A safety early warning method and system for a distribution box

Abstract

The application relates to the technical field of fault early warning systems, and discloses a safety early warning method and system for a distribution box, which comprises the following steps: acquiring multi-dimensional monitoring data collected in a continuous time window in the distribution box; constructing an isolated tree set for the multi-dimensional monitoring data; the process of constructing a single isolated tree is as follows: at a non-leaf node of the tree, generating a non-uniform selection probability distribution according to the statistical characteristics of each feature of the data samples in the node, selecting a segmentation feature based on the selection probability distribution; calculating a basic abnormality score; acquiring an exponentially weighted moving average value and a time change rate calculated from historical data points, and calculating an adjustment factor; obtaining a comprehensive risk index by using the basic abnormality score and the adjustment factor; and generating an early warning signal when the comprehensive risk index exceeds a preset risk threshold. The application can monitor early faults caused by the coordinated change of key parameters, identify potential faults that are continuously deteriorating, and reduce the risk of missed reports and false reports caused by trend judgment errors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault early warning system, in particular to a safety early warning method and system for distribution box. BACKGROUND

[0002] In the traditional safety monitoring of distribution box, a judgment method based on fixed threshold is often used. For example, for a single key operating indicator such as temperature or current, an upper limit of alarm is set in advance, and once the indicator exceeds the upper limit, an alarm is triggered. Although this method is simple to operate, it can only cover extreme abnormalities of a single indicator and is difficult to cope with scenarios of multiple indicators cooperating abnormally.

[0003] To improve detection accuracy, intelligent abnormality detection means are gradually introduced in the industry, such as using support vector machines, neural networks and other models. Such models can comprehensively analyze the multi-dimensional monitoring data of the distribution box, such as temperature, humidity, current, voltage, etc., and identify deviating abnormal states by learning normal operating patterns. Among them, the isolation forest algorithm is particularly popular in the field of distribution box abnormality detection due to its advantages of not requiring a large amount of labeled data and high computational efficiency.

[0004] However, the existing isolation forest algorithm has obvious limitations. On the one hand, when building an isolation tree, it selects the split feature in a uniform random manner, i.e., it assumes that all monitoring features such as temperature, humidity, current, and voltage have the same contribution to abnormality. However, in actual operation of the distribution box, the correlation of different features with faults is significantly different, such as the correlation of temperature abnormal rise with overload and short circuit faults is much higher than that of slight humidity fluctuation. This "averaging" feature selection reduces the accuracy of abnormality identification. On the other hand, the existing isolation forest method mainly conducts static risk assessment based on the data points of the current time or a short time window, and only outputs a momentary abnormality score. However, in actual fault development, "trend signals" are more valuable than "single high-risk points" in early warning, such as a slow rising sequence of risk index or a sudden increase in risk index change rate, which can reveal potential faults earlier. The existing method does not make full use of context information containing time dimension, which is prone to misjudgment of the fault development stage, thereby reducing the accuracy of early warning. SUMMARY

[0005] The present application provides a safety early warning method and system for distribution box to solve the problem of misjudgment of the fault development stage in the prior art, thereby reducing the accuracy of early warning.

[0006] In a first aspect, the safety early warning method for distribution box of the present application comprises the following steps:

[0007] Obtaining multi-dimensional monitoring data collected by multiple sensors in the distribution box within a continuous time window;

[0008] For the multi-dimensional monitoring data, a set of isolation trees is constructed, wherein the process of constructing a single isolation tree is as follows: at a non-leaf node of the tree, a non-uniform selection probability distribution is generated according to statistical characteristics of each feature of the data samples in the node, and a split feature is selected based on the selection probability distribution; a split point is selected in the value range of the selected split feature, and the data samples in the node are divided into two child nodes; the division is recursively performed until a preset tree height or a number of samples in the node is lower than a preset number is met;

[0009] An average path length of the data point to be warned in each isolation tree is calculated to obtain a basic anomaly score; an exponentially weighted moving average value of the basic anomaly score calculated from historical data points and a time change rate of the exponentially weighted moving average value are obtained, and an adjustment factor is calculated based on the exponentially weighted moving average value and the time change rate; a comprehensive risk index is obtained by using the basic anomaly score and the adjustment factor; when the comprehensive risk index exceeds a preset risk threshold, a power distribution box safety warning signal is generated.

[0010] Preferably, the non-uniform selection probability distribution is generated according to statistical characteristics of each feature of the data samples in the node, including: calculating the kurtosis value of the data samples in the node in each feature dimension; normalizing the kurtosis values of the features to obtain the selection probability of each feature, and the probability of a feature with a larger selection probability being selected as a split feature is higher.

[0011] Preferably, the exponentially weighted moving average value of the basic anomaly score calculated from the historical data points and the time change rate of the exponentially weighted moving average value are obtained, including: the exponentially weighted moving average value of the current time is calculated as ,

[0012] wherein, is a smoothing coefficient, is the basic anomaly score at the current time , and is the exponentially weighted moving average value at the previous time;

[0013] The time change rate of the current time is calculated as ,

[0014] .

[0015] Preferably, the comprehensive risk index is obtained by using the basic anomaly score and the adjustment factor, including:

[0016] The adjustment factor is calculated as :

[0017] ,in, and These are the preset weighting coefficients for positive constants. It is an exponential function with the natural constant e as the base. For the current moment Exponentially weighted moving average, For the current moment The rate of change over time;

[0018] Overall Risk Index Calculated using the following formula

[0019] ;in, Basic abnormal score, This is an adjustment factor.

[0020] Preferably, the risk threshold is calculated in the following manner:

[0021] Obtain historical monitoring data of the distribution box under normal operating conditions for one month, calculate the comprehensive risk index of all data points, and calculate the mean of the comprehensive risk index. and standard deviation Risk threshold equals .

[0022] Preferably, the multidimensional monitoring data includes temperature, current, voltage, humidity, and arc signal.

[0023] Secondly, the safety early warning system for the distribution box of the present invention includes the following modules:

[0024] The data acquisition module is used to acquire multi-dimensional monitoring data collected by multiple sensors in the distribution box within a continuous time window;

[0025] The isolation tree construction module is used to construct an isolation tree set for the multidimensional monitoring data. The process of constructing a single isolation tree is as follows: at the non-leaf nodes of the tree, a non-uniform selection probability distribution is generated based on the statistical characteristics of each feature of the data samples within the node, and a segmentation feature is selected based on the selection probability distribution; a segmentation point is selected within the value range of the selected segmentation feature to divide the data samples within the node into two child nodes; the division is recursively executed until a preset tree height is met or the number of samples within the node is lower than a preset number.

[0026] The early warning module is used to calculate the average path length of the data point to be warned in each isolation tree to obtain the basic anomaly score; obtain the exponentially weighted moving average and the time rate of change of the exponentially weighted moving average of the basic anomaly score calculated from historical data points, and calculate the adjustment factor based on the exponentially weighted moving average and the time rate of change; obtain the comprehensive risk index using the basic anomaly score and the adjustment factor; and generate a distribution box safety early warning signal when the comprehensive risk index exceeds the preset risk threshold.

[0027] Preferably, generating a non-uniform selection probability distribution based on the statistical characteristics of each feature of the data sample within the node includes: calculating the kurtosis value of the data sample within the node in each feature dimension; normalizing the kurtosis value of each feature to obtain the selection probability of each feature, wherein the feature with the higher selection probability has a higher probability of being used as a segmentation feature.

[0028] Preferably, obtaining the exponentially weighted moving average of the basic anomaly score calculated from historical data points and the time rate of change of the exponentially weighted moving average includes: the current time... Exponentially weighted moving average Calculated using the following formula

[0029] ,in, For smoothing coefficients, For the current moment Basic abnormal score, It is the exponentially weighted moving average of the previous time step;

[0030] Current moment rate of change over time Calculated using the following formula

[0031] .

[0032] Preferably, the step of obtaining the comprehensive risk index using the basic anomaly score and adjustment factor includes:

[0033] The adjustment factor is calculated using the following formula. :

[0034] ,in, and These are the preset weighting coefficients for positive constants. It is an exponential function with the natural constant e as the base. For the current moment Exponentially weighted moving average, For the current moment The rate of change over time;

[0035] Overall Risk Index Calculated using the following formula

[0036] ;in, Basic abnormal score, This is an adjustment factor.

[0037] The beneficial effects of this invention are as follows: By generating non-uniform selection probabilities based on the statistical characteristics of data features when constructing the isolation tree, this invention prioritizes the selection of monitoring features that are more indicative of abnormal states for segmentation, thereby enabling the detection of early faults caused by the coordinated changes of key parameters. By introducing the exponentially weighted moving average and time change rate of the basic anomaly score to construct a comprehensive risk index, the early warning decision can be combined with the development process of the equipment status, identifying potential faults that are continuously deteriorating or accelerating earlier, and reducing the risk of missed and false alarms due to misjudgment of trends. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the safety early warning method for a distribution box provided in an embodiment of the present invention. Detailed Implementation

[0039] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0040] like Figure 1 As shown, an embodiment of the safety early warning method for distribution boxes provided by the present invention includes the following steps:

[0041] S1 acquires multi-dimensional monitoring data collected by multiple sensors within the distribution box over a continuous time window.

[0042] First, temperature sensors, current transformers, voltage sensors, humidity sensors, and arc sensors are installed at key monitoring locations within the distribution box. These sensors collect core operating parameters of the distribution box in real time, including ambient temperature, three-phase line current, three-phase supply voltage, ambient humidity, and characteristic signals indicating the presence of arc discharge. This data forms a raw data stream arranged continuously in chronological order. To facilitate subsequent data integration and analysis, a sliding time window is set, for example, the most recent 10 minutes. All data points collected within this time window are integrated into a single dataset. Each data point is a multi-dimensional feature vector, containing structured data encompassing multiple monitoring dimensions such as temperature, current, voltage, humidity, and arc signal at the corresponding moment.

[0043] S2, for the multidimensional monitoring data, construct an isolation tree set, wherein the process of constructing a single isolation tree is as follows: at the non-leaf nodes of the tree, generate a non-uniform selection probability distribution based on the statistical characteristics of each feature of the data sample within the node, and select a segmentation feature based on the selection probability distribution; select a segmentation point within the value range of the selected segmentation feature, and divide the data sample within the node into two child nodes; recursively execute the division until the preset tree height is met or the number of samples within the node is lower than the preset number.

[0044] Specifically, first, an empty set of isolation trees is initialized, with a size of 100 trees. For the construction of each isolation tree in the set, a subset of samples is randomly selected from the overall training dataset and used as the dedicated training data for that isolation tree. During the construction of a single tree, when processing any non-leaf node, the variance of all data samples within that node in each feature dimension must be calculated, such as temperature, A-phase current, B-phase voltage, humidity, etc. Variance reflects the degree of data fluctuation under that feature dimension. Then, the variance of all features is normalized: by calculating the proportion of the variance of a single feature to the total variance of all features, the variance is converted into the corresponding feature selection probability. The larger the feature variance, the more significant the data difference for that feature, the stronger its ability to distinguish outliers, and therefore the higher the probability of it being selected as a segmentation feature. Finally, based on the above probability distribution, the roulette wheel selection method is used to ultimately determine the segmentation feature of the non-leaf node. In the roulette wheel selection method, features with higher probability proportions have a greater probability of being selected in random selection.

[0045] In an optional embodiment, a non-uniform selection probability distribution is generated based on the statistical characteristics of each feature of the data sample within the node, including: calculating the kurtosis value of the data sample within the node in each feature dimension; normalizing the kurtosis value of each feature to obtain the selection probability of each feature, wherein the feature with the higher selection probability has a higher probability of being used as a segmentation feature.

[0046] When constructing each node of a decision tree, feature selection must be based on the statistical characteristics of the data, with kurtosis being one of the core reference metrics. The physical meaning of kurtosis reflects the sharpness and tail thickness of the data distribution. A higher kurtosis value indicates that the data in that feature dimension contains more extreme values ​​or outliers, and these anomalous data are the core focus of anomaly detection. Therefore, features with high kurtosis are more suitable as a segmentation criterion.

[0047] For example: Suppose a node in a tree contains 100 data samples, each covering three monitoring features: temperature, current, and voltage. First, calculate the kurtosis values ​​of these 100 samples across the three feature dimensions. To convert the kurtosis into a feature probability that can be directly used for selection, these kurtosis values ​​need to be normalized: first, calculate the sum of the kurtosis values ​​of all features, then divide the kurtosis value of a single feature by the sum to obtain the selection probability of that feature. When splitting a node, the splitting feature is selected according to the probability distribution; the feature with the highest kurtosis value has the highest probability of being selected as the splitting feature. After determining the splitting feature, the next step is to select the split point. There are two ways to implement this: The first way: If temperature is selected as the splitting feature, first find the maximum and minimum temperatures of all samples within the node, then randomly select a value within this range as the split point, for example, 35.5℃. Subsequently, samples with temperatures below 35.5℃ are assigned to the left child node, and samples with temperatures ≥ 35.5℃ are assigned to the right child node. The second approach selects split points based on the local density of data samples within a node in terms of segmentation features. Regions with lower local density have a higher probability of being selected as split points. Specifically, a probability density function for this feature can be generated using kernel density estimation, making the selection probability of a split point inversely correlated with local density—that is, the lower the density, the higher the probability. This method can more accurately separate sparsely distributed outlier samples. After completing feature selection and sample partitioning for the current node, the process of feature selection-split point selection-sample partitioning needs to be repeated for the generated left and right child nodes until one of two stopping conditions is met: either the tree depth reaches a preset upper limit, or the number of samples within a node is less than a preset number. At this point, the node becomes a leaf node, and further partitioning stops.

[0048] S3, calculate the average path length of the data point to be warned in each isolation tree to obtain the basic anomaly score; obtain the exponentially weighted moving average and the time rate of change of the exponentially weighted moving average of the basic anomaly score calculated from historical data points, and calculate the adjustment factor based on the exponentially weighted moving average and the time rate of change; use the basic anomaly score and the adjustment factor to obtain the comprehensive risk index; when the comprehensive risk index exceeds the preset risk threshold, generate a power distribution box safety warning signal.

[0049] Specifically, when a newly acquired data point to be alerted is obtained, such as the current multidimensional feature vector: [temperature 45.2℃, current 15.3A, ...], this data point is input one by one into the 100 previously constructed isolation trees. Its basic anomaly score is calculated according to the following process: In each isolation tree, traversal begins from the root node. Based on the preset segmentation features and segmentation points of each node, the data point is gradually assigned to the corresponding child nodes until it finally reaches a leaf node. During this process, the path length traversed by the data point from the root node to the leaf node is recorded simultaneously, i.e., the number of edges traversed during the traversal. After traversing all 100 isolation trees, the average path length of the data point in the 100 trees is calculated. Then, according to the standard conversion formula of the isolation forest algorithm, this average path length is converted into a basic anomaly score ranging from 0 to 1. This basic anomaly score is the score for the current time. Basic abnormal score .

[0050] Specifically, at the current moment Exponentially weighted moving average Calculated using the following formula

[0051] ,in, For smoothing coefficients, For the current moment Basic abnormal score, It is the exponentially weighted moving average of the previous time step;

[0052] Current moment rate of change over time Calculated using the following formula

[0053] .

[0054] An exponentially weighted moving average method is used to calculate a smoothed sequence of scores by giving higher weights to recent data points. Assume a smoothing coefficient... With a weight of 0.3, when updating the EWMA value, the latest baseline anomaly score accounts for 30% of the weight, while the previous EWMA value accounts for 70% of the weight, which can filter out short-term noise.

[0055] In an optional embodiment, obtaining the comprehensive risk index using the baseline anomaly score and the adjustment factor includes: calculating the adjustment factor using the following formula. :

[0056] ,in, and These are the preset weighting coefficients for positive constants. It is an exponential function with the natural constant e as the base. For the current moment Exponentially weighted moving average, For the current moment The rate of change over time;

[0057] Overall Risk Index Calculated using the following formula

[0058] ;in, Basic abnormal score, This is an adjustment factor.

[0059] In an optional embodiment, the risk threshold is calculated in the following manner:

[0060] Obtain historical monitoring data of the distribution box under normal operating conditions for one month, calculate the comprehensive risk index of all data points, and calculate the mean of the comprehensive risk index. and standard deviation Risk threshold equals .

[0061] The implementation principle of the safety early warning method for distribution boxes in this invention is as follows: First, when constructing the isolation tree, this invention generates a non-uniform selection probability based on the statistical characteristics of data features. Unlike traditional methods that randomly select segmentation features, it prioritizes selecting monitoring features that are more indicative of abnormal states for data segmentation. This design can accurately capture early faults caused by the coordinated changes of key parameters, avoiding missing potential hazard signals due to ignoring core features. Second, this invention introduces an exponentially weighted moving average of the basic anomaly score and the time rate of change of this average to construct a comprehensive risk index. This design allows early warning decisions to no longer rely on static data at a single moment, but can combine the dynamic development process of equipment status, thereby identifying two types of high-risk situations earlier: one is the continuous deterioration of risk, and the other is the accelerated development of risk. Ultimately, this effectively reduces the risk of missed or false alarms caused by misjudgment of risk trends.

[0062] The safety early warning system for distribution boxes provided by this invention includes the following modules in its embodiments:

[0063] The data acquisition module is used to acquire multi-dimensional monitoring data collected by multiple sensors in the distribution box within a continuous time window;

[0064] The isolation tree construction module is used to construct an isolation tree set for the multidimensional monitoring data. The process of constructing a single isolation tree is as follows: at the non-leaf nodes of the tree, a non-uniform selection probability distribution is generated based on the statistical characteristics of each feature of the data samples within the node, and a segmentation feature is selected based on the selection probability distribution; a segmentation point is selected within the value range of the selected segmentation feature to divide the data samples within the node into two child nodes; the division is recursively executed until a preset tree height is met or the number of samples within the node is lower than a preset number.

[0065] The early warning module is used to calculate the average path length of the data point to be warned in each isolation tree to obtain the basic anomaly score; obtain the exponentially weighted moving average and the time rate of change of the exponentially weighted moving average of the basic anomaly score calculated from historical data points, and calculate the adjustment factor based on the exponentially weighted moving average and the time rate of change; obtain the comprehensive risk index using the basic anomaly score and the adjustment factor; and generate a distribution box safety early warning signal when the comprehensive risk index exceeds the preset risk threshold.

[0066] Specifically, generating a non-uniform selection probability distribution based on the statistical characteristics of each feature of the data sample within the node includes: calculating the kurtosis value of the data sample within the node in each feature dimension; normalizing the kurtosis value of each feature to obtain the selection probability of each feature, wherein the feature with the higher selection probability has a higher probability of being used as a segmentation feature.

[0067] Specifically, obtaining the exponentially weighted moving average of the basic anomaly score calculated from historical data points and the time rate of change of the exponentially weighted moving average includes: the current moment... Exponentially weighted moving average Calculated using the following formula

[0068] ,in, For smoothing coefficients, For the current moment Basic abnormal score, It is the exponentially weighted moving average of the previous time step;

[0069] Current moment rate of change over time Calculated using the following formula

[0070] .

[0071] Specifically, the method of obtaining the comprehensive risk index using the basic anomaly score and adjustment factor includes:

[0072] The adjustment factor is calculated using the following formula. :

[0073] ,in, and These are the preset weighting coefficients for positive constants. It is an exponential function with the natural constant e as the base. For the current moment Exponentially weighted moving average, For the current moment The rate of change over time;

[0074] Overall Risk Index Calculated using the following formula

[0075] ;in, Basic abnormal score, This is an adjustment factor.

[0076] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A safety early warning method for a distribution box, characterized in that, Includes the following steps: Acquire multidimensional monitoring data collected by multiple sensors within a continuous time window in the power distribution box; For the multidimensional monitoring data, an isolation tree set is constructed. The process of constructing a single isolation tree is as follows: at the non-leaf nodes of the tree, a non-uniform selection probability distribution is generated based on the statistical characteristics of each feature of the data samples within the node, and a segmentation feature is selected based on the selection probability distribution; a segmentation point is selected within the value range of the selected segmentation feature to divide the data samples within the node into two child nodes; the division is recursively executed until the preset tree height is met or the number of samples within the node is lower than the preset number. Calculate the average path length of the data point to be warned in each isolation tree to obtain the basic anomaly score; obtain the exponentially weighted moving average and the time rate of change of the exponentially weighted moving average of the basic anomaly score calculated from historical data points, and calculate the adjustment factor based on the exponentially weighted moving average and the time rate of change. The comprehensive risk index is obtained by using the baseline anomaly score and adjustment factors, including: The adjustment factor is calculated using the following formula. : ,in, and These are the preset weighting coefficients for positive constants. It is an exponential function with the natural constant e as the base. For the current moment Exponentially weighted moving average, For the current moment The rate of change over time; Overall Risk Index Calculated using the following formula ;in, Basic abnormal score, For adjustment factors; When the comprehensive risk index exceeds the preset risk threshold, a safety warning signal for the distribution box is generated.

2. The safety early warning method for a distribution box according to claim 1, characterized in that, The step of generating a non-uniform selection probability distribution based on the statistical characteristics of each feature of the data samples within the node includes: Calculate the kurtosis value of the data samples within a node on each feature dimension; The kurtosis values ​​of each feature are normalized to obtain the selection probability of each feature. The higher the selection probability, the higher the probability that the feature will be used as the segmentation feature.

3. The safety early warning method for a distribution box according to claim 1, characterized in that, The process of obtaining the exponentially weighted moving average of the basic anomaly score calculated from historical data points and the rate of change of the exponentially weighted moving average over time includes: Current moment Exponentially weighted moving average Calculated using the following formula ,in, For smoothing coefficients, For the current moment Basic abnormal score, It is the exponentially weighted moving average of the previous time step; Current moment rate of change over time Calculated using the following formula 。 4. The safety early warning method for a distribution box according to claim 1, characterized in that, The risk threshold is calculated in the following manner: Obtain historical monitoring data of the distribution box under normal operating conditions for one month, calculate the comprehensive risk index of all data points, and calculate the mean of the comprehensive risk index. and standard deviation Risk threshold equals .

5. The safety early warning method for a distribution box according to claim 1, characterized in that, The multidimensional monitoring data includes temperature, current, voltage, humidity, and arc signals.

6. A safety early warning system for a distribution box, characterized in that, Includes the following modules: The data acquisition module is used to acquire multi-dimensional monitoring data collected by multiple sensors in the distribution box within a continuous time window; The isolation tree construction module is used to construct an isolation tree set for the multidimensional monitoring data. The process of constructing a single isolation tree is as follows: at the non-leaf nodes of the tree, a non-uniform selection probability distribution is generated based on the statistical characteristics of each feature of the data samples within the node, and a segmentation feature is selected based on the selection probability distribution; a segmentation point is selected within the value range of the selected segmentation feature to divide the data samples within the node into two child nodes; the division is recursively executed until a preset tree height is met or the number of samples within the node is lower than a preset number. The early warning module is used to calculate the average path length of the data points to be warned in each isolation tree to obtain the basic anomaly score; Obtain the exponentially weighted moving average and the rate of change of the exponentially weighted moving average of the basic anomaly score calculated from historical data points, and calculate the adjustment factor based on the exponentially weighted moving average and the rate of change of the time. The comprehensive risk index is obtained by using the baseline anomaly score and adjustment factors, including: The adjustment factor is calculated using the following formula. : ,in, and These are the preset weighting coefficients for positive constants. It is an exponential function with the natural constant e as the base. For the current moment Exponentially weighted moving average, For the current moment The rate of change over time; Overall Risk Index Calculated using the following formula ;in, Basic abnormal score, For adjustment factors; When the comprehensive risk index exceeds the preset risk threshold, a safety warning signal for the distribution box is generated.

7. The safety early warning system for the distribution box according to claim 6, characterized in that, The step of generating a non-uniform selection probability distribution based on the statistical characteristics of each feature of the data samples within the node includes: Calculate the kurtosis value of the data samples within a node on each feature dimension; The kurtosis values ​​of each feature are normalized to obtain the selection probability of each feature. The higher the selection probability, the higher the probability that the feature will be used as the segmentation feature.

8. The safety early warning system for the distribution box according to claim 6, characterized in that, The process of obtaining the exponentially weighted moving average of the basic anomaly score calculated from historical data points and the rate of change of the exponentially weighted moving average over time includes: Current moment Exponentially weighted moving average Calculated using the following formula ,in, For smoothing coefficients, For the current moment Basic abnormal score, It is the exponentially weighted moving average of the previous time step; Current moment rate of change over time Calculated using the following formula 。

Images