An insulator discharge monitoring system

Abstract

The present invention relates to an insulator discharge monitoring system, which belongs to the technical field of insulators and solves the problems of low monitoring accuracy and efficiency in the prior art. The system comprises: an electrical acquisition unit, which is used to acquire leakage current and corona pulse current on the surface of the insulator to obtain electrical data of the insulator; an optical acquisition unit, which is used to acquire ultraviolet light signals of the insulator and images of the surface of the insulator to obtain optical data of the insulator; an acoustic acquisition unit, which is used to acquire acoustic signals of the insulator to obtain acoustic data of the insulator; a thermal acquisition unit, which is used to acquire thermal imaging images of the insulator to obtain thermal data of the insulator; and a monitoring unit, which is used to align the electrical data, optical data, acoustic data and thermal data of the insulator to obtain an aligned multi-dimensional time series signal, and input the aligned multi-dimensional time series signal into a trained insulator discharge prediction model to predict the discharge state of the insulator. Efficient and accurate insulator discharge monitoring is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of insulators, and in particular to an insulator discharge monitoring system. Background Art

[0002] Insulators are devices installed between conductors at different potentials or between conductors and grounded components. Their main functions are to achieve electrical insulation and mechanical fixation. There is a special physical phenomenon in the insulation system composed of vacuum and insulators: the withstand voltage of the insulator in a vacuum environment will be greatly reduced, even lower than the insulation strength of the vacuum gap or solid insulator itself at the same distance. The root cause is that a penetrating electrical breakdown process occurs on the surface of the insulator along the vacuum-insulator interface, namely, surface flashover. The surface flashover phenomenon of vacuum insulation in high-voltage vacuum equipment not only seriously restricts the withstand voltage of the equipment, affects the normal operation of related equipment, but may also cause huge economic losses. With the development of large-scale new voltage level power grid construction, higher requirements are placed on the performance of insulators. At present, there is a wide range of research on insulators. The main goal is to be able to reliably discover the abnormal state of insulators at an earlier stage. The research on surface discharge and detection and diagnosis technology of insulators is one of the main research directions.

[0003] Currently, single-dimensional data such as images or current are usually used for monitoring, which cannot fully and accurately evaluate the status of insulators, resulting in low monitoring accuracy and efficiency. Summary of the invention

[0004] In view of the above analysis, an embodiment of the present invention aims to provide an insulator discharge monitoring system to solve the problems of low accuracy and efficiency of existing monitoring.

[0005] On the one hand, an embodiment of the present invention provides an insulator discharge monitoring system, including:

[0006] An electrical data acquisition unit is used to collect leakage current and corona pulse current on the surface of the insulator to obtain electrical data of the insulator;

[0007] A light collection unit, used for collecting ultraviolet light signals of the insulator and the surface image of the insulator to obtain light data of the insulator;

[0008] A sound collection unit, used for collecting sound signals of the insulator to obtain sound data of the insulator;

[0009] A thermal collection unit, used for collecting thermal imaging images of the insulator to obtain thermal data of the insulator;

[0010] The monitoring unit is used to align the electrical data, optical data, acoustic data and thermal data of the insulator to obtain an aligned multi-dimensional time series signal, and input the aligned multi-dimensional time series signal into a trained insulator discharge prediction model to predict the discharge state of the insulator.

[0011] Based on the further improvement of the above system, the electric collection unit includes a non-contact through-core leakage current sensor and a pulse current detection device for respectively collecting the leakage current and the corona pulse current on the surface of the insulator;

[0012] The light collection unit includes an ultraviolet light signal collection circuit and a high-resolution camera, which are used to collect the ultraviolet light signal and surface image of the insulator respectively;

[0013] The heat collection unit comprises a thermal imager for collecting thermal imaging images of the insulator.

[0014] Based on the further improvement of the above system, the ultraviolet light signal acquisition circuit includes an optical lens, a photoelectric conversion circuit, a current-voltage conversion circuit and a sampling circuit;

[0015] The optical lens is used to input ultraviolet light into the photoelectric conversion circuit; the photoelectric conversion circuit converts the optical signal into a current signal;

[0016] The current-to-voltage conversion circuit is used to convert the current signal output by the photoelectric conversion circuit into a voltage signal;

[0017] The sampling circuit is used to sample the voltage signal output by the current-voltage conversion circuit to obtain an ultraviolet light signal.

[0018] Based on the further improvement of the above system, the trained insulator discharge prediction model is obtained in the following way:

[0019] Collect the electrical data, optical data, acoustic data and thermal data of the historical state of the insulator, align the electrical data, optical data, acoustic data and thermal data to obtain the aligned multi-dimensional time series signal, and annotate the discharge progress and discharge current value of the multi-dimensional time series signal to obtain a training sample set;

[0020] Each sample in the training sample set includes a multi-dimensional time series signal, a discharge progress and a discharge current value;

[0021] A multidimensional time series prediction model is constructed, and the multidimensional time series prediction model is trained based on the constructed training sample set to obtain a trained insulator discharge prediction model.

[0022] Based on the further improvement of the above system, the following formula is used to calculate the loss of the multi-dimensional time series prediction network:

[0023]

[0024] Where N represents the number of samples, z i1 represents the discharge progress of the sample, represents the discharge progress predicted by the model, z i2Indicates the peak value of the leakage current of the next discharge of the sample, Represents the peak value of the leakage current of the next discharge predicted by the model.

[0025] Based on further improvement of the above system, the multi-dimensional time series prediction model includes a feature extraction module and a prediction module, the feature extraction module is used to extract sample features; the prediction module is used to predict the discharge progress and the peak value of the leakage current of the next discharge based on the extracted sample features.

[0026] Based on the further improvement of the above system, the feature extraction module includes an input layer, multiple intermediate layers and an output layer;

[0027] The input layer is used to input sample data at each time step;

[0028] The first intermediate layer calculates the voltage and output signal of this layer according to the output data of the input layer; each of the remaining intermediate layers calculates the voltage and output signal of this layer based on the voltage and output signal of the previous intermediate layer; a feedback connection is set between the last intermediate layer and the first intermediate layer, which is used to transmit the output signal of the last intermediate layer in the previous time step to the first intermediate layer in the next time step;

[0029] The output layer is used to obtain sample features according to the output signal of the last intermediate layer.

[0030] Based on the further improvement of the above system, the output layer uses the following formula to calculate the sample features:

[0031]

[0032] Where T represents the total number of time steps of the sample, s H [t] represents the output signal of the Hth intermediate layer at the tth time step, and λ represents the leakage parameter of the neurons in the intermediate layer.

[0033] Based on the further improvement of the above system, the voltage of the first intermediate layer is calculated using the following formula:

[0034]

[0035] Among them, u 1 [t] represents the voltage of the first intermediate layer at the tth time step, represents the temporary voltage of the first intermediate layer at the tth time step, V th Represents the voltage threshold. If The i-th element in is greater than V th , then s 1 The i-th element of [t] is 1, otherwise s 1 The i-th element of [t] is 0; s1 [t] represents the output signal of the first intermediate layer at the tth time step;

[0036] The temporary voltage is calculated using the following formula:

[0037]

[0038] Among them, u 1 [t-1] represents the voltage of the first intermediate layer at the t-1th time step, s H [t-1] represents the output signal of the middle layer of the Hth layer at the t-1th time step, y[t] represents the sample data input at the tth time step, λ represents the leakage parameter of the neuron, W 1 represents the weight matrix of the feedback connection between the last intermediate layer and the first intermediate layer, F 1 Represents the weight matrix from the input layer to the first intermediate layer, V th represents the voltage threshold, b 1 represents the bias vector of the first intermediate layer, and H represents the total number of intermediate layers.

[0039] Based on the further improvement of the above system, the electrical data, optical data, acoustic data and thermal data are aligned to obtain the aligned multi-dimensional time series signal, and the discharge progress and discharge current value are annotated on the multi-dimensional time series signal to obtain a training sample set, including:

[0040] Performing dimensionality reduction processing on the electrical data, optical data and thermal data of the insulator respectively, and converting the electrical data, optical data and thermal data into one-dimensional electrical signals, one-dimensional optical signals and one-dimensional thermal signals respectively;

[0041] Respectively detecting the discharge starting time of a one-dimensional electrical signal, a one-dimensional optical signal, a one-dimensional thermal signal, and a one-dimensional acoustic signal, and aligning the one-dimensional electrical signal, the one-dimensional optical signal, the one-dimensional thermal signal, and the one-dimensional acoustic signal based on the discharge starting time to obtain a multi-dimensional time series signal;

[0042] A sliding window is used to extract T time steps of multidimensional time series signals as the input data of the sample, and the discharge progress of the data in the sliding window is calculated. The discharge progress and the peak value of the leakage current of the next discharge are used as the labels of the training samples to construct a training sample set.

[0043] Compared with the prior art, the present invention collects the electrical, optical, acoustic and thermal data of the insulator respectively through the electrical collection unit, the optical collection unit, the acoustic collection unit and the thermal collection unit, and aligns the collected electrical, optical, acoustic and thermal data through the monitoring unit to obtain multi-dimensional time series data. Then, based on the trained insulator discharge prediction model, the insulator discharge state can be quickly and accurately predicted, thereby comprehensively and accurately monitoring the insulator state, thereby ensuring the safe operation of the power grid system.

[0044] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. In the entire drawings, the same reference symbols represent the same components;

[0046] Figure 1 The block diagram of the insulator discharge monitoring system according to the embodiment of the present invention. DETAILED DESCRIPTION

[0047] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0048] A specific embodiment of the present invention discloses an insulator discharge monitoring system, such as Figure 1 As shown, including:

[0049] An electrical data acquisition unit is used to collect leakage current and corona pulse current on the surface of the insulator to obtain electrical data of the insulator;

[0050] A light collection unit, used for collecting ultraviolet light signals of the insulator and the surface image of the insulator to obtain light data of the insulator;

[0051] A sound collection unit, used for collecting sound signals of the insulator to obtain sound data of the insulator;

[0052] A thermal collection unit, used for collecting thermal imaging images of the insulator to obtain thermal data of the insulator;

[0053] The monitoring unit is used to align the electrical data, optical data, acoustic data and thermal data of the insulator to obtain an aligned multi-dimensional time series signal, and input the aligned multi-dimensional time series signal into a trained insulator discharge prediction model to predict the discharge state of the insulator.

[0054] It should be noted that predicting the discharge state of the insulator includes predicting the discharge progress of the insulator and the next discharge current amplitude.

[0055] Insulator discharge is accompanied by electrical, optical, acoustic and thermal phenomena. Compared with the prior art, the insulator discharge monitoring system provided in this embodiment collects electrical, optical, acoustic and thermal data of the insulator through an electrical collection unit, an optical collection unit, an acoustic collection unit and a thermal collection unit respectively, and aligns the collected electrical, optical, acoustic and thermal data through the monitoring unit to obtain multi-dimensional time series data. Then, based on the trained insulator discharge prediction model, the insulator discharge progress can be quickly and accurately predicted, so that the insulator status can be fully and accurately monitored, thereby ensuring the safe operation of the power grid system.

[0056] During implementation, the monitoring unit can use an industrial cloud platform or an industrial Internet platform.

[0057] The leakage current passing through the surface of the insulator working under ideal conditions is small, which will not endanger the life of the insulator and the safe operation of the equipment. As the degree of aging or electric field distortion increases, a discharge arc will be generated locally and will extend further with the increase of the external electric field. At this time, the leakage current can reach several hundred milliamperes, which may cause flashover. Therefore, collecting leakage current helps to accurately monitor the state of the insulator. The insulation resistance of the aged insulator is very low, the loop impedance is small, the number of corona pulse currents is large, and the amplitude is large. Therefore, using corona pulse current helps to judge the state of the insulator.

[0058] During implementation, the electric collection unit includes a non-contact through-core leakage current sensor and a pulse current detection device for respectively collecting leakage current and corona pulse current on the surface of the insulator.

[0059] The electrical collection unit uses electrical sensors. Since the through-core current sensor has installation positions on several connecting fittings of the insulator and the tower, it is convenient to realize live operation, so the through-core leakage current sensor is used to collect the leakage current on the surface of the insulator. The through-core leakage current sensor is set on the surface of the insulator to intercept the leakage current on the surface of the insulator. In order to make the copper ring and the insulator fully contact, conductive glue can be applied during installation, which also plays a fixing role.

[0060] During implementation, the pulse current detection device is an electrical sensor that uses a pulse current method to detect current.

[0061] Insulators generate heat when they discharge, so collecting thermal data helps determine the state of the insulators. When implemented, the thermal collection unit includes a thermal imager.

[0062] The wavelength of the light signal radiated by the insulator during discharge is in the ultraviolet band, so the insulation state is determined by detecting ultraviolet light. The light collection unit includes an ultraviolet light signal collection circuit and a high-resolution camera, which are used to collect the ultraviolet light signal and surface image of the insulator respectively.

[0063] Specifically, the ultraviolet light signal acquisition circuit includes an optical lens, a photoelectric conversion circuit, a current-voltage conversion circuit and a sampling circuit;

[0064] The optical lens is used to input the ultraviolet light into the photoelectric conversion circuit; the photoelectric conversion circuit converts the light signal into a current signal;

[0065] The current-to-voltage conversion circuit is used to convert the current signal output by the photoelectric conversion circuit into a voltage signal;

[0066] The sampling circuit is used to sample the voltage signal output by the current-voltage conversion circuit to obtain an ultraviolet light signal.

[0067] During implementation, an optical lens is used to image the ultraviolet light and send it to a photoelectric conversion circuit, where the optical lens is coated with a UV-enhanced transmission film that allows only ultraviolet light to pass through. The photoelectric conversion circuit converts the ultraviolet light signal into an electrical signal, which is then converted into a voltage and sampled to obtain the ultraviolet light signal. A high-resolution camera is used to collect image data at different angles of the insulator.

[0068] Under different discharge modes, the energy, amplitude and frequency of the acoustic emission signal of the insulator during the pollution discharge are quite different, so collecting the acoustic signal is helpful to detect the state of the insulator. During implementation, an acoustic collection unit is used to collect the acoustic signal of the insulator.

[0069] For data of different dimensions, the collection methods and equipment used are different, and the response times of various devices are difficult to be completely consistent. This will lead to differences in the representation and response time of the collected electrical, optical, acoustic and thermal data. Therefore, data alignment is performed first to obtain an aligned multi-dimensional timing signal, which facilitates accurate detection of the discharge progress of the insulator.

[0070] During implementation, the electrical data, optical data and thermal data of the insulator are respectively subjected to dimensionality reduction processing, and the electrical data, optical data and thermal data are respectively converted into one-dimensional electrical signals, one-dimensional optical signals and one-dimensional thermal signals, and then the one-dimensional acoustic signals, one-dimensional optical signals and one-dimensional thermal signals are respectively delayed according to the measured delay of the optical acquisition unit, the acoustic acquisition unit and the thermal acquisition relative to the electrical acquisition unit, so as to achieve data alignment and obtain an aligned multi-dimensional time series signal. The one-dimensional acoustic signals, one-dimensional optical signals and one-dimensional thermal signals can also be respectively delayed according to the delay of the one-dimensional acoustic signals, one-dimensional optical signals and one-dimensional thermal signals relative to the one-dimensional electrical signals calculated according to the following steps, so as to achieve data alignment and obtain an aligned multi-dimensional time series signal.

[0071] Then, the multi-dimensional signal of T time steps is extracted and input into the trained insulator discharge prediction model to predict the discharge state of the insulator.

[0072] Specifically, the trained insulator discharge prediction model is obtained in the following manner:

[0073] Collect the electrical data, optical data, acoustic data and thermal data of the historical state of the insulator, align the electrical data, optical data, acoustic data and thermal data to obtain the aligned multi-dimensional time series signal, and annotate the discharge progress and discharge current value of the multi-dimensional time series signal to obtain a training sample set;

[0074] Each sample in the training sample set includes a multi-dimensional time series signal, a discharge progress and a discharge current value;

[0075] A multidimensional time series prediction model is constructed, and the multidimensional time series prediction model is trained based on the constructed training sample set to obtain a trained insulator discharge prediction model.

[0076] During implementation, electrical data, optical data, acoustic data, and thermal data of the insulator from the start of operation until the flashover occurs are collected, and a training sample set is constructed based on the collected data for model training.

[0077] Specifically, the electrical data, optical data, acoustic data and thermal data are aligned to obtain an aligned multi-dimensional time series signal, and the discharge progress and discharge current value are annotated on the multi-dimensional time series signal to obtain a training sample set, including:

[0078] Performing dimensionality reduction processing on the electrical data, optical data and thermal data of the insulator respectively, and converting the electrical data, optical data and thermal data into one-dimensional electrical signals, one-dimensional optical signals and one-dimensional thermal signals respectively;

[0079] Respectively detecting the discharge starting time of a one-dimensional electrical signal, a one-dimensional optical signal, a one-dimensional thermal signal, and a one-dimensional acoustic signal, and aligning the one-dimensional electrical signal, the one-dimensional optical signal, the one-dimensional thermal signal, and the one-dimensional acoustic signal based on the discharge starting time to obtain a multi-dimensional time series signal;

[0080] A sliding window is used to extract T time steps of multidimensional time series signals as the input data of the sample, and the discharge progress of the data in the sliding window is calculated. The discharge progress and the peak value of the leakage current of the next discharge are used as the labels of the training samples to construct a training sample set.

[0081] For data of different dimensions, the acquisition methods and acquisition devices used are different, and the response time of various devices is difficult to be completely consistent, which will lead to differences in the representation form and response time of each signal collected, making it difficult to obtain various types of signals at the same time for comparative analysis. The existing manual alignment method is inefficient. Therefore, the present invention performs data alignment on multi-dimensional time series data based on discharge time, thereby improving the efficiency of data processing.

[0082] During implementation, the electrical data, optical data and thermal data collected at each collection moment are multiple data amounts. Therefore, the electrical data, optical data and thermal data of the insulator are respectively subjected to dimensionality reduction processing, and the electrical data, optical data and thermal data are respectively converted into one-dimensional electrical signals, one-dimensional optical signals and one-dimensional thermal signals.

[0083] For example, the leakage current signal and the corona pulse current signal of the electrical data are converted into a one-dimensional vector by principal component analysis to obtain a one-dimensional electrical signal.

[0084] For optical data, the ultraviolet pulse signal at each acquisition moment and the RGB value of the pixel in the insulator image are taken as multidimensional data, and principal component analysis is used to reduce the dimension and thus convert them into one-dimensional optical signals.

[0085] Similarly, the thermal value of the pixel point in the thermal imaging data at each acquisition moment is taken as multidimensional data, and principal component analysis is used to reduce the dimension and convert it into a one-dimensional thermal signal.

[0086] Since the acquired acoustic data is a one-dimensional signal, no processing is required.

[0087] Since the acquisition frequencies of electrical, optical, acoustic and thermal data are different, they need to be resampled and converted to the same frequency for alignment. In implementation, the acquisition frequency of electrical signals is high and the discharge moment is more obvious, so the one-dimensional signals of other dimensions can be resampled to the frequency of electrical signals and then aligned.

[0088] Specifically, the discharge start time of the one-dimensional electrical signal, the one-dimensional optical signal, the one-dimensional thermal signal and the one-dimensional acoustic signal is detected respectively, and the one-dimensional electrical signal, the one-dimensional optical signal, the one-dimensional thermal signal and the one-dimensional acoustic signal are aligned based on the discharge start time, including:

[0089] For each one-dimensional signal, a sliding window is used to extract signal segments in sequence, a fluctuation degree value and a fluctuation difference value of each signal segment are calculated, and based on the fluctuation degree value and the fluctuation difference value, it is determined whether the signal segment is a discharge signal segment;

[0090] If there are multiple continuous discharge signal segments, the multiple continuous discharge signal segments are taken as one discharge signal segment, the peak position of each discharge signal segment is determined, and the discharge start time in each discharge signal segment is determined based on the peak position and signal differential;

[0091] The delay amount of the one-dimensional optical signal, the one-dimensional thermal signal and the one-dimensional acoustic signal relative to the one-dimensional electrical signal is determined based on the discharge start time of each one-dimensional signal; and the one-dimensional optical signal, the one-dimensional thermal signal and the one-dimensional acoustic signal are adjusted based on the delay amount so that the one-dimensional optical signal, the one-dimensional thermal signal and the one-dimensional acoustic signal are aligned with the one-dimensional electrical signal.

[0092] Specifically, the fluctuation degree value is used to characterize the fluctuation degree of the signal segment compared with the reference signal segment, wherein the reference signal segment is the data when the insulator works normally and no over-discharge occurs. The reference signal is also a one-dimensional vector after the aforementioned dimensionality reduction processing. The electrical signal, optical signal, thermal signal and acoustic signal have corresponding reference signal segments respectively.

[0093] During implementation, for example, for a one-dimensional electrical signal, a sliding window is used to sequentially extract the signal segments of the one-dimensional electrical signal. During implementation, the shorter the length of the sliding window, the shorter the time interval for intercepting the timing data, the more accurate the predicted waveform alignment time will be, and the relative calculation cost will be higher. During implementation, the size of the sliding window can be determined by comprehensively considering the interference of fluctuations, the accuracy of waveform alignment, and the calculation efficiency. In order to accurately detect the start time of discharge, the sliding step length is set to a length that allows adjacent window keys to have 70% overlapping data. For the i-th step sliding, that is, the i-th signal segment, the fluctuation degree value of the signal segment is calculated by comparing it with the reference signal segment of the electrical signal.

[0094] Similarly, a sliding window is used to extract a signal segment of a one-dimensional optical signal and compare it with the reference signal segment of the optical signal to calculate the fluctuation degree value of the signal segment; a sliding window is used to extract a signal segment of a one-dimensional thermal signal and compare it with the reference signal segment of the thermal signal to calculate the fluctuation degree value of the signal segment; a sliding window is used to extract a signal segment of a one-dimensional acoustic signal and compare it with the reference signal segment of the acoustic signal to calculate the fluctuation degree value of the signal segment.

[0095] Specifically, the following formula is used to calculate the fluctuation value of each signal segment:

[0096]

[0097] Among them, x i,j represents the jth signal point of the i-th signal segment, n represents the length of the i-th signal segment, represents the jth signal point of the reference signal segment, sim min Indicates the minimum distance between all signal segments and the reference signal segment, sim max represents the maximum distance from the reference signal segment among all signal segments, ω1 and ω2 represent weight parameters, and their sum is 1.

[0098] When implementing, for a one-dimensional electrical signal, sim min is the distance between all signal segments and the reference signal segment of the electrical signal, and the minimum distance is taken as sim min , the maximum distance is sim max .

[0099] The fluctuation difference value is used to characterize the difference with adjacent signal segments. When the insulator is discharging, the collected current signal will change significantly. Therefore, calculating the fluctuation difference value helps to determine whether it is in a discharge state.

[0100] Specifically, the fluctuation difference value of each signal segment is calculated using the following formula:

[0101]

[0102] Among them, x i,j represents the jth signal point of the i-th signal segment, n represents the length of the i-th signal segment, x i-1,j represents the jth signal point of the i-1th signal segment, x i+1,j Represents the jth signal point of the i+1th signal segment.

[0103] The fluctuation degree value and the fluctuation difference value are combined to determine whether the signal segment is a discharge signal segment. During implementation, the fluctuation degree value and the fluctuation difference value are weighted and summed to determine whether the result is less than a preset threshold. If so, the signal segment is a discharge signal segment, otherwise it is not a discharge signal segment.

[0104] During implementation, after obtaining the discharge signal segment of each one-dimensional signal, it is necessary to determine the discharge start time. During implementation, there may be multiple continuous discharge signal segments, and multiple continuous discharge signal segments are regarded as one discharge signal segment. For each discharge signal segment of each one-dimensional signal, its peak position is first determined. During implementation, the peak position can be determined by the existing method for determining the peak.

[0105] Then, determining the discharge start time in each of the discharge signal segments based on the peak position and the signal differential specifically includes:

[0106] S1221, calculating a differential signal corresponding to the discharge signal segment;

[0107] S1222, setting the first sliding window at the peak position of the differential signal as the current sliding window; determining the sign mark of the current sliding window according to the differential signal of the current sliding window; sliding the current sliding window to the left by the first step length as the current sliding window;

[0108] S1223. Determine the sign mark of the current sliding window according to the differential signal of the current sliding window; if the sign mark of the current sliding window is different from the sign mark of the previous sliding window, stop sliding the window to obtain the discharge start time in the discharge signal segment; otherwise, slide the current sliding window to the left by the first step length as the current sliding window, and return to step S1223.

[0109] In implementation, for example, the i-th discharge signal segment is represented by Y i ={y i,1,y i,2 ,…,y i,m}. By (y i,1 -y0)×f is used to calculate the first signal point d1 after conversion. i,2 -y i,1 )×f is used to calculate the second signal point d2 after conversion. Similarly, the corresponding differential signal D={d1,d2,…,d m}, where m represents the total number of signal points in the discharge signal segment. It should be noted that y0 represents the signal at the previous moment of the i-th discharge signal segment. Where f represents the sampling frequency.

[0110] The differential signal corresponds to the discharge signal segment one by one, and the dynamic information of the original discharge signal segment is retained.

[0111] A first sliding window is used to slide leftward on the differential signal starting from the peak position, and a sign mark of each sliding window is determined based on the differential signal within each sliding window.

[0112] When using the first sliding window, the window size and sliding step length, i.e., the first sliding step length, must be determined first. During implementation, it can be determined based on computational efficiency and accuracy. For example, if the sliding window size is determined to be 10 sampling intervals, each step window corresponds to 10 sampling points, and the first sliding step length is 1 sampling interval, then one sampling point is slid to the left each time.

[0113] During implementation, align the right boundary of the first sliding window with the peak position as the current sliding window, then the 1st to 10th signal points on the left side of the peak are the differential signal points in the current sliding window, and the sign mark of the current sliding window is determined based on the differential signal in the current sliding window. For example, if the number of differential signals in the current sliding window that are greater than the first threshold value accounts for a proportion of the total number of differential signals in the current sliding window that exceeds the second threshold value, then the current sliding window corresponds to a positive sign, otherwise the current sliding window corresponds to a negative sign. During implementation, the first threshold value can be determined based on the fluctuation amplitude of the signal, and the first threshold value is greater than the fluctuation amplitude of the signal under normal conditions. The second threshold value can be 2 / 3.

[0114] After obtaining the positive or negative sign of the current sliding window, slide the first sliding window to the left by one step as the current sliding window; at this time, the data covered by the sliding window is the data of the 2nd to 11th sampling points on the left side of the peak, and the sign mark of the current sliding window is obtained according to the above process. If the sign mark of the current sliding window is different from the sign mark of the previous sliding window, the sliding can be stopped to obtain the discharge start time in the discharge signal segment, that is, the sampling time of the first signal point of the current sliding window, that is, the discharge start time in the discharge signal segment; if the sign mark of the current sliding window is the same as the sign mark of the previous sliding window, continue to slide to the left by one step to make a judgment until the sign marks of the two adjacent windows are different.

[0115] According to the above process, the discharge starting time in the one-dimensional electrical signal, the one-dimensional optical signal, the one-dimensional thermal signal and the one-dimensional acoustic signal is determined respectively.

[0116] The delay amount of the one-dimensional optical signal, one-dimensional thermal signal and one-dimensional acoustic signal relative to the one-dimensional electrical signal is determined based on the discharge start time of each one-dimensional signal; and then the one-dimensional optical signal, one-dimensional thermal signal and one-dimensional acoustic signal are adjusted based on the delay amount so that the one-dimensional optical signal, one-dimensional thermal signal and one-dimensional acoustic signal are aligned with the one-dimensional electrical signal.

[0117] During implementation, considering that electrical signals can more intuitively and timely reflect the state of the insulator, electrical signals are selected as the alignment standard, and other signals are aligned with the electrical signals to assist in judgment. The process of determining the delay of signals in other dimensions relative to the one-dimensional electrical signal is the same, and the following is an example of a one-dimensional optical signal.

[0118] Specifically, determining the delay of the one-dimensional optical signal relative to the one-dimensional electrical signal based on the discharge start time of the one-dimensional optical signal includes:

[0119] For each discharge start time in the one-dimensional optical signal, calculating the delay between the discharge start time and the discharge start time corresponding to the discharge start time in the one-dimensional electrical signal;

[0120] The average value of the delay amounts corresponding to all discharge start moments in the one-dimensional optical signal is the delay amount of the one-dimensional optical signal relative to the one-dimensional electrical signal.

[0121] During implementation, for each discharge start time of the one-dimensional optical signal, find the discharge start time closest to it in the one-dimensional electrical signal, calculate the difference between the two discharge start times, and use it as the delay amount corresponding to the discharge start time of the one-dimensional optical signal. Take the average of the delay amounts corresponding to all discharge start times in the one-dimensional optical signal as the delay amount of the one-dimensional optical signal relative to the one-dimensional electrical signal.

[0122] During implementation, the one-dimensional optical signal is adjusted forward according to the delay amount. For example, if the delay amount is 10 sampling points, the first 10 signal points of the one-dimensional optical signal are deleted, and 10 zeros are added at the end to serve as the aligned signal. The same process is used to align the one-dimensional thermal signal and the one-dimensional acoustic signal with the one-dimensional electrical signal. The aligned multi-dimensional time series signal is obtained.

[0123] Then, a training sample set is constructed based on the aligned multi-dimensional time series signals to train a machine learning model or a deep learning model.

[0124] Specifically, a sliding window is used to extract a multidimensional time series signal of T time steps as the input data of the sample, the discharge state of the data in the sliding window is predicted, the discharge progress and the peak value of the leakage current of the next discharge are used as the labels of the training samples, and a training sample set is constructed.

[0125] Specifically, through Determine the discharge progress of the sample, where t2 represents the moment of the last time step of the current sliding window, t1 represents the most recent discharge start moment before the last time step of the current sliding window, and Δt represents the difference between the most recent discharge start moment after the last time step of the current sliding window and the most recent discharge start moment before the last time step of the current sliding window. It should be noted that if no discharge has occurred before the last time step of the current sliding window, t1 is the initial data collection moment.

[0126] During implementation, a multidimensional time series prediction model is constructed, and the multidimensional time series prediction model is trained based on the constructed training sample set to obtain a trained insulator discharge prediction model.

[0127] During implementation, when the insulator works normally, the multi-dimensional time series signal fluctuates very little, and when discharge occurs, the data fluctuates greatly, so there is data sparsity. The constructed multi-dimensional time series prediction model includes a feature extraction module and a prediction module, wherein the feature extraction module is used to extract sample features; the prediction module is used to predict the progress to the next discharge and the peak value of the leakage current of the next discharge based on the extracted sample features.

[0128] Specifically, the feature extraction module includes an input layer, multiple intermediate layers, and an output layer;

[0129] The input layer is used to input sample data at each time step;

[0130] The first intermediate layer calculates the voltage and output signal of this layer according to the output data of the input layer; each of the remaining intermediate layers calculates the voltage and output signal of this layer based on the voltage and output signal of the previous intermediate layer; a feedback connection is set between the last intermediate layer and the first intermediate layer, which is used to transmit the output signal of the last intermediate layer in the previous time step to the first intermediate layer in the next time step;

[0131] The output layer is used to obtain sample features according to the output signal of the last intermediate layer.

[0132] Specifically, the voltage of the first middle layer is calculated using the following formula:

[0133]

[0134] Among them, u 1 [t] represents the voltage of the first intermediate layer at the tth time step, represents the temporary voltage of the first intermediate layer at the tth time step, V th Represents the voltage threshold. If The i-th element in is greater than V th , then s 1 The i-th element of [t] is 1, otherwise s 1 The i-th element of [t] is 0; s 1 [t] represents the output signal of the first intermediate layer at the tth time step;

[0135] The temporary voltage is calculated using the following formula:

[0136]

[0137] Among them, u 1 [t-1] represents the voltage of the first intermediate layer at the t-1th time step, s H [t-1] represents the output signal of the middle layer of the Hth layer at the t-1th time step, y[t] represents the sample data input at the tth time step, λ represents the leakage parameter of the neuron, W 1 represents the weight matrix of the feedback connection between the last intermediate layer and the first intermediate layer, F 1 Represents the weight matrix from the input layer to the first intermediate layer, V th represents the voltage threshold, b 1 represents the bias vector of the first intermediate layer, and H represents the total number of intermediate layers.

[0138] For neurons in the first intermediate layer, the input of the current time step includes not only the data transmitted by the input layer and the voltage of the neuron in the previous time step, but also the output signal of the last intermediate layer in the previous time step. The temporary voltage of the neuron in the current time step is calculated based on these data. If the temporary voltage is higher than the voltage threshold, the output signal of the neuron is 1, otherwise the output signal of the neuron is 0. The output signal is transmitted to the next layer of neurons.

[0139] Considering the time series correlation of insulator discharge, a feedback connection is added between the last layer and the first layer. This makes the network response delayed longer than the input sequence. The response of the feedback network constructed in this way at any time depends not only on the current input, but also on the previous input, so there is a memory effect, which makes the features extracted by the model more consistent with the characteristics of the insulator discharge process, thereby improving the accuracy of the prediction.

[0140] The voltage of the other intermediate layers except the first intermediate layer is calculated using the following formula:

[0141]

[0142] Among them, u h [t] represents the voltage of the middle layer of the hth layer at the tth time step, s h [t] represents the output signal of the h-th intermediate layer at the t-th time step, represents the temporary voltage of the middle layer of the hth layer at the tth time step, V th Represents the voltage threshold. If The i-th element in is greater than V th , then s h The i-th element of [t] is 1, otherwise s h The i-th element of [t] is 0;

[0143] The temporary voltage is calculated using the following formula:

[0144]

[0145] Among them, u h [t-1] represents the voltage of the middle layer of the hth layer at the t-1th time step, s h-1 [t] represents the output signal of the h-1th intermediate layer at the tth time step, λ represents the leakage parameter of the neuron, and F h represents the weight matrix from the middle layer of the h-1th layer to the middle layer of the hth layer, b h represents the bias vector of the h-th intermediate layer, and H represents the total number of intermediate layers.

[0146] For neurons in other intermediate layers, the input of the current time step includes the voltage of the neuron at the previous moment and the output signal of the corresponding neuron in the previous layer. The temporary voltage of the neuron at the current time step is calculated based on these data. If the temporary voltage is higher than the voltage threshold, the output signal of the neuron is 1, otherwise the output signal of the neuron is 0. The output signal is transmitted to the neurons in the next intermediate layer.

[0147] It should be noted that s H [0] is a vector of all 0s, u h[0] is a vector of all 0s, that is, at the initial moment, the voltage of each neuron is 0, and the output signal of the neurons in the last layer is 0.

[0148] The output layer encodes the output signals accumulated at all time steps of the last intermediate layer to obtain sample features.

[0149] Specifically, the output layer is based on the formula Calculate the sample characteristics.

[0150] Where T represents the total number of time steps, s H [t] represents the output signal of the Hth intermediate layer at the tth time step, and λ represents the leakage parameter of the neuron.

[0151] During implementation, the sample features output by the output layer of the feature extraction module are input into the prediction module to predict the progress to the next discharge and the peak value of the leakage current of the next discharge.

[0152] During implementation, the prediction module may adopt an existing regression network structure, such as a feedforward neural network.

[0153] The loss of the multi-dimensional time series prediction network is calculated using the following formula:

[0154]

[0155] Where N represents the number of samples, z i1 represents the discharge progress of the sample, represents the discharge progress predicted by the model, z i2 Indicates the peak value of the leakage current of the next discharge of the sample, Represents the peak value of the leakage current of the next discharge predicted by the model.

[0156] By minimizing the loss, the loss of the multidimensional time series prediction model is updated. When the loss accuracy reaches a preset threshold or reaches a preset number of iterations, the training is stopped to obtain a trained insulator discharge prediction model.

[0157] For the insulator to be detected, its electrical, optical, acoustic and thermal multi-dimensional time series data are collected and converted into one-dimensional signals respectively according to the aforementioned steps. Data alignment is performed according to the delay of the one-dimensional optical signal relative to the one-dimensional electrical signal, the delay of the one-dimensional thermal signal relative to the one-dimensional electrical signal, and the delay of the one-dimensional acoustic signal relative to the one-dimensional electrical signal. Then, the multi-dimensional time series signals of T time steps are extracted and input into the trained insulator discharge prediction model to predict the discharge progress of the insulator to be detected and the next discharge current amplitude.

[0158] According to the predicted discharge progress and the next discharge current amplitude, the insulator status can be evaluated in advance, fault prevention can be carried out in time, and pre-diagnosis and health management of insulators can be achieved, thereby ensuring the safe operation of the power grid system.

[0159] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0160] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. An insulator discharge monitoring system, characterized in that: include: An electrical data acquisition unit is used to collect leakage current and corona pulse current on the surface of the insulator to obtain electrical data of the insulator; A light collection unit, used for collecting ultraviolet light signals of the insulator and the surface image of the insulator to obtain light data of the insulator; A sound collection unit, used for collecting sound signals of the insulator to obtain sound data of the insulator; A thermal collection unit, used for collecting thermal imaging images of the insulator to obtain thermal data of the insulator; A monitoring unit is used to align the electrical data, optical data, acoustic data and thermal data of the insulator to obtain an aligned multi-dimensional time series signal, and input the aligned multi-dimensional time series signal into a trained insulator discharge prediction model to predict the discharge state of the insulator; The trained insulator discharge prediction model is obtained in the following way: Collect the electrical data, optical data, acoustic data and thermal data of the historical state of the insulator, align the electrical data, optical data, acoustic data and thermal data to obtain the aligned multi-dimensional time series signal, and annotate the discharge progress and discharge current value of the multi-dimensional time series signal to obtain a training sample set; Each sample in the training sample set includes a multi-dimensional time series signal, a discharge progress and a discharge current value; Constructing a multidimensional time series prediction model, and training the multidimensional time series prediction model based on the constructed training sample set to obtain a trained insulator discharge prediction model; The multi-dimensional time series prediction model includes a feature extraction module and a prediction module, wherein the feature extraction module is used to extract sample features; the prediction module is used to predict the discharge progress and the peak value of the leakage current of the next discharge based on the extracted sample features; The feature extraction module includes an input layer, multiple intermediate layers and an output layer; The input layer is used to input sample data at each time step; The first intermediate layer calculates the voltage and output signal of this layer based on the output data of the input layer; Each of the remaining intermediate layers calculates the voltage and output signal of the current layer based on the voltage and output signal of the previous intermediate layer; a feedback connection is set between the last intermediate layer and the first intermediate layer, which is used to transmit the output signal of the last intermediate layer in the previous time step to the first intermediate layer in the next time step; The output layer is used to obtain sample features according to the output signal of the last intermediate layer.

2. The insulator discharge monitoring system according to claim 1, characterized in that: The electric collection unit includes a non-contact through-core leakage current sensor and a pulse current detection device for respectively collecting leakage current and corona pulse current on the surface of the insulator; The light collection unit includes an ultraviolet light signal collection circuit and a high-resolution camera, which are used to collect the ultraviolet light signal and surface image of the insulator respectively; The heat collection unit comprises a thermal imager for collecting thermal imaging images of the insulator.

3. The insulator discharge monitoring system according to claim 2, characterized in that: The ultraviolet light signal acquisition circuit includes an optical lens, a photoelectric conversion circuit, a current-voltage conversion circuit and a sampling circuit; The optical lens is used to input ultraviolet light into the photoelectric conversion circuit; the photoelectric conversion circuit converts the optical signal into a current signal; The current-to-voltage conversion circuit is used to convert the current signal output by the photoelectric conversion circuit into a voltage signal; The sampling circuit is used to sample the voltage signal output by the current-voltage conversion circuit to obtain an ultraviolet light signal.

4. The insulator discharge monitoring system according to claim 1, characterized in that: The loss of the multi-dimensional time series prediction model is calculated using the following formula: Where N represents the number of samples, z i1 represents the discharge progress of the sample, represents the discharge progress predicted by the model, z i2 Indicates the peak value of the leakage current of the next discharge of the sample, Represents the peak value of the leakage current of the next discharge predicted by the model.

5. The insulator discharge monitoring system according to claim 1, characterized in that: The output layer uses the following formula to calculate the sample features: Where T represents the total number of time steps of the sample, s H [t] represents the output signal of the Hth intermediate layer at the tth time step, and λ represents the leakage parameter of the neurons in the intermediate layer.

6. The insulator discharge monitoring system according to claim 1, characterized in that: The voltage of the first middle layer is calculated using the following formula: Among them, u 1 [t] represents the voltage of the first intermediate layer at the tth time step, represents the temporary voltage of the first intermediate layer at the tth time step, V th Represents the voltage threshold. If The i-th element in is greater than V th , then s 1 The i-th element of [t] is 1, otherwise s 1 The i-th element of [t] is 0; s 1 [t] represents the output signal of the first intermediate layer at the tth time step; The temporary voltage is calculated using the following formula: Among them, u 1 [t-1] represents the voltage of the first intermediate layer at the t-1th time step, s H [t-1] represents the output signal of the middle layer of the Hth layer at the t-1th time step, y[t] represents the sample data input at the tth time step, λ represents the leakage parameter of the neuron, W 1 represents the weight matrix of the feedback connection between the last intermediate layer and the first intermediate layer, F 1 Represents the weight matrix from the input layer to the first intermediate layer, V th represents the voltage threshold, b 1 represents the bias vector of the first intermediate layer, and H represents the total number of intermediate layers.

7. The insulator discharge monitoring system according to claim 1, characterized in that: The electrical data, optical data, acoustic data and thermal data are aligned to obtain an aligned multi-dimensional time series signal. The discharge progress and discharge current value of the multi-dimensional time series signal are annotated to obtain a training sample set, including: Performing dimensionality reduction processing on the electrical data, optical data and thermal data of the insulator respectively, and converting the electrical data, optical data and thermal data into one-dimensional electrical signals, one-dimensional optical signals and one-dimensional thermal signals respectively; the acoustic data is a one-dimensional acoustic signal; Respectively detecting the discharge starting time of a one-dimensional electrical signal, a one-dimensional optical signal, a one-dimensional thermal signal, and a one-dimensional acoustic signal, and aligning the one-dimensional electrical signal, the one-dimensional optical signal, the one-dimensional thermal signal, and the one-dimensional acoustic signal based on the discharge starting time to obtain a multi-dimensional time series signal; A sliding window is used to extract T time steps of multidimensional time series signals as the input data of the sample, and the discharge progress of the data in the sliding window is calculated. The discharge progress and the peak value of the leakage current of the next discharge are used as the labels of the training samples to construct a training sample set.

Images