An index algorithm monitoring and analyzing method for a smart park

Abstract

The present application belongs to the technical field of data processing, and particularly relates to an index algorithm monitoring and analyzing method for a smart park, which comprises the following steps: performing frequency spectrum analysis on a healthy current sequence and a to-be-detected current sequence to obtain a reference current baseline vector and a current baseline vector, and calculating a current drift factor according to the difference between the two; similarly, a voltage drift factor is calculated; a logic regression model is trained by using equipment failure samples and same-working-condition data to determine a current weight and a voltage weight, and a total drift factor is obtained by weighted summation of the current drift factor and the voltage drift factor to screen equipment in a primary abnormal state; according to the healthy load level of the equipment in the primary abnormal state, a plurality of load intervals are divided; the mean value and the standard deviation of the total drift factor of each load interval are calculated to obtain an instability index; and by comparing the instability index with an instability threshold, equipment in a final abnormal state is screened. The present application improves the detection accuracy of equipment abnormalities in the park.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for monitoring and analyzing indicators in smart parks. Background Technology

[0002] As smart parks upgrade towards greater precision, real-time and accurate monitoring and analysis of the operational status of key equipment within the park, especially power equipment such as smart meters and air conditioning units, has become crucial for ensuring the safe and stable operation of the park. Existing technologies first collect multiple electrical time-series data during equipment operation through sensors, such as voltage, current, active power, and reactive power. Subsequently, through in-depth processing of these data, including frequency domain analysis and mutual information calculation, ultimate features such as harmonic energy ratio and nonlinear coupling degree between parameters are extracted. Finally, these features are fused into a single comprehensive correlation index through a preset nonlinear combination function to quantitatively assess the real-time health status of the equipment. When this index exceeds a predetermined threshold, the system will trigger an alarm.

[0003] However, in practical applications, the above methods have significant limitations: the nonlinear combination function and its internal parameters used to calculate the comprehensive correlation index, such as the weights of different features and correlation factors, are fixed models that are debugged and optimized based on specific equipment models and ideal operating conditions during the model development stage; while the environment of smart parks is complex and changeable, and the basic electrical characteristics of equipment will slowly and continuously drift during long-term operation due to natural aging, component wear, firmware upgrades or fluctuations in the external power grid environment; at this time, the fixed analysis model cannot adapt to this change, and the judgment benchmarks for normal and abnormal conditions established at the beginning gradually become invalid, leading to misjudgment of normal operating conditions or omission of early minor faults, which in turn makes it difficult to accurately monitor the indicators of smart parks. Summary of the Invention

[0004] To address the technical problem mentioned in the background that existing indicator monitoring algorithms suffer from decreased accuracy when equipment characteristics change due to the use of fixed models, this invention provides an indicator algorithm monitoring and analysis method for smart parks.

[0005] This invention provides a method for monitoring and analyzing indicators in smart parks, comprising: for equipment in the park, collecting healthy current sequences and healthy voltage sequences under healthy conditions after installation and commissioning; collecting current and voltage sequences to be detected in real time; calculating current drift factors based on the healthy current sequences and the current sequences to be detected, including: performing spectral analysis on the healthy current sequences and the current sequences to be detected, and obtaining the current drift factor based on the difference between the obtained reference current baseline vector and the current baseline vector; calculating voltage drift factors based on the healthy voltage sequences and the voltage sequences to be detected; and extracting equipment fault samples and related data from a historical database. Health data under various conditions is used to train a logistic regression model to determine current and voltage weights. The current and voltage drift factors are then weighted and summed to obtain a total drift factor, which is used to screen devices in the initial abnormal state. Based on the health load level of devices in the initial abnormal state, multiple load intervals are defined. The load interval to which a device belongs is determined based on its load level. The instability index is obtained based on the difference in distribution between the device's total drift factor and the total drift factor of its load interval. Finally, based on the comparison between the instability index and the instability threshold, devices in the ultimate abnormal state are identified, enabling monitoring and analysis of the smart park.

[0006] This invention is achieved through the following steps: First, current and voltage sequences under the health status of the equipment are collected as a benchmark, and the sequence to be detected is collected in real time. Next, current drift factor and voltage drift factor are calculated through spectrum analysis to reflect the difference between the current sequence and the health baseline. Then, a logistic regression model is trained using historical fault samples and health data to determine current weight and voltage weight, and the total drift factor is obtained by weighting to screen primary abnormal equipment. Finally, based on the health load division interval of the primary abnormal equipment, the instability index is calculated by combining the distribution difference of the total drift factor, and the ultimate abnormal equipment is determined by comparing with a threshold, thus completing the smart park monitoring and analysis. This method effectively solves the problems of difficulty in accurately identifying early anomalies and inability to judge equipment stability by combining load differences in traditional equipment monitoring. It can achieve accurate identification of all stages from primary equipment anomaly to ultimate anomaly, improving the timeliness and accuracy of equipment fault early warning in smart parks.

[0007] Preferably, the method for obtaining the reference current baseline vector includes: targeting a healthy current sequence In this process, each 10-minute time window is divided into several segments. The power spectral density of each time window is calculated using a Fast Fourier Transform (FFT). These segments are then divided into five equal frequency bands. The energy percentage of each frequency band within each time window is further calculated, and a healthy current sequence is obtained based on this calculation. The spectral eigenvector of each time window is averaged, and the average of the spectral eigenvectors of all time windows is taken as the reference current baseline vector, denoted as . .

[0008] This method effectively captures the short-term stable characteristics of healthy current sequences by dividing the time window into 10-minute intervals, avoiding random interference from single samples. Combined with Fast Fourier Transform to calculate the power spectral density and equally dividing it into 5 frequency bands, it can standardize the dimensionality of spectral feature extraction, ensure the consistency and comparability of features, and take the average value of the spectral feature vectors of all time windows as the baseline. This can further smooth out the small fluctuations in the healthy state and reduce the impact of random errors on the baseline. The final reference current baseline vector is closer to the true healthy spectral characteristics of the equipment, laying a reliable foundation for subsequent accurate calculation of the current drift factor and accurate judgment of whether the equipment deviates from the healthy state.

[0009] Preferably, obtaining the current drift factor based on the difference between the obtained reference current baseline vector and the current current baseline vector includes: ;in, It is the current drift factor; It is the current current baseline vector; It is the reference current baseline vector; This represents the Euclidean norm.

[0010] This method utilizes the Euclidean norm. The system accurately quantifies the overall difference between the current baseline vector and the reference current baseline vector, avoiding local misjudgments caused by deviations in a single frequency band. Simultaneously, normalization using the Euclidean norm of the reference current baseline vector as the denominator effectively eliminates the influence of differences in the absolute value of the healthy baseline under different equipment or operating conditions. The resulting current drift factor is a relative drift index, which is not only numerically intuitive and comparable across different systems, but also accurately reflects the deviation ratio of the current current sequence relative to the healthy state. This provides an objective and unified quantitative basis for subsequent calculation of the total drift factor in conjunction with the voltage drift factor and for accurate identification of equipment anomalies.

[0011] Preferably, calculating the voltage drift factor based on the healthy voltage sequence and the voltage sequence to be detected includes: based on the healthy voltage sequence... Obtain the reference voltage baseline vector According to the voltage sequence Obtain the current voltage baseline vector The formula for calculating the voltage drift factor is: ;in, It is the voltage drift factor; It is the current voltage baseline vector; It is the reference voltage baseline vector; This represents the Euclidean norm.

[0012] Preferably, the logistic regression model includes a current-based logistic regression model and a voltage-based logistic regression model. The training process of the logistic regression model is as follows: Input positive and negative samples corresponding to the current drift factor and voltage drift factor obtained from fault samples and health data under the same operating conditions of the equipment; output the binary classification results of each through the logistic regression model, i.e., 0 represents the normal state and 1 represents the abnormal state; integrate these four types of samples into a equipment drift state sample set, divide it into a 70% training set and a 30% validation set, train the logistic regression model using the L2 regularization algorithm, and output feature coefficients. The outputs of the two logistic regression models are the coefficients of the current drift factor, respectively. and the coefficient of voltage drift factor .

[0013] This method constructs separate models for the current drift factor and voltage drift factor, inputting corresponding positive samples representing anomalies and negative samples representing normal conditions. This allows for precise focus on the correlation between a single feature and the equipment state, avoiding discrimination bias caused by cross-interference between current and voltage features. Selecting fault samples and health data under the same operating conditions as training data effectively eliminates interference from different operating conditions on the drift factor, ensuring the model learns a true mapping between the equipment's health state and the drift factor. The 70% training set and 30% validation set partitioning ensures sufficient training data to support model learning while allowing the validation set to evaluate the model's generalization ability, avoiding overfitting. The introduction of L2 regularization suppresses excessive reliance on extreme samples and stabilizes feature coefficient output. The resulting current and voltage drift factor coefficients objectively quantify the contribution of both drift factors to the judgment of equipment anomalies, providing reliable model support for the subsequent reasonable allocation of current and voltage weights and accurate calculation of the total drift factor.

[0014] Preferably, the method for determining the current weight and voltage weight includes: based on the coefficient of the current drift factor. and the coefficient of voltage drift factor Set the current weight for a single device. and voltage weighting .

[0015] Preferably, the method for screening devices in a primary anomalous state includes: setting a total drift threshold. The following conditions must be met: When the total drift factor of the device is greater than or equal to the total drift threshold When the total drift factor of the equipment is less than the total drift threshold, the equipment is considered to be in a primary abnormal state. If the device is in normal working order, it is considered to be in a normal state.

[0016] Preferably, dividing the load into multiple load intervals based on the health load level of the device in the initial abnormal state includes: dividing the health current sequence and healthy voltage sequence The data was divided into equal 10-minute time windows. For each window, the mean of the preprocessed current values ​​was calculated as the health load level for that window. All calculated health load levels were clustered using the K-means clustering algorithm. The sum of squared errors under different numbers of clusters was calculated using the elbow method to determine the optimal number of clusters. Ultimately obtained A number of non-overlapping load ranges.

[0017] Preferably, obtaining the instability index based on the difference in distribution between the total drift factor of the device and the total drift factor of its load range includes: ;in, It is the current moment. Corresponding equipment Instability indicators; It is an index of time; It is the device index; It is the current moment. Corresponding equipment Total drift factor; It is the current moment. The mean of the total drift factor corresponding to the load interval to which the load level belongs in the corresponding time window; It is the current moment. The standard deviation of the total drift factor corresponding to the load interval to which the load level of the corresponding time window belongs; It is a preset micro value; It is the absolute value symbol.

[0018] This method effectively avoids cross-range misjudgments caused by inherent differences in the total drift factor of healthy devices under different load levels by comparing the total drift factor of the current device with the mean of the total drift factor of its load range. It accurately focuses on the degree of deviation of the current device from the health status of devices under the same load. The denominator is normalized by introducing the standard deviation of the total drift factor of the load range, which can eliminate the inherent differences in the health fluctuation amplitude of different load ranges. It makes the instability indicators of devices in different load ranges comparable horizontally, avoiding the masking of true anomalies due to different range fluctuation characteristics. The preset small value can prevent calculation failure when the standard deviation is zero. The final output instability indicator quantifies the degree of device instability with intuitive numerical values, providing an objective and unified quantitative basis for subsequent comparison with instability thresholds to determine the ultimate abnormal device. This further improves the accuracy and robustness of the entire chain of judgment from primary to ultimate anomaly in smart park equipment fault early warning.

[0019] Preferably, the step of obtaining the device in the ultimate abnormal state based on the comparison result of the instability index and the instability threshold includes: setting an instability threshold. The following conditions must be met: If at the current moment Corresponding equipment The instability index is greater than or equal to the instability threshold. At this point, a final warning needs to be triggered, requiring an immediate 24-hour special inspection and an increase in data collection frequency; if at the current moment Corresponding equipment The instability index is less than the instability threshold. At this point, the primary anomaly flag needs to be removed, the device status needs to be adjusted to the attention state, and data needs to be continuously collected at a frequency of 3 Hz and the changes in instability indicators need to be monitored.

[0020] The beneficial effects of this invention are as follows: By first collecting current and voltage sequences under the health status of the equipment to establish a reliable benchmark, and then combining this with spectrum analysis to accurately calculate the current and voltage drift factors, the subtle differences between the current electrical characteristics of the equipment and the health baseline can be effectively captured, solving the pain point of traditional monitoring being unable to identify early minor anomalies; Subsequently, a logistic regression model is trained using historical fault samples and health data under the same operating conditions to determine the weights, which not only objectively quantifies the contribution of the two types of drift factors to the anomaly through model learning, avoiding the bias of subjective weight allocation, but also makes the screening of primary abnormal equipment by the total drift factor more scientific; Furthermore, by dividing the interval according to the health load level and combining the differences in the distribution of the total drift factor to calculate the instability index, the interference of different load fluctuations on the anomaly judgment can be eliminated, preventing misjudgment of equipment status due to load differences; Finally, by comparing the instability index with the threshold, the ultimate abnormal equipment is identified, realizing full-stage monitoring from early minor anomalies to serious anomalies, significantly improving the timeliness and accuracy of equipment fault early warning in smart parks, ensuring stable equipment operation while reducing unnecessary operation and maintenance costs. Attached Figure Description

[0021] Figure 1 This is a schematic flowchart illustrating an indicator algorithm monitoring and analysis method for smart parks according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] This invention discloses a method for monitoring and analyzing indicator algorithms in smart parks, referring to... Figure 1 This includes steps S1 to S4:

[0025] S1. Preprocess the raw current data and raw voltage data collected in real time from the data acquisition terminal of the equipment in the smart park to obtain the current sequence and voltage sequence to be detected, and collect the healthy current sequence and healthy voltage sequence.

[0026] It should be noted that raw current and voltage data were chosen as the basis for analysis because: current and voltage are the most critical electrical parameters for power equipment such as smart meters and air conditioning units in the smart park, and their changes directly reflect the load status of the equipment; for example, an abnormal increase in current corresponds to overload. Energy consumption levels and potential fault precursors, such as intermittent sudden drops in current caused by poor contact, are the core data sources for subsequent calculations. Due to the influence of electromagnetic interference from sensors and packet loss in network transmission, the raw data collected in real time by the park's data acquisition terminal is prone to noise and missing values. Furthermore, the rated ranges of current and voltage vary significantly among different types of equipment. If used directly for calculation, it will lead to deviations in subsequent feature extraction. Therefore, it is necessary to preprocess the collected raw data to improve data quality and enhance the accuracy of subsequent analysis results.

[0027] Specifically, raw data from power equipment such as smart meters and air conditioning units are acquired in real time from data acquisition terminals within the smart park. The raw data includes raw current data and raw voltage data, with a single device acquisition frequency set to 1 Hz. Historical data from the initial operation of the device to the current moment is also retrieved. The historical data includes historical current data and historical voltage data. The raw data is preprocessed to obtain the preprocessed current sequence and voltage sequence to be detected.

[0028] Specifically, the preprocessing steps are as follows: First, when the missing duration is less than or equal to 2 seconds, linear interpolation is used to complete the missing original data; when the missing duration is greater than 2 seconds, it is marked as a data failure segment, and the data acquisition terminal is triggered to re-upload the original data within that time period; then, the average value of the data from a single device over a continuous hour is calculated. with standard deviation ,use The principle is to identify abnormal raw data and replace it with the average of the five raw data points before and after it. A moving average filtering method is used to remove noise, including setting the sliding window size to 5, meaning each raw data point takes the average of itself and its two adjacent raw data points to eliminate high-frequency noise caused by transient electromagnetic interference while preserving the trend of the raw data. Finally, maximum and minimum value normalization is used, based on the maximum and minimum values ​​of historical data from the initial operation of the device to the current moment, to uniformly map the noise-removed raw data to the interval [0, 1], obtaining the preprocessed current and voltage sequences to be detected, denoted as follows: , .

[0029] Furthermore, under healthy conditions after installation and commissioning, current and voltage data are collected over a complete operating cycle and preprocessed to obtain healthy current and voltage sequences, denoted as follows: , .

[0030] S2. Construct reference current and voltage baseline vectors and current current and voltage baseline vectors, calculate current drift factor and voltage drift factor; train logistic regression model to determine current weight and voltage weight, calculate total drift factor, and screen devices in primary abnormal state.

[0031] It should be noted that the core purpose of obtaining the reference current baseline vector and current drift factor is as follows: the electrical characteristics of power equipment such as smart meters and air conditioning units in smart parks will slowly drift due to factors such as equipment aging and wear. For example, wear of motor bearings will lead to an increase in specific harmonic components in the current spectrum. This drift is gradual and difficult to identify directly through time-domain changes in the original current data. The reference current baseline vector records the current spectrum characteristics of the equipment under normal operating conditions, including energy distribution and peak frequency in each frequency band, which can provide a benchmark for judging whether drift has occurred. By comparing the reference current baseline vector with the current current baseline vector to obtain the current drift factor, the degree to which the electrical characteristics of the equipment deviate from a healthy state can be accurately captured, providing data basis for subsequent dynamic adjustment analysis models.

[0032] Specifically, for healthy current sequences In this process, each 10-minute time window is divided into several segments. The power spectral density of each time window is calculated using a Fast Fourier Transform (FFT). These segments are then divided into five equal frequency bands. The energy percentage of each frequency band within each time window is further calculated, and a healthy current sequence is obtained based on this calculation. The spectral eigenvector of each time window is averaged, and the average of the spectral eigenvectors of all time windows is taken as the reference current baseline vector, denoted as . This is used to reflect the current spectrum characteristics of the device in a healthy state; similarly, based on the current sequence to be detected... Obtain the corresponding current baseline vector, denoted as In practical applications, the reference baseline vector is a non-zero vector; therefore, based on the reference current baseline vector and the current current baseline vector, the formula for calculating the current drift factor is:

[0033] ;

[0034] in, It is the current drift factor; It is the current current baseline vector; It is the reference current baseline vector; Represents the Euclidean norm; current baseline vector. With reference current baseline vector The greater the difference between them, the larger the current drift factor, reflecting how much the current spectrum characteristics of the device deviate from a healthy state.

[0035] It should be noted that using the normalized Euclidean norm formula can eliminate the influence of the amplitude of the reference baseline vector itself of different devices, making the current drift factor of different types of devices and different time periods comparable; theoretically, the current drift factor is greater than or equal to 0, and when the device is in perfect health, the current drift factor is 0.

[0036] Specifically, based on the health voltage sequence Obtain the reference voltage baseline vector According to the voltage sequence Obtain the current voltage baseline vector Therefore, based on the reference voltage baseline vector and the current voltage baseline vector, the formula for calculating the voltage drift factor is:

[0037] ;

[0038] in, It is the voltage drift factor; It is the current voltage baseline vector; It is the reference voltage baseline vector; Represents the Euclidean norm; current voltage baseline vector With reference voltage baseline vector The greater the difference between them, the larger the voltage drift factor, reflecting the more serious the deviation of the current voltage spectrum characteristics of the equipment from a healthy state.

[0039] It should be noted that the total drift factor is used to comprehensively quantify the overall electrical health status of the equipment. Compared with a single factor, it is more comprehensive and can avoid misjudgment based on a single dimension. For example, when the current drift factor of the equipment is small but the voltage drift factor is large, judging based solely on the current drift factor may result in missed detection. Moreover, most equipment faults are accompanied by coordinated changes in current and voltage characteristics. The total drift factor can more accurately capture such multi-dimensional anomalies and provide more comprehensive data for subsequent fault warnings.

[0040] Furthermore, typical equipment failure cases, such as bearing wear and precursors to winding short circuits, are selected from the historical database of the data acquisition terminal, and extracted... The number of valid fault samples can be set according to actual needs, and the following conditions must be met: For example, a positive integer. Each sample must include the time of the fault occurrence and the original current and voltage data before and after the fault. For each fault sample, the original current and voltage data for the hour before the fault occurred are extracted. During this period, the fault characteristics have appeared but have not fully erupted. After preprocessing, the corresponding current drift factor and voltage drift factor are calculated and marked as a positive fault sample with a label of 1, representing an abnormal state before the fault. At the same time, the historical data of the single device in a healthy state are extracted from the acquisition terminal, and it must meet the same time period and operating conditions as the fault sample. After preprocessing in step S1, the corresponding current drift factor and voltage drift factor are obtained based on the calculation formula of the reference baseline vector and marked as a negative healthy sample with a label of 0, representing a normal healthy state.

[0041] Specifically, the inputs are positive and negative samples corresponding to the current drift factor and voltage drift factor of the equipment samples, respectively. The logistic regression model outputs their respective binary classification results: 0 represents a normal state, and 1 represents an abnormal state. These four types of samples are integrated into a device drift state sample set, which is then divided into a 70% training set and a 30% validation set. An L2 regularization algorithm is used to train the logistic regression model, outputting the corresponding feature coefficients. The outputs of the two logistic regression models are the coefficients of the current drift factor, respectively. and the coefficient of voltage drift factor When the coefficient is larger, it reflects that the corresponding characteristic coefficient contributes more to the judgment of abnormal state before the fault, and vice versa.

[0042] It should be noted that, in this scheme, the magnitude of the current drift factor directly represents the severity of the equipment's deviation from its healthy state. Therefore, the current drift factor of a pre-fault abnormal sample labeled 1 will necessarily be greater than the current drift factor of a healthy sample labeled 0. Thus, the coefficient of the current drift factor after training... It can only be positive, and its physical meaning is that the larger the current drift factor, the higher the probability that the equipment is in a pre-fault abnormal state with tag 1; similarly, the coefficient of the voltage drift factor... It can only be positive, meaning that the larger the voltage drift factor, the higher the probability that the device is in a pre-fault abnormal state with tag 1.

[0043] Furthermore, based on the coefficient of the current drift factor... and the coefficient of voltage drift factor Set the current weight for a single device. and voltage weighting Weight the current of the device. and voltage weight The weighted sum of the calculated current drift factor and voltage drift factor is used as the total drift factor of the equipment, which reflects the severity of the equipment's current overall deviation from its healthy state.

[0044] It should be noted that the evolution of equipment from health to failure typically presents a continuous transition: under normal conditions, the equipment operates stably, the current drift factor and voltage drift factor deviate very little from the healthy baseline, and the total drift factor should be concentrated in the lower range, such as... At this point, there are no coordinated abnormalities in electrical characteristics, and no intervention is required. In the initial abnormal state, the equipment exhibits early signs of failure, such as slight bearing wear and mild chip aging, but it can still maintain basic operation, with the total drift factor showing a moderate deviation. The range precisely covers this moderate deviation range, avoiding misjudging slight fluctuations in a normal state as abnormal, and also avoiding concentrating the total drift factor in a higher range, such as... The severe state at that time is classified as a primary abnormal state.

[0045] Specifically, based on the comparison between the total drift factor and the total drift threshold, devices in a primary abnormal state are identified. The specific method is as follows: Set the total drift threshold. The following conditions must be met: The specific settings can be configured according to actual needs, for example... When the total drift factor of the device is greater than or equal to the total drift threshold When the total drift factor of the equipment is less than the total drift threshold, it indicates that the electrical health status of the equipment has significantly deviated from the baseline after installation and commissioning, but has not yet reached the level of directly triggering a fault shutdown. This situation is consistent with the early stage of equipment fault evolution: that is, the current and voltage spectrum characteristics of the equipment have shown coordinated abnormalities, such as the enhancement of current harmonics in the early stage of motor bearing wear, and the voltage sampling deviation in the early stage of meter chip aging, but the equipment can still maintain basic operating functions and has not shown obvious serious fault symptoms such as shutdown or error. Therefore, the equipment is considered to be in a primary abnormal state. If the device is in a normal state, then all devices in a primary abnormal state are obtained sequentially.

[0046] S3. Based on the healthy load level of the equipment in the primary abnormal state under the fully healthy state, divide the load intervals and calculate the mean and standard deviation of the total drift factor of the load intervals; determine the load interval to which the equipment belongs based on the load level of the equipment in the primary abnormal state; obtain its instability index based on the difference in distribution between the total drift factor of the equipment and the total drift factor of its load interval.

[0047] It should be noted that the total drift factor is affected by the current operating load of the equipment. When the operating load is high, the harmonics themselves will naturally increase, which will increase the total drift factor. However, this is a normal phenomenon. In order to eliminate misjudgment of this situation, it is necessary to distinguish between normal spectrum changes caused by load changes and abnormal changes caused by the deterioration of the equipment's own characteristics.

[0048] Specifically, obtain the health current sequence of the device in a fully healthy state. and healthy voltage sequence The time windows are divided into equal 10-minute periods. For each window, the mean value of the preprocessed current is calculated as the healthy load level for that window, and the healthy load level for each window is obtained. Then, the total drift factor for each window is calculated according to the formula for calculating the total drift factor.

[0049] Furthermore, all calculated health load levels were clustered using the K-means clustering algorithm; the sum of squared errors for different numbers of clusters was calculated using the elbow method to determine the optimal number of clusters for clustering results. This approach avoids both situations where too few clusters result in multiple healthy load levels being assigned to the same load range, and situations where too many clusters lead to redundancy in the resulting load ranges, ultimately achieving... The load intervals are non-overlapping; for each load interval, all windows where all healthy load levels fall within that load interval are identified, and the mean and standard deviation of the total drift factor for all windows are calculated.

[0050] It should be noted that the total drift factor can be calibrated using the mean and standard deviation of the total drift factor across all windows to obtain a more accurate instability index for the equipment's operating conditions. The healthy load level is calculated using the mean of the pre-processed current values ​​because current data directly reflects the actual operating load intensity of the equipment. Dividing the healthy load level into non-overlapping load intervals avoids benchmark differences in the total drift factor under different load intensities. For example, the total drift factor is generally lower in low load intervals and generally higher in high load intervals, ensuring that subsequent calibration is based solely on a healthy benchmark under the same operating conditions, thus improving the accuracy of the instability index.

[0051] Specifically, regarding the current moment The current sequence within the first 30 minutes is obtained. Following the steps S1 and S2, the mean current value within this time window is obtained as the load level for that window. The total drift factor corresponding to this time window is then obtained. Based on the load level of this time window, the load interval to which it belongs, and the mean and standard deviation of the total drift factor corresponding to that load interval, are determined. The formula for calculating the instability index is then:

[0052] ;

[0053] in, It is the current moment. Corresponding equipment Instability indicators; It is an index of time; It is the device index; It is the current moment. Corresponding equipment Total drift factor; It is the current moment. The mean of the total drift factor corresponding to the load interval to which the load level belongs in the corresponding time window; It is the current moment. The standard deviation of the total drift factor corresponding to the load interval to which the load level of the corresponding time window belongs; It is a preset tiny value, for example. This is used to prevent the standard deviation of the total drift factor from being zero. It is the absolute value symbol.

[0054] It should be noted that at the current moment Corresponding equipment Instability index The physical meaning is the current moment. Corresponding equipment The total drift factor deviates from the normal level within the corresponding load interval, i.e., the deviation is measured by how many times the standard deviation of the total drift factor. Since global data may not follow a normal distribution, dividing the data into load intervals groups data with similar total drift factor levels into one category, making the data distribution within each interval more likely to approximate a normal distribution. Therefore, the three-standard-deviation principle is more applicable. Thus, at the current moment... Corresponding equipment Instability index The value range should be within Therefore, the subsequent range for the instability threshold should also be within this range. Within the range.

[0055] S4. Based on the comparison results of instability indicators and instability thresholds, identify the devices in the ultimate abnormal state to achieve monitoring and analysis of the smart park.

[0056] Specifically, based on the current moment Corresponding equipment The instability index is compared with the instability threshold to identify devices in the ultimate abnormal state. The specific method is as follows: Set the instability threshold. The following conditions must be met: The specific settings can be configured according to actual needs, for example... This threshold strikes a balance between false alarm rate and detection sensitivity, making it suitable for most smart park devices; if at the current time Corresponding equipment The instability index is greater than or equal to the instability threshold. When this occurs, it indicates that the total drift factor of the device has significantly deviated from the average level of the healthy state in the same load range under the current load conditions. Moreover, this deviation is mainly caused by the deterioration of the device's own characteristics, rather than normal load fluctuations. At this time, the ultimate warning needs to be triggered, and a special inspection should be arranged immediately within 24 hours. The data collection frequency should also be increased, for example, from 1 Hz to 5 Hz, in order to closely monitor the abnormal state changes of the devices in the smart park.

[0057] Furthermore, if at the current moment Corresponding equipment The instability index is less than the instability threshold. When the total drift factor of the device is within the initial abnormal range under the current load conditions, it is still within the normal fluctuation range after calibration with the health benchmark in the same load range. This initial abnormality is mostly a pseudo-abnormality caused by load fluctuations, such as the increase in the total drift factor caused by short-term high load. At this time, it is necessary to remove the initial abnormality mark, adjust the device status to the attention state, continuously collect data at a frequency of 3 Hz and monitor the changes in instability indicators, without immediately starting a special inspection, to avoid ineffective consumption of operation and maintenance resources.

Claims

1. A method for monitoring and analyzing indicators in smart parks, characterized in that, include: For the equipment in the park, collect healthy current and healthy voltage sequences after the equipment has been installed, commissioned, and is in a healthy state; Real-time acquisition of the current and voltage sequences to be detected; The current drift factor is calculated based on the healthy current sequence and the current sequence to be detected, including: performing spectral analysis on the healthy current sequence and the current sequence to be detected, and obtaining the current drift factor based on the difference between the obtained reference current baseline vector and the current current baseline vector. Calculate the voltage drift factor based on the healthy voltage sequence and the voltage sequence to be detected; Fault samples and health data under the same operating conditions are extracted from the historical database to train a logistic regression model to determine current and voltage weights. The current and voltage drift factors are then weighted and summed to obtain a total drift factor, which is used to screen equipment in a primary abnormal state. The logistic regression model includes a current-based logistic regression model and a voltage-based logistic regression model. The training process for the logistic regression model is as follows: The input consists of positive and negative samples corresponding to the current drift factor and voltage drift factor, obtained from fault samples and health data under the same operating conditions. A logistic regression model outputs binary classification results for each, where 0 represents a normal state and 1 represents an abnormal state. These four types of samples are then integrated into a device drift state sample set, divided into a 70% training set and a 30% validation set. An L2 regularized algorithm is used to train the logistic regression model, outputting feature coefficients. The outputs of the two logistic regression models are the coefficients of the current drift factor. and the coefficient of voltage drift factor ; Device current weighting Voltage weighting ; Based on the health load level of devices in the initial abnormal state, multiple load intervals are divided; based on the load level of devices in the initial abnormal state, their respective load intervals are determined; based on the distribution difference between the total drift factor of the device and the total drift factor of its respective load interval, its instability index is obtained; based on the comparison result of the instability index and the instability threshold, devices in the ultimate abnormal state are identified, thereby realizing the monitoring and analysis of the smart park.

2. The indicator algorithm monitoring and analysis method for smart parks according to claim 1, characterized in that, The method for obtaining the reference current baseline vector includes: For healthy current sequences In this process, each 10-minute time window is divided into several segments. The power spectral density of each time window is calculated using a Fast Fourier Transform (FFT). These segments are then divided into five equal frequency bands. The energy percentage of each frequency band within each time window is further calculated, and a healthy current sequence is obtained based on this calculation. The spectral eigenvector of each time window is averaged, and the average of the spectral eigenvectors of all time windows is taken as the reference current baseline vector, denoted as . .

3. The indicator algorithm monitoring and analysis method for smart parks according to claim 1, characterized in that, The step of obtaining the current drift factor based on the difference between the obtained reference current baseline vector and the current current baseline vector includes: ； in, It is the current drift factor; It is the current current baseline vector; It is the reference current baseline vector; This represents the Euclidean norm.

4. The indicator algorithm monitoring and analysis method for smart parks according to claim 1, characterized in that, The step of calculating the voltage drift factor based on the healthy voltage sequence and the voltage sequence to be detected includes: According to the health voltage sequence Obtain the reference voltage baseline vector According to the voltage sequence Obtain the current voltage baseline vector The formula for calculating the voltage drift factor is: ； in, It is the voltage drift factor; It is the current voltage baseline vector; It is the reference voltage baseline vector; This represents the Euclidean norm.

5. The indicator algorithm monitoring and analysis method for smart parks according to claim 1, characterized in that, The method for screening devices in a primary abnormal state includes: Set the total drift threshold The following conditions must be met: When the total drift factor of the device is greater than or equal to the total drift threshold When the total drift factor of the equipment is less than the total drift threshold, the equipment is considered to be in a primary abnormal state. If the device is in normal working order, it is considered to be in a normal state.

6. The indicator algorithm monitoring and analysis method for smart parks according to claim 1, characterized in that, The system divides the load into multiple load ranges based on the health load level of devices in a primary abnormal state, including: healthy current sequence and healthy voltage sequence The data was divided into equal 10-minute time windows. For each window, the mean of the preprocessed current values ​​was calculated as the health load level for that window. All calculated health load levels were clustered using the K-means clustering algorithm. The sum of squared errors under different numbers of clusters was calculated using the elbow method to determine the optimal number of clusters. Ultimately obtained A number of non-overlapping load ranges.

7. The indicator algorithm monitoring and analysis method for smart parks according to claim 6, characterized in that, The method of obtaining instability indices based on the difference in distribution between the total drift factor of the equipment and the total drift factor of its load range includes: ； in, It is the current moment. Corresponding equipment Instability indicators; It is an index of time; It is the device index; It is the current moment. Corresponding equipment Total drift factor; It is the current moment. The mean of the total drift factor corresponding to the load interval to which the load level belongs in the corresponding time window; It is the current moment. The standard deviation of the total drift factor corresponding to the load interval to which the load level of the corresponding time window belongs; It is a preset micro value; It is the absolute value symbol.

8. The indicator algorithm monitoring and analysis method for smart parks according to claim 1, characterized in that, The step of determining the device in the ultimate abnormal state based on the comparison result of the instability index and the instability threshold includes: Set an instability threshold The following conditions must be met: If at the current moment Corresponding equipment The instability index is greater than or equal to the instability threshold. At this point, a final warning needs to be triggered, requiring an immediate 24-hour special inspection and an increase in data collection frequency; if at the current moment Corresponding equipment The instability index is less than the instability threshold. At this point, the primary anomaly flag needs to be removed, the device status needs to be adjusted to the attention state, and data needs to be continuously collected at a frequency of 3 Hz and the changes in instability indicators need to be monitored.

Images