Electric meter remote fault diagnosis method and system based on internet of things

Abstract

The present application relates to the field of electric meter remote monitoring, more particularly, the present application relates to the electric meter remote fault diagnosis method and system based on internet of things, the method comprises: obtaining all historical fault sequences corresponding to each fault type; the period of each historical fault sequence is calculated and the probability of each period appearing, and based on the historical fault sequence corresponding to a single period, the template sequence of the period is generated; the real-time parameter time sequence of the electric meter is collected, the similarity of the real-time parameter time sequence and the template sequence of each period corresponding to the fault type is calculated, the calculated similarity is weighted and summed according to the probability of each period appearing, and the real-time fault probability of the fault type is obtained; when the real-time fault probability is greater than the preset fault threshold, the diagnosis result of the corresponding fault type of the electric meter is output. The present application can more comprehensively and accurately describe the complex evolution law of the fault, and realize the remote and timely diagnosis of the electric meter fault.

Description

Technical Field

[0001] This invention relates to the field of remote monitoring of electricity meters. More specifically, this invention relates to a method and system for remote fault diagnosis of electricity meters based on the Internet of Things (IoT). Background Technology

[0002] With the rapid development of IoT technology in smart grids, a massive number of smart meters, as key terminals, have been connected to the IoT network, enabling remote and real-time collection of operating parameters. This provides a data foundation for the intelligent operation and maintenance of electricity meters.

[0003] Currently, the operation and maintenance of smart meters generally relies on two traditional models: regular inspections and post-fault repairs. Regular inspections are labor-intensive, costly, and lack specificity, making it difficult to effectively detect latent faults. Post-fault repairs, on the other hand, are a reactive measure, with significant delays from the occurrence of a fault to its repair, which can easily lead to inaccurate metering, power outages, and even safety accidents, affecting the reliability of the power grid and the user experience.

[0004] While the development of the Internet of Things (IoT) provides a theoretical basis for remote fault monitoring and maintenance of smart meters, existing remote fault diagnosis methods still have significant limitations in practical applications. These methods often fail to fully consider the inherent temporal characteristics of fault occurrence and development. Furthermore, their analysis dimensions are relatively singular in matching and identifying fault characteristics in remotely collected data, resulting in unsatisfactory accuracy and timeliness of predictions, making it difficult to meet the practical needs of large-scale, refined operation and maintenance.

[0005] Therefore, there is an urgent need in this field for a new method that can predict smart meter faults more accurately and in a more timely manner. Summary of the Invention

[0006] To address the technical problem that existing remote fault diagnosis methods are insufficient in capturing the dynamic behavior patterns of faults, resulting in poor predictive performance, the present invention provides solutions in the following aspects.

[0007] In the first aspect, the IoT-based remote fault diagnosis method for electricity meters includes: Obtain the historical parameter time series of the electricity meter with fault tags, and divide all historical fault sequences corresponding to each fault type based on the fault tags; For a single fault type, calculate the period of its corresponding historical fault sequence and the probability of each period occurring, and generate a template sequence for that period based on the historical fault sequence corresponding to the single period. Collect the real-time parameter time series of the meter at the current moment. For a single fault type, calculate the similarity between the real-time parameter time series and the template sequence of each period corresponding to the fault type. Based on the probability of occurrence of each period, perform a weighted summation of the calculated similarity to obtain the real-time fault probability of the real-time parameter time series under the fault type. When the calculated real-time fault probability is greater than the preset fault threshold, the output meter shows a remote diagnostic result for the corresponding fault type.

[0008] Preferably, the historical parameter time series is a multi-dimensional sequence, the dimensions of which include current, voltage and power.

[0009] Preferably, the process of obtaining the period includes: Determine the period estimate for each historical fault sequence, calculate the probability of each period estimate occurring, and accumulate all probabilities from largest to smallest. When the sum meets a preset stopping condition, stop accumulating and use the period estimate corresponding to the probability participating in the accumulation as the final determined period.

[0010] Preferably, the process of obtaining the period includes: Determine the period estimates of the historical fault sequence, cluster all period estimates to obtain multiple clusters, and use the period estimate corresponding to the center point of each cluster as the final determined period.

[0011] Preferably, the period estimate is obtained through autocorrelation analysis or fast Fourier transform.

[0012] Preferably, the similarity acquisition process includes: Calculate the Pearson correlation coefficient between the real-time parameter time series and the template sequence of any period, and use the calculated Pearson correlation coefficient as the similarity.

[0013] Preferably, the similarity acquisition process includes: The dot product of the real-time parameter time series and the corresponding parameters of the template sequence of any period is obtained by multiplying the real-time parameter time series and the template sequence. Calculate the real-time parameter time series and the modulus of the template sequence respectively, and then calculate the product of the two moduli; The similarity is obtained by dividing the dot product by the product of the two moduli.

[0014] Preferably, the similarity acquisition process includes: The real-time parameter time series and the template sequence of any period are divided according to the extreme points to obtain their respective sub-segments. For each sub-segment of the real-time parameter time series and each sub-segment of the template sequence, the DTW distance between them is calculated. The calculated DTW distance is inversely taken and normalized. The processed result is used as the weight of the edge between the sub-segment of the real-time parameter time series and the sub-segment of the template sequence. Find the maximum weight match between all segments of the real-time parameter time series and all segments of the template sequence. Sum the weights of all edges after matching to obtain the edge weight sum. Normalize the edge weight sum to obtain the similarity.

[0015] Secondly, an IoT-based remote fault diagnosis system for electricity meters includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the IoT-based remote fault diagnosis method for electricity meters is implemented as described in any one of the above.

[0016] The present invention has the following beneficial effects: 1. This invention overcomes the limitations of traditional static or single-cycle models, achieving accurate multi-mode characterization of dynamic fault behavior. Traditional methods often struggle to capture the time-varying and periodic fluctuations in fault behavior. This invention constructs a more realistic multi-cycle fault probability model by extracting multiple representative cycles and their template sequences for each fault type and introducing the probability of cycle occurrence. This is equivalent to establishing a dynamic profile containing multiple "behavioral scripts" for each fault, thereby enabling a more comprehensive and accurate description of the complex evolution of faults and achieving remote and timely diagnosis of meter faults.

[0017] 2. This invention significantly improves the robustness and reliability of remote fault diagnosis by introducing a probability-weighted multi-template similarity fusion mechanism. During the real-time remote diagnosis phase, this invention does not simply search for the most similar single template, but instead calculates the similarity with all periodic templates and performs weighted fusion based on their historical occurrence probabilities. This mechanism ensures that the diagnostic results do not depend on matching a specific periodic pattern. Even if the remotely acquired real-time parameters deviate from a major period, as long as they are highly similar to other minor periodic patterns, they can still be effectively identified, thereby reducing the risk of false alarms and false negatives and enhancing the system's stability under complex operating conditions.

[0018] 3. In the period determination stage, this invention provides a flexible strategy based on probability accumulation or clustering, which can automatically and adaptively extract the dominant and representative periodic patterns from historical parameters, rather than relying on manually setting a fixed period. This allows it to adapt to the complex operating environment of different regions and different models of electricity meters uploading data through the Internet of Things. Attached Figure Description

[0019] Figure 1 This is a flowchart of steps S1-S3 in the remote fault diagnosis method for electricity meters based on the Internet of Things according to an embodiment of the present invention.

[0020] Figure 2 This is a structural block diagram of the Internet of Things-based remote fault diagnosis system for electricity meters according to an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0022] Reference Figure 1 The IoT-based remote fault diagnosis method for electricity meters includes steps S1-S3, as detailed below: S1: Obtain the historical parameter time sequence of the electricity meter with fault tags, and divide it into all historical fault sequences corresponding to each fault type based on the fault tags.

[0023] In one embodiment, the remote monitoring platform continuously or at a fixed sampling frequency (e.g., once per minute) collects the operating parameters of the electricity meter via an Internet of Things (IoT) communication network (such as 4G / 5G, NB-IoT, etc.). These parameters typically include current, voltage, power, and other relevant parameters.

[0024] The remote monitoring platform records the above parameters and their corresponding timestamps together to form the original historical parameter time series. In other words, this historical parameter time series is a multi-dimensional sequence, whose dimensions include parameters such as current, voltage, and power.

[0025] During or after parameter acquisition, professionals (such as engineers or maintenance personnel) combine maintenance records, alarm information, or direct observation of parameter curves to assign a simplified fault label to each data segment.

[0026] For example, the fault labels mentioned above include voltage spikes, current surges, and abnormal power consumption behavior. Additionally, if all parameters are within the rated range and change smoothly without any abnormal conditions, the label can be normal.

[0027] Based on the above fault labels, all the collected parameters are filtered and classified. All parameter segments labeled as voltage spikes are used to form a historical fault sequence with voltage spike as the fault type. Similarly, historical fault sequences with current impact as the fault type and historical fault sequences with electricity consumption behavior as the fault type are obtained respectively.

[0028] By performing the above operations, we can obtain well-categorized historical fault sequences, which can lay a solid foundation for quickly locating the root cause of major faults in the future.

[0029] S2: For a single fault type, calculate the period of its corresponding historical fault sequence and the probability of each period occurring, and generate a template sequence for that period based on the historical fault sequence corresponding to the single period.

[0030] In historically collected fault sequences, the fault state often persists, resulting in excessively long sequences containing a large amount of redundant information. Directly using the full sequence length as a template would reduce analysis efficiency and the typicality of the template. To address this issue, it is necessary to extract periods from these long sequences that represent the recurrence pattern of the faults, and then construct a refined fault template based on these periods.

[0031] Furthermore, considering that historical fault sequences may be affected by various factors such as noise and random fluctuations, it is difficult to directly determine the precise period from historical fault sequences. Obtaining period estimates through autocorrelation analysis or fast Fourier transform can capture the possible periodic characteristics of historical fault sequences as a whole, and perform preliminary screening of periods.

[0032] For example, autocorrelation analysis can identify lag points with strong correlations in noisy data by calculating the correlation of sequences under different lags, and make a preliminary judgment on possible period estimates; fast Fourier transform converts time-domain signals to frequency domain and determines period estimates by finding the dominant frequency component, which can effectively deal with the interference of complex waveforms in the time domain on period judgment.

[0033] In one embodiment, the specific operation for obtaining the period estimate of a historical fault sequence based on autocorrelation analysis is as follows: For a single fault type, autocorrelation analysis is performed on each of its historical fault sequences, and then an autocorrelation function graph is plotted, where the horizontal axis represents the lag value and the vertical axis represents the autocorrelation function value.

[0034] In the autocorrelation function plot, look for the peaks of various lag values. The lag value corresponding to the first significant peak can be used as an estimate of the period of the historical fault sequence. This is because if the historical fault sequence is periodic, then at lags that are integer multiples of the period, the correlation of the historical fault sequence will be strong, and the autocorrelation function value will show a peak. The lag value corresponding to the first significant peak is a basic period of the historical fault sequence.

[0035] In another embodiment, the specific operation for obtaining the period estimate of the historical fault sequence based on the Fast Fourier Transform is as follows: For a single fault type, a Fast Fourier Transform is performed on each of its historical fault sequences to obtain the frequency domain signal of each historical fault sequence. In the frequency domain signal, the frequency component with the largest amplitude is called the dominant frequency component, which represents the most important frequency component in the signal.

[0036] Since there is a reciprocal relationship between period and frequency, the reciprocal of the above-mentioned main frequency component is used as the period estimate of the historical fault sequence.

[0037] In summary, for each historical fault sequence of a certain fault type, the period estimate of each historical fault sequence can be obtained by performing period estimation using either the autocorrelation analysis or the fast Fourier transform method mentioned above.

[0038] In the above analysis of the periodicity of historical fault sequences for a single fault type, the first method based on autocorrelation analysis directly mines the correlation characteristics of the sequence itself in the time domain, which can accurately locate a basic period of the historical fault sequence and provide key information for subsequent analysis of the repetitive patterns of the fault. The method based on fast Fourier transform, by performing frequency domain analysis on the signal, can concentrate the energy of the signal on different frequency components. Noise is usually distributed over a wide frequency range, while the periodic components of historical fault sequences are often concentrated at specific frequency points. Therefore, by selecting a frequency component with a larger amplitude as the dominant frequency component, the influence of noise can be suppressed to a certain extent, and the accuracy of the period estimation can be improved.

[0039] Even if preliminary analysis yields cycle estimates, these estimates may contain some accidental and unrepresentative cycle values. Therefore, further verification and optimization are needed to obtain more accurate cycle values.

[0040] In one embodiment, for a single fault type, the number of occurrences of the periodic estimate for each historical fault sequence is counted, and then divided by the total number of periodic estimates for all historical fault sequences to obtain the probability of occurrence of the periodic estimate for each historical fault sequence.

[0041] The calculated probabilities are then arranged in descending order and accumulated. When the accumulated sum first reaches the stopping condition, the accumulation is stopped, and the period estimate corresponding to the probability involved in the accumulation is taken as the final determined period.

[0042] For example, the accumulator is set to stop when the sum is greater than 0.95, thereby selecting period estimates that can cover most occurrences and ensuring that the final determined period has high reliability and representativeness.

[0043] In another embodiment, for a single fault type, all historical fault sequences contained therein are clustered (such as K-means clustering, hierarchical clustering, etc.). After clustering is completed, the period estimate of the cluster center point is used as the final determined period of the cluster. In addition, the proportion of the number of the final determined periods in the cluster is used as the probability of the final determined period.

[0044] The above-mentioned probability-accumulation-based period determination method focuses on high-frequency cycles, which dominate historical faults and are more likely to reflect the true periodic characteristics of faults, greatly improving the reliability of the final period determination. Meanwhile, the clustering-based period determination method can discover the hidden internal structure and patterns in the data, expanding the depth of fault cycle analysis.

[0045] In summary, for each historical fault sequence of a certain fault type, the final period is determined by using either probability accumulation or clustering methods mentioned above, thus obtaining the final determined period for each historical fault sequence.

[0046] Finally, after determining the cycle of a certain fault type, historical fault sequences within that cycle are selected. These historical fault sequences are analyzed and processed to extract representative features, forming a template sequence for that cycle. After obtaining the final determined cycle, all historical fault sequences corresponding to that cycle are acquired, aligned, and averaged point-by-point to obtain the template sequence for that cycle.

[0047] S3: Collect the real-time parameter time series of the meter at the current moment. For a single fault type, calculate the similarity between the real-time parameter time series and the template sequence of each period corresponding to the fault type. Based on the probability of occurrence of each period, perform a weighted summation of the calculated similarity to obtain the real-time fault probability of the real-time parameter time series under the fault type. When the calculated real-time fault probability is greater than the preset fault threshold, output the remote diagnostic result that the meter has the corresponding fault type.

[0048] The real-time parameter sequence of the meter is obtained by following the parameter acquisition operation in S1 described above. Meter faults typically manifest as abnormal changes in a specific parameter sequence (such as voltage fluctuations, current distortion, abnormal power factor, etc.). A large number of typical fault modes (template sequences) are accumulated in historical fault sequences. By calculating the similarity between the real-time parameter sequence and these template sequences, it can be determined whether the current operating state matches a known fault mode.

[0049] In one embodiment, for a single fault type, the Pearson correlation coefficient between the real-time parameter time series and the template sequence of any period is calculated, and the calculated Pearson correlation coefficient is used as the similarity between the real-time parameter time series and the template sequence of the period.

[0050] In another embodiment, still for a single fault type, the vector dot product method is used, specifically: the real-time parameter time series is multiplied by the parameters at the corresponding positions of the template sequence of any period and then summed to obtain the dot product of the real-time parameter time series and the template sequence; the modulus of the real-time parameter time series and the template sequence are calculated respectively, and the product of the two modulus is calculated; the calculated dot product is divided by the product of the two modulus to obtain the similarity between the real-time parameter time series and the template sequence of the period.

[0051] In another embodiment, it is considered that the Pearson correlation coefficient may be misjudged due to the difference in sequence phase caused by different sampling times (e.g., two sequences have the same overall trend, but one rises first and then falls, while the other falls first and then rises, and the Pearson correlation coefficient shows a negative correlation).

[0052] For a single fault type, both the real-time parameter time series and the template sequence for any period are divided into sub-segments based on extreme points. For example, taking the real-time parameter time series as an example, the extreme points (including maximum and minimum values) of the sequence are obtained. The extreme points represent the turning points of the sequence trend change, such as from rising to falling or from falling to rising. The extreme points are used as the segmentation boundaries to divide the real-time parameter time series into multiple sub-segments.

[0053] For each segment of the real-time parameter time series and each segment of the template sequence, calculate the DTW (Dynamic Time Warping) distance between them. Take the negative of the calculated DTW distance and normalize it (e.g., min-max normalization). The processed result is used as the weight of the edge between the segments of the real-time parameter time series and the template sequence. The smaller the DTW distance, the more similar the two segments are. After taking the negative, a larger weight value indicates higher similarity.

[0054] Furthermore, the KM algorithm (Kuhn-Munkres algorithm, also known as the Hungarian algorithm) is used to find the best match between each segment of the real-time parameter time series and each segment of the template sequence. The KM algorithm is a maximum weight matching algorithm for bipartite graphs, aiming to maximize the sum of the weights of all matching edges. This means pairing the most similar segments together, without requiring the segments to be in the same order.

[0055] After obtaining the maximum weight matching using the KM algorithm, the weights of all matching edges are normalized (e.g., minimum-maximum normalization) to obtain the real-time parameter time series and the similarity of the template sequence.

[0056] Since the parameter sequence is multidimensional, the similarity of each parameter under the same period is calculated separately, and then the average of these similarities is taken as the comprehensive similarity of that period, so as to obtain the comprehensive similarity between the real-time parameter time series and the template sequence of each period.

[0057] Furthermore, the comprehensive similarity calculated above is multiplied by the probability of the corresponding period and summed over all periods to obtain the real-time fault probability of the real-time parameter time series under this fault type.

[0058] In addition, the probability distribution of faults under normal operating conditions is collected, the mean and standard deviation are calculated, and a threshold is set based on the mean and standard deviation, or an empirical threshold is directly set based on the meter model and fault type. For example, the threshold can be set to 0.8.

[0059] When the real-time fault probability of the real-time parameter timing under a certain fault type is greater than 0.8, the system automatically generates a diagnostic conclusion and remotely outputs alarm information through the Internet of Things platform, indicating that the meter has a high risk of a certain fault type, thereby guiding maintenance personnel to take targeted measures.

[0060] This invention also provides a remote fault diagnosis system for electricity meters based on the Internet of Things. For example... Figure 2 As shown, the system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the Internet of Things-based remote fault diagnosis method for electricity meters according to the first aspect of the present invention.

[0061] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0062] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A remote fault diagnosis method for an electricity meter based on the Internet of Things, characterized in that, include: Obtain the historical parameter time series of the electricity meter with fault tags, and divide all historical fault sequences corresponding to each fault type based on the fault tags; For a single fault type, calculate the period of its corresponding historical fault sequence and the probability of each period occurring, and generate a template sequence for that period based on the historical fault sequence corresponding to the single period. Collect the real-time parameter time series of the meter at the current moment. For a single fault type, calculate the similarity between the real-time parameter time series and the template sequence of each period corresponding to the fault type. Based on the probability of occurrence of each period, perform a weighted summation of the calculated similarity to obtain the real-time fault probability of the real-time parameter time series under the fault type. When the calculated real-time fault probability is greater than the preset fault threshold, the output meter shows a remote diagnostic result for the corresponding fault type.

2. The method for remote fault diagnosis of electricity meters based on the Internet of Things according to claim 1, characterized in that, The historical parameter time series is a multi-dimensional sequence, whose dimensions include current, voltage and power.

3. The method for remote fault diagnosis of electricity meters based on the Internet of Things according to claim 1, characterized in that, The process of obtaining the period includes: Determine the period estimate for each historical fault sequence, calculate the probability of each period estimate occurring, and accumulate all probabilities from largest to smallest. When the sum meets a preset stopping condition, stop accumulating and use the period estimate corresponding to the probability participating in the accumulation as the final determined period.

4. The method for remote fault diagnosis of electricity meters based on the Internet of Things according to claim 1, characterized in that, The process of obtaining the period includes: Determine the period estimates of the historical fault sequence, cluster all period estimates to obtain multiple clusters, and use the period estimate corresponding to the center point of each cluster as the final determined period.

5. The method for remote fault diagnosis of electricity meters based on the Internet of Things according to claim 3 or 4, characterized in that, The period estimate is obtained through autocorrelation analysis or fast Fourier transform.

6. The method for remote fault diagnosis of electricity meters based on the Internet of Things according to claim 1, characterized in that, The process of obtaining the similarity includes: Calculate the Pearson correlation coefficient between the real-time parameter time series and the template sequence of any period, and use the calculated Pearson correlation coefficient as the similarity.

7. The method for remote fault diagnosis of electricity meters based on the Internet of Things according to claim 1, characterized in that, The process of obtaining the similarity includes: The dot product of the real-time parameter time series and the corresponding parameters of the template sequence of any period is obtained by multiplying the real-time parameter time series and the template sequence. Calculate the real-time parameter time series and the modulus of the template sequence respectively, and then calculate the product of the two moduli; The similarity is obtained by dividing the dot product by the product of the two moduli.

8. The method for remote fault diagnosis of electricity meters based on the Internet of Things according to claim 1, characterized in that, The process of obtaining the similarity includes: The real-time parameter time series and the template sequence of any period are divided according to the extreme points to obtain their respective sub-segments. For each sub-segment of the real-time parameter time series and each sub-segment of the template sequence, the DTW distance between them is calculated. The calculated DTW distance is inversely taken and normalized. The processed result is used as the weight of the edge between the sub-segment of the real-time parameter time series and the sub-segment of the template sequence. Find the maximum weight match between all segments of the real-time parameter time series and all segments of the template sequence. Sum the weights of all edges after matching to obtain the edge weight sum. Normalize the edge weight sum to obtain the similarity.

9. A remote fault diagnosis system for electricity meters based on the Internet of Things, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the Internet of Things-based remote fault diagnosis method for electricity meters according to any one of claims 1-8.

Images