

[0053] Figure 3 The second schematic diagram of the data analysis-based power distribution equipment operation state evaluation method provided by the embodiments of the application.

[0054] Figure 4 The schematic diagram of the first deep semantic mining provided by the embodiments of the application.

[0055] Figure 5 A schematic diagram of the multi-level focused mining provided by the embodiments of the present application.

[0056] Figure 6 A schematic diagram of the data analysis-based power distribution equipment operation state evaluation device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0057] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0058] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0059] As Figure 1 shown, the embodiments of the present application provide a data analysis-based power distribution equipment operation state evaluation system. The data analysis-based power distribution equipment operation state evaluation system can include a memory, a processor and a data analysis-based power distribution equipment operation state evaluation device.

[0060] In detail, the memory and the processor are directly or indirectly electrically connected to realize data transmission or interaction. For example, the memory and the processor can be electrically connected through one or more communication buses or signal lines. The data analysis-based power distribution equipment operation state evaluation device includes at least one software function module stored in the memory in the form of software or firmware. The processor is used to execute the executable computer programs stored in the memory, for example, the software function modules and computer programs included in the data analysis-based power distribution equipment operation state evaluation device, to realize the data analysis-based power distribution equipment operation state evaluation method provided by the embodiments of the present application.

[0061] Optionally, the memory can 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 electric erasable programmable read-only memory (EEPROM), etc. Moreover, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; and can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0062] It can be understood that, Figure 1 The structure shown is only schematic, and the data analysis-based power distribution equipment operation state evaluation system can further include more or fewer components than those shown in Figure 1 , or have a different configuration from Figure 1 , for example, it can also include a communication unit for information interaction with other devices (such as various sensors and the like).

[0063] In combination with Figure 2 , the embodiments of the present application also provide a data analysis-based power distribution equipment operation state evaluation method applicable to the above-mentioned data analysis-based power distribution equipment operation state evaluation system. The method steps defined by the flow of the data analysis-based power distribution equipment operation state evaluation method can be implemented by the data analysis-based power distribution equipment operation state evaluation system (hereinafter referred to as the evaluation system). The specific flow shown in Figure 2 will be described in detail below.

[0064] In step S110, the state data of the target power distribution equipment in multiple dimensions is obtained to obtain multiple power distribution equipment state data.

[0065] In the embodiment of the present application, the evaluation system can obtain state data of the target network device in multiple dimensions to obtain multiple network device state data. The multiple dimensions include at least two of vibration, temperature, sound, voltage, and current. Each network device state data includes state data at multiple time points, i.e., belongs to a time sequence, such as state data 1 at time point 1, state data 2 at time point 2, state data 3 at time point 3, state data 4 at time point 4, state data 5 at time point 5, and so on. In addition, the network device state data of each dimension can be detected by a corresponding sensor. For example, the network device state data of the vibration dimension can be detected by a vibration sensor, the network device state data of the temperature dimension can be detected by a temperature sensor, the network device state data of the sound dimension can be detected by a sound sensor, the network device state data of the voltage dimension can be detected by a voltage sensor, and the network device state data of the current dimension can be detected by a current sensor.

[0066] In step S120, for each network device state data, a time domain state feature of the network device state data is extracted, and the network device state data is converted into device state frequency domain data.

[0067] In the embodiment of the present application, after obtaining multiple network device state data, the evaluation system can extract a time domain state feature of each network device state data and convert the network device state data into device state frequency domain data. That is, in combination with the above-mentioned Figure 3 After obtaining the network device state data, on the one hand, the time domain network device state data can be directly subjected to feature extraction, and on the other hand, the time domain network device state data can be converted into frequency domain device state frequency domain data, such as a corresponding frequency spectrum.

[0068] In step S130, based on semantic information possessed by the time domain state feature, semantic information possessed by the device state frequency domain data is subjected to semantic mining to form a first state semantic vector.

[0069] In the embodiment of the present application, after obtaining the time domain state feature and the device state frequency domain data, the evaluation system can perform semantic mining on the semantic information possessed by the device state frequency domain data based on the semantic information possessed by the time domain state feature, to form a first state semantic vector. The first state semantic vector is used to reflect the semantic information of the target network configuration device in the time domain and the frequency domain within the dimension. That is, due to the limitation of the extracted time domain state feature, the semantic information possessed by the time domain state feature is relatively not comprehensive, and some important information may be focused on. Therefore, the time domain state feature is used as an auxiliary to assist in mining the more comprehensive and richer device state frequency domain data, so that the first state semantic vector formed can take into account the reliability and richness of the semantic information.

[0070] In step S140, the first state semantic vector corresponding to each of the network configuration device state data is fused to form a second state semantic vector.

[0071] In the embodiment of the present application, after the first state semantic vector is formed by mining, the evaluation system can fuse the first state semantic vector corresponding to each of the network configuration device state data to form a second state semantic vector. The second state semantic vector is used to reflect the global semantic information possessed by the target network configuration device in the plurality of dimensions.

[0072] In step S150, state evaluation is performed based on the second state semantic vector to obtain a running state evaluation result.

[0073] In the embodiment of the present application, after the second state semantic vector is formed by fusion, the evaluation system can perform state evaluation based on the second state semantic vector to obtain a running state evaluation result. The running state evaluation result is used to represent whether the target network configuration device is abnormal. For example, for a transformer, a change in vibration frequency, such as an increase or a change in mode, may be caused by wear, fracture or failure of internal mechanical parts, that is, there is an abnormality.

[0074] Based on the above, on the one hand, due to the semantic mining, the potential semantic information in the network configuration device state data can be utilized, and the basis for state evaluation is more sufficient, so as to guarantee the reliability of state evaluation. On the other hand, due to the correlation mining of the semantic information of the time domain and the frequency domain within the dimension, the semantic representation accuracy of the mined semantic vector can be higher, and the semantic vectors of at least two dimensions are fused, so that the basis for state evaluation can be more sufficient and reliable. Therefore, the problem that the reliability of the running state evaluation of the network configuration device in the prior art is relatively low can be improved.

[0075] The first aspect, for step S110, the specific way of obtaining the plurality of power distribution equipment state data is not limited, and can be selected according to actual needs.

[0076] For example, in an alternative embodiment, the state data collected by each sensor can be obtained in real time, thereby obtaining the plurality of power distribution equipment state data. For another example, in another alternative embodiment, the state data collected by each sensor can be stored in the corresponding storage device first, and when the state evaluation is needed, the evaluation system obtains the stored state data from the storage device, thereby obtaining the plurality of power distribution equipment state data.

[0077] The second aspect, for step S120, the specific way of extracting the time domain state feature and converting to form the equipment state frequency domain data is not limited, and can be selected according to actual needs.

[0078] For example, in an alternative embodiment, the mean value of the state data at each time point can be calculated as the time domain state feature, and the Fourier transform is performed on the power distribution equipment state data to form the equipment state frequency domain data.

[0079] For another example, in another alternative embodiment, in order to improve the representation ability of the obtained time domain state feature, the above step S120 can further include the following contents:

[0080] Firstly, for each of the power distribution equipment state data, the maximum value, the minimum value, the mean value and the standard deviation of the state data at a plurality of time points included in the power distribution equipment state data can be determined, and based on the mean value and the standard deviation, the standardized value of the fourth order central moment around the mean value (i.e. Kurtosis) of the power distribution equipment state data is determined, thereby obtaining the steepness value (i.e. the standardized value) corresponding to the power distribution equipment state data for measuring the steepness of the tail, and the maximum value, the minimum value, the mean value, the standard deviation and the steepness value are determined as the time domain state feature of the power distribution equipment state data;

[0081] Secondly, the Fourier transform can be performed on the power distribution equipment state data to obtain the equipment state frequency domain data.

[0082] The third aspect, for step S130, the specific way of performing semantic mining on the semantic information possessed by the equipment state frequency domain data is not limited, and can be selected according to actual needs.

[0083] For example, in an alternative implementation, the time domain state feature can be mapped in a vector space to form a time domain state mapping vector; and the device state frequency domain data can be convoluted to form a frequency domain state convolution vector; finally, the frequency domain state convolution vector can be subjected to attention processing based on the time domain state mapping vector to form a first state semantic vector.

[0084] For example, in another alternative implementation, in order to improve the accuracy of semantic mining, so that the semantic representation accuracy of the first state semantic vector can be higher, the above step S130 can further include steps S131, S132, S133, S134 and S135, and the specific contents of each step are as follows.

[0085] Step S131, the time domain state feature is mapped in a vector space to form a time domain state mapping vector.

[0086] In the embodiments of the present application, the time domain state feature can be mapped in a vector space to form a time domain state mapping vector. Illustratively, the time domain state feature can be embedded by a word embedding model for vector space mapping to form a time domain state mapping vector.

[0087] Step S132, the device state frequency domain data is convoluted to form a frequency domain state convolution vector.

[0088] In the embodiments of the present application, the device state frequency domain data can also be convoluted to form a frequency domain state convolution vector. Illustratively, the device state frequency domain data can be convoluted by a convolutional neural network model to form a frequency domain state convolution vector. In addition, in some implementations, the result of convolution can also be subjected to pooling and full connection processing to form a frequency domain state convolution vector. In this way, the convolutional neural network model can sequentially include a convolutional layer 1, a pooling layer 1, a convolutional layer 2, a pooling layer 2 and a full connection layer.

[0089] Step S133, based on the local mapping vector corresponding to the steepness value in the time domain state mapping vector corresponding to the device state data, the frequency domain state convolution vector is subjected to first deep semantic mining to form a first deep semantic vector.

[0090] In the embodiments of the present application, after the time domain state mapping vector and the frequency domain state convolution vector are formed, the frequency domain state convolution vector can be subjected to first deep semantic mining based on the local mapping vector corresponding to the steepness value in the time domain state mapping vector corresponding to the status data of the network device, to form a first deep semantic vector. It should be noted that the steepness of the time domain signal is a statistical index for measuring the distribution of extreme values or abnormal values in the signal, which reflects the frequency and distribution of extreme values of the signal away from the mean value in the time domain. Therefore, the steepness has a greater characterization effect on abnormal semantic information. In this way, the frequency domain state convolution vector can be subjected to first deep semantic mining based on the corresponding local mapping vector, so as to improve the accuracy of semantic mining and ensure the accuracy of the obtained first deep semantic vector.

[0091] In step S134, the frequency domain state convolution vector is subjected to second deep semantic mining based on the local mapping vectors other than the local mapping vector corresponding to the steepness value in the time domain state mapping vector, to form a second deep semantic vector.

[0092] In the embodiments of the present application, after the time domain state mapping vector and the frequency domain state convolution vector are formed, the frequency domain state convolution vector can be subjected to second deep semantic mining based on the local mapping vectors other than the local mapping vector corresponding to the steepness value in the time domain state mapping vector, to form a second deep semantic vector. Compared with the steepness value such as the maximum value, the minimum value, the mean value and the standard deviation, the status semantic information of the network device also has a certain characterization effect. Therefore, the frequency domain state convolution vector can be subjected to second deep semantic mining based on the corresponding local mapping vector, so as to improve the accuracy of semantic mining.

[0093] In step S135, the spliced vector of the first deep semantic vector and the second deep semantic vector is subjected to convolution, pooling and activation processing, to form a first status semantic vector.

[0094] In the embodiments of the present application, after the first deep semantic vector and the second deep semantic vector are formed, the spliced vector of the first deep semantic vector and the second deep semantic vector can be subjected to convolution, pooling and activation processing, to form a first status semantic vector. That is, the first deep semantic vector and the second deep semantic vector can be spliced first, to obtain a corresponding spliced vector. Then, the spliced vector can be subjected to convolution, to obtain a convolution vector. Further, the convolution vector can be subjected to pooling, to obtain a pooling vector. Finally, the pooling vector can be subjected to activation processing, to obtain a first status semantic vector.

[0095] It can be understood that the specific manner of performing the first deep semantic mining on the frequency domain state convolution vector in step S133 is not limited, for example, in an alternative embodiment, in order to improve the reliability of the first deep semantic mining, so as to sufficiently mine the semantic information that may have a correlation with the abnormal state, the above-mentioned step S133 can further include steps S133a, S133b, S133c and S133d, and the specific contents of each step are as follows.

[0096] Step S133a, from the time domain state mapping vector, extracting a local mapping vector corresponding to the steepness value for measuring the steepness of the tail corresponding to the power distribution equipment state data, to obtain a first local mapping vector.

[0097] In the embodiment of the present application, the local mapping vector corresponding to the steepness value for measuring the steepness of the tail corresponding to the power distribution equipment state data can be extracted from the time domain state mapping vector to obtain a first local mapping vector. It should be noted that when forming the time domain state mapping vector, the embedding vectors corresponding to each word in the time domain state feature are spliced to form the time domain state mapping vector, so the local mapping vector corresponding to the steepness value can be directly extracted from the time domain state mapping vector to obtain the first local mapping vector, as shown in Figure 4 .

[0098] Step S133b, performing multiple local hidden processing on the frequency domain state convolution vector to form multiple frequency domain state hidden vectors.

[0099] In the embodiment of the present application, the frequency domain state convolution vector can also be subjected to multiple local hidden processing to form multiple frequency domain state hidden vectors. Among them, the hidden regions of each two local hidden processing are at least not completely coincident (i.e. partially coincident or completely not coincident). For example, a target window can be used to slide on the frequency domain state convolution vector, so that the parameters of each selected part are hidden, such as updating the parameters to 0. Among them, in Figure 4 , the area where the "slash" is located represents the hidden area, and the area outside the "slash" represents the area that is not hidden.

[0100] Step S133c, based on the first local mapping vector, performing attention mining on each of the frequency domain state hidden vectors to form each first state attention vector.

[0101] In the embodiments of the present application, after the first local mapping vector and the plurality of frequency domain state hidden vectors are formed, each of the frequency domain state hidden vectors can be subjected to attention mining based on the first local mapping vector to form each first state attention vector. That is, by hiding the local parameters, the interference of other parameters in attention mining can be reduced to a certain extent, so that the accuracy of attention mining can be higher, and by using a plurality of frequency domain state hidden vectors, the coverage of attention mining can be improved to a certain extent, and semantic loss can be avoided.

[0102] In step S133d, the first state attention vectors are fused to form a first deep semantic vector.

[0103] In the embodiments of the present application, after each first state attention vector is formed, each first state attention vector can be fused to form a first deep semantic vector. For example, the plurality of first state attention vectors can be subjected to mean calculation to realize fusion.

[0104] It can be understood that the specific manner of performing second deep semantic mining on the frequency domain state convolution vector in the above step S134 is not limited, for example, in an alternative embodiment, in order to improve the reliability of the second deep semantic mining, so that the semantic information that can be related to the abnormal state can be sufficiently mined, the above step S134 can further include steps S134a, S134b, S134c and S134d, and the specific contents of each step are as follows.

[0105] In step S134a, other local mapping vectors except the local mapping vector corresponding to the steepness value for measuring the steepness of the tail part corresponding to the power distribution equipment state data are extracted from the time domain state mapping vector to obtain second local mapping vectors.

[0106] In the embodiments of the present application, other local mapping vectors except the local mapping vector corresponding to the steepness value for measuring the steepness of the tail part corresponding to the power distribution equipment state data can be extracted from the time domain state mapping vector to obtain second local mapping vectors. It should be noted that when the time domain state mapping vector is formed, the embedding vectors corresponding to each word in the time domain state feature are spliced to form the time domain state mapping vector, so that the local mapping vectors corresponding to the maximum value, the minimum value, the mean value and the standard deviation can be directly extracted from the time domain state mapping vector to obtain the second local mapping vectors.

[0107] In step S134b, a plurality of local hidden processing is performed on the frequency domain state convolution vector to form a plurality of frequency domain state hidden vectors.

[0108] ​In the embodiments of the present application, the frequency domain state convolution vector can also be subjected to multiple local hidden processing to form multiple frequency domain state hidden vectors. Wherein, the hidden regions of each two local hidden processing are at least not completely overlapped (i.e. partially overlapped or not completely overlapped). For example, a target window can be used to slide on the frequency domain state convolution vector to hide the parameters of each selected part, such as updating the parameters to 0. In addition, it should be noted that, since the semantic information concerned is different between the first deep semantic mining and the second deep semantic mining, the target window can have different sizes in the processes of the two deep semantic mining, for example, the size of the target window of the second deep semantic mining can be smaller than the size of the target window of the first deep semantic mining, or in other words, the size of the target window of the second deep semantic mining and the size of the target window of the first deep semantic mining can be used as network parameters of the corresponding neural network model to be determined in the training process of the neural network model, and the step length of the sliding can also be used as network parameters of the corresponding neural network model or be a fixed value, for example, equal to 5, 10, etc. in the processes of the first deep semantic mining and the second deep semantic mining.

[0109] In step S134c, attention mining is performed on each of the frequency domain state hidden vectors based on the second local mapping vector to form a second state attention vector.

[0110] In the embodiments of the present application, after the second local mapping vector and the frequency domain state hidden vector are formed, attention mining can be performed on each of the frequency domain state hidden vectors based on the second local mapping vector to form a second state attention vector. That is, by hiding the local parameters, the interference of other parameters in the attention mining can be reduced to a certain extent, so that the accuracy of the attention mining can be higher, and by using multiple frequency domain state hidden vectors, the coverage of the attention mining can be improved to a certain extent to avoid semantic loss. For example, the first frequency domain state hidden vector can be subjected to attention mining based on the second local mapping vector to form a first second state attention vector. For another example, the second frequency domain state hidden vector can be subjected to attention mining based on the second local mapping vector to form a second second state attention vector.

[0111] In step S134d, the second state attention vectors are fused to form a second deep semantic vector.

[0112] In the embodiments of the present application, after the second state attention vectors are formed, each second state attention vector can be fused to form a second deep semantic vector. For example, mean calculation can be performed on each second state attention vector to realize fusion, which can refer to that parameters at the same distribution position in multiple second state attention vectors are subjected to mean calculation to obtain parameters at the same distribution position in the second deep semantic vector.

[0113] In the fourth aspect, it needs to be explained that the specific manner of fusing each first state semantic vector corresponding to the power distribution device state data is not limited, and can be selected according to actual needs.

[0114] For example, in an alternative implementation, mean calculation can be performed on each first state semantic vector corresponding to the power distribution device state data to obtain a second state semantic vector.

[0115] For another example, in another alternative implementation, in order to sufficiently fuse each first state semantic vector so that the obtained second state semantic vector has better global semantic representation capability in multiple dimensions, the step S140 can further include steps S141, S142, S143 and S144, and the specific contents of each step are as follows.

[0116] In the step S141, two first state semantic vectors corresponding to the power distribution device state data in the vibration and sound dimensions are subjected to multi-level focused mining to form a first state focused vector.

[0117] In the embodiments of the present application, two first state semantic vectors corresponding to the power distribution device state data in the vibration and sound dimensions can be subjected to multi-level focused mining to form a first state focused vector. It needs to be explained that the power distribution device state data in the vibration and sound dimensions generally has high correlation, and therefore, the focused mining can be performed first so that the correlation can be fully utilized to guarantee the precision of fusion.

[0118] In the step S142, two first state semantic vectors corresponding to the power distribution device state data in the voltage and current dimensions are subjected to multi-level focused mining to form a second state focused vector.

[0119] In the embodiments of the present application, two first state semantic vectors corresponding to the power distribution device state data in the voltage and current dimensions can also be subjected to multi-level focused mining to form a second state focused vector. It needs to be explained that the power distribution device state data in the voltage and current dimensions also generally has high correlation, and therefore, the focused mining can be performed first so that the correlation can be fully utilized to guarantee the precision of fusion.

[0120] In the embodiment of the present application, after the first state focus vector and the second state focus vector are formed by focus mining respectively, the first state focus vector and the second state focus vector are subjected to attention mining respectively based on the first state semantic vector corresponding to the temperature dimension of the power distribution device state data, to form a first state mining vector and a second state mining vector. That is, the first state focus vector can be subjected to attention mining based on the first state semantic vector corresponding to the temperature dimension, to form the first state mining vector, and the second state focus vector can be subjected to attention mining based on the first state semantic vector corresponding to the temperature dimension, to form the second state mining vector. It should be noted that the power distribution device state data of the temperature dimension also has certain relevance with the power distribution device state data of the vibration and sound dimensions, and also has certain relevance with the power distribution device state data of the voltage and current dimensions, and therefore, the attention mining can be performed first to further capture the associated semantic information.

[0121] In the embodiment of the present application, after the first state focus vector and the second state focus vector are formed by focus mining respectively, the first state focus vector and the second state focus vector are subjected to attention mining respectively based on the first state semantic vector corresponding to the temperature dimension of the power distribution device state data, to form a first state mining vector and a second state mining vector. That is, the first state focus vector can be subjected to attention mining based on the first state semantic vector corresponding to the temperature dimension, to form the first state mining vector, and the second state focus vector can be subjected to attention mining based on the first state semantic vector corresponding to the temperature dimension, to form the second state mining vector. It should be noted that the power distribution device state data of the temperature dimension also has certain relevance with the power distribution device state data of the vibration and sound dimensions, and also has certain relevance with the power distribution device state data of the voltage and current dimensions, and therefore, the attention mining can be performed first to further capture the associated semantic information.

[0122] In the embodiment of the present application, after the first state focus vector and the second state focus vector are formed by focus mining respectively, the first state focus vector and the second state focus vector are subjected to attention mining respectively based on the first state semantic vector corresponding to the temperature dimension of the power distribution device state data, to form a first state mining vector and a second state mining vector. That is, the first state focus vector can be subjected to attention mining based on the first state semantic vector corresponding to the temperature dimension, to form the first state mining vector, and the second state focus vector can be subjected to attention mining based on the first state semantic vector corresponding to the temperature dimension, to form the second state mining vector. It should be noted that the power distribution device state data of the temperature dimension also has certain relevance with the power distribution device state data of the vibration and sound dimensions, and also has certain relevance with the power distribution device state data of the voltage and current dimensions, and therefore, the attention mining can be performed first to further capture the associated semantic information.

[0123] In the embodiment of the present application, after the first state focus vector and the second state focus vector are formed by focus mining respectively, the first state focus vector and the second state focus vector are subjected to attention mining respectively based on the first state semantic vector corresponding to the temperature dimension of the power distribution device state data, to form a first state mining vector and a second state mining vector. That is, the first state focus vector can be subjected to attention mining based on the first state semantic vector corresponding to the temperature dimension, to form the first state mining vector, and the second state focus vector can be subjected to attention mining based on the first state semantic vector corresponding to the temperature dimension, to form the second state mining vector. It should be noted that the power distribution device state data of the temperature dimension also has certain relevance with the power distribution device state data of the vibration and sound dimensions, and also has certain relevance with the power distribution device state data of the voltage and current dimensions, and therefore, the attention mining can be performed first to further capture the associated semantic information.

[0124] It can be understood that the specific manner of focusing and mining the two first state semantic vectors corresponding to the vibration and sound two-dimensional network device state data in step S141 described above is not limited, for example, in an alternative embodiment, in order to avoid over-reliance on the semantic information of the network device state data of one dimension in the process of focusing and mining, or loss of the semantic information of the network device state data of one dimension, the step S141 described above can further include steps S141a, S141b and S141c, and the specific content is as follows.

[0125] In step S141a, in the first level of focusing and mining, the first state semantic vector corresponding to the vibration dimension network device state data is mapped into the focusing weight distribution of the first state semantic vector corresponding to the sound dimension network device state data, and the first state semantic vector and the focusing weight distribution are multiplied by bit to form the first level of focusing and mining vector.

[0126] In the embodiments of the present application, in combination with Figure 5 In the first level of focusing and mining, the first state semantic vector corresponding to the vibration dimension network device state data can be mapped into the focusing weight distribution of the first state semantic vector corresponding to the sound dimension network device state data, for example, the first state semantic vector can be linearly mapped and activated, the linear mapping includes multiplying a weight matrix, and then adding the result of multiplication to a bias parameter. The activation process can be realized by a sigmoid function. And the first state semantic vector and the focusing weight distribution are multiplied by bit to form the first level of focusing and mining vector. That is, through linear mapping and activation processing, the importance distribution (i.e. focusing weight distribution) in the semantic vector of the vibration dimension can be determined, and then the semantic vector of the sound dimension is adjusted based on the importance distribution, so that in the process of mining and capturing important semantic information, the fusion of semantic information of the two dimensions of vibration and sound can also be realized. In addition, it should be noted that in order to further avoid the problem of semantic information loss, after the bit multiplication operation, the result of the bit multiplication operation and the first state semantic vector corresponding to the sound dimension network device state data can be added or averaged, thereby obtaining the first level of focusing and mining vector.

[0127] In step S141b, in the second level of focusing and mining, the first level of focusing and mining vector is mapped into the focusing weight distribution of the first state semantic vector corresponding to the vibration dimension network device state data, and the first state semantic vector and the focusing weight distribution are multiplied by bit to form the second level of focusing and mining vector.

[0128] In the embodiment of the present application, after the first-level focus mining vector is formed, in the second-level focus mining, the first-level focus mining vector can be mapped into the focus weight distribution of the first state semantic vector corresponding to the state data of the power distribution device in the vibration dimension, and the first state semantic vector and the focus weight distribution are multiplied by bit to form the second-level focus mining vector. The specific processing process can be referred to step S141a.

[0129] In step S141c, the first state focus vector is formed based on the second-level focus mining vector.

[0130] In the embodiment of the present application, after the second-level focus mining vector is formed, the first state focus vector can be formed based on the second-level focus mining vector. For example, in an alternative implementation, the second-level focus mining vector can be directly determined as the first state focus vector. For another example, in another alternative implementation, in the third-level focus mining, the second-level focus mining vector can be mapped into the focus weight distribution of the first state semantic vector corresponding to the state data of the power distribution device in the sound dimension, and the first state semantic vector and the focus weight distribution are multiplied by bit to form the third-level focus mining vector; and in the fourth-level focus mining, the third-level focus mining vector is mapped into the focus weight distribution of the first state semantic vector corresponding to the state data of the power distribution device in the vibration dimension, and the first state semantic vector and the focus weight distribution are multiplied by bit to form the fourth-level focus mining vector. In this way, the last-level focus mining vector can be obtained by analogy, and the last-level focus mining vector is determined as the first state focus vector. It should be noted that the level of focus mining can be even, and the focus weight distribution formed by mapping is alternatively corresponding to the sound dimension and the vibration dimension to balance the semantic information of the two dimensions.

[0131] It can be understood that the specific manner of the multiple levels of focus mining of the two first state semantic vectors corresponding to the state data of the power distribution device in the voltage and current dimensions in the above step S142 is not limited, for example, in an alternative implementation, in order to avoid over-reliance on the semantic information of the state data of the power distribution device in a certain dimension or loss of the semantic information of the state data of the power distribution device in a certain dimension in the process of focus mining, the above step S142 can further include steps S142a, S142b and S142c, and the specific contents are as follows.

[0132] In step S142a, in the first-level focused mining, the first state semantic vector corresponding to the power dimension distribution of the power grid equipment state data is mapped into the focused weight distribution of the first state semantic vector corresponding to the current dimension distribution of the power grid equipment state data, and the first state semantic vector and the focused weight distribution are multiplied by bit to form the first-level focused mining vector.

[0133] In the embodiment of the present application, in the first-level focused mining, the first state semantic vector corresponding to the power dimension distribution of the power grid equipment state data can be mapped into the focused weight distribution of the first state semantic vector corresponding to the current dimension distribution of the power grid equipment state data, and the first state semantic vector and the focused weight distribution are multiplied by bit to form the first-level focused mining vector. The specific processing process can be referred to the explanation of step S141a.

[0134] In step S142b, in the second-level focused mining, the first-level focused mining vector is mapped into the focused weight distribution of the first state semantic vector corresponding to the power dimension distribution of the power grid equipment state data, and the first state semantic vector and the focused weight distribution are multiplied by bit to form the second-level focused mining vector.

[0135] In the embodiment of the present application, in the second-level focused mining, the first-level focused mining vector can be mapped into the focused weight distribution of the first state semantic vector corresponding to the power dimension distribution of the power grid equipment state data, and the first state semantic vector and the focused weight distribution are multiplied by bit to form the second-level focused mining vector. The specific processing process can be referred to the explanation of step S141b.

[0136] In step S142c, the second state focused vector is formed based on the second-level focused mining vector.

[0137] In the embodiment of the present application, after the second-level focused mining vector is formed, the second state focused vector can be formed based on the second-level focused mining vector. The specific processing process can be referred to the explanation of step S141c.

[0138] In the fifth aspect, it needs to be explained that the specific manner of state evaluation based on the second state semantic vector is not limited, and can be selected according to actual needs.

[0139] For example, in an alternative implementation, the second state semantic vector can be fully connected to form a fully connected vector, wherein the size of the fully connected vector is 1*1; then, the fully connected vector can be identity mapped to obtain a corresponding state anomaly probability, and if the state anomaly probability is greater than a set threshold (such as 0.5), it is determined that there is an anomaly, and if the state anomaly probability is less than or equal to the threshold, it is determined that there is no anomaly.

[0140] For another example, in another alternative implementation, the second state semantic vector can be fully connected to form a fully connected vector, wherein the size of the fully connected vector is 1*2; then, the fully connected vector can be linearly mapped (for example, a classification function such as softmax can be used) to form a target probability distribution, wherein the size of the target probability distribution is 1*2, and the two probability values included in the target probability distribution are used to represent the probability of the target distribution network device existing an anomaly and the probability of the target distribution network device not existing an anomaly, respectively; finally, based on the size relationship between the two probability values in the target probability distribution, the running state evaluation result is determined, for example, if the probability of the target distribution network device existing an anomaly is greater than or equal to the probability of the target distribution network device not existing an anomaly, it is determined that there is an anomaly; if the probability of the target distribution network device existing an anomaly is less than the probability of the target distribution network device not existing an anomaly, it is determined that there is no anomaly.

[0141] In combination Figure 6 The embodiments of the present application also provide a data analysis-based distribution network device running state evaluation device applicable to the above-mentioned data analysis-based distribution network device running state evaluation system. The data analysis-based distribution network device running state evaluation device can include a state data acquisition module, a state data processing module, a data semantic mining module, a semantic vector fusion module, and a state evaluation module.

[0142] The state data acquisition module is configured to acquire state data of a target distribution network device in multiple dimensions to obtain a plurality of distribution network device state data, wherein the multiple dimensions include at least two of vibration, temperature, sound, voltage, and current, and each of the distribution network device state data includes state data at multiple time points. In the embodiments of the present application, the state data acquisition module can be configured to perform Figure 2 The related content of the state data acquisition module can be referred to the foregoing description of step S110.

[0143] The state data processing module is configured to extract time domain state features of each of the distribution network device state data, and convert the distribution network device state data into device state frequency domain data. In the embodiments of the present application, the state data processing module can be configured to perform Figure 2The step S120 is shown, and the related content of the state data processing module can refer to the foregoing description of the step S120.

[0144] The data semantic mining module is configured to perform semantic mining on semantic information possessed by the device state frequency domain data based on semantic information possessed by the time domain state feature, to form a first state semantic vector, wherein the first state semantic vector is used to reflect semantic information of the target network configuration device in the time domain and the frequency domain within the dimension. Figure 2 The step S130 is shown, and the related content of the data semantic mining module can refer to the foregoing description of the step S130.

[0145] The semantic vector fusion module is configured to fuse the first state semantic vector corresponding to each of the network configuration device state data, to form a second state semantic vector, wherein the second state semantic vector is used to reflect global semantic information possessed by the target network configuration device in the plurality of dimensions. Figure 2 The step S140 is shown, and the related content of the semantic vector fusion module can refer to the foregoing description of the step S140.

[0146] The state evaluation module is configured to perform state evaluation based on the second state semantic vector, to obtain a running state evaluation result, wherein the running state evaluation result is used to represent whether the target network configuration device is abnormal. Figure 2 The step S150 is shown, and the related content of the state evaluation module can refer to the foregoing description of the step S150.

[0147] To sum up, the method and system for evaluating the running state of distribution network equipment based on data analysis provided in the application first acquires a plurality of distribution network equipment state data; second, for each distribution network equipment state data, extracts the time domain state feature of the distribution network equipment state data, and converts the distribution network equipment state data into equipment state frequency domain data; then, based on the semantic information possessed by the time domain state feature, performs semantic mining on the semantic information possessed by the equipment state frequency domain data to form a first state semantic vector; further, fuses the first state semantic vector corresponding to each distribution network equipment state data to form a second state semantic vector; and finally, performs state evaluation based on the second state semantic vector to obtain a running state evaluation result. Based on the above, on the one hand, since semantic mining is performed, the potential semantic information in the distribution network equipment state data can be utilized, and the basis for state evaluation is more sufficient, so as to guarantee the reliability of state evaluation. On the other hand, since the semantic information of the time domain and the frequency domain is associated and mined within the dimension, the semantic representation accuracy of the mined semantic vector can be higher, and the semantic vectors of at least two dimensions are fused, so that the basis for state evaluation can be more sufficient and reliable. Therefore, the problem of relatively low reliability of the running state evaluation of the distribution network equipment in the prior art can be improved.

[0148] In several embodiments provided in the embodiments of the present application, it should be understood that the disclosed apparatus and method can also be implemented by other manners. The apparatus and method embodiments described above are only illustrative. For example, the flowchart and block diagram in the drawings show the possible implementation architecture, function and operation of the apparatus, method and computer program product according to the embodiments of the present application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment or a part of code, which includes one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order from that shown in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0149] In addition, each functional module in the embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0150] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product in essence or in the form of a part of the prior art or part of the technical solutions of the present application. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media. It should be noted that in this paper, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0151] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A data analysis-based distribution network equipment operation state evaluation method, characterized in that, The method comprises the following steps: acquiring state data of a target power distribution device in multiple dimensions to obtain multiple power distribution device state data, wherein the multiple dimensions include at least two of vibration, temperature, sound, voltage and current, and each of the power distribution device state data includes state data at multiple time points; for each of the power distribution device state data, extracting time domain state features of the power distribution device state data and converting the power distribution device state data into device state frequency domain data; performing vector space mapping on the time domain state features to form a time domain state mapping vector, performing convolution processing on the device state frequency domain data to form a frequency domain state convolution vector, performing first deep semantic mining on the frequency domain state convolution vector based on a local mapping vector corresponding to a kurtosis value in the time domain state mapping vector corresponding to the power distribution device state data, to form a first deep semantic vector, performing second deep semantic mining on the frequency domain state convolution vector based on other local mapping vectors except the local mapping vector corresponding to the kurtosis value in the time domain state mapping vector, to form a second deep semantic vector, and performing convolution, pooling and activation processing on a splicing vector of the first deep semantic vector and the second deep semantic vector to form a first state semantic vector; performing multi-level focused mining on two first state semantic vectors corresponding to the power distribution device state data in the vibration and sound dimensions to form a first state focused vector, performing multi-level focused mining on two first state semantic vectors corresponding to the power distribution device state data in the voltage and current dimensions to form a second state focused vector, and performing attention mining on the first state focused vector and the second state focused vector based on a first state semantic vector corresponding to the power distribution device state data in the temperature dimension, respectively, to form a first state mining vector and a second state mining vector; and performing focused mining on the first state mining vector and the second state mining vector to form a second state semantic vector; performing state evaluation based on the second state semantic vector to obtain an operation state evaluation result, wherein the operation state evaluation result is used to represent whether the target power distribution device is abnormal.

2. The data analysis-based power distribution equipment operating state evaluation method according to claim 1, characterized by, The step of extracting time domain state features of the power distribution device state data and converting the power distribution device state data into device state frequency domain data for each of the power distribution device state data comprises the following steps: for each of the power distribution device state data, determining a maximum value, a minimum value, a mean value and a standard deviation of the state data at the multiple time points included in the power distribution device state data, determining a normalized value of a fourth central moment around the mean value based on the mean value and the standard deviation to obtain a kurtosis value of the power distribution device state data for measuring the steepness of the tail, and determining the maximum value, the minimum value, the mean value, the standard deviation and the kurtosis value as the time domain state features of the power distribution device state data; performing Fourier transform on the power distribution device state data to obtain device state frequency domain data.

3. The data analysis-based power distribution equipment operating state evaluation method according to claim 1, characterized by, The step of performing first deep semantic mining on the frequency domain state convolution vector based on the local mapping vector corresponding to the steepness value for measuring the steepness of the tail part in the time domain state mapping vector corresponding to the power distribution device state data, to form a first deep semantic vector, comprises: extracting, from the time domain state mapping vector, a local mapping vector corresponding to the steepness value for measuring the steepness of the tail part in the time domain state mapping vector corresponding to the power distribution device state data, to obtain a first local mapping vector; performing multiple local hidden processing on the frequency domain state convolution vector to form multiple frequency domain state hidden vectors, wherein the hidden regions of every two local hidden processing are at least not completely coincident; performing attention mining on each of the frequency domain state hidden vectors based on the first local mapping vector to form a first state attention vector; fusing each of the first state attention vectors to form a first deep semantic vector.

4. The data analysis-based power distribution equipment operating state evaluation method according to claim 1, characterized by, The step of performing second deep semantic mining on the frequency domain state convolution vector based on the local mapping vector other than the local mapping vector corresponding to the steepness value in the time domain state mapping vector corresponding to the power distribution device state data, to form a second deep semantic vector, comprises: extracting, from the time domain state mapping vector, a local mapping vector other than the local mapping vector corresponding to the steepness value for measuring the steepness of the tail part in the time domain state mapping vector corresponding to the power distribution device state data, to obtain a second local mapping vector; performing multiple local hidden processing on the frequency domain state convolution vector to form multiple frequency domain state hidden vectors, wherein the hidden regions of every two local hidden processing are at least not completely coincident; performing attention mining on each of the frequency domain state hidden vectors based on the second local mapping vector to form a second state attention vector; fusing each of the second state attention vectors to form a second deep semantic vector.

5. The data analysis based power distribution equipment operating state evaluation method according to claim 1, characterized by, The step of performing multiple levels of focus mining on two first state semantic vectors corresponding to the power distribution device state data in the vibration and sound dimensions to form a first state focus vector, comprises: in the first level of focus mining, mapping the first state semantic vector corresponding to the power distribution device state data in the vibration dimension as a focus weight distribution of the first state semantic vector corresponding to the power distribution device state data in the sound dimension, and performing bitwise multiplication operation on the first state semantic vector and the focus weight distribution to form a first level of focus mining vector; in the second level of focus mining, mapping the first level of focus mining vector as a focus weight distribution of the first state semantic vector corresponding to the power distribution device state data in the vibration dimension, and performing bitwise multiplication operation on the first state semantic vector and the focus weight distribution to form a second level of focus mining vector; forming a first state focus vector based on the second level of focus mining vector.

6. The data analysis based power distribution equipment operating state evaluation method according to claim 1, characterized by, The step of performing multiple levels of focus mining on two first state semantic vectors corresponding to the power distribution device state data in the voltage and current dimensions to form a second state focus vector, comprises: In the first level of focused mining, the first state semantic vector corresponding to the voltage dimension of the power distribution equipment state data is mapped to the focused weight distribution of the first state semantic vector corresponding to the current dimension of the power distribution equipment state data, and the first state semantic vector and the focused weight distribution are multiplied by bit to form a first level of focused mining vector; In the second level of focused mining, the first level of focused mining vector is mapped to the focused weight distribution of the first state semantic vector corresponding to the voltage dimension of the power distribution equipment state data, and the first state semantic vector and the focused weight distribution are multiplied by bit to form a second level of focused mining vector; Based on the second level of focused mining vector, a second state focused vector is formed.

7. The data analysis based power distribution equipment operating state evaluation method according to any one of claims 1-6, characterized by, The step of performing state evaluation based on the second state semantic vector to obtain the operation state evaluation result comprises: performing full connection processing on the second state semantic vector to form a full connection vector, wherein the size of the full connection vector is 1*2; performing linear mapping on the full connection vector to form a target probability distribution, wherein the size of the target probability distribution is 1*2, and the two probability values included in the target probability distribution are respectively used to represent the probability of existence of the target power distribution equipment and the probability of non-existence of the target power distribution equipment; determining the operation state evaluation result based on the size relationship of the two probability values in the target probability distribution.

8. A data analysis-based distribution network equipment operation state evaluation system, characterized by, A device comprising a memory and a processor connected to the memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to realize the power distribution equipment operation state evaluation method based on data analysis according to any one of claims 1-7.

Images