A GRU-based power internet of things time series data anomaly detection system

Abstract

The application provides a GRU-based power Internet of Things time series data anomaly detection system, and the anomaly detection system is used for anomaly detection of cable joints in a power distribution network. The anomaly detection system comprises a processor, and the processor realizes anomaly detection through the following steps: receiving instantaneous current of the cable joint, real-time temperature of the cable joint and ambient temperature of the cable joint, and sampling the real-time temperature through a sliding window to obtain an original temperature sequence; according to a cable trench topology structure diagram of the power distribution network and the real-time temperature, obtaining a temperature difference of adjacent nodes of the cable joint and a topology weighted temperature, and further obtaining a spatial contrast feature; according to time domain features and frequency domain features of the real-time temperature and the spatial contrast feature, obtaining statistical features of the cable joint; and combining an improved GRU model, obtaining an anomaly detection result of the cable joint. Through deep fusion of physical laws and data driving, the application realizes earlier and more accurate anomaly detection than traditional methods.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power, and particularly relates to a GRU-based power Internet of Things time series data anomaly detection system. BACKGROUND

[0002] In recent years, with the large-scale development of distribution network planning projects and reconstruction projects, more and more newly manufactured cable joints are put into use. However, due to the decline in material quality and the reduction in project budget, as well as the uneven management and technical process of cable joint manufacturing quality, the failure rate of cable joints is getting higher and higher.

[0003] Therefore, it is necessary to monitor whether the cable joint is abnormal to avoid the failure of other components of the entire distribution network caused by the abnormal cable joint. SUMMARY

[0004] The present application provides a GRU-based power Internet of Things time series data anomaly detection system, which mainly aims to realize earlier and more accurate anomaly detection than traditional methods through deep fusion of physical laws and data-driven.

[0005] The present application provides a GRU-based power Internet of Things time series data anomaly detection system, which is used for anomaly detection of cable joints in a distribution network. The anomaly detection system comprises a processor, which realizes anomaly detection through the following steps:

[0006] The instantaneous current of the cable joint, the real-time temperature of the cable joint and the ambient temperature of the cable joint are received, and the real-time temperature is sampled through a sliding window to obtain an original temperature sequence;

[0007] According to the cable trench topology structure of the distribution network and the real-time temperature, the adjacent node temperature difference and the topology weighted temperature of the cable joint are obtained, and further, the spatial contrast feature is obtained;

[0008] According to the time domain feature and the frequency domain feature of the real-time temperature and the spatial contrast feature, the statistical feature of the cable joint is obtained;

[0009] According to the instantaneous current, the real-time temperature, the ambient temperature, the original temperature sequence and the statistical feature, the improved GRU model is combined to obtain the anomaly detection result of the cable joint;

[0010] The improved GRU model is a dual-path heterogeneous GRU model including a physical path and a data path, the physical path is used to obtain a first temperature prediction value of the cable joint based on a thermal dynamic equation, and the data path is used to obtain a second temperature prediction value of the cable joint based on temperature change acceleration, so that the improved GRU model obtains the anomaly detection result according to the first temperature prediction value and the second temperature prediction value.

[0011] Further, the improved GRU model includes a physical path layer, a data path layer, a fusion layer, a residual layer, and a fault classification layer, and the anomaly detection result of the cable joint is obtained according to the instantaneous current, the real-time temperature, the environmental temperature, the original temperature sequence, and the statistical characteristics, and the improved GRU model, and the steps include:

[0012] The instantaneous current and the environmental temperature are input into the physical path layer, and the first temperature prediction value is obtained by combining a thermal dynamic equation;

[0013] The original temperature sequence and the statistical characteristics are input into the data path layer, and the second temperature prediction value is obtained by combining temperature change acceleration;

[0014] The first temperature prediction value and the second temperature prediction value are input into the fusion layer to obtain a temperature prediction value;

[0015] The real-time temperature and the temperature prediction value are input into the residual layer to calculate an instantaneous residual and a cumulative residual;

[0016] The instantaneous residual and the cumulative residual are input into the fault classification layer to obtain a detection type of the cable joint, and the detection type includes normal, contact oxidation, insulation damage, and external heat source.

[0017] Further, the instantaneous current is input into the physical path layer, and the first temperature prediction value is obtained by combining a thermal dynamic equation, and the calculation formula is as follows:

[0018]

[0019] wherein, ΔT phys represents the first temperature prediction value, t represents the current time, τ represents the integral time variable, k represents the thermal resistance coefficient, I(τ) represents the instantaneous current at τ, R represents the contact resistance of the cable joint, Δt represents the time step, α represents the heat dissipation coefficient, T(τ) represents the real-time temperature of the cable joint at τ, and T env (τ) represents the environmental temperature at τ.

[0020] Further, the original temperature sequence and the statistical characteristics are input into the data path layer, and a temperature change acceleration is combined to obtain the second temperature prediction value, and the calculation formula is as follows:

[0021]

[0022] r t = σ(W r · [h t-1 , x t ] + A·T neighbor );

[0023] Wherein, z t represents the output vector of the update gate in the data path layer at time t, t represents the current time, sigma represents the sigmoid function, W z represents the trainable weight matrix of the update gate, [h t-1 , x t ] represents the hidden state vector at t-1 and the input vector at t, D z represents the physical coefficient matrix, and T represents the original temperature sequence.

[0024] r t represents the output of the reset gate at time t in the data path layer, W r represents the trainable weight matrix of the reset gate, A represents the spatial attention matrix, and T neighbor represents the adjacent joint temperature vector of the cable joint.

[0025] Further, the calculation formula of the element a ij in the i-th row and j-th column of the spatial attention matrix is as follows:

[0026]

[0027] Wherein, a ij represents the contribution degree of the jth cable joint to the state update of the ith cable joint, W represents the feature transformation of learning temperature correlation mode, [T i ||T j ] represents the concatenation vector of the instantaneous temperature of the ith cable joint and the jth cable joint, LeakyReLU represents the linear rectifier unit with leakage, and m represents the number of all adjacent joints of the ith cable joint.

[0028] Further, the calculation method of the adjacent joint temperature vector is as follows:

[0029]

[0030] The initial value of the spatial attention matrix is as follows:

[0031]

[0032] wherein, T m (t) represents the real-time temperature of the mth adjacent joint of the cable joint at time t, represents the average real-time temperature of all adjacent joints, λ represents the thermal conductivity, and L represents the cable length.

[0033] Further, the step of inputting the instantaneous residual error and the cumulative residual error into the fault classification layer to obtain the detection type of the cable joint comprises:

[0034] If the cumulative residual error is greater than the cumulative residual error threshold, and the first derivative of the instantaneous residual error is greater than the residual error change rate, it is determined that the detection type of the cable joint is contact oxidation;

[0035] If the instantaneous residual error at the current time is greater than the instantaneous residual error threshold, and the second derivative of the instantaneous residual error is greater than 0, it is determined that the detection type of the cable joint is insulation damage;

[0036] If the temperature difference between the real-time temperature of the cable joint and the real-time temperature of the adjacent joint is within 2 degrees, it is determined that the detection type of the cable joint is an external heat source;

[0037] Otherwise, it is determined that the detection type of the cable joint is normal.

[0038] Further, the processor communicates with the pre-embedded micro-Rogowski coil, platinum resistance temperature sensor, and moisture-proof temperature sensor;

[0039] The micro-Rogowski coil is pre-embedded in the interior of the cable joint and is used to collect the instantaneous current;

[0040] The platinum resistance temperature sensor is pre-embedded in the interior of the cable joint and is used to collect the real-time temperature;

[0041] The moisture-proof temperature sensor is installed in the cable trench of the cable joint and is used to collect the environmental temperature.

[0042] Further, the adjacent node temperature difference and the topological weighted temperature of the cable joint are obtained according to the cable trench topological structure diagram of the power distribution network and the real-time temperature, and the calculation formula is as follows:

[0043]

[0044] T weighted =∑w ij T j ;

[0045] wherein, ΔT neighborindicates the temperature difference of adjacent nodes, T indicates the real-time temperature of the cable joint, T adjj indicates the real-time temperature of the jth adjacent joint, m indicates the number of all adjacent joints, T weighted indicates the topology weighted temperature, w ij indicates the weight coefficient of the ith cable joint and the jth adjacent joint, T j indicates the real-time temperature of the jth cable joint.

[0046] Further, the time domain feature and the frequency domain feature of the real-time temperature are obtained by the following steps:

[0047] A plurality of historical real-time temperatures are obtained through a historical sliding window;

[0048] The time domain feature is obtained according to the mean value, sliding variance, first-order difference mean value and peak-to-peak value of the historical real-time temperature;

[0049] The frequency domain feature is obtained by extracting the main frequency component amplitude and spectrum entropy of the historical real-time temperature.

[0050] The application proposes a power Internet of Things time series data anomaly detection system based on GRU. Various types of sensors are pre-buried in newly manufactured cable joints, and the sensors send the collected data to the processor. The processor extracts the original temperature sequence and spatial contrast features according to the instantaneous current, real-time temperature and environmental temperature. Then, the statistical features are obtained according to the time domain feature, frequency domain feature and spatial contrast feature of the real-time temperature. Finally, the instantaneous current, real-time temperature, environmental temperature, original temperature sequence and statistical features are input into the improved GRU model, and the anomaly detection result of the cable joint can be obtained.

[0051] The advantages of the embodiment of the application are as follows:

[0052] 1. The GRU model is improved by using physical and data paths, so that the improved GRU model includes physical and data paths. The physical path can predict the theoretical temperature rise of the cable joint through the thermodynamic equation, the data path can capture the temperature change acceleration of the cable joint through the GRU model, and the combination of the physical path and the data path makes the improved GRU model more sensitive to the temperature change of the cable joint, so that the anomaly of the cable joint can be captured more quickly and accurately.

[0053] 2. The adjacent node temperature difference and the topology weighted temperature are calculated through the cable trench topology structure diagram, the shared information of the spatial attention mechanism is established, and the spatial attention is fused into the reset gate of the data path. When the temperature of the adjacent joint abnormally rises, the reset gate can block the propagation of false memories. Moreover, when the cable joint is abnormal, the influence on the associated node model can be automatically reduced. Therefore, the false positive rate of the cable joint anomaly detection can be reduced. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 A structure schematic diagram of a GRU-based power internet of things time series data anomaly detection system provided by an embodiment of the present application is shown in the figure.

[0055] Figure 2 An anomaly detection flowchart of a cable joint provided by an embodiment of the present application is shown in the figure.

[0056] Figure 3 A structure diagram of an improved GRU model provided by an embodiment of the present application is shown in the figure.

[0057] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0058] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below are exemplary and are only used to explain the present application, and cannot be understood as a limitation on the present application.

[0059] In order to enable those skilled in the art to better understand the scheme of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0060] In the embodiments of the present application, at least one means one or more, and multiple means two or more than two. In the description of the present application, the terms “first”, “second”, “third” and the like are only used for distinguishing the purposes of description, and cannot be understood as indicating or implying relative importance, nor can be understood as indicating or implying order. In addition, the terms “first”, “second” are only used for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited by “first”, “second” can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of “multiple” is two or more than two, unless otherwise specifically limited.

[0061] References such as “one embodiment” or “some embodiments” as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the terms “comprising,” “including,” “having,” and variations thereof, as used in this specification, mean “including, but not limited to,” unless otherwise specifically emphasized.

[0062] Cable joints in power distribution networks are high-risk locations for faults, and their temperature time-series data exhibit the following characteristics:

[0063] (1) Multi-physics coupling: Temperature is affected by multiple factors such as current, contact resistance, ambient temperature and humidity.

[0064] (2) Nonlinear hysteresis effect: Heat accumulation causes temperature changes to lag behind load changes by about 15-30 minutes.

[0065] (3) Spatial correlation: There is a conductive correlation in the temperature of multiple joints within the same cable trench.

[0066] (4) Fault mode specific: Typical anomalies include contact surface oxidation (slow temperature rise) and insulation damage (sudden temperature rise).

[0067] Figure 1 A schematic diagram of a GRU-based power Internet of Things time-series data anomaly detection system is provided as an embodiment of the present invention, as shown below. Figure 1 As shown, this anomaly detection system is used to detect anomalies in cable joints in a power distribution network. In newly manufactured cable joints, a miniature Rogowski coil can be embedded inside to collect the instantaneous current of the cable joint. In addition, a moisture-proof temperature sensor is pre-embedded in the cable trench of the cable joint to collect the air temperature within a 1m range of the cable joint, which is the ambient temperature. Finally, a platinum resistance temperature sensor is embedded in the metal conductor of the cable joint to collect the real-time temperature of the cable joint.

[0068] In this solution, the processor of the anomaly detection system can communicate with the pre-embedded miniature Rogowski coil, platinum resistance temperature sensor, and moisture-proof temperature sensor, either wirelessly or via wired communication.

[0069] Figure 2 A flowchart for abnormal detection of a cable joint is provided as an embodiment of the present invention, as follows: Figure 2 As shown, the processor implements anomaly detection through the following steps:

[0070] S210, receive the instantaneous current of the cable joint, the real-time temperature of the cable joint, and the ambient temperature of the cable joint, and sample the real-time temperature through a sliding window to obtain the original temperature sequence;

[0071] Receiving the instantaneous current, real-time temperature and ambient temperature sent by each sensor.

[0072] After obtaining the real-time temperature corresponding to each historical moment and the current moment, the obtained real-time temperature is sampled through a sliding window. For example, the length of the sliding window is 10 minutes, and sampling is performed every 10 seconds. The real-time temperature corresponding to 60 sampling points can be obtained, which is the original temperature sequence.

[0073] S220, obtaining the adjacent node temperature difference and the topological weighted temperature of the cable joint according to the cable trench topological structure diagram of the power distribution network and the real-time temperature, and further obtaining the spatial contrast feature;

[0074] The cable trench topological diagram of the power distribution network is realized through multi-source data fusion and digital processing. The cable trench topological diagram is a topological relationship diagram, which specifically includes nodes and edges. The nodes represent cable joints, and the edges represent the connection relationship between the cable joints. Each node includes four key attributes of cable_id, length, material and adj_nodes. cable_id represents the ID of the cable joint, which is used to uniquely identify the cable joint. length represents the cable length, which is used to calculate the heat conduction weight. material represents the cable joint material, which determines the thermal conductivity k. adj_nodes represents adjacent nodes, which are used for spatial attention calculation. The cable trench topological diagram is the basis for spatial correlation analysis in temperature anomaly detection, and its accuracy directly affects the calculation reliability of the adjacent node temperature difference and the topological weighted temperature.

[0075] Specifically, the adjacent node temperature difference and the topological weighted temperature calculation formula is as follows:

[0076]

[0077] T weighted =∑w ij T j ;

[0078] Where ΔT neighbor represents the adjacent node temperature difference, T represents the real-time temperature of the cable joint, T adjj represents the real-time temperature of the jth adjacent joint, m represents the number of all adjacent joints, T weighted represents the topological weighted temperature, w ij represents the weight coefficient of the ith cable joint and the jth adjacent joint, and T j represents the real-time temperature of the jth cable joint.

[0079] It should be noted that when a certain cable joint is abnormal, the maximum temperature gradient direction along the cable trench is searched, and the expected temperature rise is calculated in combination with the cable carrying capacity, so that the fault source can be located. When a sensor in the cable joint fails, the data of the failed sensor can be reconstructed based on the interpolation relationship of the cable trench interpolation graph and the adjacent node data. The embodiment of the application calculates the temperature relationship between the cable joint and the adjacent joint through the cable trench interpolation graph, obtains the adjacent node temperature difference and the interpolation weighted temperature, and thus obtains the spatial contrast feature.

[0080] S230, obtaining statistical features of the cable joint according to the time domain features and the frequency domain features of the real-time temperature and the spatial contrast feature;

[0081] Specifically, a plurality of historical real-time temperatures are obtained through a historical sliding window.

[0082] The time domain features are obtained according to the mean value, sliding variance, first-order difference mean value and peak-to-peak value of the historical real-time temperature.

[0083] The frequency domain features are obtained by extracting the main frequency component amplitude and the spectrum entropy of the historical real-time temperature.

[0084] A plurality of historical real-time temperatures are collected through a historical sliding window, and the mean value, sliding variance, first-order difference mean value and peak-to-peak value of the historical real-time temperature are calculated, which together constitute the time domain features. Wherein:

[0085] Mean value

[0086] Sliding variance

[0087] First-order difference mean value

[0088] Peak-to-peak value max(T)-min(T).

[0089] Wherein, n represents the number of historical real-time temperatures, p is a positive integer, T p represents the pth historical real-time temperature, max(T) represents the maximum value of all historical real-time temperatures, and min(T) represents the minimum value of all historical real-time temperatures.

[0090] Finally, the main frequency component amplitude and the spectrum entropy are extracted through FFT (Fast Fourier Transform), and the main frequency component amplitude and the spectrum entropy constitute the frequency domain features. The main frequency component amplitude can reflect the load fluctuation frequency, and the spectrum entropy can represent the randomness of temperature change.

[0091] S240, obtaining an abnormality detection result of the cable joint according to the instantaneous current, the real-time temperature, the environment temperature, the original temperature sequence and the statistical feature, and combining the improved GRU model;

[0092] The improved GRU model is a dual-path heterogeneous GRU model including a physical path and a data path, the physical path is used to obtain a first temperature prediction value of the cable joint based on a thermal dynamic equation, and the data path is used to obtain a second temperature prediction value of the cable joint based on temperature change acceleration, so that the improved GRU model obtains the abnormality detection result according to the first temperature prediction value and the second temperature prediction value.

[0093] The instantaneous current, the real-time temperature, the environment temperature, the original temperature sequence and the statistical feature are input into the improved GRU (Gated Recurrent Units) model, and the abnormality detection result of the cable joint can be obtained. The improved GRU model is obtained after training with samples and labels.

[0094] Specifically, Figure 3 A structure diagram of an improved GRU model provided by the embodiment of the application is shown in Figure 3 The improved GRU model includes a physical path layer, a data path layer, a fusion layer, a residual layer and a fault classification layer. The step of obtaining the abnormality detection result of the cable joint according to the instantaneous current, the real-time temperature, the environment temperature, the original temperature sequence and the statistical feature and combining the improved GRU model includes:

[0095] The instantaneous current and the environment temperature are input into the physical path layer, and the first temperature prediction value is obtained by combining a thermal dynamic equation;

[0096] The original temperature sequence and the statistical feature are input into the data path layer, and the second temperature prediction value is obtained by combining temperature change acceleration;

[0097] The first temperature prediction value and the second temperature prediction value are input into the fusion layer, and a temperature prediction value is obtained;

[0098] The real-time temperature and the temperature prediction value are input into the residual layer, and instantaneous residual and cumulative residual are calculated;

[0099] The instantaneous residual and the cumulative residual are input into the fault classification layer, and a detection type of the cable joint is obtained, and the detection type includes normal, contact oxidation, insulation damage and external heat source.

[0100] First, the instantaneous current and the ambient temperature are input to the physical channel layer, and based on the thermal dynamic equation of power, the first temperature prediction value of the cable joint is calculated based on the physical rules. Then the original temperature sequence and statistical characteristics are input into the data channel layer, a second derivative gate is introduced in the data channel layer, the acceleration of temperature change is captured through the second derivative gate, and the second temperature prediction value of the cable joint is calculated according to the acceleration of temperature change.

[0101] Among them, the data channel layer includes an update gate and a reset gate, the second derivative gate is added in the update gate to capture the acceleration of temperature change of the cable joint, and the spatial attention is introduced in the reset gate, when the temperature of the adjacent node abnormally rises, the propagation of false memory can be blocked, and when the temperature of the adjacent node is normal, the behavior of the standard GRU model is maintained, so that the false positive rate of the cable joint anomaly detection is reduced.

[0102] In the formula of the update gate , W z ·[h t-1 ,x t ] is a data-driven part, which inherits the time series modeling capability of the standard GRU model, and captures the dynamic correlation of current, temperature and other parameters; is a physical enhancement part, when the temperature rises rapidly , the output Z t of the update gate is increased , the historical memory is reduced t , and the sudden abnormality is responded faster; when the temperature change is smooth, the behavior of the standard GRU model is maintained.

[0103] The present application encodes the thermodynamic law into the neural network through D z , and the sensitivity of the second derivative term to the mutation signal is higher than that of the original temperature value, which improves the early warning capability; when the temperature is smooth, the influence of the physical term is automatically weakened to avoid overfitting of the measurement noise. The second derivative term of the update gate enhances the modeling capability of the traditional GRU model for the thermal dynamic characteristics of the power equipment through the differential equation, so that the temperature acceleration is more sensitive, and the early warning of the cable joint is realized. Compared with the traditional method, it can be detected faster.

[0104] In the reset gate formula, W r ·[h t-1 ,x t ] is a data-driven part, which inherits the time series memory management function of the standard GRU model; A·T neighbor is a spatial attention part, when the temperature of the adjacent node abnormally rises, the output r tForced forgetting of abnormal historical state—>Block false memory propagation; when the temperature of the adjacent node is normal, the behavior of the standard GRU model is maintained. Thus, false propagation of the adjacent joint being abnormally hot can be prevented, and when the cable joint is abnormal, the influence on the model of its associated node can be automatically reduced. The physical connection relationship in the power distribution network is also converted into an attention mechanism, which can reduce the false positive rate caused by heat conduction by up to 35% compared with traditional methods.

[0105] In the spatial attention matrix A, the calculation formula of each element is as follows:

[0106]

[0107] wherein a ij represents the contribution of the jth cable joint to the state update of the ith cable joint, W represents a feature transformation that learns the temperature correlation pattern, [T i ||T j represents the concatenation vector of the instantaneous temperatures of the ith cable joint and the jth cable joint, which can represent the absolute temperatures and relative temperature differences of the current node and the adjacent nodes at the same time; LeakyReLU represents a leaky rectified linear unit, which can preserve negative temperature difference information and prevent excessive suppression of normal temperature fluctuations; m represents all adjacent joints of the ith cable joint.

[0108] When T j abnormally increases—>α ij increases—>strengthen the abnormal propagation path, when T j is normal—>α ij decreases—>weaken irrelevant interference. Thus, false propagation of the adjacent node being abnormally hot can be prevented, and when a certain cable joint is abnormal, the influence of the cable joint on the model of its associated node can be automatically reduced. The physical connection relationship in the cable network can also be converted into a trainable attention mechanism.

[0109] The first temperature prediction value and the second temperature prediction value are then input to a fusion layer for weighted fusion to obtain a temperature prediction value, and the weight of the first temperature prediction value is dynamically adjusted according to the current load change rate. The real-time temperature and the temperature prediction value are input to a residual layer to calculate an instantaneous residual and a cumulative residual, wherein the instantaneous residual is the absolute difference between the real-time temperature and the temperature prediction value, and the cumulative residual is the sum of all instantaneous residuals. Finally, the instantaneous residual and the cumulative residual are input to a fault classification layer to obtain the detection type of the cable joint, which includes four types: normal, contact oxidation, insulation damage, and external heat source, all of which indicate that the cable joint is abnormal.

[0110] If the accumulated residual error is greater than the accumulated residual error threshold value, and the first derivative of the instantaneous residual error is greater than the residual error change rate, it is determined that the detection type of the cable joint is contact oxidation;

[0111] If the instantaneous residual error at the current moment is greater than the instantaneous residual error threshold value, and the second derivative of the instantaneous residual error is greater than 0, it is determined that the detection type of the cable joint is insulation damage;

[0112] If the temperature difference between the real-time temperature of the cable joint and the real-time temperature of the adjacent joint is within 2 degrees, it is determined that the detection type of the cable joint is external heat source;

[0113] Otherwise, it is determined that the detection type of the cable joint is normal.

[0114] Table 1

[0115]

[0116] Table 1 is a comparison table of fault type detection effects between the improved GRU model and the traditional standard GRU model in the embodiment of the present application. As shown in Table 1, the accuracy of the improved GRU model in the present scheme for the three abnormal types of increased contact resistance, insulation partial discharge and external heat source interference is increased, and the improved GRU model has better effect.

[0117] Table 2

[0118]

[0119] Table 2 is a comparison table of comprehensive detection effects between the improved GRU model and the traditional standard GRU model in the embodiment of the present application. As shown in Table 2, it can be seen that the present scheme has better performance effect in detection speed, false alarm rate and fault classification.

[0120] In summary, the present scheme realizes the following effects through deep coupling of physical laws and data-driven:

[0121] Earlier: physical equations provide prior knowledge to amplify weak signals more accurately;

[0122] More accurate: spatial-time double verification suppresses false alarms;

[0123] More intelligent: adaptive mechanism reduces manual intervention.

[0124] The above-mentioned various modules in the GRU-based power Internet of Things time series data anomaly detection system can be realized by software, hardware and their combinations. The above-mentioned various modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to call and execute the operations corresponding to the above-mentioned various modules by the processor.

[0125] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0126] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of functional units and modules is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the above-described functions.

[0127] The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, but not limit it. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features. The modification or replacement does not make the essence of the corresponding technical solution deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A power Internet of Things (IoT) time-series data anomaly detection system based on GRU, characterized in that, The anomaly detection system is used to detect anomalies in cable joints in the power distribution network. The anomaly detection system includes a processor, which performs anomaly detection through the following steps: The instantaneous current of the cable joint, the real-time temperature of the cable joint, and the ambient temperature of the cable joint are received, and the original temperature sequence is obtained by sampling the real-time temperature through a sliding window. Based on the cable trench topology diagram of the power distribution network and the real-time temperature, the temperature difference between adjacent nodes and the topology-weighted temperature of the cable joint are obtained, and spatial comparison features are further obtained. Based on the time-domain and frequency-domain characteristics of the real-time temperature and the spatial comparison characteristics, the statistical characteristics of the cable joint are obtained. Based on the instantaneous current, the real-time temperature, the ambient temperature, the original temperature sequence, and the statistical characteristics, combined with the improved GRU model, the abnormal detection results of the cable joint are obtained. The improved GRU model is a dual-path heterogeneous GRU model that includes a physical path and a data path. The physical path is used to obtain a first temperature prediction value of the cable joint based on the thermal dynamic equation, and the data path is used to obtain a second temperature prediction value of the cable joint based on the temperature change acceleration, so that the improved GRU model can obtain the anomaly detection result based on the first temperature prediction value and the second temperature prediction value. The temperature difference between adjacent nodes and the topology-weighted temperature of the cable joint are obtained based on the cable trench topology diagram of the power distribution network and the real-time temperature. The calculation formula is as follows: T weighted =∑w ij T j ; Where, ΔT neighbor T represents the temperature difference between adjacent nodes, and T represents the real-time temperature of the cable joint. adjj T represents the real-time temperature of the j-th adjacent joint, m represents the total number of adjacent joints, and T represents the temperature of the j-th adjacent joint. weighted The topological weighted temperature, w ij T represents the weighting coefficient between the i-th cable joint and the j-th adjacent joint. j This represents the real-time temperature of the j-th cable connector.

2. The power IoT time-series data anomaly detection system based on GRU according to claim 1, characterized in that, The improved GRU model includes a physical path layer, a data path layer, a fusion layer, a residual layer, and a fault classification layer. The steps for obtaining the anomaly detection result of the cable joint based on the instantaneous current, the real-time temperature, the ambient temperature, the original temperature sequence, and the statistical characteristics, combined with the improved GRU model, include: The instantaneous current and the ambient temperature are input into the physical pathway layer, and combined with the thermal dynamics equation, the first temperature prediction value is obtained. The original temperature sequence and the statistical features are input into the data path layer, and combined with the temperature change acceleration, the second temperature prediction value is obtained. The first temperature prediction value and the second temperature prediction value are input into the fusion layer to obtain the temperature prediction value; The real-time temperature and the predicted temperature are input into the residual layer to calculate the instantaneous residual and the cumulative residual. The instantaneous residual and the cumulative residual are input into the fault classification layer to obtain the detection type of the cable joint. The detection type includes normal, contact oxidation, insulation damage, and external heat source.

3. The power IoT time-series data anomaly detection system based on GRU according to claim 2, characterized in that, The instantaneous current is input into the physical path layer, and combined with the thermal dynamics equation, the first temperature prediction value is obtained. The calculation formula is as follows: Where, ΔT phys Let represent the first predicted temperature value, t represent the current time, τ represent the integration time variable, k represent the thermal resistance coefficient, I(τ) represent the instantaneous current at time τ, R represent the contact resistance of the cable joint, Δt represent the time step, α represent the heat dissipation coefficient, and T(τ) represent the real-time temperature of the cable joint at time τ. env (τ) represents the ambient temperature at time τ.

4. The power IoT time-series data anomaly detection system based on GRU according to claim 2, characterized in that, The original temperature sequence and the statistical features are input into the data path layer, and combined with the temperature change acceleration, the second temperature prediction value is obtained. The calculation formula is as follows: r t =σ(W r ·[h t-1 ,x t ]+A·T neighbor ); Among them, z t This represents the output vector of the update gate in the data path layer at time t, where t represents the current time, σ represents the sigmoid function, and W... z Let [h] represent the trainable weight matrix of the update gate. t-1 ,x t ] represents the concatenation of the hidden state vector at time t-1 and the input vector at time t, D z represents the physical coefficient matrix, and T represents the original temperature sequence; r t W represents the output of the reset gate at time t in the data path layer. r Let A represent the trainable weight matrix of the reset gate, and let T represent the spatial attention matrix. neighbor This represents the adjacent joint temperature vector of the cable joint.

5. The power IoT time-series data anomaly detection system based on GRU according to claim 4, characterized in that, The element a in the i-th row and j-th column of the spatial attention matrix ij The calculation formula is as follows: Among them, a ij The value of W represents the contribution of the j-th cable joint to the state update of the i-th cable joint, and W represents the feature transformation of the learned temperature correlation pattern. i ||T j ] represents the spliced ​​vector of the instantaneous temperatures of the i-th cable connector and the j-th cable connector, LeakyReLU represents the linear rectifier unit with leakage, and m represents the number of all adjacent connectors of the i-th cable connector.

6. The power IoT time-series data anomaly detection system based on GRU according to claim 4, characterized in that, The temperature vector of the adjacent joint is calculated as follows: Among them, T m (t) represents the real-time temperature of the m-th adjacent connector of the cable joint at time t. This represents the average real-time temperature of all adjacent joints.

7. The power IoT time-series data anomaly detection system based on GRU according to claim 2, characterized in that, The step of inputting the instantaneous residual and the cumulative residual into the fault classification layer to obtain the detection type of the cable joint includes: If the cumulative residual is greater than the cumulative residual threshold, and the first derivative of the instantaneous residual is greater than the residual change rate, then the detection type of the cable joint is determined to be contact oxidation. If the instantaneous residual is greater than the instantaneous residual threshold and the second derivative of the instantaneous residual is greater than 0, then the detection type of the cable joint is determined to be insulation failure. The instantaneous residual is the absolute difference between the real-time temperature at the current moment and the predicted temperature value. If the temperature difference between the real-time temperature of the cable joint and the real-time temperature of the adjacent joint is within 2 degrees, then the detection type of the cable joint is determined to be an external heat source. Otherwise, the detection type of the cable connector is determined to be normal.

8. The power IoT time-series data anomaly detection system based on GRU according to claim 1, characterized in that, The processor communicates with the pre-embedded miniature Rogowski coil, platinum resistance temperature sensor, and moisture-proof temperature sensor. The miniature Rogowski coil is embedded inside the cable connector and is used to collect the instantaneous current; The platinum resistance temperature sensor is embedded inside the cable connector and is used to collect the real-time temperature. The moisture-proof temperature sensor is installed in the cable trench of the cable joint to collect the ambient temperature.

9. The power Internet of Things time-series data anomaly detection system based on GRU according to claim 1, characterized in that, The time-domain and frequency-domain characteristics of the real-time temperature are obtained through the following steps: Multiple historical real-time temperatures can be obtained through a historical sliding window; The time-domain features are obtained based on the mean, sliding variance, first-order difference mean, and peak-to-peak value of the historical real-time temperatures. Extract the amplitude of the dominant frequency component and the spectral entropy of the historical real-time temperature to obtain the frequency domain features.

Images