Electricity utilization safety index evaluation method, device and equipment

Abstract

The invention provides an electricity utilization safety index evaluation method, device and equipment, and relates to the technical field of data analysis. In the application, firstly, target power utilization condition distribution is constructed based on the power utilization safety event frequency and the equipment damage degree of each time sub-interval in a target time interval; secondly, performing semantic coding on the target power consumption condition distribution to form a power consumption condition coding vector; then, guiding deep semantic mining of the power consumption condition coding vector based on abnormal data in the target power consumption condition distribution to form a power consumption abnormal coding vector; and finally, semantic decoding is carried out on the electricity utilization abnormity coding vector to form an electricity utilization safety evaluation result. Based on the above content, the problem that the reliability of power utilization safety index evaluation is relatively low in the prior art can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to a power consumption safety index evaluation method, device and equipment. BACKGROUND

[0002] In modern society, the power system is the basis for supporting the operation of various types of equipment, and the safe operation of power equipment is crucial to protect people's daily life, production activities and public safety. The problem of power consumption safety has always been a key problem that needs to be solved in the operation of the power system, especially in the context of the increasing complexity of power consumption environment and the increasing number of equipment types, how to effectively and accurately evaluate the power consumption safety situation has become an important technical challenge. The existing power consumption safety evaluation method mainly uses traditional monitoring means and simple equipment failure statistics. For example, some methods evaluate safety based on the real-time state and failure frequency of equipment, but these methods often fail to fully consider the potential semantic information of the data. In addition, the current power consumption safety evaluation technology mostly relies on experience rules or simple statistical models, and fails to fully utilize big data technology and deep learning methods to effectively analyze massive power consumption data, such as threshold judgment only. Therefore, how to use advanced data analysis technology to conduct more comprehensive and accurate safety evaluation based on relevant data and improve the reliability and accuracy of the evaluation results has become a problem to be solved in the field of power consumption safety management. SUMMARY

[0003] Therefore, the purpose of the present application is to provide a power consumption safety index evaluation method, device and equipment to improve the problem of relatively low reliability of power consumption safety index evaluation in the prior art.

[0004] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows: A power consumption safety index evaluation method, comprising: Based on the number of power consumption safety events and the degree of equipment damage in each time sub-interval within the target time interval, a target power consumption situation distribution is constructed, wherein the size of the target power consumption situation distribution is N*M*2, N corresponds to the number of time sub-intervals, and M corresponds to the maximum value of the number of power consumption safety events and the maximum value of the degree of equipment damage; Semantic encoding is performed on the target power consumption situation distribution to form a power consumption situation encoding vector; Based on the abnormal data in the target power consumption situation distribution, the deep semantic mining of the power consumption situation encoding vector is guided to form a power consumption abnormality encoding vector; The power consumption abnormality encoding vector is subjected to semantic decoding to form a power consumption safety evaluation result, wherein the power consumption safety evaluation result is used to reflect the safety degree of power consumption equipment.

[0005] In a preferred embodiment of this application, in the above-mentioned method for evaluating electricity safety indicators, the step of guiding deep semantic mining of the electricity consumption encoding vector based on abnormal data in the target electricity consumption distribution to form an abnormal electricity consumption encoding vector includes: The number of each power safety event in the target power consumption distribution that is less than a first threshold is hidden, and the degree of damage of each device in the target power consumption distribution that is less than a second threshold is hidden, thus forming an abnormal data distribution. The abnormal data distribution is semantically encoded to form an abnormal data encoding vector; Based on the abnormal data encoding vector, deep semantic mining of the electricity consumption encoding vector is guided to form an abnormal electricity consumption encoding vector.

[0006] In a preferred embodiment of this application, in the above-mentioned method for evaluating electricity safety indicators, the step of guiding deep semantic mining of the electricity consumption status encoding vector based on the abnormal data encoding vector to form an abnormal electricity consumption encoding vector includes: In the first deep semantic mining process, the abnormal data encoding vector is subjected to first pooling and second pooling respectively to form a first-depth first abnormal pooling vector and a first-depth second abnormal pooling vector. Based on the first abnormal pooling vector and the second abnormal pooling vector, the first-depth compressed vector of the electricity consumption encoding vector is subjected to gating mapping and cross-attention processing. The result vector of gating mapping and the result vector of cross-attention processing are averaged or weighted averaged to form the first-depth electricity consumption anomaly mining vector. In each subsequent deep semantic mining process, the first and second abnormal pooling vectors of the previous depth are respectively subjected to first pooling and second pooling to form the first abnormal pooling vector and the second abnormal pooling vector of the current depth. Based on the first and second abnormal pooling vectors, the compressed vector of the electricity consumption anomaly mining vector of the previous depth of the electricity consumption encoding vector is subjected to gating mapping and cross-attention processing. The result vector of gating mapping and the result vector of cross-attention processing are averaged or weighted averaged to form the electricity consumption anomaly mining vector of the current depth. The power consumption anomaly encoding vector is determined based on the power consumption anomaly mining vector from the last depth.

[0007] In a preferred embodiment of this application, the step of determining the electricity anomaly encoding vector based on the electricity anomaly mining vector at the last depth in the above-mentioned electricity safety index assessment method includes: Based on the size of the electricity consumption encoding vector, an upsampling operation is performed on the electricity consumption anomaly mining vector at the last depth to form an electricity consumption anomaly upsampling vector. The power consumption anomaly upsampling vector and the power consumption status encoding vector are concatenated to form the power consumption anomaly encoding vector.

[0008] In a preferred embodiment of this application, the step of semantically encoding the abnormal data distribution to form an abnormal data encoding vector in the above-mentioned electricity safety index assessment method includes: Perform a three-dimensional convolution on the abnormal data distribution to form an abnormal three-dimensional convolution vector; Two-dimensional convolution is performed on the two channels of the abnormal data distribution to form a first abnormal convolution vector and a second abnormal convolution vector, wherein the first abnormal convolution vector corresponds to the number of electrical safety events and the second abnormal convolution vector corresponds to the degree of equipment damage. Based on the first abnormal convolution vector and the second abnormal convolution vector, semantic enhancement is performed on the abnormal three-dimensional convolution vector to form an abnormal data encoding vector.

[0009] In a preferred embodiment of this application, in the above-mentioned method for evaluating electricity safety indicators, the step of semantically enhancing the abnormal three-dimensional convolutional vector based on the first abnormal convolutional vector and the second abnormal convolutional vector to form an abnormal data encoding vector includes: Based on the first abnormal convolution vector, cross-attention processing is performed on the second abnormal convolution vector to form the first abnormal attention vector; Based on the second abnormal convolution vector, the first abnormal convolution vector is subjected to cross-attention processing to form the second abnormal attention vector; In the channel direction, the first abnormal attention vector and the second abnormal attention vector are concatenated to form a three-dimensional abnormal concatenation vector; The three-dimensional anomaly stitching vector is linearly mapped and nonlinearly activated to form an anomaly focusing parameter distribution, wherein each focusing parameter in the anomaly focusing parameter distribution is used to reflect the importance of the corresponding position; Based on the distribution of the anomaly focusing parameters, the anomaly 3D convolution vector is weighted and mapped to form an anomaly data encoding vector.

[0010] In a preferred embodiment of this application, the step of semantically encoding the target electricity consumption distribution to form an electricity consumption encoding vector in the above-mentioned electricity safety index assessment method includes: The target electricity consumption distribution is subjected to three-dimensional convolution to form a three-dimensional convolution vector of electricity consumption. Two-dimensional convolution is performed on the two channels of the target power consumption distribution to form a first power consumption convolution vector and a second power consumption convolution vector, wherein the first power consumption convolution vector corresponds to the number of power safety events and the second power consumption convolution vector corresponds to the degree of equipment damage. Based on the first and second electricity consumption convolution vectors, semantic enhancement is performed on the three-dimensional electricity consumption convolution vectors to form an electricity consumption status encoding vector.

[0011] In a preferred embodiment of this application, in the above-mentioned method for evaluating electricity safety indicators, the step of semantically enhancing the three-dimensional convolutional vector of electricity consumption based on the first and second convolutional vectors of electricity consumption to form an electricity consumption encoding vector includes: Based on the first power consumption convolution vector, the second power consumption convolution vector is subjected to cross-attention processing to form the first power consumption attention vector; Based on the second power consumption convolution vector, the first power consumption convolution vector is subjected to cross-attention processing to form the second power consumption attention vector; In the channel direction, the first power consumption attention vector and the second power consumption attention vector are concatenated to form a three-dimensional power consumption concatenation vector; The three-dimensional power consumption splicing vector is linearly mapped and nonlinearly activated to form a power consumption focusing parameter distribution, wherein each focusing parameter in the power consumption focusing parameter distribution is used to reflect the importance of the corresponding position; Based on the power consumption focusing parameter distribution, the power consumption three-dimensional convolution vector is weighted and mapped to form a power consumption status encoding vector.

[0012] This application also provides an electrical safety index assessment device, including: The situation distribution construction module is used to construct a target power consumption situation distribution based on the number of power safety events and the degree of equipment damage in each time sub-interval within the target time interval. The size of the target power consumption situation distribution is N*M*2, where N corresponds to the number of time sub-intervals and M corresponds to the maximum value of the number of power safety events and the maximum value of the degree of equipment damage. The semantic encoding module is used to perform semantic encoding on the target electricity consumption distribution to form an electricity consumption encoding vector; The deep mining module is used to guide the deep semantic mining of the electricity consumption encoding vector based on the abnormal data in the target electricity consumption distribution, and form an abnormal electricity consumption encoding vector. The semantic decoding module is used to perform semantic decoding on the power consumption anomaly encoding vector to form a power consumption safety assessment result, wherein the power consumption safety assessment result is used to reflect the safety level of the electrical equipment.

[0013] Based on the above, this application also provides an electronic device, including: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the above-mentioned method for evaluating electrical safety indicators.

[0014] The electricity safety indicator assessment method, apparatus, and equipment provided in this application first construct a target electricity consumption distribution based on the number of electricity safety events and the degree of equipment damage in each time sub-interval within a target time interval; secondly, semantically encode the target electricity consumption distribution to form an electricity consumption encoding vector; then, based on abnormal data in the target electricity consumption distribution, guide deep semantic mining of the electricity consumption encoding vector to form an electricity consumption anomaly encoding vector; finally, semantically decode the electricity consumption anomaly encoding vector to form an electricity safety assessment result. Based on the above, on the one hand, because a target electricity consumption distribution is constructed, the correlation between data can be fully utilized during the semantic encoding process, thereby ensuring that the formed encoding vector has high semantic representation capabilities. On the other hand, because the deep semantic mining process is guided based on abnormal data, the deep semantic mining process focuses on mining semantic information related to anomalies. Therefore, the semantic representation accuracy of the formed electricity consumption anomaly encoding vector in the dimensions of safety assessment and anomaly identification is relatively high, thereby ensuring the reliability of the electricity safety assessment result formed by semantic decoding based on this. Therefore, it can improve the problem of relatively low reliability in the assessment of electrical safety indicators in existing technologies. Attached Figure Description

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0016] Figure 1 A structural block diagram of an electronic device provided in an embodiment of this application.

[0017] Figure 2 This is a flowchart illustrating the method for evaluating electricity safety indicators provided in an embodiment of this application.

[0018] Figure 3 A schematic diagram illustrating semantic encoding provided in an embodiment of this application.

[0019] Figure 4 A schematic diagram illustrating semantic enhancement provided in an embodiment of this application.

[0020] Figure 5 This is a schematic diagram illustrating the guidance provided for deep semantic mining in an embodiment of this application.

[0021] Figure 6 This is a block diagram of an electrical safety index assessment device provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0024] like Figure 1 As shown in the illustration, this application provides an electronic device. The electronic device may include a memory, a processor, and an electrical safety indicator assessment device.

[0025] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, the memory and the processor can be electrically connected via one or more communication buses or signal lines. The electrical safety indicator assessment device includes at least one software functional module stored in the memory in the form of software or firmware. The processor is used to execute executable computer programs stored in the memory, such as the software functional modules and computer programs included in the electrical safety indicator assessment device, to implement the electrical safety indicator assessment method provided in this application embodiment.

[0026] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0027] Optionally, 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, or discrete hardware components.

[0028] Understandable. Figure 1 The structure shown is for illustrative purposes only; the electronic device may also include components that are more advanced than those shown. Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices.

[0029] Combination Figure 2 This application also provides a method for evaluating electrical safety indicators applicable to the aforementioned electronic devices. The method steps defined in the process of the electrical safety indicator evaluation method can be implemented by the electronic devices.

[0030] The following will be about Figure 2 The specific process shown will be explained in detail.

[0031] Step S110: Based on the number of power safety events and the degree of equipment damage in each time sub-interval within the target time interval, construct the target power consumption distribution.

[0032] In this embodiment, the electronic device can construct a target power consumption distribution based on the number of power safety events and the degree of equipment damage in each time sub-interval within the target time interval. The target power consumption distribution has a size of N*M*2 (i.e., a three-dimensional matrix with a resolution of N*M and 2 channels, corresponding to the number of power safety events and the degree of equipment damage, respectively; the degree of equipment damage can be determined based on the ratio between repair costs and the total cost of the equipment). N corresponds to the number of time sub-intervals, and M corresponds to the maximum value of the number of power safety events and the maximum value of the degree of equipment damage; for example, it can be a normalized value, such as 1. Furthermore, the specific granularity of the target time interval and the time sub-intervals is not limited; for example, the target time interval can be a year, a month, etc., and the time sub-intervals can be a day, a week, a month, etc.

[0033] Step S120: Semantically encode the target electricity consumption distribution to form an electricity consumption encoding vector.

[0034] In this embodiment, after obtaining the target electricity consumption distribution, the electronic device can semantically encode the target electricity consumption distribution to form an electricity consumption encoding vector. That is, it can extract potential semantic information from the target electricity consumption distribution and represent it in vector form, thereby obtaining the electricity consumption encoding vector. The electricity consumption encoding vector is used to represent the global semantic information in the target electricity consumption distribution.

[0035] Step S130: Based on the abnormal data in the target electricity consumption distribution, guide the deep semantic mining of the electricity consumption encoding vector to form an abnormal electricity consumption encoding vector.

[0036] In this embodiment, after obtaining the electricity consumption encoding vector, the electronic device can guide deep semantic mining of the electricity consumption encoding vector based on abnormal data in the target electricity consumption distribution to form an electricity consumption anomaly encoding vector. That is, since the electricity consumption encoding vector is used to represent global semantic information, its accuracy for electricity safety assessment is relatively low. Therefore, on the one hand, further deep semantic mining can be performed to provide semantic representation capabilities and obtain high-level, abstract semantic features. On the other hand, during the deep semantic mining process, corresponding mining guidance can be performed based on abnormal data in the target electricity consumption distribution, focusing on representing anomaly-related semantic information. Thus, the accuracy of electricity safety assessment can be further improved.

[0037] Step S140: Semantically decode the electricity consumption anomaly encoding vector to form an electricity safety assessment result.

[0038] In this embodiment, after obtaining the power consumption anomaly encoding vector, the electronic device can perform semantic decoding on the power consumption anomaly encoding vector to form a power consumption safety assessment result. The power consumption safety assessment result reflects the safety level of the electrical equipment, such as 0-1, where a larger value indicates a higher safety level, and a smaller value indicates a lower safety level. For example, the semantic decoding process may include: performing a fully connected operation on the power consumption anomaly encoding vector to obtain a 1*1 fully connected vector; then, performing an identity mapping (e.g., y=x) or nonlinear activation (e.g., using a sigmoid function) on the fully connected vector to obtain the power consumption safety assessment result.

[0039] Based on the above, on the one hand, the construction of the target electricity consumption distribution allows for full utilization of data correlations during semantic encoding, ensuring that the resulting encoded vector has high semantic representation capabilities. On the other hand, the deep semantic mining process is guided by anomaly data, focusing on mining semantic information related to anomalies. Therefore, the resulting electricity anomaly encoded vector exhibits relatively high semantic representation accuracy in the dimensions of safety assessment and anomaly identification, thus ensuring the reliability of the electricity safety assessment results derived from semantic decoding. Therefore, this addresses the problem of relatively low reliability in electricity safety indicator assessments in existing technologies.

[0040] Firstly, regarding step S120, it should be noted that the specific method of semantically encoding the target electricity consumption distribution is not limited and can be selected according to actual needs.

[0041] For example, in an alternative implementation, the target electricity consumption distribution can be directly convolved in three dimensions to form a corresponding electricity consumption encoding vector.

[0042] For example, in another alternative implementation, in order to fully extract the global semantic information in the target electricity consumption distribution, the above step S120 may further include steps S121, S122 and S123, wherein the specific contents of each step are as follows.

[0043] Step S121: Perform a three-dimensional convolution on the target electricity consumption distribution to form a three-dimensional convolution vector of electricity consumption.

[0044] In the embodiments of this application, combined with Figure 3 The target electricity consumption distribution can be subjected to three-dimensional convolution (that is, the corresponding convolution kernel can be three-dimensional) to form a three-dimensional convolution vector of electricity consumption.

[0045] Step S122: Perform two-dimensional convolution on the two channels of the target electricity consumption distribution to form a first electricity consumption convolution vector and a second electricity consumption convolution vector.

[0046] In this embodiment, two-dimensional convolution can be performed on the two channels of the target electricity consumption distribution to form a first electricity consumption convolution vector and a second electricity consumption convolution vector. The first electricity consumption convolution vector corresponds to the number of electricity safety events, and the second electricity consumption convolution vector corresponds to the degree of equipment damage. That is, the target electricity consumption distribution can include two matrices: a first matrix reflecting the number of electricity safety events in each time sub-interval within the target time interval, and a second matrix reflecting the degree of equipment damage in each time sub-interval within the target time interval. Thus, a two-dimensional convolution can be performed on the first matrix (e.g., the corresponding convolution kernel can be two-dimensional) to obtain the first electricity consumption convolution vector, and a two-dimensional convolution can be performed on the second matrix to obtain the second electricity consumption convolution vector. Based on this, the three-dimensional electricity consumption convolution vector can be used to represent the data of the two channels as a whole, while the first and second electricity consumption convolution vectors can represent the data of their respective channels individually.

[0047] Step S123: Based on the first power consumption convolution vector and the second power consumption convolution vector, perform semantic enhancement on the three-dimensional power consumption convolution vector to form a power consumption status encoding vector.

[0048] In this embodiment of the application, after obtaining the three-dimensional convolutional vector of electricity consumption, the first convolutional vector of electricity consumption, and the second convolutional vector of electricity consumption, semantic enhancement can be performed on the three-dimensional convolutional vector of electricity consumption based on the first convolutional vector of electricity consumption and the second convolutional vector of electricity consumption to form an electricity consumption status encoding vector, thereby making the accuracy of the electricity consumption status encoding vector higher.

[0049] It is understood that the specific method of semantic enhancement of the three-dimensional convolutional vector in step S123 above is not limited. For example, in an alternative implementation, in order to ensure higher reliability of semantic enhancement, step S123 above may further include steps S123a, S123b, S123c, S123d and S123e, wherein the specific contents of each step are as follows.

[0050] Step S123a: Based on the first power consumption convolution vector, perform cross-attention processing on the second power consumption convolution vector to form a first power consumption attention vector.

[0051] In the embodiments of this application, combined with Figure 4Based on the first power consumption convolution vector, cross-attention processing can be performed on the second power consumption convolution vector to form a first power consumption attention vector. For example, the first power consumption convolution vector can be mapped to a query vector, and the second power consumption convolution vector can be mapped to a key vector and a value vector. In this way, semantic information related to the first power consumption convolution vector can be extracted from the second power consumption convolution vector, that is, the first power consumption attention vector.

[0052] Step S123b: Based on the second power consumption convolution vector, perform cross-attention processing on the first power consumption convolution vector to form a second power consumption attention vector.

[0053] In this embodiment, the first power consumption convolution vector can be subjected to cross-attention processing based on the second power consumption convolution vector to form a second power consumption attention vector. For example, the second power consumption convolution vector can be mapped to a query vector, and the first power consumption convolution vector can be mapped to a key vector and a value vector. In this way, semantic information related to the second power consumption convolution vector can be extracted from the first power consumption convolution vector, i.e., the second power consumption attention vector.

[0054] Step S123c: In the channel direction, the first power consumption attention vector and the second power consumption attention vector are spliced ​​together to form a three-dimensional power consumption splicing vector.

[0055] In this embodiment of the application, after obtaining the first power consumption attention vector and the second power consumption attention vector, the first power consumption attention vector and the second power consumption attention vector can be concatenated in the channel direction to form a three-dimensional power consumption concatenated vector, wherein the size of the three-dimensional power consumption concatenated vector can be equal to the size of the power consumption three-dimensional convolution vector.

[0056] Step S123d involves linear mapping and nonlinear activation of the three-dimensional power splicing vector to form a power focusing parameter distribution.

[0057] In this embodiment, after obtaining the three-dimensional power supply splicing vector, a linear mapping and nonlinear activation can be performed on the vector to form a power supply focusing parameter distribution. Each focusing parameter in this distribution reflects the importance of its corresponding location. It should be noted that the linear mapping can be implemented using linear functions such as Y=Ax+B, where A is the weight matrix and B is the bias parameter. Furthermore, the linear mapping does not change the size of the vector. Nonlinear activation can be implemented using functions such as sigmoid. In this way, the three-dimensional power supply splicing vector can be mapped into a parameter distribution that can characterize importance.

[0058] Step S123e: Based on the power consumption focusing parameter distribution, perform weight mapping on the three-dimensional convolutional vector of power consumption to form a power consumption status encoding vector.

[0059] In this embodiment, after obtaining the power consumption focusing parameter distribution, the power consumption three-dimensional convolutional vector can be weighted based on the power consumption focusing parameter distribution to form a power consumption status encoding vector. For example, the power consumption focusing parameter distribution and the power consumption three-dimensional convolutional vector can be multiplied bitwise, that is, the parameters at corresponding positions in the power consumption three-dimensional convolutional vector are weighted based on the first power consumption convolutional vector and the second power consumption convolutional vector to achieve constraints. It should be noted that before performing weight mapping, mutual cross-attention processing and concatenation are performed first, which can represent the important semantic information within each part. In this way, after concatenation, linear mapping and nonlinear activation can ensure that the formed power consumption focusing parameter distribution can effectively represent important semantic information, that is, avoid the problem that positions corresponding to invalid semantic information are given special attention, which would lead to low reliability of subsequent weight mapping. In addition, it should also be noted that after performing weight mapping, the result of weight mapping can be expanded to obtain the power consumption status encoding vector.

[0060] Secondly, regarding step S130, it should be noted that the specific method for guiding the deep semantic mining of the electricity consumption encoding vector is not limited and can be selected according to actual needs.

[0061] For example, in an alternative implementation, in order to ensure that the semantic representation accuracy of the formed electricity consumption anomaly coding vector is higher, the above step S130 may further include steps S131, S132 and S133, wherein the specific contents of each step are as follows.

[0062] Step S131: Hide the number of each power safety event in the target power consumption distribution that is less than the first threshold, and hide the degree of damage of each device in the target power consumption distribution that is less than the second threshold, to form an abnormal data distribution.

[0063] In this embodiment, the number of electricity safety events below a first threshold and the degree of equipment damage below a second threshold in the target electricity consumption distribution can be hidden to form an abnormal data distribution. That is, the number of electricity safety events below the first threshold and the degree of equipment damage below the second threshold can be determined as normal. Furthermore, hiding can refer to updating the corresponding parameters to 0, indicating that no corresponding safety event has occurred. Thus, in the formed abnormal data distribution, special attention can be paid to abnormal or serious safety events.

[0064] Step S132: Semantically encode the abnormal data distribution to form an abnormal data encoding vector.

[0065] In this embodiment of the application, after obtaining the abnormal data distribution, the abnormal data distribution can be semantically encoded to form an abnormal data encoding vector. That is, the potential semantic information (i.e., abnormal-related semantic information) in the abnormal data distribution can be mined and represented in vector form, thus obtaining the abnormal data encoding vector.

[0066] Step S133: Based on the abnormal data encoding vector, guide the deep semantic mining of the electricity consumption encoding vector to form an abnormal electricity consumption encoding vector.

[0067] In this embodiment, after obtaining the abnormal data encoding vector, deep semantic mining of the electricity consumption encoding vector can be guided based on the abnormal data encoding vector to form an electricity consumption anomaly encoding vector. Therefore, since the abnormal data encoding vector can fully represent the semantic information related to the anomaly, after guidance during deep semantic mining, an encoding vector focusing on representing the semantic information related to the anomaly can be obtained.

[0068] It is understood that the specific method of semantically encoding the abnormal data distribution in step S132 above is not limited. For example, in an alternative implementation, in order to fully extract the global semantic information in the abnormal data distribution, step S132 above may further include steps S132a, S132b and S132c, wherein the specific contents of each step are as follows.

[0069] Step S132a: Perform three-dimensional convolution on the abnormal data distribution to form an abnormal three-dimensional convolution vector.

[0070] In this embodiment of the application, the abnormal data distribution can be subjected to three-dimensional convolution (that is, the corresponding convolution kernel can be three-dimensional) to form an abnormal three-dimensional convolution vector.

[0071] Step S132b: Perform two-dimensional convolution on the two channels of the abnormal data distribution to form a first abnormal convolution vector and a second abnormal convolution vector.

[0072] In this embodiment, two-dimensional convolution can be performed on the two channels of the abnormal data distribution to form a first abnormal convolution vector and a second abnormal convolution vector. The first abnormal convolution vector corresponds to the number of electrical safety events, and the second abnormal convolution vector corresponds to the degree of equipment damage. That is, the abnormal data distribution can include two matrices: a first matrix reflecting the number of electrical safety events in each time sub-interval within the target time interval, and a second matrix reflecting the degree of equipment damage in each time sub-interval within the target time interval. Thus, a two-dimensional convolution can be performed on the first matrix (e.g., the corresponding convolution kernel can be two-dimensional) to obtain the first abnormal convolution vector, and a two-dimensional convolution can be performed on the second matrix to obtain the second abnormal convolution vector. Based on this, the abnormal three-dimensional convolution vector can be used to represent the data of the two channels as a whole, while the first and second abnormal convolution vectors can represent the abnormal data of their respective channels individually.

[0073] Step S132c: Based on the first abnormal convolution vector and the second abnormal convolution vector, perform semantic enhancement on the abnormal three-dimensional convolution vector to form an abnormal data encoding vector.

[0074] In this embodiment of the application, after obtaining the abnormal 3D convolution vector, the first abnormal convolution vector, and the second abnormal convolution vector, semantic enhancement can be performed on the abnormal 3D convolution vector based on the first abnormal convolution vector and the second abnormal convolution vector to form an abnormal data encoding vector, thereby making the abnormal data encoding vector more accurate.

[0075] It is understood that the specific method of semantic enhancement of the abnormal three-dimensional convolution vector in step S132c above is not limited. For example, in an alternative implementation, in order to ensure higher reliability of semantic enhancement, step S132c above may further include steps c1, c2, c3, c4 and c5, wherein the specific contents of each step are as follows.

[0076] Step c1: Based on the first abnormal convolution vector, perform cross-attention processing on the second abnormal convolution vector to form a first abnormal attention vector.

[0077] In this embodiment, the second abnormal convolution vector can be subjected to cross-attention processing based on the first abnormal convolution vector to form a first abnormal attention vector. For example, the first abnormal convolution vector can be mapped to a query vector, and the second abnormal convolution vector can be mapped to a key vector and a value vector. In this way, semantic information related to the first abnormal convolution vector can be extracted from the second abnormal convolution vector, i.e., the first abnormal attention vector.

[0078] Step c2: Based on the second abnormal convolution vector, perform cross-attention processing on the first abnormal convolution vector to form a second abnormal attention vector.

[0079] In this embodiment, the first abnormal convolutional vector can be subjected to cross-attention processing based on the second abnormal convolutional vector to form a second abnormal attention vector. For example, the second abnormal convolutional vector can be mapped to a query vector, and the first abnormal convolutional vector can be mapped to a key vector and a value vector. In this way, semantic information related to the second abnormal convolutional vector can be extracted from the first abnormal convolutional vector, i.e., the second abnormal attention vector.

[0080] Step c3: In the channel direction, the first abnormal attention vector and the second abnormal attention vector are concatenated to form a three-dimensional abnormal concatenation vector.

[0081] In this embodiment of the application, after obtaining the first abnormal attention vector and the second abnormal attention vector, the first abnormal attention vector and the second abnormal attention vector can be concatenated in the channel direction to form a three-dimensional abnormal concatenation vector, wherein the size of the three-dimensional abnormal concatenation vector can be equal to the size of the abnormal three-dimensional convolution vector.

[0082] Step c4 involves linear mapping and nonlinear activation of the three-dimensional anomaly splicing vector to form an anomaly focusing parameter distribution.

[0083] In this embodiment, after obtaining the three-dimensional anomaly splicing vector, linear mapping and nonlinear activation can be performed on the vector to form an anomaly focusing parameter distribution. Each focusing parameter in this distribution reflects the importance of its corresponding location. It should be noted that the linear mapping can be implemented using linear functions such as Y=Ax+B, where A is the weight matrix and B is the bias parameter. Furthermore, the linear mapping does not change the size of the vector. Nonlinear activation can be implemented using functions such as sigmoid. In this way, the three-dimensional anomaly splicing vector can be mapped into a parameter distribution that can characterize importance.

[0084] Step c5: Based on the distribution of the abnormal focusing parameters, perform weight mapping on the abnormal three-dimensional convolution vector to form an abnormal data encoding vector.

[0085] In this embodiment, after obtaining the anomaly focusing parameter distribution, weight mapping can be performed on the anomaly 3D convolution vector based on the anomaly focusing parameter distribution to form an anomaly data encoding vector. For example, the anomaly focusing parameter distribution and the anomaly 3D convolution vector can be multiplied bitwise, that is, the parameters at corresponding positions in the anomaly 3D convolution vector are weighted based on the first anomaly convolution vector and the second anomaly convolution vector to achieve constraints. It should be noted that before performing weight mapping, mutual cross-attention processing and concatenation are performed first, which can represent the important semantic information within each part. In this way, after concatenation, linear mapping and nonlinear activation can ensure that the formed anomaly focusing parameter distribution can effectively represent important semantic information, that is, avoid the problem that positions corresponding to invalid semantic information are given special attention, which would lead to low reliability of subsequent weight mapping.

[0086] It is understood that in step S133 above, the specific method of guiding the deep semantic mining of the electricity consumption encoding vector based on the abnormal data encoding vector is not limited. For example, in an alternative implementation, in order to make the corresponding encoding guidance more sufficient by deeply fusing the abnormal data encoding vector and the electricity consumption encoding vector, step S133 above may further include steps S133a, S133b and S133c, wherein the specific contents of each step are as follows.

[0087] Step S133a: In the first deep semantic mining process, the abnormal data encoding vector is subjected to first pooling and second pooling respectively to form a first-depth first abnormal pooling vector and a first-depth second abnormal pooling vector. Based on the first abnormal pooling vector and the second abnormal pooling vector, the compressed vector of the first depth of the electricity consumption encoding vector is subjected to gating mapping and cross-attention processing respectively. The result vector of gating mapping and the result vector of cross-attention processing are averaged or weighted averaged to form the first-depth electricity consumption anomaly mining vector.

[0088] In the embodiments of this application, combined with Figure 5In the first deep semantic mining process, the abnormal data encoding vector is subjected to a first pooling (such as mean pooling) and a second pooling (such as max pooling) to form a first-depth abnormal pooling vector and a first-depth abnormal pooling vector. Based on the first and second abnormal pooling vectors, the first-depth compressed vector of the electricity consumption encoding vector is subjected to gating mapping (e.g., gating mapping is performed on the compressed vector based on the first abnormal pooling vector, wherein nonlinear activation is performed on the first abnormal pooling vector to obtain the corresponding focusing parameter distribution, and then the focusing parameter distribution and the compressed vector are multiplied bitwise) and cross-attention processing (e.g., cross-attention processing is performed on the compressed vector based on the second abnormal pooling vector). The result vector of the gating mapping and the result vector of the cross-attention processing are averaged or weighted averaged to form the first-depth electricity consumption anomaly mining vector.

[0089] In step S133b, during each subsequent deep semantic mining process, the first and second abnormal pooling vectors of the previous depth are respectively subjected to first pooling and second pooling to form the first abnormal pooling vector and the second abnormal pooling vector of the current depth. Based on the first and second abnormal pooling vectors, the compressed vector of the electricity consumption anomaly mining vector of the previous depth of the electricity consumption encoding vector is subjected to gating mapping and cross-attention processing. The result vector of gating mapping and the result vector of cross-attention processing are averaged or weighted averaged to form the electricity consumption anomaly mining vector of the current depth.

[0090] In the embodiments of this application, during each of the second and subsequent deep semantic mining processes, the first and second abnormal pooling vectors of the previous depth are respectively subjected to first pooling and second pooling to form the first abnormal pooling vector and the second abnormal pooling vector of the current depth. Based on the first and second abnormal pooling vectors, the compressed vector of the electricity consumption anomaly mining vector of the previous depth of the electricity consumption encoding vector (for example, in the second deep semantic mining process, the electricity consumption anomaly mining vector of the first depth can be processed by convolution or pooling to compress the vector size; in the third deep semantic mining process, the electricity consumption anomaly mining vector of the second depth can be processed by convolution or pooling to compress the vector size) is subjected to gating mapping and cross-attention processing. The result vector of gating mapping and the result vector of cross-attention processing are averaged or weighted averaged to form the electricity consumption anomaly mining vector of the current depth.

[0091] Step S133c: Determine the power consumption anomaly encoding vector based on the power consumption anomaly mining vector of the last depth.

[0092] In this embodiment of the application, after obtaining the power consumption anomaly mining vector at the last depth, the power consumption anomaly encoding vector can be determined based on the power consumption anomaly mining vector at the last depth. For example, the power consumption anomaly mining vector at the last depth can be determined as the power consumption anomaly encoding vector.

[0093] It is understood that the specific method for determining the electricity consumption anomaly encoding vector in step S133c above is not limited. For example, in an alternative implementation, in order to improve the semantic representation capability of the determined electricity consumption anomaly encoding vector, such as avoiding the problem of partial semantic loss due to excessive depth during deep mining, step S133c above may include: First, based on the size of the electricity consumption encoding vector, an upsampling operation (such as transpose convolution, which increases the size of the vector) can be performed on the electricity consumption anomaly mining vector at the last depth to form an electricity consumption anomaly upsampling vector, wherein the size of the electricity consumption anomaly upsampling vector can be equal to the size of the electricity consumption encoding vector; Secondly, the power consumption anomaly upsampling vector and the power consumption status encoding vector can be concatenated (e.g., spliced, averaged, or weighted averaged) to form the power consumption anomaly encoding vector.

[0094] Combination Figure 6 This application also provides an electrical safety indicator assessment device applicable to the aforementioned electronic devices. The electrical safety indicator assessment device may include a situation distribution construction module, a semantic encoding module, a deep mining module, and a semantic decoding module.

[0095] The situation distribution construction module is used to construct a target electricity consumption situation distribution based on the number of electricity safety events and the degree of equipment damage in each time sub-interval within the target time interval. The target electricity consumption situation distribution has a size of N*M*2, where N corresponds to the number of time sub-intervals, and M corresponds to the maximum value of the number of electricity safety events and the maximum value of the degree of equipment damage. In this embodiment, the situation distribution construction module can be used to execute... Figure 2 The relevant content regarding the situation distribution construction module shown in step S110 can be found in the previous description of step S110.

[0096] The semantic encoding module is used to semantically encode the target electricity consumption distribution to form an electricity consumption encoding vector. In this embodiment, the semantic encoding module can be used to perform... Figure 2The relevant content regarding the semantic encoding module in step S120 shown can be found in the previous description of step S120.

[0097] The deep mining module is used to guide deep semantic mining of the electricity consumption encoding vector based on abnormal data in the target electricity consumption distribution, thereby forming an electricity consumption anomaly encoding vector. In this embodiment, the deep mining module can be used to perform... Figure 2 The relevant content regarding the deep mining module in step S130 shown can be found in the previous description of step S130.

[0098] The semantic decoding module is used to perform semantic decoding on the electricity consumption anomaly encoding vector to form an electricity safety assessment result, wherein the electricity safety assessment result is used to reflect the safety level of the electrical equipment. In this embodiment of the application, the semantic decoding module can be used to execute... Figure 2 The relevant content regarding the semantic decoding module in step S140 shown can be found in the preceding description of step S140.

[0099] In summary, the electricity safety indicator assessment method, apparatus, and equipment provided in this application first construct a target electricity consumption distribution based on the number of electricity safety events and the degree of equipment damage in each time sub-interval within a target time interval; second, semantically encode the target electricity consumption distribution to form an electricity consumption encoding vector; then, based on abnormal data in the target electricity consumption distribution, guide deep semantic mining of the electricity consumption encoding vector to form an electricity consumption anomaly encoding vector; finally, semantically decode the electricity consumption anomaly encoding vector to form an electricity safety assessment result. Based on the above, on the one hand, because a target electricity consumption distribution is constructed, the correlation between data can be fully utilized during the semantic encoding process, thereby ensuring that the formed encoding vector has high semantic representation capabilities. On the other hand, because the deep semantic mining process is guided based on abnormal data, the deep semantic mining process focuses on mining semantic information related to anomalies. Therefore, the semantic representation accuracy of the formed electricity consumption anomaly encoding vector in the dimensions of safety assessment and anomaly identification is relatively high, thus ensuring the reliability of the electricity safety assessment result formed by semantic decoding based on this. Therefore, it can improve the problem of relatively low reliability in the assessment of electrical safety indicators in existing technologies.

[0100] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0101] In addition, the functional modules in the various embodiments of this 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.

[0102] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0103] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for evaluating an electrical safety indicator, characterized by, The method comprises the steps of: constructing a target power consumption condition distribution based on the number of power consumption safety events and the degree of damage of equipment in each time sub-interval within a target time interval, wherein the size of the target power consumption condition distribution is N*M*2, N corresponds to the number of time sub-intervals, and M corresponds to the maximum value of the number of power consumption safety events and the maximum value of the degree of damage of equipment; performing semantic coding on the target power consumption condition distribution to form a power consumption condition coding vector; based on the abnormal data in the target power consumption condition distribution, guiding the deep semantic mining of the power consumption condition coding vector to form a power consumption anomaly coding vector; performing semantic decoding on the power consumption anomaly coding vector to form a power consumption safety evaluation result, wherein the power consumption safety evaluation result is used to reflect the safety degree of the power consumption equipment.

2. The method of claim 1, wherein, The step of guiding the deep semantic mining of the power consumption condition coding vector based on the abnormal data in the target power consumption condition distribution to form a power consumption anomaly coding vector comprises: hiding each power consumption safety event number in the target power consumption condition distribution that is less than a first threshold value, and hiding each degree of damage of equipment in the target power consumption condition distribution that is less than a second threshold value to form an abnormal data distribution; performing semantic coding on the abnormal data distribution to form an abnormal data coding vector; based on the abnormal data coding vector, guiding the deep semantic mining of the power consumption condition coding vector to form a power consumption anomaly coding vector.

3. The method of claim 2, wherein, The step of guiding the deep semantic mining of the power consumption condition coding vector based on the abnormal data coding vector to form a power consumption anomaly coding vector comprises: in the process of the first deep semantic mining, respectively performing first pooling and second pooling on the abnormal data coding vector to form a first abnormal pooling vector of the first depth and a second abnormal pooling vector of the first depth, and respectively based on the first abnormal pooling vector and the second abnormal pooling vector, performing gated mapping and cross-attention processing on the compressed vector of the first depth of the power consumption condition coding vector, and performing mean value or weighted mean value on the result vector of the gated mapping and the result vector of the cross-attention processing to form a power consumption anomaly mining vector of the first depth; in the process of each deep semantic mining from the second to the subsequent, respectively performing first pooling and second pooling on the first abnormal pooling vector and the second abnormal pooling vector of the previous depth to form a first abnormal pooling vector of the current depth and a second abnormal pooling vector of the current depth, and respectively based on the first abnormal pooling vector and the second abnormal pooling vector, performing gated mapping and cross-attention processing on the compressed vector of the power consumption anomaly mining vector of the previous depth of the power consumption condition coding vector, and performing mean value or weighted mean value on the result vector of the gated mapping and the result vector of the cross-attention processing to form a power consumption anomaly mining vector of the current depth; determining the power consumption anomaly coding vector based on the power consumption anomaly mining vector of the last depth.

4. The method of claim 3, wherein, The step of determining the power consumption anomaly coding vector based on the power consumption anomaly mining vector of the last depth comprises: Based on the size of the power consumption situation encoding vector, the last depth power consumption anomaly mining vector is up-sampled to form a power consumption anomaly up-sampled vector; The power consumption anomaly up-sampled vector and the power consumption situation encoding vector are connected to form a power consumption anomaly encoding vector.

5. The method of claim 2, wherein, The step of performing semantic encoding on the abnormal data distribution to form an abnormal data encoding vector comprises: Performing three-dimensional convolution on the abnormal data distribution to form an abnormal three-dimensional convolution vector; Respectively performing two-dimensional convolution on two channel data of the abnormal data distribution to form a first abnormal convolution vector and a second abnormal convolution vector, wherein the first abnormal convolution vector corresponds to the number of power consumption safety events, and the second abnormal convolution vector corresponds to the degree of equipment damage; Based on the first abnormal convolution vector and the second abnormal convolution vector, performing semantic enhancement on the abnormal three-dimensional convolution vector to form an abnormal data encoding vector.

6. The method of claim 5, wherein, The step of performing semantic enhancement on the abnormal three-dimensional convolution vector based on the first abnormal convolution vector and the second abnormal convolution vector to form an abnormal data encoding vector comprises: Based on the first abnormal convolution vector, performing cross-attention processing on the second abnormal convolution vector to form a first abnormal attention vector; Based on the second abnormal convolution vector, performing cross-attention processing on the first abnormal convolution vector to form a second abnormal attention vector; In the channel direction, the first abnormal attention vector and the second abnormal attention vector are spliced to form a three-dimensional abnormal splicing vector; Performing linear mapping and nonlinear activation on the three-dimensional abnormal splicing vector to form an abnormal focus parameter distribution, wherein each focus parameter in the abnormal focus parameter distribution is used to reflect the importance of the corresponding position; Based on the abnormal focus parameter distribution, performing weight mapping on the abnormal three-dimensional convolution vector to form an abnormal data encoding vector.

7. The method of claim 1-6, wherein, The step of performing semantic encoding on the target power consumption situation distribution to form a power consumption situation encoding vector comprises: Performing three-dimensional convolution on the target power consumption situation distribution to form a power consumption three-dimensional convolution vector; Respectively performing two-dimensional convolution on two channel data of the target power consumption situation distribution to form a first power consumption convolution vector and a second power consumption convolution vector, wherein the first power consumption convolution vector corresponds to the number of power consumption safety events, and the second power consumption convolution vector corresponds to the degree of equipment damage; Based on the first power consumption convolution vector and the second power consumption convolution vector, performing semantic enhancement on the power consumption three-dimensional convolution vector to form a power consumption situation encoding vector.

8. The method of claim 7, wherein, The step of performing semantic enhancement on the power consumption three-dimensional convolution vector based on the first power consumption convolution vector and the second power consumption convolution vector to form a power consumption situation encoding vector comprises: Based on the first power consumption convolution vector, performing cross-attention processing on the second power consumption convolution vector to form a first power consumption attention vector; Based on the second power consumption convolution vector, performing cross-attention processing on the first power consumption convolution vector to form a second power consumption attention vector; The first power consumption attention vector and the second power consumption attention vector are spliced in a channel direction to form a three-dimensional power consumption spliced vector; The three-dimensional power consumption spliced vector is linearly mapped and nonlinearly activated to form a power consumption focus parameter distribution, wherein each focus parameter in the power consumption focus parameter distribution is used to reflect the importance of a corresponding position; Based on the power consumption focus parameter distribution, the power consumption three-dimensional convolution vector is weight-mapped to form a power consumption situation encoding vector.

9. An electric power safety index evaluation device characterized by comprising: Comprise: A situation distribution construction module is configured to construct a target power consumption situation distribution based on the number of power consumption safety events and the degree of equipment damage in each time sub-interval within a target time interval, wherein the size of the target power consumption situation distribution is N*M*2, N corresponds to the number of time sub-intervals, and M corresponds to the maximum value of the number of power consumption safety events and the maximum value of the degree of equipment damage; A semantic encoding module is configured to perform semantic encoding on the target power consumption situation distribution to form a power consumption situation encoding vector; A deep mining module is configured to guide deep semantic mining of the power consumption situation encoding vector based on abnormal data in the target power consumption situation distribution to form a power consumption anomaly encoding vector; A semantic decoding module is configured to perform semantic decoding on the power consumption anomaly encoding vector to form a power consumption safety evaluation result, wherein the power consumption safety evaluation result is used to reflect the safety degree of power consumption equipment.

10. An electronic device, comprising: Comprise: A memory is configured to store a computer program; A processor connected with the memory is configured to execute the computer program stored in the memory to implement the power consumption safety index evaluation method of any one of claims 1-8.

Images