Multi-source data fusion system operation monitoring system

Abstract

The invention relates to the technical field of industrial measurement and control, in particular to a multi-source data fusion system operation monitoring system which comprises a jump behavior recognition module, a confidence coefficient screening module, a behavior consistency extraction module, a power recovery anomaly detection module and an operation state monitoring output module. According to the method, the sudden change trend in the current and voltage signals of the operation equipment is extracted, the response dislocation condition between the signals is combined, the behavior change of the equipment can be accurately described, potential abnormal fragments are identified, the signal fragments are screened by means of the multi-dimensional confidence factor, and low-reliability data content is removed; the method comprises the following steps: establishing a characteristic time section with representativeness through time interlacing characteristics between triggering behaviors and behavior type consistency judgment, and realizing accurate marking of abnormal behaviors in a power recovery process of equipment by combining trend delay and fluctuation amplitude change of a power signal in a voltage rising process.

Description

Technical Field

[0001] The present invention relates to the field of industrial measurement and control technology, and in particular to a system operation monitoring system that integrates multi-source data. Background Art

[0002] The field of industrial measurement and control technology refers to a type of engineering technology system that continuously monitors and adjusts the operating processes of key links during industrial production and operation to ensure the safety, reliability and efficiency of the system. It regulates production behavior through specific control logic and makes timely responses and adjustments based on real-time on-site conditions.

[0003] Among them, traditional station power system operation monitoring refers to methods used to observe and record the operating conditions of internal power supply systems such as substations and distribution rooms, so as to determine whether various power-consuming units in the station are in normal condition, and whether there are power anomalies, energy consumption deviations or operation deviations.

[0004] Traditional station power system operation monitoring usually relies on the basic status recording and manual judgment of current and voltage. It cannot effectively distinguish the difference between behavioral dislocations and normal fluctuations of equipment in situations such as sudden switching and power disturbances. During complex load switching or voltage recovery, it is difficult to identify situations where equipment power has not been restored in time or there is a response lag. Especially in scenarios where multiple devices are operating in coordination or with high frequency start and stop, a single data source is prone to information misjudgment and abnormal omissions, resulting in delayed system status identification or untimely early warning response. For example, when multiple devices operate continuously in a short period of time, signal interference may be superimposed and mask the abnormal performance of a key device, thereby affecting the operational stability and troubleshooting efficiency of the overall power supply system. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a system operation monitoring system with multi-source data fusion.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a system operation monitoring system of multi-source data fusion includes: The jump behavior identification module obtains the current sampling data and voltage on-off signals of the running equipment in the station power system, identifies the jump trend behavior of each device, and obtains the behavior response dislocation data segment; A confidence screening module calculates the direction score and amplitude deviation of the signal segment corresponding to the device jump trend behavior trigger point based on the behavior response misalignment data segment, and screens the data source segment whose confidence needs to be lowered; A behavior consistency extraction module removes data source segments whose confidence needs to be lowered, constructs interleaved time intervals consisting of adjacent trigger points based on the retained device jump trend behavior trigger points, analyzes the overlap of trigger points within the intervals, and extracts behavior consistency feature segments; A power recovery anomaly detection module extracts power sampling data of load devices in the station power system based on the staggered time intervals of the behavioral consistency feature segments, determines the power recovery delay during the voltage recovery process, and marks the power recovery anomaly segment; The operation status monitoring output module generates an abnormal entry sequence in chronological order based on the power recovery abnormality segment and outputs the operation status monitoring result of the station power system.

[0007] As a further solution of the present invention, the behavior response misalignment data segment includes the trigger point time difference, the adjacent trigger point misalignment identifier, and the current mutation segment; the data source segment whose confidence needs to be lowered includes the direction score value, the amplitude deviation value, and the low confidence label; the behavior consistency feature segment includes the staggered time index, the repeated trigger frequency, and the behavior event type; the power recovery abnormality segment includes the power recovery starting point, the power trend starting point, and the power slope fluctuation; the operation status monitoring result includes the abnormal equipment identifier, the abnormal data entry, and the output time series.

[0008] As a further solution of the present invention, the jump behavior recognition module includes: The electrical parameter acquisition structured submodule obtains current sampling data and voltage on / off signals from the equipment running in the station's power system. It calculates the rate of change sequence between consecutive sampling points for the current sampling data, extracts the sudden change segments, and combines them with the edge jump time points in the voltage on / off signals to perform a unified period normalization process on the two types of signal sampling information to obtain a structured electrical parameter change sequence. The behavior trajectory synchronization submodule calls the current change rate segment and voltage jump point sequence in the structured electrical parameter change sequence to synchronize the position of each device behavior trajectory, and pairs each change rate peak with the corresponding voltage change point for detection to obtain a corresponding trajectory group of the device behavior; The response offset identification submodule determines whether the time difference information of each pair of signals in the trajectory group corresponding to the device behavior exceeds the set time offset threshold. For the behavior segments that meet the conditions, the time interval, device number and signal alignment status information are extracted to obtain the behavior response misalignment data segment.

[0009] As a further solution of the present invention, the confidence screening module includes: A response segment extraction submodule extracts, based on the behavior response misalignment data segment, a signal change segment of each jump trend behavior trigger point in the current change rate sequence and a direction change segment in the voltage on / off signal to obtain a jump signal alignment segment group; The offset calculation submodule calls the jump signal to align each group of current change rate segments and voltage direction segments in the segment group, constructs a time matching vector group, calculates the cosine value of the angle between the time matching vector group and the direction score, extracts the maximum value and the mean value of the reference segment in each group of current change segments, calculates the ratio of the two as the amplitude deviation, and obtains a combination of the direction score and the amplitude deviation; The confidence value determination submodule determines whether the direction score and amplitude deviation combination exceeds the preset offset score threshold and amplitude deviation threshold based on the score and deviation items, filters the signal segments that meet the dual conditions, extracts the device number and time index to which they belong, marks the data content as having confidence that does not meet the requirements, and obtains the data source fragment whose confidence needs to be lowered.

[0010] As a further solution of the present invention, the behavior consistency extraction module includes: The behavior point integration submodule removes the data source segments whose confidence needs to be lowered, extracts the time index and device number information of the retained device jump trend behavior trigger point, and obtains the jump behavior interleaved sequence group; The interval frequency aggregation submodule calls the jump behavior interleaved sequence group, matches the event types defined in the behavior annotation set, counts the occurrence frequency of the event type in each combination, and filters the jump behavior concentrated segment group; The coupling feature screening submodule performs a conditional independence test on the relationship between the behavior frequency and the on / off state of the corresponding device based on the concentrated segment group of the jump behavior, calculates the chi-square value and screens the statistical correlation degree of the segment, and retains the behavior consistency feature fragments.

[0011] As a further solution of the present invention, the power recovery abnormality detection module includes: The power data extraction submodule calls each staggered time interval in the behavior consistency feature segment, obtains the power sampling sequence of the load device in the corresponding time period, and constructs a load power time series structure set; The recovery feature calculation submodule extracts the time index of the voltage recovery starting point and the power trend starting change point based on the load power time series structure set, calculates the response interval between the two, obtains the power slope sequence within a fixed period after the recovery, determines the fluctuation amplitude between each slope in the continuous segment, and obtains the power recovery response feature group; The delayed segment marking submodule, based on the power recovery response feature group, filters out segments whose response interval exceeds the recovery offset threshold and whose slope fluctuation amplitude exceeds the stability limit, extracts the corresponding time range and load device identification information, marks them as abnormal state intervals in the power recovery process, and obtains the power recovery abnormal segment.

[0012] As a further solution of the present invention, the operating status monitoring output module includes: The abnormality indicator extraction submodule calls each data structure in the power recovery abnormality segment, extracts the recorded load device identification, response delay time and power change slope fluctuation information, combines the three indicators according to the device dimension, and generates an abnormality indicator combination set; An entry sequence construction submodule arranges the status information entries corresponding to all abnormal devices in chronological order according to the time index of each group of data in the abnormal indicator combination set to generate a sequence of abnormal behavior time series entries; The monitoring result output submodule writes the sorted content in the abnormal behavior time sequence entry sequence into the station power system operation status record table to obtain the station power system operation status monitoring result.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by extracting the mutation trend in the current and voltage signals of the running equipment and combining the response misalignment between the signals, it is possible to accurately characterize the changes in equipment behavior and identify potential abnormal segments. The signal segments are screened with the help of multi-dimensional confidence factors to eliminate low-reliability data content. Then, by judging the time interleaving characteristics and behavior type consistency between triggering behaviors, a representative characteristic time segment is constructed. Combined with the trend delay and fluctuation amplitude change of the power signal during the voltage recovery process, accurate marking of abnormal behaviors in the equipment power recovery process is achieved, and finally a sequence of abnormal states arranged in chronological order is formed, realizing high-timeliness monitoring of the operating status. This processing logic significantly improves the robustness and recognition accuracy of data analysis, and enhances the early warning capability and fault response efficiency of the station power system in actual application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a system flow chart of the present invention; Figure 2 This is a flow chart of the jump behavior recognition module of the present invention; Figure 3 This is a flow chart of the confidence screening module of the present invention; Figure 4 This is a flow chart of the behavior consistency extraction module of the present invention; Figure 5 This is a flow chart of the power recovery anomaly detection module of the present invention; Figure 6 This is a flow chart of the operating status monitoring output module of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0016] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0017] See also Figure 1 ,The system operation monitoring system of multi-source data fusion includes: The jump behavior identification module obtains the current sampling data and voltage on-off signals of the running equipment in the station power system, identifies the jump trend behavior of each device, and obtains the behavior response dislocation data segment; The confidence screening module calculates the direction score and amplitude deviation of the signal segment corresponding to the device jump trend behavior trigger point based on the behavioral response misalignment data segment, and screens the data source segments whose confidence needs to be lowered; The behavior consistency extraction module removes data source segments whose confidence needs to be lowered. Based on the retained device jump trend behavior trigger points, it constructs interleaved time intervals consisting of adjacent trigger points, analyzes the overlap of trigger points within the interval, and extracts behavior consistency feature segments. The power recovery anomaly detection module extracts power sampling data from load devices in the station's power system based on the staggered time intervals of behavioral consistency feature segments, determines the power recovery delay during voltage recovery, and marks power recovery anomaly segments. The operation status monitoring output module generates a sequence of abnormal entries in chronological order based on the power recovery abnormality segment and outputs the operation status monitoring results of the station power system; The behavioral response dislocation data segment includes the trigger point time difference, the adjacent trigger point dislocation identifier, and the current mutation segment. The data source segment whose confidence needs to be lowered includes the direction score value, the amplitude deviation value, and the low-confidence label. The behavioral consistency feature segment includes the staggered time index, the repeated trigger frequency, and the behavioral event type. The power recovery anomaly segment includes the power recovery starting point, the power trend starting point, and the power slope fluctuation. The operating status monitoring results include the abnormal equipment identifier, abnormal data entry, and the output time series.

[0018] See also Figure 2 , the jump behavior recognition module includes: The electrical parameter acquisition structured submodule obtains current sampling data and voltage on / off signals from the equipment running in the station's power system. It calculates the rate of change sequence between consecutive sampling points for the current sampling data, extracts the sudden change segments, and combines them with the edge jump time points in the voltage on / off signals to perform a unified period normalization process on the two types of signal sampling information to obtain a structured electrical parameter change sequence. Obtain the current sampling data and voltage on-off signals of the running equipment in the station power system. First, calculate the change rate between consecutive sampling points of the current sampling data. Assuming that 5 current data points are collected every 1 second, and the current of a certain device is 10A, 10.2A, 11A, 9.5A, and 8A respectively, then its change rate is: the second point is , point 3 is , point 4 is , point 5 is , these change rates are formed into a change rate sequence, and then the mutation segment is found in the sequence. The change threshold can be set to When the rate of change of two consecutive points exceeds the threshold, it is marked as a mutation segment. Therefore, points 3 to 5 are marked as mutations. After the mutation extraction is completed, the edge jump time point in the voltage on-off signal is extracted. Assuming that the voltage state changes from "on" to "off" at the 3rd second, the edge jump point time is identified as the 3rd second. The current mutation segment and the voltage jump time are normalized to a unified cycle. Assuming the sampling period is 1 second and the normalized period length is 5 seconds, the current change rate sequence and the voltage jump point information are synchronized to the cycle. For example, if the jump occurs in the 3rd second and coincides with the current change mutation segment, the structured electrical parameter change sequence constructed in the cycle is: change rate peak position 3, voltage jump time point 3, and a complete structured sample is formed after normalization.

[0019] The behavior trajectory synchronization submodule calls the current change rate segment and voltage jump point sequence in the structured electrical parameter change sequence to synchronize the position of each device behavior trajectory, and pairs each change rate peak with the corresponding voltage change point for detection to obtain the corresponding trajectory group of the device behavior; Call the current change rate segment and voltage jump point sequence in the structured electrical parameter change sequence to synchronize the behavior trajectory position of each device. First, based on the time point of the change rate peak, such as the 3rd second mentioned above, set this moment as the key point in the current change trajectory of the device, and then find whether there are time points that coincide or are close in the voltage jump time point. Set the time matching threshold to Seconds, that is, between 2.8 seconds and 3.2 seconds. If the voltage jump occurs at 3.1 seconds, it is successfully paired with the current change peak. The pair of peaks and jump points are matched into a set of trajectory points. Then, according to the sampling device number and time point, a behavior trajectory group is constructed. For example, if the device ID is D001, the trajectory group record is [D001, t=3.0, current mutation, voltage jump]. The above process is performed for each device to obtain the trajectory group corresponding to the device behavior. Each trajectory group corresponds to a synchronous behavior sample composed of the electrical parameter changes of a device within a specific time.

[0020] The response offset identification submodule determines whether the time difference information of each pair of signals in the device behavior corresponding trajectory group exceeds the set time offset threshold. For the behavior segments that meet the conditions, the time interval, device number and signal alignment status information are extracted to obtain the behavior response misalignment data segment. According to the time difference of each pair of signals in the trajectory group corresponding to the device behavior, it is judged whether it exceeds the set time offset threshold. The offset threshold is set to 0.5 seconds. If the current mutation time in a trajectory group is 3 seconds and the voltage jump time is 3.7 seconds, then the time difference is Seconds, because , is determined to be a response misalignment behavior, and is extracted. The time interval of the segment is recorded as 0.7 seconds, the device number is D001, and the alignment status is "unaligned". If the current of a device suddenly changes at 4 seconds and the voltage jumps at 4.3 seconds, the difference is If the threshold condition is met, it will be ignored. Similarly, all response offset fragments are extracted. Each record includes the time interval (in seconds), device number (such as D002), and signal alignment status (such as misaligned).

[0021] See also Figure 3 , the confidence screening module includes: The response segment extraction submodule extracts the signal change segments of each jump trend behavior trigger point in the current change rate sequence and the direction change segments in the voltage on-off signal based on the behavior response misalignment data segments, and obtains the jump signal alignment segment group; Based on the behavioral response misalignment data segment, the signal change segment of each jump trend behavior trigger point in the current change rate sequence and the direction change segment in the voltage on-off signal are extracted. First, the trigger point of the voltage jump trend is located. Assume that the jump from on to off occurs at the 5th second. According to the offset record, the current change rate of the device suddenly changes at the 4.6th second. The current change rate data within the range of 1 second before and after the jump point is extracted to form the signal change segment of the trigger point. If the sampling frequency in this period is 5Hz, the current change rate segment consists of 10 points. Then the corresponding direction change is extracted from the voltage on-off signal. The voltage is determined to be high (e.g., 220V) before the transition and low (e.g., 0V) after the transition. The direction change can be defined as a change from 1 to 0. The direction change segment is constructed as a set of direction vectors [1, 0]. The above current change rate segment is combined with the voltage direction change segment to form a transition signal alignment segment group. If the device number is D015, the segment group is recorded as [D015, 5th second, current change rate segment, 4.0–5.0 seconds; voltage direction segment, 5.0–6.0 seconds]. This method is executed for all behavioral response misaligned data segments, and the aligned transition segment groups are extracted one by one.

[0022] The offset calculation submodule uses the jump signal to align each current change rate segment with the voltage direction segment in the segment group, constructs a time matching vector group, calculates the cosine value of the angle between the time matching vector group and the direction score, extracts the maximum value and the mean value of the reference segment in each group of current change segments, and calculates the ratio of the two as the amplitude deviation, thus obtaining a combination of the direction score and the amplitude deviation. Call the transition signal to align each current change rate segment and voltage direction segment in the segment group to construct a time matching vector group. First, extract the sampling sequence of the current change rate of each segment from each transition event. Assume that the current change rate segments of a device are 0.02, 0.04, 0.08, -0.01, and -0.03, and the corresponding voltage direction changes from "on" to "off". The voltage direction vector can be formed as 1, 0, 0, 0, 0, that is, only the first bit reflects the direction jump. The time matching vector group is: Current rate of change vector: ; Voltage direction change vector: ; Next, the current rate of change vector Perform maximum absolute value normalization, the maximum value is 0.08, and the normalized result is: .

[0023] Then, the direction score is calculated and the similarity between the two vectors is calculated by the cosine angle between them. The formula is as follows: ; in, : the angle between two vectors; : Direction score, which indicates the degree of consistency of the directions of two vectors. Direction score is an indicator used to measure the consistency (similarity) between a certain section of current change rate signal and the voltage direction jump. This indicator is used to identify whether the current change rate and voltage direction show consistent physical behavior at the moment of jump, so as to assist in judging whether the event is a valid or typical power system jump behavior. If the score value is close to 1, it means that the change direction of the jump current is highly consistent with the jump direction of the voltage; if the score value is close to 0, it means that there is almost no correlation between the two; if the score value is negative, it means that the two are in opposite directions. In power system event analysis, a high direction score means that the event is very likely to be a physical jump behavior with consistent voltage and current; : Normalized current rate of change vector; : voltage direction change vector; : Vector dot product, which represents the projection product of two vectors in the same direction; :vector The modulus of , indicating its size; :vector The modulus of , indicating its size.

[0024] The specific numerical calculation is as follows: Dot product: ; Mould length: ; Mould length: ; Final direction score: .

[0025] Then calculate the amplitude deviation and extract the maximum value in the current change rate segment , and then extract the mean change rate of the reference segment (such as the previous second) , calculate the ratio of the two as follows: ; in, : Amplitude deviation, which indicates the amplification of the signal mutation amplitude relative to the background change; : The maximum value in the current current change rate segment; : The average rate of change of the reference segment (period without mutation).

[0026] Finally, the result of this segment is a direction score of 0.209 and an amplitude deviation of 4.0. The combined record is [direction score = 0.209, amplitude deviation = 4.0]. The above processing is performed on all jump signal alignment segment groups to generate a complete direction score and amplitude deviation combination data group.

[0027] The confidence value determination submodule determines whether the direction score and amplitude deviation combination exceeds the preset offset score threshold and amplitude deviation threshold based on the score and deviation items. It then selects signal segments that meet these two conditions, extracts the device number and time index, marks the data content as not meeting the confidence requirements, and obtains the data source segments whose confidence needs to be lowered. According to the score and deviation items in the combination of direction score and amplitude deviation, determine whether they exceed the preset offset score threshold and amplitude deviation threshold. First, set the offset score threshold to 0.3 and the amplitude deviation threshold to 2.5. When the direction score of a data is 0.209, which is less than 0.3, and the amplitude deviation is 4.0, which is greater than 2.5, it is identified as an abnormal signal segment because it meets both the conditions of insufficient direction score and excessive amplitude deviation. Further extract the device number and time index to which the segment belongs. Assuming that the segment belongs to device D037 and the time index is the 12th second, it is marked as data content with unsatisfactory confidence and recorded as [D037, 12 seconds, direction score = 0.209, amplitude deviation = 4.0, insufficient confidence]. Similarly, perform conditional screening on each combination of data. Only when the direction score is And the amplitude deviation When the confidence level of a data source segment is lowered, it is recorded as a data source segment whose confidence level needs to be lowered. Otherwise, it is not processed, and eventually a group of abnormal data segments with insufficient labeled confidence levels are formed.

[0028] See also Figure 4 ,The behavior consistency extraction module includes: The behavior point integration submodule removes data source segments whose confidence needs to be lowered, extracts the time index and device number information of the retained device jump trend behavior trigger point, and obtains the jump behavior interleaved sequence group; Device transition behavior refers to sudden, significant changes in current sampling data or voltage on / off signals during device operation, reflecting the transient behavior of the device as it transitions from one stable operating state to another (such as startup, shutdown, or switching). These changes typically manifest as signal patterns such as a sudden increase or decrease in the rate of change of current or a voltage transition edge. These changes exhibit recognizable mutation trends and serve as key indicators of device behavioral state transitions. Eliminate data source segments whose confidence needs to be lowered, and extract the time index and device number information of the retained device jump trend behavior trigger points. First, exclude data records marked as insufficiently confident from all identified jump behavior segments. For example, the previously identified segment [D037, 12 seconds] will be eliminated, and the remaining segments will be retained as valid behavior points. Then traverse the retained segments and extract the time index and device number of each, such as [D012, 10 seconds], [D015, 15 seconds], and [D026, 18 seconds]. These valid behavior trigger information are integrated and sorted in chronological order to form a jump behavior interleaved sequence group. If multiple devices have jump behaviors at similar times, they are merged into sequence nodes of multiple device interleaved behaviors on the timeline, such as [D012, 10 seconds], [D015, 10.2 seconds], [D012, 11 seconds], and [D026, 11.3 seconds], forming a continuous device behavior event stream.

[0029] The interval frequency aggregation submodule calls the jump behavior interleaved sequence group, matches the event types defined in the behavior annotation set, counts the occurrence frequency of the event type in each combination, and filters the segment groups in the jump behavior set; Call the transition behavior interleaved sequence group, match the event types defined in the behavior annotation set, and count the occurrence frequency of the event type in each combination. First, set the behavior annotation set, such as "switching operation", "power switching", "equipment abnormality" and other event labels, and match the trigger characteristics of the transition sequence according to the behavior pattern. For example, if a device has three transitions within 5 seconds, which meets the "switching operation" behavior pattern, then mark the behavior as a switching event and write the matching type into the behavior sequence node. After matching all transition behaviors, count the occurrence of each event type in each fixed time interval, such as every 10 seconds. For example, "switching operation" appears twice in [0-10 seconds], "power switching" appears once, and "equipment abnormality" appears three times in [10-20 seconds]. The frequency statistics results are organized into an interval event frequency table. The cases where a certain type of event occurs intensively in a certain time period are screened out, and this time period is extracted as the transition behavior concentration segment group.

[0030] The coupling feature screening submodule performs a conditional independence test on the relationship between the behavior frequency and the on / off status of the corresponding device based on the concentrated segment group of the jump behavior, calculates the chi-square value, and screens the statistical correlation degree of the segment to retain the behavioral consistency feature fragments; Based on the concentrated segment groups of jump behaviors, a conditional independence test was performed on the relationship between the behavior frequency and the on / off status of the corresponding equipment. The chi-square value was calculated to screen the statistical correlation degree of the segments and retain the behavioral consistency feature segments. First, for each concentrated segment of jump behaviors, the joint distribution of the device jump frequency and on / off status was statistically analyzed. For example, within a 10-second segment, the jump behavior of device D012 occurred 3 times, the power-on state occurred 4 times, and the power-off state occurred 1 time; the jump behavior of device D015 occurred 1 time, the power-on state occurred 2 times, and the power-off state occurred 3 times. The observation frequency table is then formed and presented as follows: when the jump behavior is high (≥3), the power is on 4 times and the power is off 1 time; when the jump behavior is low (≤1), the power is on 2 times and the power is off 3 times, and the total number of records is 10 times.

[0031] The formula used to calculate the chi-square value is as follows: ; in, : Chi-square statistic, used to evaluate the degree of deviation between observed data and expected data; : No. The frequency of this type of behavior is The frequency of observations in the on and off states; : No. The expected frequency of the item represents the theoretical frequency if the behavior frequency is independent of the on-off state. The calculation formula is as follows: , : No. The total frequency of the row (i.e., the total number of occurrences of a certain behavior frequency category); : No. The total frequency of the column (i.e., the total number of times the power-on or power-off state occurs); : Total sample size, that is, the sum of the frequencies of all observation data.

[0032] Assume high frequency behavior (3 times) in the power-on state: , power-off state: , low frequency behavior (1st time) in power-on state: , power-off state: , row sum: , ; Column sum: , ; Total number of samples: .

[0033] Calculate the expected frequency: , , , .

[0034] Substituting into the formula: .

[0035] The judgment threshold is set to 3.84 (degree of freedom 1, confidence level 0.05), because , so there is no significant statistical relationship between the behavior frequency and the on-off state in this section, which does not constitute a consistent coupling feature and needs to be eliminated. , the behavior frequency is determined to have a statistical correlation with the on / off state and is considered a consistent segment and retained. This calculation process is repeated for all segments. Behavioral consistency feature segments are signal segments that show a significant statistical correlation between the frequency of the transition behavior and the device's on / off state. Based on the image content, to determine whether a segment is a behavioral consistency segment, a joint frequency table of the device transition frequency and the on / off state within the segment is calculated, and a chi-square test of independence is performed. If the calculated chi-square value χ² is greater than the set threshold of 3.84 (with 1 degree of freedom and a confidence level of 0.05), it indicates a significant statistical correlation between the transition behavior and the device's on / off state, and the segment is considered a behavioral consistency feature segment and retained. Conversely, if χ² is less than 3.84, there is no significant correlation between the two, the segment does not constitute a consistent coupling feature, and should be eliminated. Ultimately, all segments that meet the statistically significant criteria are identified as behavioral consistency feature segments, ultimately selecting a set of statistically significant feature segments that meet the behavioral consistency criteria.

[0036] See also Figure 5 , the power recovery anomaly detection module includes: The power data extraction submodule calls each staggered time interval in the behavior consistency feature segment, obtains the power sampling sequence of the load device in the corresponding period, and constructs the load power time series structure set; Each interleaved time interval in the behavioral consistency feature segment is called to obtain the power sampling sequence of the load device in the corresponding period, and a load power time series structure set is constructed. First, a data extraction operation is performed on each retained jump behavior segment. Assume that a certain interleaved time interval is from the 30th to the 40th second, and the devices involved include D011 and D014. The system performs power sampling at a frequency of once per second. Then, 10 sampling points will be extracted for each device in this segment. For example, the power values ​​of D011 are 120W, 123W, 128W, etc., indicating the load state change trend in this segment. At the same time, the power values ​​collected by D014 in the same interval are 98W, 101W, 105W, etc., forming a time series set of power changes over time. The power sampling sequence corresponding to each device is matched with the time index to form a complete load power time series structure set. The recording format includes device number, start and end time index, and sampling value sequence.

[0037] The recovery feature calculation submodule extracts the time index of the voltage recovery starting point and the power trend starting point based on the load power time series structure set, calculates the response interval between the two, obtains the power slope sequence within a fixed period after the recovery, determines the fluctuation amplitude between each slope in the continuous segment, and obtains the power recovery response feature set; According to the load power time series structure set, the time index of the voltage recovery starting point and the power trend starting change point is extracted, the response interval between the two is calculated, the power slope sequence within the fixed period after recovery is obtained, the fluctuation amplitude between the slopes in the continuous segment is determined, and the power recovery response feature group is obtained. First, the recovery point is identified. For example, the voltage of device D011 recovers from 0V to 220V at the 35th second, which is the starting point of voltage recovery. Observe its power sampling data. From the 36th second, the power increases from 120W to 135W. Define 36 seconds as the starting point of the power trend, the response interval is 1 second, and then the fixed period is set to 5 Seconds, extract power data from 36 seconds to 40 seconds, calculate the power change per second, and obtain four slope values. For example, from 120W to 126W is 6W / s, from 126W to 132W is 6W / s, and to 135W is 3W / s. Calculate the difference between adjacent slopes one by one. For example, the difference between 6W / s and 6W / s is 0, and the difference between 6W / s and 3W / s is 3. This forms a slope fluctuation sequence. Further analyze its fluctuation amplitude, record the maximum fluctuation value of 3W / s, and finally form the response feature group of the device within the specified period, including the response interval of 1 second, the maximum slope fluctuation of 3W / s, etc.

[0038] The delay segment marking submodule, based on the power recovery response feature group, filters out segments where the response interval exceeds the recovery offset threshold and the slope fluctuation amplitude exceeds the stability limit. It then extracts the corresponding time range and load device identification information, marking them as abnormal state intervals during the power recovery process, and obtaining the power recovery abnormal segment. Based on the power recovery response feature set, we screen segments where the response interval exceeds the recovery offset threshold and the slope fluctuation amplitude exceeds the stability limit. We then extract the corresponding time range and load device identification information and mark them as abnormal state intervals during the power recovery process. This results in abnormal power recovery segments. First, we set the recovery offset threshold to 2 seconds and the stability limit to a slope fluctuation of no more than 2.5 W / s. When a device's response interval is 3 seconds, exceeding the 2-second threshold, and the power slope fluctuation amplitude is 3 W / s, also exceeding the fluctuation upper limit, both conditions meet the abnormality criteria. Therefore, the power recovery segment of this device is marked as an abnormal segment. Its device number, such as D011, and the time range from 35 to 40 seconds are recorded and written into the abnormal data structure. For example, another device, D018, has a response interval of 1 second but a slope fluctuation of 4 W / s. Although the response interval is normal, the fluctuation exceeds the limit and is also identified as an abnormal recovery segment. Key information such as the time range and device number are extracted to form a complete set of abnormal power recovery segments.

[0039] See also Figure 6 , the operating status monitoring output module includes: The abnormal indicator extraction submodule calls the data structure of each segment in the power recovery abnormal segment, extracts the recorded load device identification, response delay time and power change slope fluctuation information, and combines the three indicators according to the device dimension to generate an abnormal indicator combination set; The data structure of each segment in the power recovery anomaly segment is called to extract the recorded load device identification, response delay time, and power change slope fluctuation information. These three indicators are combined by device dimension to generate an abnormal indicator combination set. First, the corresponding field content is parsed from each recovery segment marked as abnormal. For example, in the segment from 50 to 55 seconds, the response delay of device D021 is 3 seconds and the maximum power slope fluctuation is 3.8 W / s. These two indicators are combined with the device number to form an abnormal data entry. Then, the response delay of device D034 recorded in the segment from 60 to 65 seconds is 2.5 seconds and the slope fluctuation is 4.1 W / s. This data entry is also generated. All entries are organized by device number dimension. The device identification, response delay, and slope fluctuation are packaged into an abnormal indicator combination set with a structure such as [D021, 3 seconds, 3.8 W / s] and [D034, 2.5 seconds, 4.1 W / s]. The combined results of all devices are used for sorting and record construction.

[0040] The entry sequence construction submodule arranges the status information entries corresponding to all abnormal devices in chronological order according to the time index of each group of data in the abnormal indicator combination set, and generates a sequence of abnormal behavior time series entries; Based on the time index of each set of data in the abnormal indicator combination set, the status information entries corresponding to all abnormal devices are arranged in chronological order to generate a sequence of abnormal behavior time series entries. First, the time index range contained in each combination entry is extracted. For example, the abnormal period of D021 is 50 to 55 seconds, D034 is 60 to 65 seconds, and D011 is 52 to 58 seconds. All entries are sorted by the starting time, that is, 50 seconds, 52 seconds, and 60 seconds in sequence. Then, based on the sorting results, the corresponding device identification, response delay, and slope fluctuation information are output in sequence to form a time-arranged abnormal behavior time series sequence. For example, the first entry is [D021, 3 seconds, 3.8 W / s], the second entry is [D011, 2 seconds, 3.2 W / s], and the third entry is [D034, 2.5 seconds, 4.1 W / s]. This sorting method ensures the consistency of the time axis of the abnormal information and ensures that the system has a logical order of event evolution when reading the abnormality later.

[0041] The monitoring result output submodule writes each entry structure into the station power system operation status record table based on the sorted content in the abnormal behavior time sequence entry sequence to obtain the station power system operation status monitoring result; Based on the sorted content in the abnormal behavior time series entry sequence, the structure of each entry is written into the station power system operation status record table to obtain the operation status monitoring results. First, the record row content is generated for each time series entry. The fields include equipment number, abnormal start time, response delay time, maximum power slope fluctuation value and other information. For example, the first entry [D021, 50 seconds, 3 seconds, 3.8W / s] is written to the first row of the record table, and then the second entry [D011, 52 seconds, 2 seconds, 3.2W / s] is written to the second row. Each record is bound to the time index and equipment number. At the same time, the system will mark the status of each record as "power recovery abnormality" and attach a number for query use. Finally, all entries are written into the station power system monitoring data file in time sequence to generate a complete operation status monitoring result table for the operation and maintenance system to read, compare and analyze.

[0042] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. The system operation monitoring system of multi-source data fusion is characterized by: The system comprises: The jump behavior identification module obtains the current sampling data and voltage on-off signals of the running equipment in the station power system, identifies the jump trend behavior of each device, and obtains the behavior response dislocation data segment; A confidence screening module calculates the direction score and amplitude deviation of the signal segment corresponding to the device jump trend behavior trigger point based on the behavior response misalignment data segment, and screens the data source segment whose confidence needs to be lowered; A behavior consistency extraction module removes data source segments whose confidence needs to be lowered, constructs interleaved time intervals consisting of adjacent trigger points based on the retained device jump trend behavior trigger points, analyzes the overlap of trigger points within the intervals, and extracts behavior consistency feature segments; A power recovery anomaly detection module extracts power sampling data of load devices in the station power system based on the staggered time intervals of the behavioral consistency feature segments, determines the power recovery delay during the voltage recovery process, and marks the power recovery anomaly segment; The operation status monitoring output module generates an abnormal entry sequence in chronological order based on the power recovery abnormality segment and outputs the operation status monitoring result of the station power system.

2. The system operation monitoring system of multi-source data fusion according to claim 1 is characterized in that: The behavior response misalignment data segment includes the trigger point time difference, the adjacent trigger point misalignment identifier, and the current mutation segment; the data source segment whose confidence needs to be lowered includes the direction score value, the amplitude deviation value, and the low confidence label; the behavior consistency feature segment includes the staggered time index, the repeated trigger frequency, and the behavior event type; the power recovery abnormality segment includes the power recovery starting point, the power trend starting point, and the power slope fluctuation; the operation status monitoring result includes the abnormal equipment identifier, the abnormal data entry, and the output time series.

3. The system operation monitoring system of multi-source data fusion according to claim 1 is characterized in that: The jump behavior recognition module includes: The electrical parameter acquisition structured submodule obtains current sampling data and voltage on / off signals from the equipment running in the station's power system. It calculates the rate of change sequence between consecutive sampling points for the current sampling data, extracts the sudden change segments, and combines them with the edge jump time points in the voltage on / off signals to perform a unified period normalization process on the two types of signal sampling information to obtain a structured electrical parameter change sequence. The behavior trajectory synchronization submodule calls the current change rate segment and voltage jump point sequence in the structured electrical parameter change sequence to synchronize the position of each device behavior trajectory, and pairs each change rate peak with the corresponding voltage change point for detection to obtain a corresponding trajectory group of the device behavior; The response offset identification submodule determines whether the time difference information of each pair of signals in the trajectory group corresponding to the device behavior exceeds the set time offset threshold. For the behavior segments that meet the conditions, the time interval, device number and signal alignment status information are extracted to obtain the behavior response misalignment data segment.

4. The system operation monitoring system of multi-source data fusion according to claim 3 is characterized in that: The confidence screening module includes: A response segment extraction submodule extracts, based on the behavior response misalignment data segment, a signal change segment of each jump trend behavior trigger point in the current change rate sequence and a direction change segment in the voltage on / off signal to obtain a jump signal alignment segment group; The offset calculation submodule calls the jump signal to align each group of current change rate segments and voltage direction segments in the segment group, constructs a time matching vector group, calculates the cosine value of the angle between the time matching vector group and the direction score, extracts the maximum value and the mean value of the reference segment in each group of current change segments, calculates the ratio of the two as the amplitude deviation, and obtains a combination of the direction score and the amplitude deviation; The confidence value determination submodule determines whether the direction score and amplitude deviation combination exceeds the preset offset score threshold and amplitude deviation threshold based on the score and deviation items, filters the signal segments that meet the dual conditions, extracts the device number and time index to which they belong, marks the data content as having confidence that does not meet the requirements, and obtains the data source fragment whose confidence needs to be lowered.

5. The system operation monitoring system of multi-source data fusion according to claim 4 is characterized in that: The behavior consistency extraction module includes: The behavior point integration submodule removes the data source segments whose confidence needs to be lowered, extracts the time index and device number information of the retained device jump trend behavior trigger point, and obtains the jump behavior interleaved sequence group; The interval frequency aggregation submodule calls the jump behavior interleaved sequence group, matches the event types defined in the behavior annotation set, counts the occurrence frequency of the event type in each combination, and filters the jump behavior concentrated segment group; The coupling feature screening submodule performs a conditional independence test on the relationship between the behavior frequency and the on / off state of the corresponding device based on the concentrated segment group of the jump behavior, calculates the chi-square value and screens the statistical correlation degree of the segment, and retains the behavior consistency feature fragments.

6. The system operation monitoring system of multi-source data fusion according to claim 5 is characterized in that: The power recovery abnormality detection module includes: The power data extraction submodule calls each staggered time interval in the behavior consistency feature segment, obtains the power sampling sequence of the load device in the corresponding time period, and constructs a load power time series structure set; The recovery feature calculation submodule extracts the time index of the voltage recovery starting point and the power trend starting change point based on the load power time series structure set, calculates the response interval between the two, obtains the power slope sequence within a fixed period after the recovery, determines the fluctuation amplitude between each slope in the continuous segment, and obtains the power recovery response feature group; The delayed segment marking submodule, based on the power recovery response feature group, filters out segments whose response interval exceeds the recovery offset threshold and whose slope fluctuation amplitude exceeds the stability limit, extracts the corresponding time range and load device identification information, marks them as abnormal state intervals in the power recovery process, and obtains the power recovery abnormal segment.

7. The system operation monitoring system of multi-source data fusion according to claim 6 is characterized in that: The operating status monitoring output module includes: The abnormality indicator extraction submodule calls each data structure in the power recovery abnormality segment, extracts the recorded load device identification, response delay time and power change slope fluctuation information, combines the three indicators according to the device dimension, and generates an abnormality indicator combination set; An entry sequence construction submodule arranges the status information entries corresponding to all abnormal devices in chronological order according to the time index of each group of data in the abnormal indicator combination set to generate a sequence of abnormal behavior time series entries; The monitoring result output submodule writes the sorted content in the abnormal behavior time sequence entry sequence into the station power system operation status record table to obtain the station power system operation status monitoring result.

Images