Semi-solid-state circuit breaker with intelligent collection function of full power data

Abstract

The present invention relates to the technical field of power data analysis, and in particular to a semi-solid circuit breaker with a full-electricity data intelligent acquisition function. The semi-solid circuit breaker includes: a data acquisition unit, which is used to collect different types of full-electricity data within a preset time period; a noise possibility analysis unit, which is used to obtain the same-position data corresponding to each data, and calculate the noise possibility of each data; a data initial segmentation unit, which is used to obtain the turning moment possibility of each data, and then segment the full-electricity data to obtain the initial data segment; a data segmentation correction unit, which is used to calculate the periodic regularity similarity between two initial data segments, and then merge the initial data segments to obtain the final data segment; a missing data completion unit, which is used to set a sliding window, use the sliding window to obtain the influence weight of each data in other final data segments, and complete the missing data. The present application can accurately complete the missing data in the full-electricity data.

Description

Technical Field

[0001] The present invention relates to the technical field of power data analysis, and in particular to a semi-solid circuit breaker with a full electric quantity data intelligent collection function. Background Art

[0002] Semi-solid-state circuit breakers are a new type of power protection device that combines traditional mechanical and electronic switches. They combine the high interrupting capacity of traditional circuit breakers with the high-speed response characteristics of solid-state devices. Full power data acquisition involves high-precision, synchronous, and continuous collection of multi-dimensional data such as current, voltage, power, harmonics, and temperature during circuit breaker operation. This collected data enables rapid fault identification and response, predicts and maintains circuit anomalies, and analyzes and optimizes power quality, improving power supply quality and safety.

[0003] Semi-solid-state circuit breakers rely on data collection and analysis for functions such as circuit anomaly prediction and fault response. However, issues such as poor connector contact, electromagnetic interference, and communication anomalies during the data collection process often lead to discontinuous data collection and missing data. Because data varies depending on the circuit's operating state, traditional data interpolation methods cannot incorporate these data variations. The interpolated results may differ significantly from the true value, affecting the accuracy of circuit anomaly prediction. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a semi-solid circuit breaker with a full power data intelligent acquisition function. The technical solution adopted is as follows:

[0005] An embodiment of the present invention provides a semi-solid circuit breaker with a full power data intelligent acquisition function, the semi-solid circuit breaker comprising:

[0006] A data collection unit, used to collect different types of full power data within a preset time period;

[0007] A noise possibility analysis unit is used to obtain the corresponding co-location data of each data within the window centered on each data; and calculate the noise possibility of a data based on the corresponding co-location data of the data;

[0008] The data initial segmentation unit is used to obtain the turning point probability of the data based on the slope value corresponding to two adjacent data and the noise probability of the data; the full power data is segmented according to the turning point probability of the data at the same time in different types of full power data to obtain initial data segments;

[0009] A data segment correction unit is used to calculate the periodic regularity similarity based on the time length difference and data difference between two initial data segments; and to merge the initial data segments according to the periodic regularity similarity between each two initial data segments to obtain a final data segment;

[0010] The missing data completion unit is used to set a sliding window based on the length of the final data segment where a missing data is located in a type of full power data, and use the sliding window to slide on other final data segments to obtain the influence weight of each data in other final data segments; and complete the missing data based on the data in other final data segments and their influence weights.

[0011] Preferably, obtaining the same-position data corresponding to each data from the data in the window centered on each data includes:

[0012] In a type of full power data, the sum of the absolute values ​​of the differences between the data other than the data itself in a window corresponding to one data and the data at the corresponding position in the window of another data is obtained, and then added to the hyperparameter and inverted to obtain the inverted result. The inverted result is normalized to obtain the possibility that the other data is the data at the same position in a different period of the data; if the possibility that the other data is the data at the same position in a different period of the data is greater than a first reference threshold, then the other data is the data at the same position corresponding to the data.

[0013] Preferably, calculating the noise probability of a piece of data based on the co-located data corresponding to the piece of data includes:

[0014] The noise probability of the data is obtained by obtaining the average of the absolute values ​​of the differences between a data and its corresponding data at the same position and normalizing them.

[0015] Preferably, obtaining the turning point possibility of a data based on the slope values ​​corresponding to two adjacent data and the noise possibility of the data includes:

[0016] For a type of full power data, a time domain graph corresponding to the full power data is obtained. In the time domain graph, the absolute value of the difference in slopes corresponding to two adjacent data before and after the data is obtained and normalized to obtain a slope difference. The slope difference is multiplied by the difference between a preset value and the noise probability of the data to obtain the turning moment probability of the data.

[0017] Preferably, the full power data is segmented to obtain initial data segments according to the likelihood of transition moments of data at the same moment in different types of full power data, including:

[0018] Calculate the mean of the turning point possibilities of the data at the same moment in different types of full power data, and record it as the turning point possibility average value at that moment; if the turning point possibility average value at a moment is greater than the second reference threshold, then the moment is a suspected working state change moment, and the suspected working state change moment is used to segment the full power data to obtain the initial data segment.

[0019] Preferably, calculating the periodic regularity similarity based on the time length difference and data difference between the two initial data segments includes:

[0020] The absolute value of the difference in time length between the two initial data segments is added to the hyperparameter and the inverse is taken to obtain the time length similarity eigenvalue; the absolute value of the difference in a type of full power data at a moment between the two initial data segments is obtained, recorded as the difference of the type of full power data at that moment, the absolute value of the difference between the type of full power data at each two adjacent moments is summed up, and then added to the hyperparameter and the inverse is taken to obtain the data similarity eigenvalue between the two initial data segments corresponding to the type of full power data, and the sum of the data similarity eigenvalues ​​between the two initial data segments corresponding to each type of full power data is obtained to obtain the comprehensive data similarity eigenvalue between the two initial data segments; the time length similarity eigenvalue and the comprehensive data similarity eigenvalue are multiplied and normalized to obtain the periodic regularity similarity between the two initial data segments.

[0021] Preferably, merging the initial data segments to obtain the final data segment according to the similarity of the periodic regularity between each two initial data segments comprises:

[0022] Starting from the first initial data segment, obtain the initial data segment whose periodic regularity similarity with the first initial data segment is greater than the third reference threshold, and record it as a suspected similar data segment; arrange the suspected similar data segments in time sequence to obtain a first sequence, and obtain the time interval between the starting moments of two adjacent suspected similar data segments in the first sequence as the adjacent interval between the two adjacent suspected similar data segments; if the adjacent interval between every two adjacent suspected similar data segments in the first sequence is less than or equal to the time interval threshold, then all the suspected similar data segments in the first sequence are the periodic data segments corresponding to the first initial data segment; if the adjacent interval is greater than ... Adjacent suspected similar data segments with an interval threshold are selected, and the adjacent intervals between such adjacent suspected similar data segments are recorded as discontinuous intervals. The first discontinuous interval is located in chronological order. At this time, the suspected similar data segment before the first discontinuous interval is the periodic data segment corresponding to the first initial data segment; and so on, starting from the first initial data segment in the remaining initial data segments in chronological order, the periodic data segment corresponding to the first initial data segment in the remaining initial data segments is searched until all the periodic data segments are obtained; the two adjacent periodic data segments in the initial data segments that are periodic data segments are merged in chronological order to obtain the final data segment.

[0023] Preferably, the sliding window is used to slide on other final data segments to obtain the influence weight of each data in other final data segments, including:

[0024] A sliding window is used to slide on the final data segment to which a data belongs, with the sliding step being a preset value. The average of the absolute values ​​of the differences between each data in a sliding window and each corresponding data in the final data segment to which a missing data belongs is obtained, and the inverse is calculated to obtain the overlap corresponding to the sliding window. The maximum value of the overlap corresponding to each sliding window is obtained as the maximum overlap. The maximum overlap is multiplied by the difference between the preset value and the noise possibility of the data to obtain the influence weight of the data.

[0025] Preferably, the missing data is completed based on the data in other final data segments and their influence weights, including:

[0026] Obtain the influence weight of each data in other final data segments except the data in the final data segment where the missing data is located, multiply the influence weight of each data in other final data segments by each data in other final data segments respectively to obtain the updated data corresponding to each data in other final data segments; complete the missing data according to the updated data corresponding to each data in other final data segments and the LSTM algorithm.

[0027] Embodiments of the present invention have at least the following beneficial effects: the present application collects different types of full power data, analyzes the co-located data of each data type, and then obtains the noise probability of each data, which is used to eliminate the influence of noise when interpolating missing data, thereby improving the accuracy of missing data interpolation. Furthermore, the turning point probability of each data is obtained by using the slope value corresponding to two adjacent data and the noise probability of the data, and then the full power data is segmented to obtain initial data segments. Then, the similarity of the periodicity between each two initial data segments is calculated to merge the initial data segments to obtain final data segments. Segmenting the data taking into account both data changes and periodic changes can effectively avoid mistakenly identifying data changes caused by accidental factors as the change pattern of the data segment, making the final data segment obtained after segmentation more accurate. Finally, by setting a sliding window with the same length as the final data segment where the missing data is located, it is used to identify data in other final data segments with similar change patterns as the missing data, and thereby adjust the influence weight of the data in other final data segments on the missing data interpolation results, which can effectively improve the accuracy of data interpolation and facilitate improving the effectiveness of anomaly identification and early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0029] Figure 1 This is a unit block diagram of a semi-solid circuit breaker with full electrical data intelligent acquisition function provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] To further illustrate the technical means and effectiveness of the present invention in achieving its intended objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effectiveness of a semi-solid-state circuit breaker with intelligent full-power data acquisition capabilities proposed in accordance with the present invention. In the following description, references to different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0031] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0032] The following describes in detail a specific solution of a semi-solid circuit breaker with a full electric quantity data intelligent collection function provided by the present invention with reference to the accompanying drawings.

[0033] Example:

[0034] The main application scenario of the present invention is: during the data collection process of the semi-solid circuit breaker, the collected data may be missing due to factors such as poor contact and transmission interference, which in turn affects the identification and early warning of abnormal conditions in the current monitoring circuit and affects the safety of the circuit operation.

[0035] See also Figure 1 , which shows a unit block diagram of a semi-solid circuit breaker with a full power data intelligent acquisition function provided by an embodiment of the present invention, the semi-solid circuit breaker includes the following units:

[0036] The data collection unit is used to collect different types of full power data within a preset time period.

[0037] The full electrical data includes multiple types of data. One embodiment of the present application collects five types of full electrical data, including current data, voltage data, power data, harmonic data, and temperature data. Therefore, it is necessary to set a current transformer and a resistor divider in the semi-solid circuit breaker to collect current data and voltage data respectively. The power data is obtained by multiplying the current and voltage in the controller, and the harmonic data is obtained using the FFT module. Because some abnormal performance of the circuit will also be clearly reflected in the temperature, it is also necessary to set a temperature sensor to monitor the temperature data. In order for the circuit breaker to be able to identify and respond to circuit abnormalities more quickly, the monitoring frequency of current, voltage, power, and harmonic data is 10kHz. Since the temperature changes relatively slowly, the temperature data is monitored every 1s.

[0038] Furthermore, to facilitate analysis, it is necessary to have a one-to-one correspondence between the monitoring moments of different types of data. Since temperature data is monitored at a low frequency, the temperature data at one monitoring moment represents the temperature value within 1 second after monitoring a regular customer. This is to achieve the purpose of data filling and align each type of full power data. Since the value ranges of different types of full power data are different, to facilitate calculation and comparison, the data obtained in each type of full power data is normalized, and subsequent calculations are performed on the normalized data values. The time for collecting full power data is a preset period, which is one day. The different types of full power data analyzed in this application are data collected in the most recent day.

[0039] The noise possibility analysis unit is used to obtain the same-position data corresponding to each data within the window centered on each data; and calculate the noise possibility of the data based on the same-position data corresponding to a data.

[0040] Due to interference factors such as electromagnetic interference, noise data, representing sudden data changes, may appear in the collected full power data of various types. This noise data will significantly deviate from the normal fluctuation signal and differ significantly from the actual data, making it unreliable. Therefore, we can first calculate the probability that the historical data is noise data based on the degree of conformity between the historical data and the data variation patterns within the time period. Because data variation characteristics vary depending on the circuit's operating state, the data variation patterns of historical data in the same operating state as the missing data are more consistent with the data characteristics at the location of the missing data. Furthermore, the closer the historical data is to the missing data, the closer the circuit state in that time period is. Therefore, based on the probability that the historical data and the current missing data were in similar operating states, combined with the probability that the historical data is noise data, we can calculate the degree of influence of the historical data on the current missing data interpolation results.

[0041] First, we analyze the data from different types of full power data to determine the probability that each data point is noise data. Under normal conditions, the monitored data changes slightly and follows a regular trend. However, noise data often exhibits sharp, sudden changes. Therefore, we calculate the probability that each historical data point is noise data based on its consistency with the data change characteristics within the time period.

[0042] In order to facilitate the analysis of the data change characteristics at the corresponding moment of each data, a window is established with each data as the center according to the data monitoring frequency and the frequency of data changes. Preferably, the time length of the window in the present invention is 1s. The implementer can adjust the time length of the window according to actual conditions and analyze the characteristics of the data located at the center of the window with the data in the window.

[0043] Due to the physical characteristics of the power grid monitored by circuit breakers, the data they monitor typically exhibits significant periodicity. Therefore, when unaffected by noise, the values ​​of data at the same location within different periods of the full power data should be similar. Therefore, data at the same location within different periods of the data to be analyzed can be identified based on the proximity of the same data point within the window corresponding to each data point and the data to be analyzed.

[0044] Next, the data within the window centered on each data point is used to obtain the corresponding data at the same location. Specifically, for a full battery data set, the sum of the absolute values ​​of the differences between the data in the window corresponding to a data point, excluding the data point itself, and the data at the corresponding location in the window of another data point is obtained. This sum is then added to the hyperparameter and inverted to obtain the inverted result. The inverted result is then normalized to determine the probability that the other data point is the same location data in a different period of the data point.

[0045] Furthermore, a first reference threshold is set, with a value of 0.8. Implementers can adjust the first reference threshold based on data statistics from multiple experiments. If the probability that another data point in a certain type of full power data point is the same position data in a different period is greater than the first reference threshold, then the other data point is the data at the same position as the data point. This allows you to obtain the data at the same position corresponding to each data point in the same type of full power data point.

[0046] Another calculation model for the possibility of data at the same position in different periods of a data is:

[0047] ,

[0048] in, It indicates the possibility that the jth data in the cth type of full power data is the data at the same position in different periods of the ith data. norm is the normalization function. Indicates the number of data contained in the window corresponding to the i-th data (excluding the i-th data itself); Indicates the value of the oth data in the window corresponding to the i-th data in the c-th type of full power data (the oth data cannot be the i-th data itself). Indicates the value of the oth data in the window corresponding to the jth data in the cth data. It represents the absolute value of the difference between the data at the same position in the window corresponding to the i-th data and the data at the same position in the window corresponding to the j-th data. This means that the smaller the sum of the absolute values ​​of the differences, the closer the data changes within the two data windows, that is, the more likely they are in the same position across different periods. ε is a hyperparameter used to ensure that the denominator of the fraction is not zero and is set to 0.01 here.

[0049] If a data point is not noise data, then its value should be close to the value of the corresponding data at the same location obtained according to the above process. Otherwise, it means that the data value deviates significantly from the variation pattern of the current data, that is, it may be noise data. Therefore, the noise probability of a data point is calculated based on the corresponding data at the same location.

[0050] Specifically, the average of the absolute values ​​of the differences between a data and its corresponding data at the same position is obtained and normalized to obtain the noise probability of the data.

[0051] The specific calculation model of noise possibility is:

[0052] ,

[0053] in, represents the noise possibility of the i-th data in the c-th full power data, Indicates the number of data at the same position corresponding to the i-th data, Indicates the value of the i-th data in the c-th type of full power data, Indicates the value of the jth data at the same position in the data at the same position corresponding to the i-th data in the c-th type of full power data; Indicates the absolute value of the difference between the i-th data and its corresponding data at the same position, This represents the average difference between the i-th data point and the data at the same period position (data at the same location). The larger the average value, the more the i-th data point deviates from the normal data variation pattern, that is, the more likely it is noise data. This can be used to obtain the noise probability of each data point.

[0054] The data initial segmentation unit is used to obtain the turning moment probability of the data based on the slope value corresponding to two adjacent data and the noise probability of the data; and to segment the full power data according to the turning moment probability of the data at the same time in different types of full power data to obtain initial data segments.

[0055] Because data interpolation using the LSTM algorithm predicts missing data based on the variation patterns of historical data, and data variation patterns can vary across different circuit operating states, predicting missing data using historical data with similar circuit operating states at the time corresponding to the missing data yields more accurate results. Because the probability of the historical data and the currently missing data being in similar operating states must be calculated based on the similarity between the data variation patterns of the historical data's time period and the missing data's time period, the historical data must first be divided into time periods, each of which must be in the same operating state. Furthermore, because data in different states may exhibit similar behavior over a short period of time, while exhibiting significant overall variation, the time period division should cover the entire data variation cycle to avoid misidentifying data in different operating states as identical. Since changes in a circuit also alter its data variation patterns, the closer the temporal proximity between the historical data and the missing data, the less likely the circuit has changed; that is, the more likely they are to share the same data variation pattern. Therefore, the probability of each historical data point being in a similar operating state to the currently missing data point is calculated based on the similarity between the data variation patterns of the historical data and the currently missing data's time period, combined with the time interval between the historical data and the currently missing data.

[0056] Because the data change trend under the same working state is relatively stable, the slope value of each data in its corresponding time domain diagram is obtained (The slope value at the i-th data point in the time-domain graph of the c-th type of full power data). A significant change in the slope value indicates a significant change in the data trend at that moment, indicating a possible change in the circuit's operating state. Therefore, the probability that each data point corresponds to a change in operating state is calculated. Considering that the slope of data before and after noise data can also significantly change, historical data with a high probability of being noise data should be excluded.

[0057] The turning point probability of a data set is then obtained based on the slope values ​​corresponding to two adjacent data points and the noise probability of the data. Specifically, within a type of full power data, a time domain graph corresponding to the full power data set is obtained. In the time domain graph, the absolute value of the difference between the slopes of two adjacent data points before and after the data point is obtained and normalized to obtain the slope difference. The slope difference is then multiplied by the difference between a preset value and the noise probability of the data point to obtain the turning point probability of the data point.

[0058] The specific calculation model for the possibility of a turning point is: ,

[0059] in, is the turning point possibility of the i-th data in the c-th type of full power data, indicating the possibility that the corresponding moment of the i-th data in the c-th type of full power data is the moment of working state change. Indicates the slope value corresponding to the data at the moment after the i-th data, Indicates the slope value corresponding to the data at the moment before the i-th data, It represents the difference between the data change trend before and after the corresponding moment of the i-th data. The larger the difference, the more likely it is the moment when the working status changes. and Respectively represent the maximum and minimum slope values ​​corresponding to each data in the cth type of full power data, which are used to Normalization, represents the noise possibility of the i-th data, The higher the possibility that the i-th data is noise data, the lower the possibility that the moment corresponding to the data is the real working status change moment.

[0060] Because when the working status changes, it often affects the data change patterns in multiple dimensions. In order to avoid identifying an abnormal fluctuation in a certain data as the moment of working status change, it is necessary to integrate the possibility that the corresponding moment of data in different dimensions is the moment of working status change.

[0061] The full power data is segmented to obtain initial data segments based on the likelihood of a turning moment for data at the same moment in different types of full power data. Specifically, the average of the likelihood of a turning moment for data at the same moment in different types of full power data is calculated and recorded as the average likelihood of a turning moment at that moment. If the average likelihood of a turning moment at a moment is greater than a second reference threshold, that moment is considered a suspected working state change moment, and the suspected working state change moment is used to segment the full power data to obtain initial data segments.

[0062] The calculation model of the possible turning point average is as follows: ,

[0063] in, Indicates the possible average value of the turning point at the moment (i-th moment) corresponding to the i-th data, Indicates the number of types of total power data. Indicates the probability of a turning point for the i-th data point in the c-th type of full power data. The second reference threshold is for reference only and should be adjusted by implementers based on actual conditions. After segmenting the full power data based on the suspected operating state change moments, an initial data segment corresponds to multiple types of full power data within the corresponding time period of that initial data segment.

[0064] The data segmentation correction unit is used to calculate the periodic regularity similarity based on the time length difference and data difference between two initial data segments; and merge the initial data segments according to the periodic regularity similarity between each two initial data segments to obtain the final data segment.

[0065] Because circuit monitoring data typically changes periodically, segmenting the data into different data segments based solely on the moment when the data trend changes may separate continuous data under the same operating state, hindering analysis of the overall data variation patterns under different operating states. Therefore, it is necessary to merge the initial data segments obtained in the above process so that each segment in the final segmentation result can relatively completely cover the monitoring data under the current operating state.

[0066] According to the periodic change characteristics of the data, the similarity of the data change trend between each initial data segment and other initial data segments and the similarity of the corresponding time length of the data segments are calculated, that is, the periodic regularity similarity is calculated based on the time length difference and data difference between the two initial data segments.

[0067] Specifically, the absolute value of the difference in time length between the two initial data segments is added to the hyperparameter and the inverse is taken to obtain the time length similarity eigenvalue; the absolute value of the difference in a type of full power data at a moment between the two initial data segments is obtained, recorded as the difference of the type of full power data at the moment, the absolute value of the difference in the type of full power data at each two adjacent moments is summed up, added to the hyperparameter and the inverse is taken to obtain the data similarity eigenvalue between the two initial data segments corresponding to the type of full power data, the sum of the data similarity eigenvalues ​​between the two initial data segments corresponding to each type of full power data is obtained to obtain the comprehensive data similarity eigenvalue between the two initial data segments; the time length similarity eigenvalue and the comprehensive data similarity eigenvalue are multiplied and normalized to obtain the periodic regularity similarity between the two initial data segments.

[0068] The calculation model of periodic regularity similarity is as follows:

[0069] ,

[0070] in, is the similarity of the periodicity between the u-th initial data segment and the v-th initial data segment, indicating the possibility that the u-th initial data segment and the v-th initial data segment are in the same periodic variation law. Indicates the time length corresponding to the u-th initial data segment, Indicates the time length corresponding to the vth initial data segment, Indicates the closeness between the time length of the u-th initial data segment and the v-th initial data segment, ε is a hyperparameter used to ensure that the fraction is meaningful. is the similarity eigenvalue of time length, where Indicates the number of data contained in the u-th initial data segment, It represents the absolute value of the difference between the data in the cth type of full power data corresponding to the i-th moment in the u-th initial data segment and the v-th initial data segment, that is, the difference between the cth type of full power data at the i-th moment, This represents the absolute difference between the cth type of full power data corresponding to the i+1th time in the uth initial data segment and the vth initial data segment, i.e., the difference between the cth type of full power data at the i+1th time. It should be noted that if the two initial data segments are of unequal length, any misaligned positions are padded with zeros. The absolute value of the difference between the cth type of full power data at two adjacent moments (i-th moment and i+1-th moment) (the smaller the absolute value, the closer it is), sum it up and add it to the hyperparameter, and then invert it to get , represents the data similarity feature value between the two initial data segments corresponding to the c-th type of full power data. The closer the difference between the adjacent position data is, the more similar the data change trends of the two initial data segments are, that is, the more likely they are in the same periodic change law.

[0071] Furthermore, the initial data segments are merged according to the similarity of the periodic regularity between each two initial data segments to obtain the final data segment. A third reference threshold and a time interval threshold are set, and starting from the first initial data segment, the initial data segments whose periodic regularity similarity with the first initial data segment is greater than the third reference threshold are obtained and recorded as suspected similar data segments; the suspected similar data segments are arranged in time sequence to obtain a first sequence, and the time interval between the starting moments of two adjacent suspected similar data segments in the first sequence is obtained as the adjacent interval between the two adjacent suspected similar data segments; if the adjacent interval between every two adjacent suspected similar data segments in the first sequence is less than or equal to the time interval threshold, then all the suspected similar data segments in the first sequence are the periodic data segments corresponding to the first initial data segment; if there is a For adjacent suspected similar data segments whose adjacent intervals are greater than the time interval threshold, the adjacent intervals between such adjacent suspected similar data segments are recorded as discontinuous intervals, and the first discontinuous interval is located in chronological order. At this time, the suspected similar data segment before the first discontinuous interval is the periodic data segment corresponding to the first initial data segment; and so on, starting from the first initial data segment in the remaining initial data segments in chronological order, the periodic data segment corresponding to the first initial data segment in the remaining initial data segments is searched until all the periodic data segments are obtained; the two adjacent periodic data segments in the initial data segments that are periodic data segments are merged in chronological order to obtain the final data segment.

[0072] It should be noted that the reference values ​​of the third reference threshold and the time interval threshold are 0.6 and 5, and the time interval threshold is 5 moments. Among all the initial data segments, some initial data segments are not periodic data segments. In this case, these segments are not merged and are directly used as the final data segments. There may also be independent periodic data segments. In this case, they are not merged and are directly used as the final data segment. If there are three consecutive periodic data segments, the first two are merged in order, and the last one is not merged and is directly used as the final data segment.

[0073] In this way, the final data segment division result can be obtained.

[0074] The missing data completion unit is used to set a sliding window based on the length of the final data segment where a missing data is located in a type of full power data, and use the sliding window to slide on other final data segments to obtain the influence weight of each data in other final data segments; and complete the missing data based on the data in other final data segments and their influence weights.

[0075] Based on the final data segment division results obtained in the above steps, the data change characteristics of the final data segment containing the missing data are compared with those of the other final data segments. The higher the degree of overlap between the final data segment containing the missing data and the other final data segments, the more consistent the operating state of the circuit in the final data segment with that of the missing data, that is, the greater the influence of the historical data on the current missing data interpolation results. A sliding window is set based on the time length of the final data segment containing the missing data. The time length of the sliding window is the same as the time length of the final data segment containing the missing data. The sliding window is slid across each final data segment, and the degree of overlap between the final data segment containing the missing data and the data within the sliding window is calculated in sequence. The maximum degree of overlap is the degree of overlap between the data segment to be analyzed and the data segment containing the missing data.

[0076] However, if a data point is noise data, even if its final data segment has a high degree of overlap with the final data segment of the missing data, this data point still cannot effectively reflect the correct interpolation result of the missing data. Therefore, considering the possibility that the data point is noise and the degree of overlap between the final data segment of the data point and the final data segment of the current missing data point, the degree of influence of each data point on the current missing data interpolation result is calculated.

[0077] A sliding window is used to slide on other final data segments to obtain the influence weight of each data in other final data segments; specifically, a sliding window is used to slide on the final data segment to which a data belongs, with a sliding step size of a preset value, to obtain the average of the absolute values ​​of the differences between each data in a sliding window and each corresponding data in the final data segment to which a missing data belongs, and the inverse is calculated to obtain the overlap corresponding to the sliding window; the maximum value of the overlap corresponding to each sliding window is obtained as the maximum overlap; the maximum overlap is multiplied by the difference between the preset value and the noise possibility of the data to obtain the influence weight of the data.

[0078] The calculation model of the data's impact weight is as follows:

[0079] ,

[0080] in, is the influence degree of the i-th data in the c-th type of full electricity data on the interpolation result of a missing data in the c-th type of full electricity data, that is, the influence weight of the i-th data in the c-th type of full electricity data. is the noise probability of the i-th data in the c-th full power data, 1 is the preset value, It means that when the sliding window slides on the final data segment to which the i-th data belongs, the j-th data in the w-th sliding window is, Indicates the jth data in the final data segment where the missing data is located, It indicates the degree of overlap between the data in the sliding window and the data change in the final data segment where the missing data is located. The smaller the absolute value of the difference, the higher the overlap, that is, the closer the circuit working state is. The average overlap is used to represent the overlap between the data in the sliding window and the data in the final data segment where the missing data is located. max represents the maximum value operation, which takes the maximum value of the overlap corresponding to each sliding window to represent the overlap between the final data segment where the i-th data is located and the final data segment where the missing data is located.

[0081] Furthermore, the influence weight of each data point in the final data segment other than the data point in the final data segment containing the missing data point is obtained. The influence weight of each data point in the other final data segment is multiplied by each data point in the other final data segment to obtain the updated data corresponding to each data point in the other final data segment. The missing data complement is calculated based on the updated data corresponding to each data point in the other final data segment and the LSTM algorithm. Because the data values ​​used for analysis are normalized, the interpolation results must also be restored to the range of the corresponding data category.

[0082] In summary, this application compares various historical full-power data segments to determine the data variation patterns for different operating states. Based on the similarity of the operating states between the historical data and the missing data, as well as the temporal proximity of the historical data and the missing data, the application adjusts the weight of the historical data's influence on the missing value interpolation results. Finally, using the LSTM algorithm, combined with the weight of the historical data's influence, the interpolation results for the missing value locations are calculated. This effectively solves the aforementioned problems, improves the accuracy of the collected data, and facilitates further analysis of circuit anomalies.

[0083] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0084] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A semi-solid circuit breaker with full power data intelligent acquisition function, characterized in that: The semi-solid circuit breaker includes: A data collection unit, used to collect different types of full power data within a preset time period; A noise possibility analysis unit is used to obtain the corresponding co-location data of each data within the window centered on each data; and calculate the noise possibility of a data based on the corresponding co-location data of the data; The data initial segmentation unit is used to obtain the turning point probability of the data based on the slope value corresponding to two adjacent data and the noise probability of the data; the full power data is segmented according to the turning point probability of the data at the same time in different types of full power data to obtain initial data segments; A data segment correction unit is used to calculate the periodic regularity similarity based on the time length difference and data difference between two initial data segments; and to merge the initial data segments according to the periodic regularity similarity between each two initial data segments to obtain a final data segment; A missing data completion unit is configured to set a sliding window based on the length of a final data segment containing a missing data in a type of full power data, slide the sliding window over other final data segments to obtain the influence weights of each data in the other final data segments, and complete the missing data based on each data in the other final data segments and their influence weights; The calculating of the periodic regularity similarity based on the time length difference and data difference between the two initial data segments includes: The absolute value of the difference in time length between the two initial data segments is added to the hyperparameter and the inverse is taken to obtain the time length similarity eigenvalue; the absolute value of the difference in a type of full power data at a moment between the two initial data segments is obtained, recorded as the difference of the type of full power data at that moment, the absolute value of the difference between the type of full power data at each two adjacent moments is summed up, and then added to the hyperparameter and the inverse is taken to obtain the data similarity eigenvalue between the two initial data segments corresponding to the type of full power data, and the sum of the data similarity eigenvalues ​​between the two initial data segments corresponding to each type of full power data is obtained to obtain the comprehensive data similarity eigenvalue between the two initial data segments; the time length similarity eigenvalue and the comprehensive data similarity eigenvalue are multiplied and normalized to obtain the periodic regularity similarity between the two initial data segments.

2. A semi-solid circuit breaker with full electric quantity data intelligent collection function according to claim 1, characterized in that: The data in the window centered on each data is used to obtain the corresponding co-location data of each data, including: In a type of full power data, the sum of the absolute values ​​of the differences between the data other than the data itself in a window corresponding to one data and the data at the corresponding position in the window of another data is obtained, and then added to the hyperparameter and inverted to obtain the inverted result. The inverted result is normalized to obtain the possibility that the other data is the data at the same position in a different period of the data; if the possibility that the other data is the data at the same position in a different period of the data is greater than a first reference threshold, then the other data is the data at the same position corresponding to the data.

3. The semi-solid circuit breaker with full power data intelligent acquisition function according to claim 1 is characterized in that: The calculating the noise possibility of a piece of data according to the co-located data corresponding to the piece of data includes: The noise probability of the data is obtained by obtaining the average of the absolute values ​​of the differences between a data and its corresponding data at the same position and normalizing them.

4. The semi-solid circuit breaker with full electric quantity data intelligent collection function according to claim 1 is characterized in that: The method of obtaining the turning point possibility of the data based on the slope values ​​corresponding to two adjacent data and the noise possibility of the data includes: For a type of full power data, a time domain graph corresponding to the full power data is obtained. In the time domain graph, the absolute value of the difference in slopes corresponding to two adjacent data before and after the data is obtained and normalized to obtain a slope difference. The slope difference is multiplied by the difference between a preset value and the noise probability of the data to obtain the turning moment probability of the data.

5. The semi-solid circuit breaker with full electric quantity data intelligent acquisition function according to claim 1 is characterized in that: The step of segmenting the full power data to obtain initial data segments according to the likelihood of transition moments of data at the same moment in different types of full power data includes: Calculate the mean of the turning point possibilities of the data at the same moment in different types of full power data, and record it as the turning point possibility average value at that moment; if the turning point possibility average value at a moment is greater than the second reference threshold, then the moment is a suspected working state change moment, and the suspected working state change moment is used to segment the full power data to obtain the initial data segment.

6. The semi-solid circuit breaker with full electric quantity data intelligent collection function according to claim 1 is characterized in that: The step of merging the initial data segments according to the periodic regularity similarity between each two initial data segments to obtain the final data segment includes: Starting from the first initial data segment, obtain the initial data segment whose periodic regularity similarity with the first initial data segment is greater than the third reference threshold, and record it as a suspected similar data segment; arrange the suspected similar data segments in time sequence to obtain a first sequence, and obtain the time interval between the starting moments of two adjacent suspected similar data segments in the first sequence as the adjacent interval between the two adjacent suspected similar data segments; if the adjacent interval between every two adjacent suspected similar data segments in the first sequence is less than or equal to the time interval threshold, then all the suspected similar data segments in the first sequence are the periodic data segments corresponding to the first initial data segment; if the adjacent interval is greater than ... Adjacent suspected similar data segments with an interval threshold are selected, and the adjacent intervals between such adjacent suspected similar data segments are recorded as discontinuous intervals. The first discontinuous interval is located in chronological order. At this time, the suspected similar data segment before the first discontinuous interval is the periodic data segment corresponding to the first initial data segment; and so on, starting from the first initial data segment in the remaining initial data segments in chronological order, the periodic data segment corresponding to the first initial data segment in the remaining initial data segments is searched until all the periodic data segments are obtained; the two adjacent periodic data segments in the initial data segments that are periodic data segments are merged in chronological order to obtain the final data segment.

7. The semi-solid circuit breaker with full electric quantity data intelligent collection function according to claim 1 is characterized in that: The method of using the sliding window to slide on other final data segments to obtain the influence weight of each data in other final data segments includes: A sliding window is used to slide on the final data segment to which a data belongs, with the sliding step being a preset value. The average of the absolute values ​​of the differences between each data in a sliding window and each corresponding data in the final data segment to which a missing data belongs is obtained, and the inverse is calculated to obtain the overlap corresponding to the sliding window. The maximum value of the overlap corresponding to each sliding window is obtained as the maximum overlap. The maximum overlap is multiplied by the difference between the preset value and the noise possibility of the data to obtain the influence weight of the data.

8. The semi-solid circuit breaker with full electric quantity data intelligent collection function according to claim 1 is characterized in that: The method of completing the missing data based on the data in other final data segments and their influence weights includes: Obtain the influence weight of each data in other final data segments except the data in the final data segment where the missing data is located, multiply the influence weight of each data in other final data segments by each data in other final data segments respectively to obtain the updated data corresponding to each data in other final data segments; complete the missing data according to the updated data corresponding to each data in other final data segments and the LSTM algorithm.

Images