Bus duct temperature and humidity abnormity monitoring system

Abstract

The invention relates to the technical field of power system monitoring, in particular to a bus duct temperature and humidity abnormity monitoring method, which comprises the following steps of S1, periodically acquiring temperature and humidity data through a temperature and humidity sensor, and performing data preprocessing to obtain processed data; s2, mutation detection and trend analysis are carried out according to the processed data, and mutation data and trend data are obtained respectively; s3, according to the abrupt change data and the trend data, a weighting strategy is adjusted in a self-adaptive mode; when the method is used, the credibility of data can be improved, the noise and errors of the sensor can be effectively reduced, the adaptability of the system can be improved, the data subjected to denoising and abnormal value elimination can be used as high-quality input of mutation detection, trend analysis and time sequence prediction, the accuracy of the whole temperature and humidity abnormity monitoring is improved, and the reliability of the system is improved. And short-time window detection and long-time window detection are integrated, so that the accuracy of anomaly monitoring is improved, false alarm caused by purely based on mutation detection is avoided, and the response speed of the method is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and in particular to a bus duct temperature and humidity anomaly monitoring system. Background Art

[0002] Busbars play a vital role in power systems, primarily for efficiently and safely distributing and transmitting high-current electrical energy. Busbars typically consist of metal strips (such as copper or aluminum) encased in insulating material and installed within a closed metal enclosure, forming a distribution channel. Because busbars typically carry high current loads, any overheating or excessive humidity can cause electrical failures or even serious accidents such as fires. Therefore, implementing effective temperature and humidity monitoring is crucial to preventing potential risks.

[0003] Patent publication number CN118640975A states in its specification that "the present invention discloses a method for monitoring busbar duct temperature and humidity anomalies, relating to the field of power system monitoring technology. The method collects regional values of the busbar duct, including the busbar duct body, components within the busbar duct, and the busbar duct connector. Based on the regional values, a monitoring layout is established for the busbar duct, and the monitoring layout is divided into temperature monitoring and humidity monitoring. The two monitoring values are labeled to construct an anomaly matrix. The monitoring values obtained in the anomaly matrix are compared with preset alarm rules. When the triggering alarm rules are met, an alarm message is generated, thereby improving the accuracy and reliability of busbar duct temperature and humidity monitoring, reducing false alarms or missed alarms, and notifying personnel of abnormal conditions in a more intuitive and convenient manner to quickly locate and handle abnormal conditions, thereby achieving multiple judgments on temperature and humidity, and further improving monitoring accuracy." Although the above-mentioned technology improves the stability and accuracy of monitoring data through sliding average filtering and multi-indicator comparison, the use of sliding average filtering to process data may cause a delay in system response. In the event of rapid temperature rise or sudden humidity changes, the system may not respond immediately, affecting safety.

[0004] In summary, the development of a bus duct temperature and humidity anomaly monitoring system is still a key issue that needs to be urgently addressed in the field of power system monitoring technology. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem in the prior art that although the above-mentioned technology improves the stability and accuracy of monitoring data through sliding average filtering and multi-indicator comparison, the use of sliding average filtering to process data may cause a lag in system response. When the temperature rises rapidly or the humidity suddenly changes, the system may not be able to respond immediately.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention provides a method for monitoring abnormal temperature and humidity in a bus duct, comprising the following steps:

[0008] S1. Periodically collect temperature and humidity data through temperature and humidity sensors, and perform data preprocessing to obtain processed data;

[0009] S2. performing mutation detection and trend analysis on the processed data to obtain mutation data and trend data, respectively;

[0010] S3, adaptively adjust the weighting strategy based on mutation data and trend data;

[0011] S4. Use time series forecasting models to analyze historical data and trend data and provide forecast data;

[0012] S5. Perform anomaly detection and abnormality alarm and early warning by real-time monitoring of the mutation data, trend data and forecast data.

[0013] Furthermore, in step S1, the temperature and humidity data are periodically collected by the temperature and humidity sensor, and the data is preprocessed to obtain the processed data as follows:

[0014] The temperature and humidity sensors are installed at key locations of the bus duct, including the connection end, closed conductor, and places where water easily accumulates in a humid environment. The sensors collect temperature and humidity data X(t) in a periodic T sampling manner. At time t, the temperature and humidity data are X(t) respectively. W (t) and X S (t), expression formula: where Q W (t),Q S (t) is the real temperature and humidity signal, E W (t),E S (t) is random noise, which is set to have a mean of 0 and a variance of σ 2 Gaussian white noise, expression formula: E W (t),E S (t)~ζ(0,σ 2 ), the data preprocessing includes using Kalman filtering to denoise the temperature and humidity data, and the denoised temperature and humidity data are respectively and Use Grubbs test to remove outliers in temperature and humidity data, and define temperature data X W The mean and standard deviation of (t) are expressed as:

[0015] in Represents the average value of temperature data in a time window, N represents the total number of data points in the statistical time window, represents the denoised temperature value at the i-th time point, represents the sum of all data points, Indicates the degree of fluctuation of temperature data, represents the difference between the i-th temperature value and the average temperature, is the square of the temperature deviation from the mean, is the sum of the squared deviations of all data points, is an unbiased estimate used in computing the mean square error, The result is squared and the Grubbs statistic is calculated using the formula: The threshold value of Grubbs test is: where t β / (2N),N-2 is the critical value of the t distribution with N-2 degrees of freedom, when R>R ctil The data point is determined to be an outlier and is removed. The method for removing outliers in humidity data is the same.

[0016] Furthermore, in step S2, mutation detection and trend analysis are performed based on the processed data to obtain mutation data and trend data respectively as follows:

[0017] The mutation detection includes using short-term window processing to calculate the temperature and humidity change trend within the last 1 second to 3 seconds. At time t, temperature and humidity data within O (1 second to 3 seconds) are selected and the change rate of the short-term window is calculated. The expression formula is: Among them U W (t) is the temperature change rate, U S (t) is the humidity change rate, O is the short-term window (1 second to 3 seconds), ψ i is the short-time window weight coefficient. The dynamic threshold method is used to set the adaptive mutation threshold using the historical mean and standard deviation. The discrete wavelet transform is used to perform multi-scale analysis on the temperature and humidity signals, extract the high-frequency components, and identify the mutation points. The expression formula is: where Y W (a,b),Y S (a,b) are the wavelet transform coefficients, a is the scale factor (used to separate high and low frequencies), b is the translation factor, and χ * (·) is the mother wavelet (Daubechies wavelet).

[0018] Furthermore, in step S2, mutation detection and trend analysis are performed based on the processed data to obtain mutation data and trend data respectively as follows:

[0019] The trend analysis includes using long-term window processing to calculate the temperature and humidity change trend in the last 10 seconds to 60 seconds, selecting data within the past A (10 seconds to 60 seconds), and calculating the weighted sliding average to smooth the trend. The expression formula is: Where W ted (t) is the temperature trend value, S ted (t) is the humidity trend value, A is the trend analysis window length (within the past 10 seconds to 60 seconds), δ i is the long-term window weight coefficient, satisfying ∑δ i =1, weight coefficient expression formula: Where ε is the speed at which the weight decays. is an exponential decay weight, is the normalization factor.

[0020] Furthermore, in step S3, the method of adaptively adjusting the weighting strategy according to the mutation data and trend data is as follows:

[0021] The weighting strategy includes increasing the short-term window weight if any one of the two items, temperature sudden increase ≥ 5°C and humidity sudden increase ≥ 10%RH, is included in the mutation data, and defining the mutation response factor D jmp (t), expression formula: Where ΔW=W(t)-W(tO) is the short-term temperature change, ΔS=S(t)-S(tO) is the short-term humidity change, jmp (t) ≥ 1, the mutation is triggered and the short-term window weight is updated. The expression formula is:

[0022] where φ w The short-term window adjustment rate is 0.1-0.3, ψ i (t) is the short-term window weight at time t, ψ i (t-1) is the short-term window weight of the previous moment, (1-ψ i (t-1)) is the complement of the short-time window, ζ i (t)=1-ψ i (t) is the long-term window weight used for long-term trend analysis. When the short-term window weight increases, the long-term window weight decreases accordingly, and vice versa.

[0023] Furthermore, in step S3, the method of adaptively adjusting the weighting strategy according to the mutation data and trend data is as follows:

[0024] The weighting strategy includes the following two items: if the temperature continues to rise ≥3℃ / min or the humidity continues to rise ≥5%RH / min in the trend data, it is determined that there is a trend anomaly, the short-term window weight is reduced, the trend data analysis is strengthened, and the trend anomaly factor D is defined. ted (t), expression formula: in is the long-term temperature change rate, is the long-term humidity change rate, in D ted(t) ≥ 1, the trend anomaly is triggered, and the long-term window weight is adjusted. The expression formula is:

[0025] where φ s The trend adjustment coefficient is 0.05-0.2, ζ i (t) is the long-term window weight at the current time t, ζ i (t-1) is the long-term window weight of the previous moment, D ted (t) is the trend detection factor, (1-ζ i (t-1)) is the adjustment factor, and the weight is updated using the exponential smoothing method, expressed as: in Use a value of 0.2-0.5 for smooth changes.

[0026] Furthermore, in step S4, a method for using a time series forecasting model to analyze historical data and trend data and provide forecast data is as follows:

[0027] Collect temperature and humidity data X(t) and perform time series modeling for short-term prediction. The expression formula is: Where X(t) is the temperature and humidity data at the current time t, c represents any one of the average level and long-term trend of the temperature and humidity data, and ∈(t) represents the unpredictable noise at the current time point. represents how past prediction errors affect the current temperature and humidity prediction, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. G represents how many past prediction errors are used for prediction, and η k It measures the impact of past prediction errors on future predictions and uses long-term short-term memory networks for long-term predictions. The output hidden state is used to predict the temperature and humidity data in the next second. The expression formula is:

[0028] in is the temperature and humidity data at the predicted time t+n, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. X(t+ni) represents the temperature and humidity value at time (t+ni), G represents the number of past prediction errors used for prediction, and η k It is a measure of the impact of past forecast errors on future forecasts. Represents the error between the predicted value and the true value at a past time point.

[0029] Furthermore, in step S4, a method for using a time series forecasting model to analyze historical data and trend data and provide forecast data is as follows:

[0030] The weighted fusion method is used to combine the temperature and humidity data predicted by the long short-term memory network and the time series prediction model, and the expression formula is:

[0031] where ι is the weight factor, Represents the final prediction result at time t+n, is the prediction result of the time series prediction model at time t+n. It is the prediction result of the long short-term memory network at time t+n, which depends on the error between the long short-term memory network and the time series prediction model. The expression formula is: Where ι is the weight factor, h is the long short-term memory network, κ is the time series prediction model, e is the base of the natural logarithm, and MSE is the mean squared error.

[0032] Furthermore, in step S5, the method for performing anomaly detection and abnormality alarm and early warning by real-time monitoring of the mutation data, trend data and prediction data is as follows:

[0033] The anomaly detection is based on the triple threshold judgment method to calculate the anomaly score of each data source, expressed as follows: λ(t) = μ1·λ v (t)+μ2·λ θ (t)+μ3·λ κ (t), where μ1, μ2, and μ3 are weighted coefficients satisfying μ1+μ2+μ3=1, which are used to adjust the contribution of different data sources, and λ v (t) reflects the degree of sudden change in temperature and humidity, λ θ (t) reflects the long-term trend anomaly of temperature and humidity, λ κ (t) reflects the abnormal degree of predicted temperature and humidity. The degree of mutation of temperature and humidity can be calculated by the rate of change within a short time window, expressed as: in is the temperature and humidity change rate within the short-term window, The standard deviation of historical temperature and humidity changes is used for normalization. The abnormality of trend data is calculated by the temperature and humidity change rate in the past A (10 seconds to 60 seconds), expressed as: where λ θ (t) is the overall change trend of the temperature and humidity data at time t, A is the long-term window size, is the standard deviation of the trend data, The trend change at time ti represents the rate of change of temperature and humidity data. The predicted anomaly score measures the deviation between the predicted temperature and humidity and the current data. The expression formula is: where λ κ (t) is the predicted anomaly score that measures the degree of deviation between the predicted value of temperature and humidity and the current measured value. is the standard deviation of the predicted data used to normalize the predicted scores, is the predicted temperature and humidity value at time t+n′, is the actual measured value of temperature and humidity at the current time t, is the absolute error between the predicted value and the current actual measured value. According to the comprehensive anomaly score λ(t), the adaptive alarm threshold Θ is set. bj (t), expression formula: where Θ bj (t) is the adaptive alarm threshold, is the mean of the historical anomaly scores, is the standard deviation of historical anomaly scores, and r′ is the adaptive coefficient that determines the alarm sensitivity. The alarm sensitivity includes level one warning, level two warning and level three warning.

[0034] On the other hand, the present invention also provides a bus duct temperature and humidity anomaly monitoring system, comprising:

[0035] The data acquisition module periodically collects temperature and humidity data through temperature and humidity sensors, and performs data preprocessing to obtain processed data;

[0036] a detection and analysis module, performing mutation detection and trend analysis based on the processed data to obtain mutation data and trend data respectively;

[0037] The adjustment strategy module adaptively adjusts the weighting strategy based on mutation data and trend data;

[0038] The time series forecasting module uses a time series forecasting model to analyze historical data and trend data and provide forecast data;

[0039] The abnormality alarm module detects abnormalities and issues abnormality alarms and early warnings by monitoring the mutation data, trend data and forecast data in real time;

[0040] Furthermore, the temperature and humidity data are collected periodically by the temperature and humidity sensor, and the data is pre-processed to obtain the processed data. The operation process includes: the temperature and humidity sensor is installed at the key parts of the bus duct, which include the connection end, the closed conductor, and the place where water easily accumulates in the humid environment. The sensor collects the temperature and humidity data X(t) in a periodic T sampling manner. At the time t, the temperature and humidity data are X(t) respectively. W (t) and X S (t), expression formula: where Q W (t),Q S (t) is the real temperature and humidity signal, E W (t),E S(t) is random noise, which is set to have a mean of 0 and a variance of σ 2 Gaussian white noise, expression formula: E W (t),E S (t)~ζ(0,σ 2 ), the data preprocessing includes using Kalman filtering to denoise the temperature and humidity data, and the denoised temperature and humidity data are respectively and Use Grubbs test to remove outliers in temperature and humidity data, and define temperature data X W The mean and standard deviation of (t) are expressed as: in Represents the average value of temperature data in a time window, N represents the total number of data points in the statistical time window, represents the denoised temperature value at the i-th time point, represents the sum of all data points, Indicates the degree of fluctuation of temperature data, represents the difference between the i-th temperature value and the average temperature, is the square of the temperature deviation from the mean, is the sum of the squared deviations of all data points, is an unbiased estimate used in computing the mean square error, The result is squared and the Grubbs statistic is calculated using the formula: The threshold value of Grubbs test is: where t β / (2N),N-2 is the critical value of the t distribution with N-2 degrees of freedom, when R>R ctil The data point is determined to be an outlier and is removed. The method for removing outliers in humidity data is the same;

[0041] Furthermore, mutation detection and trend analysis are performed on the processed data to obtain mutation data and trend data respectively. The mutation detection includes calculating the temperature and humidity change trend within the last 1 second to 3 seconds using short-term window processing. At time t, temperature and humidity data within O (1 second to 3 seconds) are selected to calculate the change rate of the short-term window. The expression formula is: Among them U W (t) is the temperature change rate, U S (t) is the humidity change rate, O is the short-term window (1 second to 3 seconds), ψ i is the short-time window weight coefficient. The dynamic threshold method is used to set the adaptive mutation threshold using the historical mean and standard deviation. The discrete wavelet transform is used to perform multi-scale analysis on the temperature and humidity signals, extract the high-frequency components, and identify the mutation points. The expression formula is: where Y W(a,b),Y S (a,b) are the wavelet transform coefficients, a is the scale factor (used to separate high and low frequencies), b is the translation factor, and χ * (·) is the mother wavelet (Daubechies wavelet).

[0042] The trend analysis includes using long-term window processing to calculate the temperature and humidity change trend in the last 10 seconds to 60 seconds, selecting data within the past A (10 seconds to 60 seconds), and calculating the weighted sliding average to smooth the trend. The expression formula is: Where W ted (t) is the temperature trend value, S ted (t) is the humidity trend value, A is the trend analysis window length (within the past 10 seconds to 60 seconds), δ i is the long-term window weight coefficient, satisfying Σδ i =1, weight coefficient expression formula: Where ε is the speed at which the weight decays. is an exponential decay weight, is the normalization factor;

[0043] Furthermore, the operation process of adaptively adjusting the weighting strategy according to the mutation data and trend data includes: if the weighting strategy includes any one of the two items of temperature sudden increase ≥5°C and humidity sudden increase ≥10%RH in the mutation data, the short-term window weight is increased, and the mutation response factor D is defined. jmp (t), expression formula: Where ΔW=W(t)-W(tO) is the short-term temperature change, ΔS=S(t)-S(tO) is the short-term humidity change, jmp (t) ≥ 1, the mutation is triggered and the short-term window weight is updated. The expression formula is: where φ w The short-term window adjustment rate is 0.1-0.3, ψ i (t) is the short-term window weight at time t, ψ i (t-1) is the short-term window weight of the previous moment, (1-ψ i (t-1)) is the complement of the short-time window, ζ i (t)=1-ψ i (t) is the long-term window weight used for long-term trend analysis. When the short-term window weight increases, the long-term window weight decreases accordingly, and vice versa.

[0044] The weighting strategy includes the following two items: if the temperature continues to rise ≥3℃ / min or the humidity continues to rise ≥5%RH / min in the trend data, it is determined that there is a trend anomaly, the short-term window weight is reduced, the trend data analysis is strengthened, and the trend anomaly factor D is defined. ted (t), expression formula: in is the long-term temperature change rate, is the long-term humidity change rate, in D ted (t) ≥ 1, the trend anomaly is triggered, and the long-term window weight is adjusted. The expression formula is:

[0045] where φ s The trend adjustment coefficient is 0.05-0.2, ζ i (t) is the long-term window weight at the current time t, ζ i (t-1) is the long-term window weight of the previous moment, D ted (t) is the trend detection factor, (1-ζ i (t-1)) is the adjustment factor, and the weight is updated using the exponential smoothing method, expressed as: in Use a value of 0.2-0.5 for smooth changes;

[0046] Furthermore, the time series forecasting model is used to analyze historical data and trend data and provide forecast data. The operation process includes: collecting temperature and humidity data X(t) and performing time series modeling for short-term forecasting. The expression formula is:

[0047] Where X(t) is the temperature and humidity data at the current time t, c represents any one of the average level and long-term trend of the temperature and humidity data, and ∈(t) represents the unpredictable noise at the current time point. represents how past prediction errors affect the current temperature and humidity prediction, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. G represents how many past prediction errors are used for prediction, and η k It measures the impact of past prediction errors on future predictions and uses long-term short-term memory networks for long-term predictions. The output hidden state is used to predict the temperature and humidity data in the next second. The expression formula is:

[0048] in is the temperature and humidity data at the predicted time t+n, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. X(t+ni) represents the temperature and humidity value at time (t+ni), G represents the number of past prediction errors used for prediction, and η k It is a measure of the impact of past forecast errors on future forecasts. Represents the error between the predicted value and the true value at a past time point.

[0049] The weighted fusion method is used to combine the temperature and humidity data predicted by the long short-term memory network and the time series prediction model, and the expression formula is:

[0050] where ι is the weight factor, Represents the final prediction result at time t+n, is the prediction result of the time series prediction model at time t+n. It is the prediction result of the long short-term memory network at time t+n, which depends on the error between the long short-term memory network and the time series prediction model. The expression formula is: Where ι is the weight factor, h is the long short-term memory network, κ is the time series prediction model, e is the base of the natural logarithm, and MSE is the mean square error;

[0051] Furthermore, by real-time monitoring of the mutation data, trend data and prediction data, anomaly detection and anomaly alarm and early warning operation process includes: the anomaly detection is based on the triple threshold judgment method, and the anomaly score of each data source is calculated, which is expressed as follows: λ(t) = μ1·λ v (t)+μ2·λ θ (t)+μ3·λ κ (t), where μ1, μ2, and μ3 are weighted coefficients satisfying μ1+μ2+μ3=1, which are used to adjust the contribution of different data sources, and λ v (t) reflects the degree of sudden change in temperature and humidity, λ θ (t) reflects the long-term trend anomaly of temperature and humidity, λ κ (t) reflects the abnormal degree of predicted temperature and humidity. The degree of mutation of temperature and humidity can be calculated by the rate of change within a short time window, expressed as: in is the temperature and humidity change rate within the short-term window, The standard deviation of historical temperature and humidity changes is used for normalization. The abnormality of trend data is calculated by the temperature and humidity change rate in the past A (10 seconds to 60 seconds), expressed as: where λ θ (t) is the overall change trend of the temperature and humidity data at time t, A is the long-term window size, is the standard deviation of the trend data, The trend change at time ti represents the rate of change of temperature and humidity data. The predicted anomaly score measures the deviation between the predicted temperature and humidity and the current data. The expression formula is: where λ κ (t) is the predicted anomaly score that measures the degree of deviation between the predicted value of temperature and humidity and the current measured value. is the standard deviation of the predicted data used to normalize the predicted scores, is the predicted temperature and humidity value at time t+n′, is the actual measured value of temperature and humidity at the current time t, is the absolute error between the predicted value and the current actual measured value. According to the comprehensive anomaly score λ(t), the adaptive alarm threshold Θ is set. bj (t), expression formula: where Θ bj (t) is the adaptive alarm threshold, is the mean of the historical anomaly scores, is the standard deviation of historical anomaly scores, and r′ is the adaptive coefficient that determines the alarm sensitivity. The alarm sensitivity includes level one warning, level two warning and level three warning.

[0052] Beneficial effects

[0053] Compared with the known public technology, the technical solution provided by the present invention has the following advantages:

[0054] Beneficial effects:

[0055] When used, the present invention is conducive to improving the credibility of data. This method can effectively reduce sensor noise and errors, improve system adaptability, and the data after denoising and outlier removal can be used as high-quality input for mutation detection, trend analysis and time series prediction, thereby improving the accuracy of the entire temperature and humidity anomaly monitoring. It integrates short-term window and long-term window detection to improve the accuracy of anomaly monitoring, which is conducive to avoiding false alarms caused by mutation detection alone and improving the response speed of the present invention.

[0056] When in use, the present invention adopts an adaptive weighting strategy, which can dynamically adjust the weights of short-term and long-term windows according to the real-time changes of temperature and humidity, which is conducive to avoiding the misjudgment problem that may be caused by a fixed weight scheme. The adaptive adjustment of the weights of the two can not only quickly respond to anomalies, but also accurately predict trend hazards, thereby improving the stability and accuracy of the overall monitoring. The triple threshold judgment method combines short-term, long-term and predicted data to facilitate the reduction of false alarms and missed alarms, improve detection accuracy, adopt an adaptive alarm threshold, dynamically adjust the alarm standard, adapt to different environmental changes, and avoid misjudgment caused by fixed thresholds. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a flow chart of a method for monitoring abnormal temperature and humidity in a bus duct according to the present invention;

[0058] Figure 2 This is a system diagram of a bus duct temperature and humidity anomaly monitoring system of the present invention. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0060] It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the numbers used in this manner are interchangeable where appropriate so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatuses.

[0061] The present invention is described in further detail below with reference to the accompanying drawings:

[0062] Example 1:

[0063] like Figure 1 As shown, the present invention provides a bus duct temperature and humidity anomaly monitoring method, comprising the following steps: S1, periodically collecting temperature and humidity data through a temperature and humidity sensor, and performing data preprocessing to obtain processed data;

[0064] Furthermore, in step S1, the temperature and humidity data are periodically collected by the temperature and humidity sensor, and the data is preprocessed to obtain the processed data as follows:

[0065] The temperature and humidity sensors are installed at key locations of the bus duct, including the connection end, closed conductor, and places where water easily accumulates in a humid environment. The sensors collect temperature and humidity data X(t) in a periodic T sampling manner. At time t, the temperature and humidity data are X(t) respectively. W (t) and X S (t), expression formula: where Q W (t),Q S (t) is the real temperature and humidity signal, E W (t),E S (t) is random noise, which is set to have a mean of 0 and a variance of σ 2 Gaussian white noise, expression formula: EW (t),E S (t)~ζ(0,σ 2 ), the data preprocessing includes using Kalman filtering to denoise the temperature and humidity data, and the denoised temperature and humidity data are respectively and Use Grubbs test to remove outliers in temperature and humidity data, and define temperature data X W The mean and standard deviation of (t) are expressed as:

[0066] in Represents the average value of temperature data in a time window, N represents the total number of data points in the statistical time window, represents the denoised temperature value at the i-th time point, represents the sum of all data points, Indicates the degree of fluctuation of temperature data, represents the difference between the i-th temperature value and the average temperature, is the square of the temperature deviation from the mean, is the sum of the squared deviations of all data points, is an unbiased estimate used in computing the mean square error, The result is squared and the Grubbs statistic is calculated using the formula: The threshold value of Grubbs test is: where t β / (2N),N-2 is the critical value of the t distribution with N-2 degrees of freedom, when R>R ctil The data point is determined to be an outlier and is removed. The method for removing outliers in humidity data is the same.

[0067] In this embodiment, Kalman filtering can effectively remove Gaussian noise, make temperature and humidity data smoother, reduce errors, and help improve the reliability of monitoring. Grubbs test can identify and eliminate outliers, avoid interference of extreme data on overall trend judgment, and help improve the credibility of data. This method can effectively reduce sensor noise and errors, improve system adaptability, and the data after denoising and outlier removal can be used as high-quality input for mutation detection, trend analysis and time series prediction, thereby improving the accuracy of the entire temperature and humidity anomaly monitoring.

[0068] S2. performing mutation detection and trend analysis on the processed data to obtain mutation data and trend data, respectively;

[0069] Furthermore, in step S2, mutation detection and trend analysis are performed based on the processed data to obtain mutation data and trend data respectively as follows:

[0070] The mutation detection includes using short-term window processing to calculate the temperature and humidity change trend within the last 1 second to 3 seconds. At time t, temperature and humidity data within O (1 second to 3 seconds) are selected and the change rate of the short-term window is calculated. The expression formula is: Among them U W (t) is the temperature change rate, U S (t) is the humidity change rate, O is the short-term window (1 second to 3 seconds), ψ i is the short-time window weight coefficient. The dynamic threshold method is used to set the adaptive mutation threshold using the historical mean and standard deviation. The discrete wavelet transform is used to perform multi-scale analysis on the temperature and humidity signals, extract the high-frequency components, and identify the mutation points. The expression formula is: where Y W (a,b),Y S (a,b) are the wavelet transform coefficients, a is the scale factor (used to separate high and low frequencies), b is the translation factor, and χ * (·) is the mother wavelet (Daubechies wavelet).

[0071] Furthermore, in step S2, mutation detection and trend analysis are performed based on the processed data to obtain mutation data and trend data respectively as follows:

[0072] The trend analysis includes using long-term window processing to calculate the temperature and humidity change trend in the last 10 seconds to 60 seconds, selecting data within the past A (10 seconds to 60 seconds), and calculating the weighted sliding average to smooth the trend. The expression formula is: Where W ted (t) is the temperature trend value, S ted (t) is the humidity trend value, A is the trend analysis window length (within the past 10 seconds to 60 seconds), δ i is the long-term window weight coefficient, satisfying ∑δ i =1, weight coefficient expression formula: Where ε is the speed at which the weight decays. is an exponential decay weight, is the normalization factor.

[0073] In this embodiment, short-term window analysis combined with dynamic thresholds and wavelet transforms can effectively identify short-term anomalies such as local overheating of the bus duct and sudden increases in humidity. Long-term window analysis uses weighted sliding average and smooth attenuation weights to avoid the interference of short-term fluctuations on long-term trend judgment, improve prediction accuracy, and integrate short-term window and long-term window detection to improve the accuracy of anomaly monitoring, which is conducive to avoiding false alarms caused by mutation detection alone and improving the response speed of the present invention.

[0074] S3, adaptively adjust the weighting strategy based on mutation data and trend data;

[0075] Furthermore, in step S3, the method of adaptively adjusting the weighting strategy according to the mutation data and trend data is as follows:

[0076] The weighting strategy includes increasing the short-term window weight if any one of the two items, temperature sudden increase ≥ 5°C and humidity sudden increase ≥ 10%RH, is included in the mutation data, and defining the mutation response factor D jmp (t), expression formula: Where ΔW=W(t)-W(tO) is the short-term temperature change, ΔS=S(t)-S(tO) is the short-term humidity change, jmp (t) ≥ 1, the mutation is triggered and the short-term window weight is updated. The expression formula is:

[0077] where φ w The short-term window adjustment rate is 0.1-0.3, ψ i (t) is the short-term window weight at time t, ψ i (t-1) is the short-term window weight of the previous moment, (1-ψ i (t-1)) is the complement of the short-time window, ζ i (t)=1-ψ i (t) is the long-term window weight used for long-term trend analysis. When the short-term window weight increases, the long-term window weight decreases accordingly, and vice versa.

[0078] Furthermore, in step S3, the method of adaptively adjusting the weighting strategy according to the mutation data and trend data is as follows:

[0079] The weighting strategy includes the following two items: if the temperature continues to rise ≥3℃ / min or the humidity continues to rise ≥5%RH / min in the trend data, it is determined that there is a trend anomaly, the short-term window weight is reduced, the trend data analysis is strengthened, and the trend anomaly factor D is defined. ted (t), expression formula: in is the long-term temperature change rate, is the long-term humidity change rate, in D ted (t) ≥ 1, the trend anomaly is triggered, and the long-term window weight is adjusted. The expression formula is:

[0080] where φ s The trend adjustment coefficient is 0.05-0.2, ζ i (t) is the long-term window weight at the current time t, ζ i (t-1) is the long-term window weight of the previous moment, D ted (t) is the trend detection factor, (1-ζ i(t-1)) is the adjustment factor, and the weight is updated using the exponential smoothing method, expressed as: in Use a value of 0.2-0.5 for smooth changes.

[0081] In this embodiment, an adaptive weighting strategy is adopted to dynamically adjust the weights of short-term and long-term windows according to the real-time changes in temperature and humidity, which is conducive to avoiding the misjudgment problem that may be caused by a fixed weight scheme. When a sudden change occurs, the weight of the short-term window is quickly increased to improve the detection capability of sudden anomalies. For example, the sudden temperature rise at the bus duct connection point or the rapid identification of water accumulation in a humid environment are only short-term fluctuations and will not be misjudged as an anomaly. Instead, more accurate judgments are made with the support of trend data, thereby reducing false alarms and missed alarms. The short-term window is used to detect sudden changes, and the long-term window is used to analyze the long-term operating trend of the equipment. The weights of the two are adaptively adjusted, which can not only quickly respond to anomalies, but also accurately predict trend hazards, thereby improving the stability and accuracy of the overall monitoring.

[0082] S4. Use time series forecasting models to analyze historical data and trend data and provide forecast data;

[0083] Furthermore, in step S4, a method for using a time series forecasting model to analyze historical data and trend data and provide forecast data is as follows:

[0084] Collect temperature and humidity data X(t) and perform time series modeling for short-term prediction. The expression formula is: Where X(t) is the temperature and humidity data at the current time t, c represents any one of the average level and long-term trend of the temperature and humidity data, and ∈(t) represents the unpredictable noise at the current time point. represents how past prediction errors affect the current temperature and humidity prediction, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. G represents how many past prediction errors are used for prediction, and η k It measures the impact of past prediction errors on future predictions and uses long-term short-term memory networks for long-term predictions. The output hidden state is used to predict the temperature and humidity data in the next second. The expression formula is:

[0085] in is the temperature and humidity data at the predicted time t+n, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. X(t+ni) represents the temperature and humidity value at time (t+ni), G represents the number of past prediction errors used for prediction, and η kIt is a measure of the impact of past forecast errors on future forecasts. Represents the error between the predicted value and the true value at a past time point.

[0086] Furthermore, in step S4, a method for using a time series forecasting model to analyze historical data and trend data and provide forecast data is as follows:

[0087] The weighted fusion method is used to combine the temperature and humidity data predicted by the long short-term memory network and the time series prediction model, and the expression formula is:

[0088] where ι is the weight factor, Represents the final prediction result at time t+n, is the prediction result of the time series prediction model at time t+n. It is the prediction result of the long short-term memory network at time t+n, which depends on the error between the long short-term memory network and the time series prediction model. The expression formula is: Where ι is the weight factor, h is the long short-term memory network, κ is the time series prediction model, e is the base of the natural logarithm, and MSE is the mean squared error.

[0089] In this embodiment, through short-term prediction, mutation trends can be discovered a few seconds before temperature and humidity anomalies occur. The long short-term memory network can predict the overall trend of temperature and humidity in the bus duct through long-term time-dependent learning, which helps to plan equipment maintenance in advance, avoid unnecessary alarms, and improve early warning accuracy. The weights are adaptively adjusted according to real-time data changes, and the most appropriate prediction method is automatically selected, which is conducive to improving prediction accuracy.

[0090] S5. Perform anomaly detection and abnormality alarm and early warning by real-time monitoring of the mutation data, trend data, and forecast data;

[0091] Furthermore, in step S5, the method for performing anomaly detection and abnormality alarm and early warning by real-time monitoring of the mutation data, trend data and prediction data is as follows:

[0092] The anomaly detection is based on the triple threshold judgment method to calculate the anomaly score of each data source, expressed as follows: λ(t) = μ1·λ v (t)+μ2·λ θ (t)+μ3·λ κ (t), where μ1, μ2, and μ3 are weighted coefficients satisfying μ1+μ2+μ3=1, which are used to adjust the contribution of different data sources, and λ v (t) reflects the degree of sudden change in temperature and humidity, λ θ (t) reflects the long-term trend anomaly of temperature and humidity, λ κ(t) reflects the abnormal degree of predicted temperature and humidity. The degree of mutation of temperature and humidity can be calculated by the rate of change within a short time window, expressed as: in is the temperature and humidity change rate within the short-term window, The standard deviation of historical temperature and humidity changes is used for normalization. The abnormality of trend data is calculated by the temperature and humidity change rate in the past A (10 seconds to 60 seconds), expressed as: where λ θ (t) is the overall change trend of the temperature and humidity data at time t, A is the long-term window size, is the standard deviation of the trend data, The trend change at time ti represents the rate of change of temperature and humidity data. The predicted anomaly score measures the deviation between the predicted temperature and humidity and the current data. The expression formula is: where λ κ (t) is the predicted anomaly score that measures the degree of deviation between the predicted value of temperature and humidity and the current measured value. is the standard deviation of the predicted data used to normalize the predicted scores, is the predicted temperature and humidity value at time t+n′, is the actual measured value of temperature and humidity at the current time t, is the absolute error between the predicted value and the current actual measured value. According to the comprehensive anomaly score λ(t), the adaptive alarm threshold Θ is set. bj (t), expression formula: where Θ bj (t) is the adaptive alarm threshold, is the mean of the historical anomaly scores, is the standard deviation of historical anomaly scores, and r′ is the adaptive coefficient that determines the alarm sensitivity. The alarm sensitivity includes level one warning, level two warning and level three warning.

[0093] In this embodiment, the first-level warning includes the system recording abnormal points, but does not trigger external alarms and is only used for internal monitoring. The second-level warning includes further increase in the abnormality score, triggering notification to operation and maintenance personnel to conduct inspections. The third-level alarm includes serious exceeding of the threshold, immediately triggering sound and light alarms, and linking equipment protection measures. The triple threshold judgment method combines short-term, long-term and predicted data to facilitate the reduction of false alarms and missed alarms, improve detection accuracy, and adopts adaptive alarm thresholds to dynamically adjust the alarm standards to adapt to changes in different environments and avoid misjudgments caused by fixed thresholds.

[0094] Example 2:

[0095] like Figure 2 As shown, embodiment 2 provides a bus duct temperature and humidity anomaly monitoring system, including:

[0096] The data acquisition module periodically collects temperature and humidity data through temperature and humidity sensors, and performs data preprocessing to obtain processed data;

[0097] a detection and analysis module, performing mutation detection and trend analysis based on the processed data to obtain mutation data and trend data respectively;

[0098] The adjustment strategy module adaptively adjusts the weighting strategy based on mutation data and trend data;

[0099] The time series forecasting module uses a time series forecasting model to analyze historical data and trend data and provide forecast data;

[0100] The abnormality alarm module detects abnormalities and issues abnormality alarms and early warnings by monitoring the mutation data, trend data and forecast data in real time;

[0101] Furthermore, the temperature and humidity data are collected periodically by the temperature and humidity sensor, and the data is pre-processed to obtain the processed data. The operation process includes: the temperature and humidity sensor is installed at the key parts of the bus duct, which include the connection end, the closed conductor, and the place where water easily accumulates in the humid environment. The sensor collects the temperature and humidity data X(t) in a periodic T sampling manner. At the time t, the temperature and humidity data are X(t) respectively. W (t) and X S (t), expression formula: where Q W (t),Q S (t) is the real temperature and humidity signal, E W (t),E S (t) is random noise, which is set to have a mean of 0 and a variance of σ 2 Gaussian white noise, expression formula: E W (t),E S (t)~ζ(0,σ 2 ), the data preprocessing includes using Kalman filtering to denoise the temperature and humidity data, and the denoised temperature and humidity data are respectively and Use Grubbs test to remove outliers in temperature and humidity data, and define temperature data X W The mean and standard deviation of (t) are expressed as: in Represents the average value of temperature data in a time window, N represents the total number of data points in the statistical time window, represents the denoised temperature value at the i-th time point, represents the sum of all data points, Indicates the degree of fluctuation of temperature data, represents the difference between the i-th temperature value and the average temperature, is the square of the temperature deviation from the mean, is the sum of the squared deviations of all data points, is an unbiased estimate used in computing the mean square error, The result is squared and the Grubbs statistic is calculated using the formula: The threshold value of Grubbs test is: where t β / (2N),N-2 is the critical value of the t distribution with N-2 degrees of freedom, when R>R ctil The data point is determined to be an outlier and is removed. The method for removing outliers in humidity data is the same;

[0102] Furthermore, mutation detection and trend analysis are performed on the processed data to obtain mutation data and trend data respectively. The mutation detection includes calculating the temperature and humidity change trend within the last 1 second to 3 seconds using short-term window processing. At time t, temperature and humidity data within O (1 second to 3 seconds) are selected to calculate the change rate of the short-term window. The expression formula is: Among them U W (t) is the temperature change rate, U S (t) is the humidity change rate, O is the short-term window (1 second to 3 seconds), ψ i is the short-time window weight coefficient. The dynamic threshold method is used to set the adaptive mutation threshold using the historical mean and standard deviation. The discrete wavelet transform is used to perform multi-scale analysis on the temperature and humidity signals, extract the high-frequency components, and identify the mutation points. The expression formula is: where Y W (a,b),Y S (a,b) are the wavelet transform coefficients, a is the scale factor (used to separate high and low frequencies), b is the translation factor, and χ * (·) is the mother wavelet (Daubechies wavelet).

[0103] The trend analysis includes using long-term window processing to calculate the temperature and humidity change trend in the last 10 seconds to 60 seconds, selecting data within the past A (10 seconds to 60 seconds), and calculating the weighted sliding average to smooth the trend. The expression formula is: Where W ted (t) is the temperature trend value, S ted (t) is the humidity trend value, A is the trend analysis window length (within the past 10 seconds to 60 seconds), δ i is the long-term window weight coefficient, satisfying ∑δ i =1, weight coefficient expression formula: Where ε is the speed at which the weight decays. is an exponential decay weight, is the normalization factor;

[0104] Furthermore, the operation process of adaptively adjusting the weighting strategy according to the mutation data and trend data includes: if the weighting strategy includes any one of the two items of temperature sudden increase ≥5°C and humidity sudden increase ≥10%RH in the mutation data, the short-term window weight is increased, and the mutation response factor D is defined. jmp (t), expression formula: Where ΔW=W(t)-W(tO) is the short-term temperature change, ΔS=S(t)-S(tO) is the short-term humidity change, jmp (t) ≥ 1, the mutation is triggered and the short-term window weight is updated. The expression formula is: where φ w The short-term window adjustment rate is 0.1-0.3, ψ i (t) is the short-term window weight at time t, ψ i (t-1) is the short-term window weight of the previous moment, (1-ψ i (t-1)) is the complement of the short-time window, ζ i (t)=1-ψ i (t) is the long-term window weight used for long-term trend analysis. When the short-term window weight increases, the long-term window weight decreases accordingly, and vice versa.

[0105] The weighting strategy includes the following two items: if the temperature continues to rise ≥3℃ / min or the humidity continues to rise ≥5%RH / min in the trend data, it is determined that there is a trend anomaly, the short-term window weight is reduced, the trend data analysis is strengthened, and the trend anomaly factor D is defined. ted (t), expression formula: in is the long-term temperature change rate, is the long-term humidity change rate, in D ted (t) ≥ 1, the trend anomaly is triggered, and the long-term window weight is adjusted. The expression formula is:

[0106] where φ s The trend adjustment coefficient is 0.05-0.2, ζ i (t) is the long-term window weight at the current time t, ζ i (t-1) is the long-term window weight of the previous moment, D ted (t) is the trend detection factor, (1-ζ i (t-1)) is the adjustment factor, and the weight is updated using the exponential smoothing method, expressed as: in Use a value of 0.2-0.5 for smooth changes;

[0107] Furthermore, the time series forecasting model is used to analyze historical data and trend data and provide forecast data. The operation process includes: collecting temperature and humidity data X(t) and performing time series modeling for short-term forecasting. The expression formula is:

[0108] Where X(t) is the temperature and humidity data at the current time t, c represents any one of the average level and long-term trend of the temperature and humidity data, and ∈(t) represents the unpredictable noise at the current time point. represents how past prediction errors affect the current temperature and humidity prediction, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. G represents how many past prediction errors are used for prediction, and η k It measures the impact of past prediction errors on future predictions and uses long-term short-term memory networks for long-term predictions. The output hidden state is used to predict the temperature and humidity data in the next second. The expression formula is:

[0109] in is the temperature and humidity data at the predicted time t+n, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. X(t+ni) represents the temperature and humidity value at time (t+ni), G represents the number of past prediction errors used for prediction, and η k It is a measure of the impact of past forecast errors on future forecasts. Represents the error between the predicted value and the true value at a past time point.

[0110] The weighted fusion method is used to combine the temperature and humidity data predicted by the long short-term memory network and the time series prediction model, and the expression formula is:

[0111] where ι is the weight factor, Represents the final prediction result at time t+n, is the prediction result of the time series prediction model at time t+n. It is the prediction result of the long short-term memory network at time t+n, which depends on the error between the long short-term memory network and the time series prediction model. The expression formula is: Where ι is the weight factor, h is the long short-term memory network, κ is the time series prediction model, e is the base of the natural logarithm, and MSE is the mean square error;

[0112] Furthermore, by real-time monitoring of the mutation data, trend data and prediction data, anomaly detection and anomaly alarm and early warning operation process includes: the anomaly detection is based on the triple threshold judgment method, and the anomaly score of each data source is calculated, which is expressed as follows: λ(t) = μ1·λ v (t)+μ2·λ θ (t)+μ3·λ κ (t), where μ1, μ2, and μ3 are weighted coefficients satisfying μ1+μ2+μ3=1, which are used to adjust the contribution of different data sources, and λ v (t) reflects the degree of sudden change in temperature and humidity, λ θ (t) reflects the long-term trend anomaly of temperature and humidity, λ κ (t) reflects the abnormal degree of predicted temperature and humidity. The degree of mutation of temperature and humidity can be calculated by the rate of change within a short time window, expressed as: in is the temperature and humidity change rate within the short-term window, The standard deviation of historical temperature and humidity changes is used for normalization. The abnormality of trend data is calculated by the temperature and humidity change rate in the past A (10 seconds to 60 seconds), expressed as: where λ θ (t) is the overall change trend of the temperature and humidity data at time t, A is the long-term window size, is the standard deviation of the trend data, The trend change at time ti represents the rate of change of temperature and humidity data. The predicted anomaly score measures the deviation between the predicted temperature and humidity and the current data. The expression formula is: where λ κ (t) is the predicted anomaly score that measures the degree of deviation between the predicted value of temperature and humidity and the current measured value. is the standard deviation of the predicted data used to normalize the predicted scores, is the predicted temperature and humidity value at time t+n′, is the actual measured value of temperature and humidity at the current time t, is the absolute error between the predicted value and the current actual measured value. According to the comprehensive anomaly score λ(t), the adaptive alarm threshold Θ is set. bj (t), expression formula: where Θ bj (t) is the adaptive alarm threshold, is the mean of the historical anomaly scores, is the standard deviation of historical anomaly scores, and r′ is the adaptive coefficient that determines the alarm sensitivity. The alarm sensitivity includes level one warning, level two warning and level three warning.

[0113] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for monitoring abnormal temperature and humidity of a bus duct, characterized in that: The following steps are involved: S1. Periodically collect temperature and humidity data through temperature and humidity sensors, and perform data preprocessing to obtain processed data; S2. performing mutation detection and trend analysis on the processed data to obtain mutation data and trend data, respectively; S3, adaptively adjust the weighting strategy based on mutation data and trend data; S4. Use time series forecasting models to analyze historical data and trend data and provide forecast data; S5. Perform anomaly detection and abnormality alarm and early warning by real-time monitoring of the mutation data, trend data and forecast data.

2. A bus duct temperature and humidity abnormality monitoring method according to claim 1, characterized in that: In step S1, the temperature and humidity data are periodically collected by the temperature and humidity sensor, and the data is preprocessed to obtain the processed data as follows: The temperature and humidity sensors are installed at key locations of the bus duct, including the connection end, closed conductor, and places where water easily accumulates in a humid environment. The sensors collect temperature and humidity data X(t) in a periodic T sampling manner. At time t, the temperature and humidity data are X(t) respectively. W (t) and X S (t), expression formula: where Q W (t),Q S (t) is the real temperature and humidity signal, E W (t),E S (t) is random noise, which is set to have a mean of 0 and a variance of σ 2 Gaussian white noise, expression formula: E W (t),E S (t)~ζ(0,σ 2 ), the data preprocessing includes using Kalman filtering to denoise the temperature and humidity data, and the denoised temperature and humidity data are respectively and Use Grubbs test to remove outliers in temperature and humidity data, and define temperature data X W The mean and standard deviation of (t) are expressed as: in Represents the average value of temperature data in a time window, N represents the total number of data points in the statistical time window, represents the denoised temperature value at the i-th time point, represents the sum of all data points, Indicates the degree of fluctuation of temperature data, represents the difference between the i-th temperature value and the average temperature, is the square of the temperature deviation from the mean, is the sum of the squared deviations of all data points, is an unbiased estimate used in computing the mean square error, The result is squared and the Grubbs statistic is calculated using the formula: The threshold value of Grubbs test is: where t β / (2N),N-2 is the critical value of the t distribution with N-2 degrees of freedom, when R>R ctil The data point is determined to be an outlier and is removed. The method for removing outliers in humidity data is the same.

3. The method for monitoring abnormal temperature and humidity of a bus duct according to claim 2, characterized in that: In step S2, mutation detection and trend analysis are performed based on the processed data to obtain mutation data and trend data respectively as follows: The mutation detection includes using short-term window processing to calculate the temperature and humidity change trend within the last 1 second to 3 seconds. At time t, temperature and humidity data within O (1 second to 3 seconds) are selected and the change rate of the short-term window is calculated. The expression formula is: Among them U W (t) is the temperature change rate, U S (t) is the humidity change rate, O is the short-term window (1 second to 3 seconds), ψ i is the short-time window weight coefficient. The dynamic threshold method is used to set the adaptive mutation threshold using the historical mean and standard deviation. The discrete wavelet transform is used to perform multi-scale analysis on the temperature and humidity signals, extract the high-frequency components, and identify the mutation points. The expression formula is: where Y W (a,b),Y S (a,b) are the wavelet transform coefficients, a is the scale factor (used to separate high and low frequencies), b is the translation factor, and χ * (·) is the mother wavelet (Daubechies wavelet).

4. A bus duct temperature and humidity abnormality monitoring method according to claim 3, characterized in that: In step S2, mutation detection and trend analysis are performed based on the processed data to obtain mutation data and trend data respectively as follows: The trend analysis includes using long-term window processing to calculate the temperature and humidity change trend in the last 10 seconds to 60 seconds, selecting data within the past A (10 seconds to 60 seconds), and calculating the weighted sliding average to smooth the trend. The expression formula is: Where W ted (t) is the temperature trend value, S ted (t) is the humidity trend value, A is the trend analysis window length (within the past 10 seconds to 60 seconds), δ i is the long-term window weight coefficient, satisfying ∑δ i =1, weight coefficient expression formula: Where ε is the speed of weight decay, l -εi is an exponential decay weight, is the normalization factor.

5. The method for monitoring abnormal temperature and humidity of a bus duct according to claim 4, characterized in that: In step S3, the method for adaptively adjusting the weighting strategy according to the mutation data and trend data is as follows: The weighting strategy includes the sudden increase in temperature ≥5°C and the sudden increase in humidity ≥ 10%RH Any of the two items will increase the short-term window weight and define the mutation response factor D jmp (t), expression formula: Where ΔW=W(t)-W(tO) is the short-term temperature change, ΔS=S(t)-S(tO) is the short-term humidity change, jmp (t) ≥ 1, the mutation is triggered and the short-term window weight is updated. The expression formula is: where φ w The short-term window adjustment rate is 0.1-0.3, ψ i (t) is the short-term window weight at time t, ψ i (t-1) is the short-term window weight of the previous moment, (1-ψ i (t-1)) is the complement of the short-time window, ζ i (t)=1-ψ i (t) is the long-term window weight used for long-term trend analysis. When the short-term window weight increases, the long-term window weight decreases accordingly, and vice versa.

6. A bus duct temperature and humidity abnormality monitoring method according to claim 5, characterized in that: In step S3, the method for adaptively adjusting the weighting strategy according to the mutation data and trend data is as follows: The weighting strategy includes the following two items: if the temperature continues to rise ≥3℃ / min or the humidity continues to rise ≥5%RH / min in the trend data, it is determined that there is a trend anomaly, the short-term window weight is reduced, the trend data analysis is strengthened, and the trend anomaly factor D is defined. ted (t), expression formula: in is the long-term temperature change rate, is the long-term humidity change rate, in D ted (t) ≥ 1, the trend anomaly is triggered, and the long-term window weight is adjusted. The expression formula is: where φ s The trend adjustment coefficient is 0.05-0.2, ζ i (t) is the long-term window weight at the current time t, ζ i (t-1) is the long-term window weight of the previous moment, D ted (t) is the trend detection factor, (1-ζ i (t-1)) is the adjustment factor, and the weight is updated using the exponential smoothing method, expressed as: in Use a value of 0.2-0.5 for smooth changes.

7. A bus duct temperature and humidity abnormality monitoring method according to claim 6, characterized in that: In step S4, a method for using a time series forecasting model to analyze historical data and trend data and provide forecast data is as follows: Collect temperature and humidity data X(t) and perform time series modeling for short-term prediction. The expression formula is: Where X(t) is the temperature and humidity data at the current time t, c represents any one of the average level and long-term trend of the temperature and humidity data, and ∈(t) represents the unpredictable noise at the current time point. represents how past prediction errors affect the current temperature and humidity prediction, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. G represents how many past prediction errors are used for prediction, and η k It measures the impact of past prediction errors on future predictions and uses long-term short-term memory networks for long-term predictions. The output hidden state is used to predict the temperature and humidity data in the next second. The expression formula is: in is the temperature and humidity data at the predicted time t+n, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. X(t+ni) represents the temperature and humidity value at time (t+ni), G represents the number of past prediction errors used for prediction, and η k It is a measure of the impact of past forecast errors on future forecasts. Represents the error between the predicted value and the true value at a past time point.

8. The method for monitoring abnormal temperature and humidity of a bus duct according to claim 7, characterized in that: In step S4, a method for analyzing historical data and trend data using a time series forecasting model and providing forecast data is as follows: The weighted fusion method is used to combine the temperature and humidity data predicted by the long short-term memory network and the time series prediction model, and the expression formula is: where ι is the weight factor, Represents the final prediction result at time t+n, is the prediction result of the time series prediction model at time t+n. It is the prediction result of the long short-term memory network at time t+n, which depends on the error between the long short-term memory network and the time series prediction model. The expression formula is: Where ι is the weight factor, h is the long short-term memory network, κ is the time series prediction model, e is the base of the natural logarithm, and MSE is the mean squared error.

9. The method for monitoring abnormal temperature and humidity of a bus duct according to claim 8, characterized in that: In step S5, the method of performing anomaly detection and abnormality alarm and early warning by real-time monitoring of the mutation data, trend data and forecast data is as follows: The anomaly detection is based on the triple threshold judgment method to calculate the anomaly score of each data source, expressed as follows: λ(t) = μ1·λ v (t)+μ2·λ θ (t)+μ3·λ κ (t), where μ1, μ2, and μ3 are weighted coefficients satisfying μ1+μ2+μ3=1, which are used to adjust the contribution of different data sources, and λ v (t) reflects the degree of sudden change in temperature and humidity, λ θ (t) reflects the long-term trend anomaly of temperature and humidity, λ κ (t) reflects the abnormal degree of predicted temperature and humidity. The degree of mutation of temperature and humidity can be calculated by the rate of change within a short time window, expressed as: in is the temperature and humidity change rate within the short-term window, The standard deviation of historical temperature and humidity changes is used for normalization. The abnormality of trend data is calculated by the temperature and humidity change rate in the past A (10 seconds to 60 seconds), expressed as: where λ θ (t) is the overall change trend of the temperature and humidity data at time t, A is the long-term window size, is the standard deviation of the trend data, The trend change at time ti represents the rate of change of temperature and humidity data. The predicted anomaly score measures the deviation between the predicted temperature and humidity and the current data. The expression formula is: where λ κ (t) is the predicted anomaly score that measures the degree of deviation between the predicted value of temperature and humidity and the current measured value. is the standard deviation of the predicted data used to normalize the predicted scores, is the predicted temperature and humidity value at time t+n′, is the actual measured value of temperature and humidity at the current time t, is the absolute error between the predicted value and the current actual measured value. According to the comprehensive anomaly score λ(t), the adaptive alarm threshold Θ is set. bj (t), expression formula: where Θ bj (t) is the adaptive alarm threshold, is the mean of the historical anomaly scores, is the standard deviation of historical anomaly scores, and r′ is the adaptive coefficient that determines the alarm sensitivity. The alarm sensitivity includes level one warning, level two warning and level three warning.

10. A bus duct temperature and humidity anomaly monitoring system, based on a bus duct temperature and humidity anomaly monitoring method according to any one of claims 1 to 9, characterized in that: include: The data acquisition module periodically collects temperature and humidity data through temperature and humidity sensors, and performs data preprocessing to obtain processed data; a detection and analysis module, performing mutation detection and trend analysis based on the processed data to obtain mutation data and trend data respectively; The adjustment strategy module adaptively adjusts the weighting strategy based on mutation data and trend data; The time series forecasting module uses a time series forecasting model to analyze historical data and trend data and provide forecast data; The abnormality alarm module detects abnormalities and issues abnormality alarms and early warnings by monitoring the mutation data, trend data and forecast data in real time; Furthermore, the temperature and humidity data are collected periodically by the temperature and humidity sensor, and the data is pre-processed to obtain the processed data. The operation process includes: the temperature and humidity sensor is installed at the key parts of the bus duct, which include the connection end, the closed conductor, and the place where water easily accumulates in the humid environment. The sensor collects the temperature and humidity data X(t) in a periodic T sampling manner. At the time t, the temperature and humidity data are X(t) respectively. W (t) and X S (t), expression formula: where Q W (t),Q S (t) is the real temperature and humidity signal, E W (t),E S (t) is random noise, which is set to have a mean of 0 and a variance of σ 2 Gaussian white noise, expression formula: E W (t),E S (t)~ζ(0,σ 2 ), the data preprocessing includes using Kalman filtering to denoise the temperature and humidity data, and the denoised temperature and humidity data are respectively and Use Grubbs test to remove outliers in temperature and humidity data, and define temperature data X W The mean and standard deviation of (t) are expressed as: in Represents the average value of temperature data in a time window, N represents the total number of data points in the statistical time window, represents the denoised temperature value at the i-th time point, represents the sum of all data points, Indicates the degree of fluctuation of temperature data, represents the difference between the i-th temperature value and the average temperature, is the square of the temperature deviation from the mean, is the sum of the squared deviations of all data points, is an unbiased estimate used in computing the mean square error, The result is squared and the Grubbs statistic is calculated using the formula: The threshold value of Grubbs test is: where t β / (2N),N-2 is the critical value of the t distribution with N-2 degrees of freedom, when R>R ctil The data point is determined to be an outlier and is removed. The method for removing outliers in humidity data is the same; Furthermore, mutation detection and trend analysis are performed on the processed data to obtain mutation data and trend data respectively. The mutation detection includes calculating the temperature and humidity change trend within the last 1 second to 3 seconds using short-term window processing. At time t, temperature and humidity data within O (1 second to 3 seconds) are selected to calculate the change rate of the short-term window. The expression formula is: Among them U W (t) is the temperature change rate, U S (t) is the humidity change rate, O is the short-term window (1 second to 3 seconds), ψ i is the short-time window weight coefficient. The dynamic threshold method is used to set the adaptive mutation threshold using the historical mean and standard deviation. The discrete wavelet transform is used to perform multi-scale analysis on the temperature and humidity signals, extract the high-frequency components, and identify the mutation points. The expression formula is: where Y W (a,b),Y S (a,b) are the wavelet transform coefficients, a is the scale factor (used to separate high and low frequencies), b is the translation factor, and χ * (·) is the mother wavelet (Daubechies wavelet). The trend analysis includes using long-term window processing to calculate the temperature and humidity change trend in the last 10 seconds to 60 seconds, selecting data within the past A (10 seconds to 60 seconds), and calculating the weighted sliding average to smooth the trend. The expression formula is: Where W ted (t) is the temperature trend value, S ted (t) is the humidity trend value, A is the trend analysis window length (within the past 10 seconds to 60 seconds), δ i is the long-term window weight coefficient, satisfying ∑δ i =1, weight coefficient expression formula: Where ε is the speed at which the weight decays, e -εi is an exponential decay weight, is the normalization factor; Furthermore, the operation process of adaptively adjusting the weighting strategy according to the mutation data and trend data includes: if the weighting strategy includes any one of the two items of temperature sudden increase ≥5°C and humidity sudden increase ≥10%RH in the mutation data, the short-term window weight is increased, and the mutation response factor D is defined. jmp (t), expression formula: Where ΔW=W(t)-W(tO) is the short-term temperature change, ΔS=S(t)-S(tO) is the short-term humidity change, jmp (t) ≥ 1, the mutation is triggered and the short-term window weight is updated. The expression formula is: where φ w The short-term window adjustment rate is 0.1-0.3, ψ i (t) is the short-term window weight at time t, ψ i (t-1) is the short-term window weight of the previous moment, (1-ψ i (t-1)) is the complement of the short-time window, ζ i (t)=1-ψ i (t) is the long-term window weight used for long-term trend analysis. When the short-term window weight increases, the long-term window weight decreases accordingly, and vice versa. The weighting strategy includes the following two items: if the temperature continues to rise ≥3℃ / min or the humidity continues to rise ≥5%RH / min in the trend data, it is determined that there is a trend anomaly, the short-term window weight is reduced, the trend data analysis is strengthened, and the trend anomaly factor D is defined. ted (t), expression formula: in is the long-term temperature change rate, is the long-term humidity change rate, in D ted (t) ≥ 1, the trend anomaly is triggered, and the long-term window weight is adjusted. The expression formula is: where φ s The trend adjustment coefficient is 0.05-0.2, ζ i (t) is the long-term window weight at the current time t, ζ i (t-1) is the long-term window weight of the previous moment, D ted (t) is the trend detection factor, (1-ζ i (t-1)) is the adjustment factor, and the weight is updated using the exponential smoothing method, expressed as: in Use a value of 0.2-0.5 for smooth changes; Furthermore, the time series forecasting model is used to analyze historical data and trend data and provide forecast data. The operation process includes: collecting temperature and humidity data X(t) and performing time series modeling for short-term forecasting. The expression formula is: Where X(t) is the temperature and humidity data at the current time t, c represents any one of the average level and long-term trend of the temperature and humidity data, and ∈(t) represents the unpredictable noise at the current time point. represents how past prediction errors affect the current temperature and humidity prediction, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. G represents how many past prediction errors are used for prediction, and η k It measures the impact of past prediction errors on future predictions and uses long-term short-term memory networks for long-term predictions. The output hidden state is used to predict the temperature and humidity data in the next second. The expression formula is: in is the temperature and humidity data at the predicted time t+n, F represents how many past temperature and humidity data are used for prediction, and γ i It measures the impact of past temperature and humidity data on future predictions. X(t+ni) represents the temperature and humidity value at time (t+ni), G represents the number of past prediction errors used for prediction, and η k It is a measure of the impact of past forecast errors on future forecasts. Represents the error between the predicted value and the true value at a past time point. The weighted fusion method is used to combine the temperature and humidity data predicted by the long short-term memory network and the time series prediction model, and the expression formula is: where ι is the weight factor, Represents the final prediction result at time t+n, is the prediction result of the time series prediction model at time t+n. It is the prediction result of the long short-term memory network at time t+n, which depends on the error between the long short-term memory network and the time series prediction model. The expression formula is: Where ι is the weight factor, h is the long short-term memory network, κ is the time series prediction model, e is the base of the natural logarithm, and MSE is the mean square error; Furthermore, the operation process of performing anomaly detection and abnormal alarm and early warning by real-time monitoring of the mutation data, trend data and prediction data includes: the anomaly detection is based on a triple threshold judgment method to calculate the anomaly score of each data source, expressed as follows: λ(t)=μ1·λ v (t)+μ2·λ θ (t)+μ3·λ κ (t), where μ1, μ2, and μ3 are weighted coefficients satisfying μ1+μ2+μ3=1, which are used to adjust the contribution of different data sources, and λ v (t) reflects the degree of sudden change in temperature and humidity, λ θ (t) reflects the long-term trend anomaly of temperature and humidity, λ κ (t) reflects the abnormal degree of predicted temperature and humidity. The degree of mutation of temperature and humidity can be calculated by the rate of change within a short time window, expressed as: in is the temperature and humidity change rate within the short-term window, The standard deviation of historical temperature and humidity changes is used for normalization. The abnormality of trend data is calculated by the temperature and humidity change rate in the past A (10 seconds to 60 seconds), expressed as: where λ θ (t) is the overall change trend of the temperature and humidity data at time t, A is the long-term window size, is the standard deviation of the trend data, The trend change at time ti represents the rate of change of temperature and humidity data. The predicted anomaly score measures the deviation between the predicted temperature and humidity and the current data. The expression formula is: where λ κ (t) is the predicted anomaly score that measures the degree of deviation between the predicted value of temperature and humidity and the current measured value. is the standard deviation of the predicted data used to normalize the predicted scores, is the predicted temperature and humidity value at time t+n′, is the actual measured value of temperature and humidity at the current time t, is the absolute error between the predicted value and the current actual measured value. According to the comprehensive anomaly score λ(t), the adaptive alarm threshold Θ is set. bj (t), expression formula: where Θ bj (t) is the adaptive alarm threshold, is the mean of the historical anomaly scores, θ is the standard deviation of the historical anomaly scores, and r′ is the adaptive coefficient that determines the alarm sensitivity The alarm sensitivity includes level one warning, level two warning and level three warning.

Images