A data collection and processing method based on SF6 pressure gauge in substation

Abstract

The present invention discloses a data collection and processing method based on the SF6 pressure gauge of a substation, which relates to the field of data processing technology, including collecting substation environmental data in real time through SF6 pressure sensors, temperature sensors and humidity sensors; inputting the substation environmental data into a multi-sensor fusion algorithm, performing temperature compensation and humidity correction, and generating a corrected multidimensional data set; inputting the extracted key features into an artificial intelligence model for fault prediction, and outputting a fault warning signal; triggering an alarm mechanism and notifying the operation and maintenance personnel according to the fault warning signal, and feeding back the actual fault processing results to a central processing platform. The present invention effectively solves the problem in the prior art that SF6 pressure data is easily affected by ambient temperature and humidity through multi-sensor fusion technology and an adaptive filtering algorithm, and significantly improves the accuracy of pressure monitoring.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, in particular to a data collection and processing method based on a SF6 pressure gauge in a transformer substation. Background Art

[0002] With the widespread application of power systems, substations, as the core hub of power transmission and distribution, have a direct impact on the overall stability of the power system due to their operating status. Among them, SF6 (sulfur hexafluoride) gas in high-voltage equipment is widely used in circuit breakers, disconnectors and other equipment in substations due to its excellent insulation and arc extinguishing capabilities. However, the pressure change of SF6 gas directly affects the working state of the equipment, especially SF6 leakage may cause the insulation performance of the equipment to deteriorate, and in severe cases may also cause electrical accidents. Therefore, how to monitor the pressure state of SF6 gas in real time and accurately is an important issue to ensure the safe operation of substations.

[0003] Existing SF6 pressure monitoring technologies mainly rely on mechanical pressure gauges or single sensor systems. These methods cannot effectively cope with complex environmental conditions, especially the impact of temperature and humidity on pressure measurement results. Since the pressure of SF6 gas is very sensitive to changes in temperature and humidity, it is difficult to provide accurate monitoring data without temperature compensation and humidity correction solutions. At the same time, existing technologies have limited data processing capabilities and usually only perform simple data collection and storage. There is a lack of in-depth analysis of the data and noise suppression, resulting in low accuracy of the measured data. In addition, traditional monitoring systems mostly rely on manual inspections, with low levels of automation and intelligence, and cannot provide timely warnings of the operating status of the equipment, increasing the risk of equipment operation and maintenance. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a data collection and processing method based on the SF6 pressure gauge of the substation to solve the problem that the SF6 pressure data in the traditional monitoring system is easily affected by the ambient temperature and humidity, resulting in insufficient monitoring accuracy.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for collecting and processing data based on a substation SF6 pressure gauge, which includes collecting substation environmental data in real time through an SF6 pressure sensor, a temperature sensor, and a humidity sensor;

[0008] Input substation environmental data into the multi-sensor fusion algorithm to perform temperature compensation and humidity correction to generate a corrected multidimensional data set;

[0009] Use adaptive filters to suppress noise and smooth data on the corrected multidimensional data set to generate high-precision pressure monitoring data;

[0010] The high-precision pressure monitoring data is wirelessly transmitted to the central processing platform through the Internet of Things for storage, and data cleaning and key feature extraction are performed;

[0011] The extracted key features are input into the artificial intelligence model for fault prediction and fault warning signals are output;

[0012] The alarm mechanism is triggered according to the fault warning signal and the operation and maintenance personnel are notified. At the same time, the actual fault handling results are fed back to the central processing platform.

[0013] As a preferred solution of the data collection and processing method based on the substation SF6 pressure gauge of the present invention, the real-time collection of substation environmental data comprises the following specific steps:

[0014] Install the SF6 pressure sensor near the SF6 gas pressure vessel;

[0015] The temperature sensors are installed on the equipment surface of the substation, and the humidity sensors are arranged in the air circulation area;

[0016] Use a pressure calibration pump to apply a standard pressure value to the sensor, and correct the sensor output value and the pressure value of the calibration pump through linear calibration;

[0017] Insert the temperature sensor into the temperature calibration furnace, record the output resistance value of the sensor, and compare it with the temperature resistance table to make calibration corrections;

[0018] Use a humidity calibration chamber, place the humidity sensor in the calibration chamber, record the sensor output, and compare it with the reference humidity value of the calibration chamber to make calibration corrections;

[0019] Set the sampling frequency of the data acquisition equipment, synchronize with the external time server through the network time protocol, and output the substation environment data in real time.

[0020] As a preferred solution of the data collection and processing method based on the substation SF6 pressure gauge of the present invention, the specific steps of generating the corrected multidimensional data set are as follows:

[0021] Adaptive Kalman filter is used to denoise the substation environmental data collected by sensors;

[0022] Based on the nonlinear effect of temperature on pressure and humidity, temperature compensation is performed through an adaptive nonlinear model, and the expression is:

[0023] ;

[0024] in, Indicates the pressure value after temperature compensation. Indicates the real-time pressure value. represents the temperature compensation coefficient, Indicates the real-time temperature value. Indicates the reference temperature value, Indicates the absolute difference between the real-time temperature value and the reference temperature value. represents the coefficient of exponential decay, represents the time constant;

[0025] Based on the pressure value after temperature compensation, a nonlinear correction model based on humidity is constructed through the nonlinear influence of humidity on pressure and temperature values. The expression is:

[0026] ;

[0027] in, Indicates the pressure value after humidity correction. Indicates the real-time humidity value. Indicates the linear coefficient of real-time humidity value correction, Indicates the nonlinear coefficient of real-time humidity value correction;

[0028] The pressure value after nonlinear temperature compensation and humidity correction is integrated with the real-time temperature value and humidity value to form a corrected multidimensional data set.

[0029] As a preferred solution of the data collection and processing method based on the substation SF6 pressure gauge of the present invention, the specific steps of generating high-precision pressure monitoring data are as follows:

[0030] The correlation between temperature, humidity and pressure in the multidimensional data set was analyzed by Pearson correlation coefficient;

[0031] According to the analysis results of the correlation between temperature, humidity and pressure, the state transfer equation of the Kalman filter is established;

[0032] The Kalman filter uses the state transfer equation and the current state estimate, combined with real-time temperature, humidity and pressure measurements, to predict the state and its covariance at the next moment;

[0033] At each time step, the measurement error is calculated by comparing the Kalman filter prediction with the actual measurement;

[0034] The real-time temperature value and real-time humidity values Introduced as observation variables into the state space model of the Kalman filter to monitor the temperature and humidity in each time step;

[0035] Set the threshold of temperature and humidity change rate ;

[0036] When the real-time temperature value Or real-time humidity value , exceeds the set threshold When , the filter will increase the smoothing strength of the pressure value;

[0037] When the real-time temperature value Or real-time humidity value , below the threshold When , the filter will reduce the smoothing strength;

[0038] By combining the measurement error and the rate of change of temperature and humidity at each time step, the parameters of the Kalman filter are adjusted in real time;

[0039] Apply the extended and adjusted Kalman filter to dynamically calculate the Kalman gain and optimize the Kalman filter based on the measurement error and covariance at each time step;

[0040] Integrate real-time pressure values ​​after Kalman filter optimization , Real-time temperature value and real-time humidity values , generating high-precision pressure monitoring data.

[0041] As a preferred solution of the data collection and processing method based on the substation SF6 pressure gauge of the present invention, the data cleaning and key feature extraction are specifically performed as follows:

[0042] Pack high-precision pressure monitoring data into standardized data packets and encrypt the data using the AES-256 symmetric encryption algorithm;

[0043] A low-power micro-computing platform is deployed near the substation to receive encrypted sensor data packets and store them in a time series database;

[0044] Verify the integrity of the data packet in the check sequence database;

[0045] The interquartile range method was used to detect and eliminate outliers, and the Z-score method was used to further detect outliers;

[0046] Linear interpolation is used to handle missing values ​​in sensor data;

[0047] Using the Bayesian fusion algorithm, data from different sensors are fused to build a global perception state;

[0048] Based on the global perception state, the hidden features of multimodal data are automatically extracted through the autoencoder;

[0049] A reinforcement learning algorithm is introduced to adjust the frequency and content of data transmission according to the changes in hidden features.

[0050] As a preferred solution of the data collection and processing method based on the SF6 pressure gauge of the substation described in the present invention, wherein: the extracted key features are input into the artificial intelligence model for fault prediction, and a fault warning signal is output. The specific steps are as follows:

[0051] The hidden features of the extracted multimodal data are normalized by Min-Max normalization to scale the data to the same numerical range;

[0052] The normalized hidden features are organized into a three-dimensional tensor according to the time series structure and input into the time series convolutional neural network for fault prediction;

[0053] In the time series convolutional neural network, the fault risk model is introduced to perform fault risk scoring, and the expression is:

[0054] ;

[0055] in, Indicates time Failure risk score at the moment, Indicates The weight of the feature, Indicates Features The nonlinear mapping function is is the total number of feature variables, An index variable representing a feature;

[0056] Failure Risk Score Will accumulate over time to set a failure risk threshold ;

[0057] when When , a fault warning signal is output, indicating that a fault is about to occur;

[0058] when When , it is in normal state and no warning is needed;

[0059] For the training of the fault risk model, the cross entropy loss function will be used, and the Adam optimizer will be used for parameter update.

[0060] As a preferred solution of the data collection and processing method based on the SF6 pressure gauge of the substation described in the present invention, the training of the fault risk model will adopt the cross entropy loss function and use the Adam optimizer to update the parameters. The specific steps are as follows:

[0061] The training loss function expression of the fault risk model is:

[0062] ;

[0063] in, represents the length of the time series, represents the value of the loss function, represents the real fault label, Represents the predicted value of the convolutional neural network model;

[0064] Input historical sensor data and corresponding fault labels into the fault risk model and divide it into training set, validation set and test set;

[0065] Use the validation set to tune the hyperparameters of the failure risk model. After each epoch, use the validation set to evaluate the accuracy, AUC, and F1 score of the model.

[0066] Select the optimal hyperparameters based on the performance of the validation set;

[0067] After training and tuning are completed, the final performance of the failure risk model is evaluated using the accuracy, AUC, and F1 score metrics in the test set.

[0068] As a preferred solution of the data collection and processing method based on the SF6 pressure gauge of the substation described in the present invention, wherein: the alarm mechanism is triggered according to the fault warning signal and the operation and maintenance personnel are notified, and the actual fault processing results are fed back to the central processing platform. The specific steps are:

[0069] After the fault warning signal is triggered, a fault warning report is generated;

[0070] Push fault warning reports to the monitoring terminals of operation and maintenance personnel through the IoT platform;

[0071] After receiving the fault warning signal, the operation and maintenance personnel will go to the site for investigation based on the key characteristic data in the warning report;

[0072] After troubleshooting, the operation and maintenance personnel record the actual troubleshooting results and enter the troubleshooting results through the operation and maintenance terminal;

[0073] The fault handling results are uploaded to the central processing platform in real time through the Internet of Things platform. The central processing platform stores the newly collected fault handling data in the database as label data for subsequent model training to iteratively optimize the artificial intelligence model.

[0074] In a second aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for collecting and processing data based on the substation SF6 pressure gauge as described in the first aspect of the present invention is implemented.

[0075] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the method for collecting and processing data based on the substation SF6 pressure gauge as described in the first aspect of the present invention is implemented.

[0076] The beneficial effects of the present invention are as follows: the present invention effectively solves the problem in the prior art that SF6 pressure data is easily affected by ambient temperature and humidity through multi-sensor fusion technology and adaptive filtering algorithm, and significantly improves the accuracy of pressure monitoring. At the same time, through real-time compensation and correction of temperature and humidity, the present invention can maintain high-precision monitoring of SF6 pressure data under complex environmental conditions; in addition, the remote transmission of data is realized by combining the Internet of Things technology, and fault prediction and early warning are performed through artificial intelligence models, which greatly improves the intelligent operation and maintenance level of substation equipment and reduces the occurrence rate of equipment failures and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0078] Figure 1 This is a flow chart of the data collection and processing method based on the substation SF6 pressure gauge in Example 1.

[0079] Figure 2 This is a flowchart for iteratively optimizing the artificial intelligence model in Example 1. DETAILED DESCRIPTION

[0080] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0081] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0082] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0083] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a method for collecting and processing data based on a substation SF6 pressure gauge, comprising the following steps:

[0084] S1. Collect substation environmental data in real time through SF6 pressure sensor, temperature sensor and humidity sensor.

[0085] Furthermore, the SF6 pressure sensor is installed near the SF6 gas pressure vessel, as close as possible to the key components of the equipment to ensure the accuracy of the pressure value;

[0086] The temperature sensor is installed on the equipment surface of the substation, and the humidity sensor is arranged in the air circulation area to prevent local conditions from affecting the humidity measurement and ensure accurate measurement of the ambient humidity;

[0087] Use a pressure calibration pump to apply a standard pressure value to the sensor, and correct the sensor output value and the pressure value of the calibration pump through linear calibration;

[0088] The "standard pressure value" refers to the known and highly accurate pressure value used in the calibration process, which is usually generated by calibration equipment (such as a pressure calibration pump or a pressure calibrator). These pressure values ​​are precisely controlled and measured and used to calibrate the pressure sensor to ensure that the sensor can provide accurate readings in actual applications.

[0089] Insert the temperature sensor into the temperature calibration furnace, record the output resistance value of the sensor, and compare it with the temperature resistance table to make calibration corrections;

[0090] It should be noted that the output of the PT1000 temperature sensor is a resistance value, and the temperature calibration furnace used for calibration should have high-precision temperature control. During calibration, the temperature is set at a known temperature point (such as 0°C, 50°C, 100°C, etc.), and then the resistance output of the sensor at these points is recorded and compared with a standard PT1000 temperature resistance table. If the output value deviates, calibration correction is performed.

[0091] Use a humidity calibration chamber, place the humidity sensor in the calibration chamber, record the sensor output, and compare it with the reference humidity value of the calibration chamber to make calibration corrections;

[0092] Set the sampling frequency of the data acquisition device, for example 1Hz, that is, collect data once per second, depending on the real-time requirements of the collected data, synchronize with the external time server through the Network Time Protocol (NTP), and output the substation environment data in real time.

[0093] Preferably, the Network Time Protocol (NTP) is used to ensure that the timestamp of the data acquisition device is consistent with the external time. Especially in multi-sensor systems, time synchronization is crucial to the accuracy of the data.

[0094] S2. Input the substation environmental data into the multi-sensor fusion algorithm to perform temperature compensation and humidity correction to generate a corrected multidimensional data set.

[0095] Furthermore, an adaptive Kalman filter is used to denoise the substation environmental data collected by the sensor;

[0096] Preferably, the Kalman filter can recursively suppress sensor noise to ensure a smooth output signal.

[0097] The output of the pressure sensor is usually affected by temperature changes, and this effect is often nonlinear. In order to compensate for the measurement error caused by temperature, it is necessary to establish a nonlinear model that can reflect the effect of temperature on pressure.

[0098] The absolute difference between the real-time temperature value and the reference temperature value is defined as: ;

[0099] in, Indicates the real-time temperature value. Indicates the reference temperature value;

[0100] In order to accurately describe the nonlinear effect of temperature on pressure, a combination of logarithmic function and exponential function is selected to establish a compensation model:

[0101] Logarithmic function It is used to describe the characteristic that the pressure error increases slowly when the temperature difference increases;

[0102] Exponential decay function Used to describe the rapid attenuation characteristics of temperature difference to pressure error;

[0103] Introduction coefficient , and , adjust the flexibility of the model, where The temperature compensation coefficient representing the logarithmic term controls the influence of the logarithmic function on the pressure compensation; The temperature compensation coefficient of the exponential term controls the influence of the exponential function on the pressure compensation. represents the time constant, which determines the rate of exponential decay;

[0104] Combining the above factors, an adaptive nonlinear model is constructed, which is expressed as:

[0105] ;

[0106] in, Indicates the pressure value after temperature compensation. Indicates the original pressure value, represents the temperature compensation coefficient, Indicates the real-time temperature value, that is, the current temperature read from the temperature sensor. Indicates the reference temperature value, which is usually taken as a standard value (such as 20°C or the average temperature of the environment), indicating a normal reference temperature. Indicates the absolute difference between the real-time temperature value and the reference temperature value. represents the coefficient of exponential decay, Represents the time constant, a parameter used to control the decay rate. It determines the rate at which the compensation amount decreases as the temperature deviation increases. Usually, The value of is determined by experiment or system characterization;

[0107] While considering the nonlinear effect of temperature on pressure and humidity, humidity changes will also have an impact on the output of the pressure sensor, which may include linear and nonlinear components;

[0108] The linear and quadratic terms of humidity are used to represent the nonlinear effect of humidity on pressure. The expression is:

[0109] ;

[0110] in, Indicates the linear coefficient of real-time humidity value correction, Represents the nonlinear coefficient of real-time humidity value correction, Indicates the real-time humidity value;

[0111] Apply the humidity correction factor to the temperature-compensated pressure value to obtain the humidity-corrected pressure value:

[0112] ;

[0113] in, Indicates the pressure value after humidity correction;

[0114] The pressure value after temperature compensation and humidity correction With real-time temperature value and humidity value Integrate to form a corrected multidimensional data set The expression is: ;

[0115] It should be noted that this solution, by introducing a nonlinear dynamic compensation model, can maintain a high correction accuracy when the temperature and humidity fluctuate rapidly. Compared with traditional linear compensation methods, nonlinear compensation can better adapt to the nonlinear characteristics of the environment, especially under extreme temperature and humidity conditions. The effect is remarkable; through historical data and environmental changes, the compensation and correction parameters are dynamically adjusted to ensure the applicability of the model under different environmental conditions, avoiding the limitations of fixed parameters in traditional models; through nonlinear dynamic compensation and humidity correction, the correction accuracy of the data can be significantly improved. Compared with traditional linear models, the error can be reduced by up to 10%-15%, especially under extreme environmental conditions (such as drastic changes in temperature and humidity), the stability and accuracy of the data can be better maintained.

[0116] S3. Use an adaptive filter to suppress noise and smooth data on the corrected multidimensional data set to generate high-precision pressure monitoring data.

[0117] Furthermore, we analyzed the multidimensional datasets using the Pearson correlation coefficient. The correlation between temperature, humidity and pressure is used to determine the degree of influence of temperature and humidity changes on pressure values;

[0118] According to the analysis results of the correlation between temperature, humidity and pressure, the state transfer equation of the Kalman filter is established;

[0119] The Kalman filter uses the state transition equation and the current state estimate, combined with real-time temperature, humidity and pressure measurements, to predict the system state (such as pressure) and its covariance at the next moment;

[0120] Among them, the state transfer equation usually expresses the dynamic change of pressure value with a linear or approximate linear model.

[0121] At each time step, the measurement error is calculated by comparing the Kalman filter prediction with the actual measurement;

[0122] The real-time temperature value and real-time humidity values It is introduced into the state space model of the Kalman filter as an observation variable to monitor the temperature and humidity in each time step, and provide coordinated input for the filter adjustment of the pressure value according to the severity of the temperature and humidity changes;

[0123] Set the threshold of temperature and humidity change rate , which is used to determine whether the temperature and humidity change dramatically, so as to dynamically adjust the filtering strength of the filter. The specific steps are as follows:

[0124] From the Cube Collect the temperature over a period of time and humidity data;

[0125] The expression for the rate of change of temperature is:

[0126] ;

[0127] The expression for the rate of change of humidity is:

[0128] ;

[0129] in, Indicates a time interval;

[0130] Perform statistical analysis on the rate of change of temperature and humidity and calculate the mean temperature , humidity mean and temperature standard deviation , humidity standard deviation ;

[0131] According to the statistical results, the threshold expression for setting the temperature change rate is:

[0132]

[0133] The threshold expression for setting the temperature change rate is:

[0134] ;

[0135] in, Usually 2 or 3 are used to cover 95%-99% of the data variation;

[0136] When the real-time temperature value Or real-time humidity value , exceeds the set threshold When , the filter will increase the smoothing strength of the pressure value;

[0137] When the real-time temperature value Or real-time humidity value , below the threshold When , the filter will reduce the smoothing strength and retain more details;

[0138] By combining the measurement error and the rate of change of temperature and humidity at each time step, the parameters of the Kalman filter are adjusted in real time;

[0139] Apply the extended and adjusted Kalman filter to dynamically calculate the Kalman gain based on the measurement error and covariance at each time step to adjust the weight between the predicted value and the observed value, thereby optimizing the effect of the filter;

[0140] By dynamically calculating the Kalman gain, the filter can process real-time data more accurately and adaptively adjust the filtering strength under different conditions;

[0141] Integrate real-time pressure values ​​after Kalman filter optimization , Real-time temperature value and real-time humidity values , generating high-precision pressure monitoring data, this dataset has a good smoothing effect and effectively suppresses environmental noise.

[0142] S4. The high-precision pressure monitoring data is wirelessly transmitted to the central processing platform for storage via the Internet of Things, and data cleaning and key feature extraction are performed.

[0143] Furthermore, the high-precision pressure monitoring data is packaged into standardized data packets and encrypted using the AES-256 symmetric encryption algorithm;

[0144] A low-power micro-computing platform is deployed near the substation to receive encrypted sensor data packets and store them in a time series database;

[0145] Use the checksum to verify the integrity of the data packet in the check sequence database to ensure that the data is not damaged during transmission. The checksum is an error detection method that generates a short fixed-length check code by processing the data with a specific algorithm and appends it to the end of the data. The receiver processes the data with the same algorithm and compares it with the appended check code to detect whether there is an error in the data transmission process.

[0146] The interquartile range (IQR) method was used to detect and eliminate outliers, and the Z-score method was used to further detect outliers;

[0147] Linear interpolation is used to handle missing values ​​in sensor data;

[0148] Using the Bayesian fusion algorithm, the data from different sensors (pressure, temperature, humidity) are fused to construct a global perception state;

[0149] Based on the global perception state, the hidden features of multimodal data are automatically extracted through the autoencoder;

[0150] The reinforcement learning algorithm is introduced to adjust the frequency and content of data transmission according to the changes in hidden features. The expression is:

[0151] ;

[0152] in, is the state-action value (Q value), indicating that in the state Next action The expected cumulative reward when is the current state, indicating the current state of the agent. is the current action, indicating that the agent is in state The action performed , is the learning rate, which controls the degree of mixing of new knowledge and old knowledge. is the immediate reward, indicating that the agent performs an action in the current state After receiving immediate feedback, is the discount factor, which indicates the decay coefficient of future rewards, is the next state, indicating that the agent is in the current state Execute an action After that, the new state is transferred ;

[0153] Learning Rate The value range is ,when When it is larger, the agent “trusts” new experiences more; when When it is smaller, the agent relies more on existing experience;

[0154] It should be noted that the transmission optimization of the reinforcement learning algorithm can reduce unnecessary data transmission, save bandwidth, and ensure that the data transmission frequency is accelerated at critical moments (such as when anomalies are detected). This feedback loop ensures the efficiency and flexibility of the system and also improves the overall intelligence level of monitoring.

[0155] S5. Input the extracted key features into the artificial intelligence model for fault prediction and output fault warning signals.

[0156] Furthermore, the hidden features of the extracted multimodal data are subjected to Min-Max normalization data normalization to scale the data to the same numerical range;

[0157] The normalized hidden features (such as pressure, temperature, humidity, etc.) are organized into three-dimensional tensors according to the time series structure and input into the time series convolutional neural network for fault prediction;

[0158] In the time series convolutional neural network, the fault risk model is introduced to perform fault risk scoring, and the expression is:

[0159] ;

[0160] in, Indicates time Failure risk score at the moment, Indicates The weight of the feature, Indicates Features The nonlinear mapping function is is the total number of feature variables, Index variables representing features (such as pressure, temperature, humidity, etc.);

[0161] The nonlinear mapping function is defined as:

[0162] ;

[0163] in, represents the scaling factor of the feature, For the The mean of the features;

[0164] Failure Risk Score Will accumulate over time to set a failure risk threshold ;

[0165] when When , a fault warning signal is output, indicating that a fault is about to occur;

[0166] when When , it is in a normal state and no warning is needed. The fault risk scoring mechanism can smooth short-term fluctuations and capture long-term abnormal trends through the integration operation of the time series, thereby improving the accuracy of fault prediction;

[0167] It should be noted that setting the failure risk threshold It is a systematic process from data analysis to model optimization, involving multiple links and gradual optimization. The specific process is as follows:

[0168] First, collect a large amount of historical sensor data and fault labels to calculate the fault risk score at each time point By analyzing the risk score trend before and after the failure, the preliminary threshold is determined. , which is usually set using the risk score distribution (such as the 95% quantile) when the failure occurs.

[0169] Use historical data to train the failure prediction model and calculate the risk score in the validation set , according to the performance of the validation set, adjust , so that the accuracy, recall rate, F1 score and other indicators of the early warning can be optimally balanced. The results of this step can further optimize the thresholds initially set. .

[0170] Based on specific business requirements, such as fault sensitivity and cost trade-offs If the cost of failure is high, the threshold can be appropriately lowered to allow the system to issue an early warning. This ensures a balance between the sensitivity of the warning and the false alarm rate.

[0171] The fault warning and processing results in actual operation will be fed back to the system to further supplement the historical data. Use this new data to retrain the model and dynamically adjust Through iterative optimization, the threshold is more accurate and adapts to changes in different operating conditions.

[0172] Finally, the failure risk threshold It can be combined with other key indicators (such as temperature and humidity changes, pressure fluctuations) for dynamic adjustment to build a multi-dimensional early warning mechanism to ensure the accuracy and flexibility of fault prediction.

[0173] For the training of the fault risk model, the cross entropy loss function will be used, and the Adam optimizer will be used for parameter update.

[0174] The training loss function expression of the fault risk model is:

[0175] ;

[0176] in, represents the length of the time series, represents the value of the loss function, represents the true fault label (0 or 1), Represents the predicted value of the TCN model;

[0177] Input historical sensor data and corresponding fault labels into the fault risk model and divide it into training set, validation set and test set;

[0178] Use the validation set to tune model hyperparameters (such as learning rate, regularization parameters, etc.). At the same time, after each epoch, use the validation set to evaluate the model's accuracy, AUC, F1 and other score indicators;

[0179] Select the optimal hyperparameters or model structure based on the performance of the validation set, such as preventing overfitting through early stopping;

[0180] After training and tuning, use the test set for final model evaluation. At this point, the data model of the test set has never seen it before, and the results of the test set can reflect the generalization ability of the model in the real environment;

[0181] Evaluate the final performance of the fault risk model using multiple metrics such as accuracy, AUC, and F1 score, and generate a confusion matrix to further understand the classification effect of the model;

[0182] For the selection of failure risk model, the best model is usually determined based on the AUC or F1 score of the validation set. Especially in the case of data imbalance, AUC and F1 scores can better reflect the model performance than simple accuracy.

[0183] For model tuning methods, grid search or random search can be used to find the optimal hyperparameter combination (such as learning rate, regularization coefficient, etc.), and the selection can be made based on the performance of the validation set.

[0184] S6. Trigger the alarm mechanism based on the fault warning signal and notify the operation and maintenance personnel, and at the same time feed back the actual fault handling results to the central processing platform.

[0185] Furthermore, after the fault warning signal is triggered, a fault warning report is generated, wherein the report includes the following contents: Fault risk score , the current value of key characteristics (pressure, temperature, humidity, etc.), the warning timestamp, the reason why the warning was triggered (such as which characteristic is out of the normal range), and the recommended action (such as checking a certain component);

[0186] The fault warning report is pushed to the monitoring terminal of the operation and maintenance personnel through the Internet of Things platform. The specific push channel can be pushed via SMS, email or a dedicated operation and maintenance APP to ensure that the operation and maintenance personnel can obtain the fault warning information in the first time.

[0187] After receiving the fault warning signal, the operation and maintenance personnel will go to the site to conduct troubleshooting based on the key characteristic data (such as pressure, temperature, humidity, etc.) provided in the warning report and the suggestions provided by the system;

[0188] The operation and maintenance personnel handle the fault, which may include: replacing damaged equipment or components, adjusting system parameters, repairing sensors or communication equipment;

[0189] After handling the fault, the operation and maintenance personnel record the actual fault handling results, including: whether an actual fault occurred, the specific cause of the fault, the details of the fault handling (such as repair time, handling method), and whether further follow-up is required;

[0190] Feedback the troubleshooting results of the operation and maintenance personnel to the central processing platform, and the feedback data will be used for model retraining;

[0191] The operation and maintenance personnel enter the fault handling results through the operation and maintenance terminal. The key fields of the handling results include: whether the fault occurred (1 means it occurred, 0 means it did not occur), the actual cause of the fault (such as equipment aging, sensor failure, etc.), the fault handling result (such as: replacing equipment, repairing sensors, etc.), and the fault handling time.

[0192] The fault handling results are uploaded to the central processing platform in real time through the IoT platform. The central processing platform stores the newly collected fault handling data in the database as label data for subsequent model training to iteratively optimize the artificial intelligence model.

[0193] By feeding back the actual fault handling results to the central processing platform, the system can use this new data to iteratively optimize the artificial intelligence model. The specific steps are as follows:

[0194] Combine the newly added fault processing data (including sensor feature data and fault labels) with historical data to build a new training data set;

[0195] Retrain the failure risk model based on the new training data set;

[0196] The training steps are similar to the previous steps: data preprocessing (normalization and time series construction); model training using a temporal convolutional neural network (TCN); and model parameter optimization using a cross entropy loss function.

[0197] After each model training, use the validation set to evaluate the model's performance, including accuracy, AUC, F1 score and other indicators, to ensure that the model's performance is improved with the addition of new data;

[0198] Perform hyperparameter tuning based on the results of the validation set. In particular, when new data causes model performance fluctuations, use grid search or random search to adjust the model's learning rate, regularization coefficient and other parameters.

[0199] After the model optimization is completed, the new version of the model will be deployed to the system to replace the old version of the model. At the same time, the new model will continue to predict the risk of failure. After each fault warning, the system will still iterate and optimize based on the new sensor data and fault handling results, forming a closed loop of continuous learning.

[0200] This embodiment also provides a computer device, which is suitable for the case of a data collection and processing method based on a substation SF6 pressure gauge, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the data collection and processing method based on a substation SF6 pressure gauge as proposed in the above embodiment.

[0201] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0202] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the data collection and processing method based on the SF6 pressure gauge of the substation is implemented as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0203] In summary, the present invention effectively solves the problem in the prior art that SF6 pressure data is easily affected by ambient temperature and humidity through multi-sensor fusion technology and adaptive filtering algorithm, and significantly improves the accuracy of pressure monitoring. At the same time, through real-time compensation and correction of temperature and humidity, the present invention can maintain high-precision monitoring of SF6 pressure data under complex environmental conditions; in addition, the remote transmission of data is realized by combining the Internet of Things technology, and fault prediction and early warning are performed through artificial intelligence models, which greatly improves the intelligent operation and maintenance level of substation equipment and reduces the occurrence rate of equipment failures and maintenance costs.

[0204] Example 2, referring to Table 1, is the second example of the present invention. To further verify the technical solution of the present invention, experimental simulation data based on the data collection and processing method of the SF6 pressure gauge in the substation is provided.

[0205] The experimental environment selected the SF6 gas pressure monitoring system of a substation, including pressure sensors, temperature sensors and humidity sensors, and connected to the central processing platform through the Internet of Things wireless transmission module. In order to compare the effectiveness of the existing technology and the method of the present invention, two groups were set up in the experiment: the control group used the traditional SF6 pressure gauge monitoring method, and the experimental group used the multi-sensor fusion algorithm and artificial intelligence fault prediction model of the present invention.

[0206] First, the experimental group and the control group used the same specifications of pressure sensor (resolution 0.01Pa), PT1000 platinum resistance temperature sensor and capacitive humidity sensor. The sensor installation position followed the standard layout requirements to ensure the accuracy of data collection.

[0207] Secondly, before the experiment begins, all sensors need to be strictly calibrated. The pressure sensor is calibrated by a pressure calibration pump; the temperature sensor is calibrated by a temperature calibration furnace; the humidity sensor is calibrated by a humidity calibration chamber to ensure that the sensor output is consistent with the actual environmental conditions.

[0208] Next, the data acquisition device set the sampling frequency to 1Hz and synchronized with the external time server through the Network Time Protocol (NTP) to ensure the data consistency of each sensor. At the same time, the experimental group used an adaptive Kalman filter to denoise the data and corrected it based on the nonlinear model of temperature and humidity to generate a multidimensional data set. The control group used traditional data acquisition and processing methods without multi-sensor fusion and advanced filtering.

[0209] Furthermore, the experimental group collected environmental data of the substation in real time through multiple sensors, including the pressure, external temperature and humidity of SF6 gas. Subsequently, an adaptive filter was used to suppress noise and smooth the data to generate high-precision pressure monitoring data. The control group only collected SF6 pressure data and used a simple linear correction method to compensate for temperature and humidity; the experimental group uploaded the processed multidimensional data set to the central processing platform through the IoT wireless transmission module, and performed further data cleaning and key feature extraction. The control group uploaded SF6 pressure data without complex processing through the traditional monitoring platform; the experimental group input the extracted key features into the artificial intelligence model for fault prediction and generated a fault warning signal; the control group monitored the changes in SF6 pressure through manually set thresholds, and triggered an alarm when the monitoring value exceeded the threshold.

[0210] The details are shown in Table 1 below:

[0211] Table 1 Comparison of key parameters between experimental group and control group

[0212]

[0213] It can be seen from the experimental data that the data collection and processing method based on the substation SF6 pressure gauge of the present invention has significant advantages in multiple key performance indicators. The experimental data is analyzed in detail as follows:

[0214] Comparison of pressure monitoring accuracy: The pressure monitoring accuracy of the experimental group reached 0.005Pa, which is about 4 times higher than the 0.02Pa of the control group. High-precision pressure monitoring benefits from the multi-sensor fusion algorithm in the present invention, especially the temperature compensation and humidity correction model, which greatly reduces the impact of environmental factors on the pressure value. Traditional monitoring methods only use simple linear compensation, which cannot effectively cope with complex environmental changes and leads to large errors.

[0215] Comparison of data noise suppression rate: The data noise suppression rate of the experimental group reached 98.5%, significantly higher than 72.3% of the control group. This advantage is mainly due to the adaptive Kalman filter used in the present invention, which can adjust the smoothing strength in real time and dynamically respond to changes in temperature and humidity, thereby effectively suppressing noise while maintaining data details. The control group only used traditional filtering methods and could not dynamically adjust the filtering parameters, resulting in poor noise suppression effect.

[0216] Comparison of fault prediction success rate: The fault prediction success rate of the experimental group was 95.6%, while that of the control group was only 80.2%. The present invention greatly improves the accuracy of fault prediction by inputting multimodal data into the artificial intelligence model and using a time series convolutional neural network for fault prediction. In contrast, the control group relies on a fixed threshold for fault judgment, which is prone to false positives or false negatives, resulting in a low success rate of fault prediction.

[0217] Comparison of fault warning response time: The fault warning response time of the experimental group was 2.3 seconds, significantly lower than the 5.8 seconds of the control group. Through real-time key feature extraction and rapid response of the artificial intelligence model, the present invention can quickly issue a warning signal when a fault is about to occur, thereby buying valuable time for operation and maintenance personnel to troubleshoot the fault. In contrast, the warning response of the control group was relatively slow and could not respond to sudden faults in a timely manner.

[0218] Comparison of data transmission packet loss rate: The data packet loss rate of the experimental group was only 0.1%, while that of the control group was 0.8%. The present invention uses AES-256 encryption and low-power IoT devices to ensure the integrity and security of data transmission. Since the control group uses traditional transmission methods, data is easily disturbed during transmission, resulting in a higher packet loss rate.

[0219] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A data collection and processing method based on SF6 pressure gauge in a substation, characterized in that: include, Collect substation environmental data in real time through SF6 pressure sensor, temperature sensor and humidity sensor; Input substation environmental data into the multi-sensor fusion algorithm to perform temperature compensation and humidity correction to generate a corrected multidimensional data set; Use adaptive filters to suppress noise and smooth data on the corrected multidimensional data set to generate high-precision pressure monitoring data; The high-precision pressure monitoring data is wirelessly transmitted to the central processing platform through the Internet of Things for storage, and data cleaning and key feature extraction are performed; The extracted key features are input into the artificial intelligence model for fault prediction and fault warning signals are output; Trigger the alarm mechanism based on the fault warning signal and notify the operation and maintenance personnel, and feed back the actual fault handling results to the central processing platform; The real-time collection of substation environmental data comprises the following specific steps: Install the SF6 pressure sensor near the SF6 gas pressure vessel; The temperature sensors are installed on the equipment surface of the substation, and the humidity sensors are arranged in the air circulation area; Use a pressure calibration pump to apply a standard pressure value to the sensor, and correct the sensor output value and the pressure value of the calibration pump through linear calibration; Insert the temperature sensor into the temperature calibration furnace, record the output resistance value of the sensor, and compare it with the temperature resistance table to make calibration corrections; Use a humidity calibration chamber, place the humidity sensor in the calibration chamber, record the sensor output, and compare it with the reference humidity value of the calibration chamber to make calibration corrections; Set the sampling frequency of the data acquisition equipment, synchronize with the external time server through the network time protocol, and output the substation environment data in real time; The specific steps of generating the corrected multidimensional data set are: Adaptive Kalman filter is used to denoise the substation environmental data collected by sensors; Based on the nonlinear effect of temperature on pressure and humidity, temperature compensation is performed through an adaptive nonlinear model, and the expression is: ； in, Indicates the pressure value after temperature compensation. Indicates the real-time pressure value. represents the temperature compensation coefficient, Indicates the real-time temperature value. Indicates the reference temperature value, Indicates the absolute difference between the real-time temperature value and the reference temperature value. represents the coefficient of exponential decay, represents the time constant; Based on the pressure value after temperature compensation, a nonlinear correction model based on humidity is constructed through the nonlinear influence of humidity on pressure and temperature values. The expression is: ； in, Indicates the pressure value after humidity correction. Indicates the real-time humidity value. Indicates the linear coefficient of real-time humidity value correction, Indicates the nonlinear coefficient of real-time humidity value correction; Integrate the pressure value after nonlinear temperature compensation and humidity correction with the real-time temperature value and humidity value to form a corrected multidimensional data set; The specific steps of generating high-precision pressure monitoring data are as follows: The correlation between temperature, humidity and pressure in the multidimensional data set was analyzed by Pearson correlation coefficient; According to the analysis results of the correlation between temperature, humidity and pressure, the state transfer equation of the Kalman filter is established; The Kalman filter uses the state transfer equation and the current state estimate, combined with real-time temperature, humidity and pressure measurements, to predict the state and its covariance at the next moment; At each time step, the measurement error is calculated by comparing the Kalman filter prediction with the actual measurement; The real-time temperature value and real-time humidity values Introduced as observation variables into the state space model of the Kalman filter to monitor the temperature and humidity in each time step; Set the threshold of temperature and humidity change rate ; When the real-time temperature value Or real-time humidity value , exceeds the set threshold When , the filter will increase the smoothing strength of the pressure value; When the real-time temperature value Or real-time humidity value , below the threshold When , the filter will reduce the smoothing strength; By combining the measurement error and the rate of change of temperature and humidity at each time step, the parameters of the Kalman filter are adjusted in real time; Apply the extended and adjusted Kalman filter to dynamically calculate the Kalman gain and optimize the Kalman filter based on the measurement error and covariance at each time step; Integrate real-time pressure values ​​after Kalman filter optimization , Real-time temperature value and real-time humidity values , generate high-precision pressure monitoring data; The data cleaning and key feature extraction are specifically performed in the following steps: Pack high-precision pressure monitoring data into standardized data packets and encrypt the data using the AES-256 symmetric encryption algorithm; A low-power micro-computing platform is deployed near the substation to receive encrypted sensor data packets and store them in a time series database; Verify the integrity of the data packet in the check sequence database; The interquartile range method was used to detect and eliminate outliers, and the Z-score method was used to further detect outliers; Linear interpolation is used to handle missing values ​​in sensor data; Using the Bayesian fusion algorithm, data from different sensors are fused to build a global perception state; Based on the global perception state, the hidden features of multimodal data are automatically extracted through the autoencoder; Introducing reinforcement learning algorithms to adjust the frequency and content of data transmission according to changes in hidden features; The extracted key features are input into the artificial intelligence model for fault prediction and fault warning signals are output. The specific steps are: The hidden features of the extracted multimodal data are normalized by Min-Max normalization to scale the data to the same numerical range; The normalized hidden features are organized into a three-dimensional tensor according to the time series structure and input into the time series convolutional neural network for fault prediction; In the time series convolutional neural network, the fault risk model is introduced to perform fault risk scoring, and the expression is: ； in, Indicates time Failure risk score at the moment, Indicates The weight of the feature, Indicates Features The nonlinear mapping function is is the total number of feature variables, An index variable representing a feature; Failure Risk Score Will accumulate over time to set a failure risk threshold ; when When , a fault warning signal is output, indicating that a fault is about to occur; when When , it is in normal state and no warning is needed; For the training of the fault risk model, the cross entropy loss function will be used, and the Adam optimizer will be used for parameter update.

2. The method for collecting and processing data based on the SF6 pressure gauge of a substation according to claim 1, characterized in that: The training of the fault risk model will adopt the cross entropy loss function and use the Adam optimizer to update the parameters. The specific steps are as follows: The training loss function expression of the fault risk model is: ； in, represents the length of the time series, represents the value of the loss function, represents the actual fault label, Represents the predicted value of the convolutional neural network model; Input historical sensor data and corresponding fault labels into the fault risk model and divide it into training set, validation set and test set; Use the validation set to tune the hyperparameters of the failure risk model. After each epoch, use the validation set to evaluate the accuracy, AUC, and F1 score of the model. Select the optimal hyperparameters based on the performance of the validation set; After training and tuning are completed, the final performance of the failure risk model is evaluated using the accuracy, AUC, and F1 score metrics in the test set.

3. The data collection and processing method based on the SF6 pressure gauge of the substation according to claim 2 is characterized in that: The alarm mechanism is triggered according to the fault warning signal and the operation and maintenance personnel are notified, and the actual fault handling results are fed back to the central processing platform. The specific steps are: After the fault warning signal is triggered, a fault warning report is generated; Push fault warning reports to the monitoring terminals of operation and maintenance personnel through the IoT platform; After receiving the fault warning signal, the operation and maintenance personnel will go to the site for investigation based on the key characteristic data in the warning report; After troubleshooting, the operation and maintenance personnel record the actual troubleshooting results and enter the troubleshooting results through the operation and maintenance terminal; The fault handling results are uploaded to the central processing platform in real time through the Internet of Things platform. The central processing platform stores the newly collected fault handling data in the database as label data for subsequent model training to iteratively optimize the artificial intelligence model.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for collecting and processing data based on the substation SF6 pressure gauge according to any one of claims 1 to 3 are implemented.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the method for collecting and processing data based on the substation SF6 pressure gauge according to any one of claims 1 to 3 are implemented.

Images