A self-correcting and self-optimizing method for frequency modulation model of steam turbine of thermal power unit

Abstract

The application provides a frequency modulation model self-correction and self-optimization method for a steam turbine of a thermal power unit, which comprises the following steps: constructing a data set, inputting the data set into a PyCT-PsLSTM prediction network model, outputting the calculation error of the actual measured values of the main steam pressure, the valve opening degree and the unit load, and prewarning and starting PID parameter optimization correction when the error is greater than a threshold value; screening the current power PID parameters without tampering from the data set, inputting the parameters into a GPR model, obtaining a predicted mean value, a standard deviation and a confidence interval, and determining tampering when the parameters are greater than an interval; based on the mean value and the standard deviation, an uncertainty coefficient is calculated, the global search is performed by using a particle swarm algorithm when the coefficient is greater than a threshold value, the local optimization is performed when the coefficient is less than or equal to the threshold value, the optimal solution is obtained, the data set is updated and the tampered parameters are fine-tuned, the actual value is compared with the GPR output, and it is determined whether the solution is removed or kept in prewarning. By comparing the predicted PID controller parameters and the confidence interval thereof, it is determined whether the prewarning is ended or kept, and the modeling ability of the time sequence characteristics of the key parameters is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent frequency modulation and automatic control of thermal power generating units, and in particular to a self-correction and self-optimization method for a frequency modulation model of a steam turbine of a thermal power generating unit. BACKGROUND

[0002] In the frequency modulation control of modern thermal power generating units, the dynamic characteristics of the steam turbine body and the parameter identification technology of various types of actuators are relatively mature, and the current modeling and parameter acquisition means can meet the actual engineering requirements. However, there is still a great uncertainty in the setting of PID parameters in the governor regulation system. Due to the performance evaluation or economic optimization of each power plant, the PID controller parameters are often adjusted privately, which may lead to a decrease in the response ability of the frequency modulation system to power grid disturbances, and even cause low-frequency oscillation, system oscillation and unstable operation, etc., which may cause hidden dangers to the safety of the power system.

[0003] The existing dispatching system generally lacks real-time supervision and effective constraints on the PID parameters, and cannot timely detect and correct abnormal operating conditions caused by parameter tampering. The traditional parameter setting method mostly relies on artificial experience or offline experiments, and is difficult to adapt to the dynamic changes of power plant working conditions and complex disturbance environment, and lacks real-time warning and self-adaptive correction mechanism for parameter changes.

[0004] In view of the above industry pain points, the present application constructs and trains a data-driven model by real-time collection of unit frequency, main steam pressure, load, speed and other operating data, realizes online prediction and supervision of the PID parameters of the governor regulation system, and when the actual parameters deviate from the model prediction value by more than a set threshold, the system can automatically trigger an alarm and correct the parameters, effectively preventing the decline of frequency modulation performance and the risk of power grid stability caused by human tampering, and improving the intelligentization and automation level of power plant frequency modulation control. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a self-correction and self-optimization method for a frequency modulation model of a steam turbine of a thermal power generating unit, which aims to solve the problems in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a self-correction and self-optimization method for a frequency modulation model of a steam turbine of a thermal power generating unit, comprising:

[0007] Step S1: constructing a standardized time domain data set;

[0008] Step S2: Input the standardized time-domain dataset into the PyCT-PsLSTM prediction network model for processing, and calculate the difference between the output main steam pressure, valve opening and unit load and the actual measured values ​​at the current time. When the error of the difference calculation exceeds the preset error threshold, an early warning is triggered and the PID controller parameter prediction optimization correction mechanism is started.

[0009] Step S3: In the prediction optimization and correction mechanism, the current power PID controller parameters that have not been tampered with are selected from the standardized time-domain dataset and input into the GPR model for processing. The predicted mean, predicted standard deviation and confidence interval are output. The current power PID controller parameters are compared with the confidence interval. If the current power PID controller parameters fall outside the confidence interval, it is determined that the current power PID controller parameters have been illegally tampered with, and a correction command is issued.

[0010] Step S4: After issuing the correction command, first calculate the uncertainty coefficient based on the predicted mean and predicted standard deviation, compare the uncertainty coefficient with the preset coefficient threshold. If it is greater than the coefficient threshold, use the particle swarm optimization algorithm to perform a global search path for the predicted mean and predicted standard deviation. If it is less than or equal to the coefficient threshold, use the particle swarm optimization algorithm to perform local optimization exploration for the predicted mean and predicted standard deviation to obtain the current optimal solution. Then, update the standardized time domain dataset based on the current optimal solution, and fine-tune and correct the illegally tampered current power PID controller parameters based on the updated standardized time domain dataset.

[0011] Step S5: By periodically comparing the actual measured values ​​with the predicted mean, predicted standard deviation, and confidence interval output by the GPR model, determine whether to lift the warning or keep the warning triggered.

[0012] Furthermore, the specific process of step S1 is as follows:

[0013] The operating data of primary frequency regulation of thermal power units, including main steam pressure, is obtained based on sensors. Rotation speed Unit load Valve opening Current power PID controller parameters and target power value Key parameters; current power PID controller parameters include proportional gain. Integral gain Differential gain ;

[0014] A time-domain dataset is constructed based on the operating data of primary frequency regulation of thermal power units;

[0015] By normalizing, sliding window slicing, and cleaning the time-domain dataset, standardized time-series samples are obtained.

[0016] The operating data of the thermal power unit under different time points and load conditions are collected. The operating data includes operating conditions covering high load, medium load, low load, frequency disturbance, and load step.

[0017] Appropriate time tolerance windows are used for data points with different acquisition frequencies in the operating data of thermal power units. Processing is performed to allow for a reasonable time tolerance window. Data points collected at different frequencies are considered valid data at the same time.

[0018] All valid data timestamps are uniformly sorted and then distributed according to a set reasonable time tolerance window. The time stamp is iterated along the sorted timeline, and during the iteration, the timestamp is placed within the current unified and reasonable time tolerance window. Data points within the specified range are grouped together; within a reasonable time tolerance window During the grouping process, synchronously identify and remove timestamps that exhibit abnormal drift or exceed the reasonable time tolerance window. The discrete data at the threshold is used to obtain grouped data, and a linear interpolation function is applied to the grouped data to fill a reasonable time tolerance window. For missing values, Min-Max normalization is used to map the data after filling in the missing values ​​to the [0,1] interval, thus obtaining multi-source data;

[0019] Wherein, the linear interpolation function and normalization represent:

[0020] ;

[0021] ;

[0022] In the formula, This represents the estimated operating parameters of the thermal power unit at the missing positions within the time tolerance window after processing by the linear interpolation function; that is, the missing operating data values. and These represent the values ​​of two adjacent known data points in the operating data of a thermal power unit; and These represent the operating data of thermal power units. and timestamp; The original operating data of thermal power units to be normalized; This represents the minimum value of the operating data for thermal power units. This represents the maximum value of the operating data for the thermal power unit. This refers to the normalized multi-source data;

[0023] The final standardized time series samples and multi-source data form a standardized time-domain data set containing load, main steam pressure, valve opening, and rotating speed ; is the standardized time-domain data set at time t; is the main steam pressure at time t; is the rotating speed at time t; is the unit load at time t; is the valve opening at time t; is the proportional gain at time t; is the integral gain at time t; is the differential gain at time t; is the power target value at time t.

[0024] Further, the processing procedure of the PyCT-PsLSTM prediction network model is as follows:

[0025] The PyCT-PsLSTM prediction network model comprises a channel independence unfolding module, a patching module, a feature extraction module, a channel-time collaborative attention module, and a long short-term memory network module.

[0026] The standardized time-domain data set is input into the channel independence unfolding module for processing to obtain the output of the channel independence unfolding module.

[0027] The output of the channel independence unfolding module is input into the patching module, which divides the time series in each channel of the output of the channel independence unfolding module into fixed-length time segments Patch. Each fixed-length time segment Patch is mapped to a unified feature dimension through a linear embedding layer to capture local dependency information and cross-time segment association in the time series of the feature dimension, thereby obtaining the output of the patching module.

[0028] The output of the patching module is input into the double-path stacked feature extraction module, which comprises two structurally parallel sub-paths: the first structurally parallel sub-path is a local path composed of two one-dimensional convolution-maximum pooling units connected in series, and the first sub-path output is obtained by extracting short-term local features in the time series of the output of the patching module; the second structurally parallel sub-path is a global path composed of a downsampling-large convolution kernel-upsampling three-unit, and the second sub-path output is obtained by extracting global trend information of the time span in the output of the patching module; the first sub-path output and the second sub-path output are fused by element-wise addition to form a fusion feature tensor ;

[0029] The channel-time collaborative attention module processes the fusion feature tensor The time dimension and the space dimension are globally averaged and pooled, and then linearly transformed and passed through a sigmoid function to generate the channel attention weight ;

[0030] The channel dimension and the space dimension of the fused feature tensor are maximally pooled to obtain the time attention weight , and are element-wise multiplied to obtain the attention-weighted feature output , wherein represents element-wise multiplication;

[0031] After being flattened, the is input into the long short-term memory network module as an input vector for processing, and the main steam pressure, the valve opening and the load are obtained as the predicted values.

[0032] Further, the specific process of calculating the difference between the output main steam pressure, valve opening and unit load in step S2 and the actual measured value at the current time is as follows:

[0033] When the absolute error between the predicted value and the actual measured value exceeds the set error threshold , the inference of the GPR model is triggered and a warning is given, and the PID controller parameter prediction optimization correction process is started; when the error threshold is not exceeded , safety is displayed; is the mean absolute error of the network, is the redundant error threshold, and the actual measured value is the main steam pressure , the unit load and the valve opening in step S1.

[0034] Further, the specific process of outputting the predicted mean and the predicted standard deviation and the confidence interval in step S3 is as follows:

[0035] When the absolute error between the predicted value and the actual measured value exceeds the set error threshold , the current power PID controller parameters without tampering are selected from the standardized time domain data set, and the real-time process variables and extended statistical features are extracted from the current power PID controller parameters without tampering as the input of the GPR model training;

[0036] The real-time process variables are the valve opening , the main steam pressure p and the unit load at each sampling time;

[0037] The extended statistical features include the moving average mean( ​), the standard deviation of the main steam pressure std(p) and the valve opening change trend( ) calculated by linear fitting;

[0038] The normalized time-domain data set is proportionally divided into a training data set and a test data set in chronological order;

[0039] The training data set is , wherein represents the i-th feature vector, is the true value of the current power PID controller parameter at the i-th corresponding moment; is the number of samples;

[0040] The radial basis kernel is used to capture the nonlinear coupling between the valve opening and the main steam pressure p, the linear kernel is used to represent the steady-state association between the unit load and the valve opening , and the periodic kernel is introduced to describe the day-night regularity of the unit load fluctuation, respectively, to obtain , while satisfies the normalization constraint , and is adjusted through adaptive optimization , specifically:

[0041] ;

[0042] ;

[0043] ;

[0044] ;

[0045] ;

[0046] In the formula, is the covariance matrix; is the weight coefficient corresponding to the radial basis kernel; is the radial basis kernel function; is the weight coefficient corresponding to the linear kernel; is the linear kernel function; is the weight coefficient corresponding to the periodic kernel; is the periodic kernel function; is the variance of the observation noise; is the unit matrix; is the weight coefficient of the i-th kernel function; ; is the transpose of the i-th weight vector; is the feature vector; is the transpose of the jth weight vector; is the log probability; is the log probability is the partial derivative of ; is the log probability is the partial derivative of ; is the is the partial derivative of ; is the weight vector at the t+1th iteration; is the weight vector at the tth iteration; is the gradient of the log probability with respect to the weight vector w; is the weight change at the tth iteration; is the learning rate, is the momentum coefficient;

[0047] Adding a noise variance term in the covariance matrix K , the complete covariance matrix is defined as , where is the N x 6-dimensional training data matrix;

[0048] Based on the Bayesian framework, the optimization hyperparameter set is set by maximizing the marginal likelihood function estimate , the optimized hyperparameter set is obtained, the GPR model is optimized, the GPR model with the optimized hyperparameter set is obtained, and the hyperparameter set is adjusted based on adaptive optimization and the complete covariance matrix is constructed, where is the length scale of the radial basis kernel, is the period parameter of the periodic kernel, is the smoothing coefficient of the periodic kernel, is the amplitude coefficient of the periodic kernel, is the noise standard deviation, is the standard deviation of the radial basis kernel;

[0049] The GPR model with the optimized hyperparameter set is calculated on the test data set to obtain the predicted mean value and the observation noise variance of the current power PID controller parameters, which are represented as:

[0050] ;

[0051] ;

[0052] In the formula, is the predicted mean value of the test data set using the GPR model with the optimized hyperparameter set; k is the covariance function; is the test data set, This represents the true values ​​of the current power PID controller parameters at the corresponding moment; The standard deviation of the predictions for the test dataset;

[0053] Final output predicted mean Includes parameter values and confidence zone ; This is the predicted mean of the proportional term; This is the predicted mean of the integral term; This is the predicted mean of the differential term.

[0054] Furthermore, the specific process of obtaining the current optimal solution in step S4 is as follows:

[0055] After issuing the correction command, it is first based on the predicted mean. and the predicted standard deviation Calculate the uncertainty coefficient. The coefficient threshold is 0.15, when When, execute the global search path; when At that time, local optimization exploration is carried out;

[0056] when Perform a global search path; then in the global parameter space In the process, a certain number of particles are randomly generated. As corresponding candidate parameters; It is defined in the global parameter space One of the variables;

[0057] The error between the predicted value and the actual measured value and the predicted mean The sum of errors with the candidate parameters constitutes the fitness function. A particle swarm optimization algorithm is used to find the optimal particle parameters until the iteration reaches the specified number of iterations or the fitness converges. The optimal position of the particle is then taken as the current optimal solution. The formula is as follows:

[0058] ;

[0059] ;

[0060] In the formula, For the first Fitness function for each particle; for Weighting coefficients; For the first The error between the predicted value and the actual measured value of each particle; for Weighting coefficients; For the first Error of the current power PID controller parameter of the particle and the candidate parameter is the inertia weight is the change of the position of the particle in the jth parameter dimension in the ith iteration is the velocity of the particle in the jth parameter dimension in the ith iteration is the optimal position of the particle in the jth parameter dimension is the inertia weight is the change of the position of the particle in the jth parameter dimension in the ith iteration is the velocity of the particle in the jth parameter dimension in the ith iteration is the optimal position of the particle in the jth parameter dimension is the current position of the particle in the jth parameter dimension in the ith iteration is the individual learning factor is a random number with a value range of [0, 1], which adjusts the random intensity in learning is the individual learning factor is the optimal position of the particle in the jth parameter dimension is the current position of the particle in the jth parameter dimension in the ith iteration is the individual learning factor is a random number with a value range of [0, 1], which adjusts the random intensity in learning is the optimal position of the particle in the jth parameter dimension is the current position of the particle in the jth parameter dimension in the ith iteration is the social learning factor is a random number with a value range of [0, 1], which adjusts the random intensity in learning is the social learning factor is a random number with a value range of [0, 1], which adjusts the random intensity in learning is the global optimal position of the entire particle swarm in the jth parameter dimension is the current position of the particle in the jth parameter dimension in the ith iteration is the global optimal position of the entire particle swarm in the jth parameter dimension is the current position of the particle in the jth parameter dimension in the ith iteration is the global optimal position of the entire particle swarm in the jth parameter dimension is the global optimal position of the entire particle swarm in the jth parameter dimension

[0061] When , local optimization is performed; in the global parameter space , the error of the predicted value and the actual measured value and the error of the predicted mean value and the candidate parameter are summed as the fitness function, and the velocity and position of the particle are constantly updated until the number of iterations meets or the fitness converges, and finally the optimal position of the particle is taken as the current optimal solution ;

[0062] wherein the change of the current optimal solution , indicates:

[0063] .

[0064] Further, the specific process of fine-tuning and correcting the illegally tampered current power PID controller parameter in step S4 is as follows:

[0065] ​​The confidence bias threshold is set to 0.1, and the current optimal solution in the neighborhood of the current optimal solution is stored together with the feature vector as a new training sample, which is expanded to the standardized time domain data set , and the hyperparameter set of the illegally tampered current power PID controller parameters is fine-tuned and corrected using the expanded standardized time domain data set by the maximum marginal likelihood function. A new training sample is added to the standardized time domain data set. The feature vector composed of the real-time process variable and the extended statistical features

[0066] The change of the current optimal solution exceeds the set confidence bias threshold , the change of the current optimal solution is marked as an abnormal PID controller parameter. For the PID controller parameters in the abnormal state, the candidate parameter with the smallest error is selected from the global parameter space as the optimal control parameter

[0067] The feature vector and the optimal control parameter are combined into a sample combination, and the sample combination is added to the standardized time domain data set , that is , to obtain an updated standardized time domain data set; and based on the updated standardized time domain data set, the hyperparameter set of the illegally tampered current power PID controller parameters is fine-tuned and corrected by the maximum marginal likelihood function, to obtain the corrected current power PID controller parameters. Further, the specific process of step S5 is as follows:

[0068] After executing the current power PID controller parameter correction instruction, the current power PID controller parameter marked as illegally tampered is continuously supervised and dynamically judged by the current power PID controller parameter optimization correction mechanism, specifically as follows:

[0069] After the current optimal solution

[0070] deviates from the confidence interval, it enters the continuous observation stage: when the absolute error between the predicted value and the actual measured value is less than the set error threshold , it indicates that the adjustment of the illegally tampered current power PID controller parameter is effective, and the early warning is released, and when the absolute error between the predicted value and the actual measured value is greater than the set error threshold , it enters the next cycle judgment process and remains warning.

[0071] ​​Based on the real-time operation data of the primary frequency modulation of the thermal power generating unit obtained by the sensor, the GPR model is trained through the real-time operation data, the real-time fitting update is carried out on the corrected current power PID controller parameter operation data and the real-time operation data based on the trained GPR model, and the uncertainty coefficient of the current time step is calculated according to the real-time fitting update , the uncertainty coefficient of the current time step , which indicates that the current power PID controller parameter supporting illegal tampering is fine-tuned and corrected.

[0072] Compared with the prior art, the present application has the following beneficial effects:

[0073] (1) The present application proposes a feature extraction module based on a PyCT-PsLSTM prediction network model, which combines local time structure blocks and local dependency information to improve the time series feature modeling capability of key parameters such as main steam pressure, valve opening, unit load, etc. in the thermal power generating unit operation data; the GPR model is introduced for nonlinear parameter mapping, which has the ability to output prediction variance, so that the GPR model can not only predict key operating parameters, but also provide confidence information to enhance interpretability and prediction reliability.

[0074] (2) The present application designs a double-trigger self-correction mechanism based on the output prediction value error and uncertainty coefficient of the PyCT-PsLSTM prediction network model, which can identify and reject unreasonable parameters that may be tampered with by humans when the current power PID controller parameter deviates from the prediction value or confidence interval; based on the confidence interval of the PyCT-PsLSTM prediction network model, the optimal non-known PID parameter is found out, and the optimal parameter is found out by using the optimization ability of the particle swarm algorithm and the prediction fitting of the thermal power generating unit operation state to update the standardized time domain data set and the micro-batch training GPR model, which can improve the prediction accuracy of the PyCT-PsLSTM prediction network model for the thermal power generating unit operation state, optimize the operation performance of the thermal power generating unit, reduce fuel consumption and pollutant emissions, and ensure the safe and stable operation of the unit. BRIEF DESCRIPTION OF DRAWINGS

[0075] Figure 1 The present application provides a method flowchart. DETAILED DESCRIPTION

[0076] As shown in Figure 1 , the present application provides a technical solution: a thermal power generating unit steam turbine frequency modulation model self-correction and self-optimization method, comprising:

[0077] Step S1: Construct a standardized time domain data set;

[0078] Step S2: input the standardized time domain data set into the PyCT-PsLSTM prediction network model for processing, output the main steam pressure, valve opening and unit load, and calculate the difference value with the actual measurement value at the current time, when the error of the difference value calculation exceeds the preset error threshold, the warning is triggered and the PID controller parameter prediction optimization correction mechanism is started;

[0079] Step S3: in the prediction optimization correction mechanism, the current power PID controller parameters without tampering are selected from the standardized time domain data set and input into the GPR model for processing, and the predicted mean and predicted standard deviation and confidence interval are output; compare the current power PID controller parameters with the confidence interval, if the current power PID controller parameters fall outside the confidence interval range, it is determined that the current power PID controller parameters are illegally tampered, and a correction instruction is issued;

[0080] Step S4: after issuing the correction instruction, first calculate the uncertainty coefficient based on the predicted mean and predicted standard deviation, compare the uncertainty coefficient with the preset coefficient threshold, if greater than the coefficient threshold, use the particle swarm algorithm to perform global search path on the predicted mean and predicted standard deviation, if less than or equal to the coefficient threshold, use the particle swarm algorithm to perform local optimization exploration on the predicted mean and predicted standard deviation, obtain the current optimal solution, and update the standardized time domain data set based on the current optimal solution, and fine-tune and correct the illegally tampered current power PID controller parameters based on the updated standardized time domain data set;

[0081] Step S5: compare the actual measurement value with the predicted mean and predicted standard deviation and confidence interval output by the GPR model periodically to determine whether to release the warning or keep the warning triggered.

[0082] The specific process of step S1 is as follows:

[0083] The operation data of the thermal power unit primary frequency modulation based on the sensor includes the main steam pressure , the rotating speed , the unit load , the valve opening , the current power PID controller parameters and the power target value key parameters; the current power PID controller parameters include the proportional gain , the integral gain , and the differential gain .

[0084] Based on the operation data of the thermal power unit primary frequency modulation, a time domain data set is constructed.

[0085] The time series samples after normalization, sliding window slicing and cleaning are obtained through the time domain data set.

[0086] The operating data of the thermal power unit under different time points and load conditions are collected. The operating data includes operating conditions covering high load, medium load, low load, frequency disturbance, and load step.

[0087] Appropriate time tolerance windows are used for data points with different acquisition frequencies in the operating data of thermal power units. Processing is performed to allow for a reasonable time tolerance window. Data points collected at different frequencies are considered valid data at the same time.

[0088] All valid data timestamps are uniformly sorted and then distributed according to a set reasonable time tolerance window. The time stamp is iterated along the sorted timeline, and during the iteration, the timestamp is placed within the current unified and reasonable time tolerance window. Data points within the specified range are grouped together; within a reasonable time tolerance window During the grouping process, synchronously identify and remove timestamps that exhibit abnormal drift or exceed the reasonable time tolerance window. The discrete data at the threshold is used to obtain grouped data, and a linear interpolation function is applied to the grouped data to fill a reasonable time tolerance window. For missing values, Min-Max normalization is used to map the data after filling in the missing values ​​to the [0,1] interval, thus obtaining multi-source data;

[0089] Wherein, the linear interpolation function and normalization represent:

[0090] (1);

[0091] (2);

[0092] In the formula, This represents the estimated operating parameters of the thermal power unit at the missing positions within the time tolerance window after processing by the linear interpolation function; that is, the missing operating data values. and These represent the values ​​of two adjacent known data points in the operating data of a thermal power unit; and These represent the operating data of thermal power units. and timestamp; The original operating data of thermal power units to be normalized; This represents the minimum value of the operating data for thermal power units. This represents the maximum value of the operating data for the thermal power unit. This refers to the normalized multi-source data;

[0093] The final standardized time-series samples and multi-source data form a standardized time-domain dataset containing load, main steam pressure, valve opening, and speed. ; This is the standardized time-domain dataset at time t; Let be the main steam pressure at time t; Let be the rotational speed at time t; Let be the unit load at time t; Let be the valve opening at time t; The proportional gain at time t; Let be the integral gain at time t; Let be the differential gain at time t; Let t be the target power value at time t.

[0094] The PyCT-PsLSTM prediction network model processing procedure is as follows:

[0095] The PyCT-PsLSTM prediction network model includes a channel-independent unrolling module, a patching module, a feature extraction module, a channel-temporal collaborative attention module, and a long short-term memory (LSTM) network module.

[0096] The standardized time-domain dataset is fed into the channel independence expansion module for processing, and the output of the channel independence expansion module is obtained.

[0097] The output of the channel independence expansion module is input into the patching module. The patching module divides the time series within each channel in the output of the channel independence expansion module into fixed-length time segments (P), for example, into N segments of length P, thereby obtaining a local time structure block of the form N×P. Each fixed-length time segment P is then mapped to a unified feature dimension through a linear embedding layer to capture the local dependency information and cross-time segment association in the time series of the feature dimension, thus obtaining the output of the patching module.

[0098] The output of the patching module is fed into a dual-path stacked feature extraction module, which consists of two parallel sub-paths: the first parallel sub-path is a local path, composed of two cascaded one-dimensional convolution-max pooling units, used to extract short-term local features from the time series output of the patching module to obtain the output of the first sub-path; the second parallel sub-path is a global path, composed of downsampling-large convolutional kernel-upsampling units, used to extract global trend information over a longer time span from the output of the patching module to obtain the output of the second sub-path; the outputs of the first and second sub-paths are then fused element-wise to form a fused feature tensor. ;

[0099] Channel-temporal collaborative attention module for fusion feature tensors The time dimension and the space dimension are globally averaged pooled GAP, and then linearly transformed and Sigmoid function, to finally generate channel attention weight ;

[0100] The channel dimension and the space dimension of the fusion feature tensor are maximally pooled to obtain time attention weight Element-wise multiplication is performed on , and to finally obtain attention weighted feature output , wherein ⊙ represents element-wise multiplication;

[0101] After being flattened, the is input as an input vector to the long short-term memory network module for processing, to obtain the main steam pressure, valve opening and load as prediction values, denoted as:

[0102] (3);

[0103] (4);

[0104] (5);

[0105] (6);

[0106] (7);

[0107] (8);

[0108] (9);

[0109] In the formula, is the output value of the output gate at time step t; is a sigmoid activation function; is the output gate; is the input vector at the current time step t; is the recurrent weight matrix of the output gate; is the hidden state at time step t-1; is the bias term of the output gate; is the output value of the update gate at time step t; is an activation function; is the weight matrix of the update gate; is the recurrent weight matrix of the update gate; is the bias term of the update gate; is the output value of the input gate at time step t; is an exponential function; is a weight matrix of the input gate; is a recurrent weight matrix of the input gate; is a bias term of the input gate; is an output value of the forget gate at time step t; is an input weight matrix of the forget gate; is a recurrent weight matrix of the forget gate; is a bias term of the forget gate; is the main steam pressure at the current time t; is the main steam pressure at the previous time t-1; is the valve opening at time step t; is the valve opening at time step t-1; is the load at time step t.

[0110] The specific process of step S2 of outputting the main steam pressure, the valve opening and the unit load and calculating the difference with the actual measured value at the current time is as follows:

[0111] When the absolute error between the predicted value and the actual measured value exceeds the set error threshold , it indicates that the current power PID controller parameters are inconsistent between the predicted value and the actual measured value, triggering the inference of the GPR model and giving a warning, and starting the PID controller parameter prediction optimization correction process; when the error threshold is not exceeded , it shows safety. is the average absolute error of the network, is the redundancy error threshold, and the actual measured values are the main steam pressure , the unit load , and the valve opening .

[0112] The specific process of step S3 of outputting the predicted mean and the predicted standard deviation and the confidence interval is as follows:

[0113] When the absolute error between the predicted value and the actual measured value exceeds the set error threshold , the current power PID controller parameters without tampering are selected from the standardized time domain data set, and the real-time process variables and the extended statistical features are extracted from the current power PID controller parameters without tampering as the input of the GPR model training;

[0114] The real-time process variables are the valve opening , the main steam pressure p and the unit load at each sampling time;

[0115] The extended statistical features include the moving average mean( ), the standard deviation of the main steam pressure std(p) and the valve opening change trend( ) calculated by linear fitting;

[0116] The normalized time-domain data set is divided in time sequence in a ratio of 8:2, with the first 80% as the training data set and the remaining 20% data as the test data set;

[0117] The training data set is wherein represents the i-th feature vector, is the true value of the current power PID controller parameter at the i-th corresponding moment; is the number of samples;

[0118] The radial basis kernel is used to capture the nonlinear coupling between the valve opening and the main steam pressure p, the linear kernel is used to represent the steady-state association between the unit load and the valve opening , and the periodic kernel is introduced to describe the day-night regularity of the unit load fluctuation, respectively, to obtain , while satisfies the normalization constraint , and is adjusted through adaptive optimization , specifically:

[0119] (10);

[0120] (11);

[0121] (12);

[0122] (13);

[0123] (14);

[0124] In the formula, is the covariance matrix; is the weight coefficient corresponding to the radial basis kernel; is the radial basis kernel function; is the weight coefficient corresponding to the linear kernel; is the linear kernel function; is the weight coefficient corresponding to the periodic kernel; is the periodic kernel function; is the variance of the observation noise; is the unit matrix; is the weight coefficient of the i-th kernel function; ; is the transpose of the i-th weight vector; For feature vectors; This is the transpose of the j-th weight direction; Logarithmic probability; Logarithmic probability right The partial derivatives; Logarithmic probability right The partial derivatives; for right The partial derivatives; This is the weight vector at the (t+1)th iteration; Let be the weight vector at the t-th iteration; The gradient of the logarithmic probability with respect to the weight vector w; Let be the weight change in the t-th iteration; For learning rate, The momentum coefficient;

[0125] Display the added noise variance term in the covariance matrix K. To eliminate sensor acquisition interference; the complete covariance matrix is ​​defined as follows: ,in The training data matrix is ​​N×6 dimensional;

[0126] Based on the Bayesian framework, the optimal hyperparameter set is set by estimating the maximum marginal likelihood function. The optimized hyperparameter set is used to optimize the GPR model, resulting in a GPR model with the optimized hyperparameter set. The hyperparameter set is constructed based on formulas (10)-(14) and the complete covariance matrix, where... The length scale of the radial base. For the periodic parameters of the periodic kernel, The smoothing coefficient of the periodic kernel. The amplitude coefficient of the periodic kernel. The standard deviation of noise. The standard deviation of the radial basis kernel;

[0127] The optimized hyperparameter set's GPR model is used to calculate the predicted mean and observed noise variance of the PID controller parameters on the test dataset, which are expressed as follows:

[0128] (15);

[0129] (16);

[0130] In the formula, To use the GPR model with the optimized hyperparameter set on the test dataset The predicted mean; is a covariance function; is a test data set, is a true value of the current power PID controller parameter at the corresponding moment; is a prediction standard deviation of the test data set;

[0131] the final output prediction mean contains parameter values and a 95% confidence interval ; is a prediction mean of the proportional term; is a prediction mean of the integral term; is a prediction mean of the differential term.

[0132] The specific process of obtaining the current optimal solution in step S4 is as follows:

[0133] After the correction instruction is issued, firstly, the uncertainty coefficient is calculated based on the prediction mean and the prediction standard deviation , the uncertainty coefficient , and the coefficient threshold is 0.15; when , it indicates that the load working condition is fluctuating sharply under the current running state or the GPR model has not fully learned the load working condition under the current running state, and the global search path is executed; when , it is considered that the prediction is reliable, and local optimization is performed;

[0134] When , the global search path is executed; then, a certain number of particles are randomly generated in the global parameter space as corresponding candidate parameters; is a variable defined in the global parameter space , indicating the value range for executing global search in the case of high uncertainty;

[0135] The sum of the error between the prediction value and the actual measurement value and the error between the prediction mean and the candidate parameter is the fitness function, and the particle swarm algorithm is used to find the optimal particle parameter, until the iteration number or the fitness converges, and the optimal position of the particle is taken as the current optimal solution , and the formula is as follows:

[0136] (17);

[0137] (18);

[0138] (19);

[0139] In the formula, fitness function of the first particle; weight coefficient; error of the first particle predicted value and actual measured value; weight coefficient of the error; error of the first particle predicting current power PID controller parameters and candidate parameters; change amount of the first particle in the first parameter dimension at the first iteration; velocity of the first particle in the first parameter dimension at the first iteration; individual learning factor; random number with value range in [0, 1], adjusting random intensity in learning; optimal position of the first particle in the first parameter dimension; social learning factor; random number with value range in [0, 1], adjusting random intensity in learning; global optimal position of the entire particle group in the first parameter dimension;

[0140] when , local optimization exploration is performed; in the global parameter space , the sum of the error of the predicted value and the actual measured value and the error of the predicted mean value and the candidate parameters is taken as the fitness function, the velocity and position of the particle are constantly updated, until the iteration number meets or the fitness converges, and finally the optimal position of the particle is taken as the current optimal solution ;

[0141] ​​​​​​​​​​​​​​​​​​​​​​​​​Wherein, the current optimal solution The change amount , indicates:

[0142] (20).

[0143] Wherein, the specific process of fine-tuning and correcting the illegally tampered current power PID controller parameters in step S4 is:

[0144] The set confidence bias threshold is 0.1, and the 10% neighborhood of the current optimal solution , the current optimal solution and the feature vector are stored together as a new training sample, and are expanded to The expanded standardized time domain data set is used to fine-tune and correct the hyperparameter set of the illegally tampered current power PID controller parameters by using the maximum marginal likelihood function; A new training sample is added to the standardized time domain data set; The feature vector is composed of real-time process variables and extended statistical characteristics;

[0145] When the change amount of the current optimal solution exceeds the set confidence bias threshold , the change amount of the current optimal solution is marked as an abnormal PID controller parameter;

[0146] For the PID controller parameter in the abnormal state, the candidate parameter with the smallest error is selected from the global parameter space as the optimal control parameter The feature vector and the optimal control parameter are combined into a sample combination, and the sample combination is added to the standardized time domain data set , that is , to obtain an updated standardized time domain data set; and based on the updated standardized time domain data set, the hyperparameter set of the illegally tampered current power PID controller parameters is fine-tuned and corrected by using the maximum marginal likelihood function, to obtain the corrected current power PID controller parameters, and complete the closed-loop self-evolution of the GPR model.

[0147] Wherein, the specific process of step S5 is:

[0148] After executing the current power PID controller parameter correction instruction, the current power PID controller parameter marked as illegally tampered is continuously supervised and dynamically judged by using the current power PID controller parameter optimization correction mechanism, and the specific process is:

[0149] Current optimal solution After the confidence interval deviates, enter the continuous observation stage: when the absolute error between the predicted value and the actual measured value is less than the set error threshold , the current power PID controller parameter adjustment indicating illegal tampering is effective, the pre-warning is released, and when the absolute error between the predicted value and the actual measured value is greater than the set error threshold , enter the next cycle judgment process and keep the warning;

[0150] In the actual operation of thermal power units, slow fouling, blade erosion and sensor drift will cause the input-output characteristics of the PyCT-PsLSTM prediction network model to deviate from the original training distribution; To this end, the micro-batch enhanced training is introduced to the GPR model after training, and the tracking ability of the GPR model to slow degradation factors such as fouling is significantly improved. At the same time, the corrected current power PID controller parameter operation data is updated in real time; Specifically:

[0151] Based on the real-time operation data of the sensor to obtain the primary frequency modulation of the thermal power unit, the GPR model is trained through the real-time operation data, the GPR model is trained based on the corrected current power PID controller parameter operation data and the real-time operation data, and the uncertainty coefficient of the current time step , the uncertainty coefficient of the current time step , indicating that the current power PID controller parameter supporting illegal tampering is adjusted to correct this judgment.

[0152] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for self-correction and self-optimization of a frequency modulation model of a steam turbine of a thermal power unit, characterized in that, The method comprises the following steps: Step S1: constructing a standardized time domain data set; Step S2: inputting the standardized time domain data set into a PyCT-PsLSTM prediction network model for processing, and outputting the main steam pressure, valve opening and unit load, and then performing difference calculation on the actual measured values at the current time, when the error of the difference calculation exceeds a preset error threshold, triggering a warning and starting a PID controller parameter prediction optimization correction mechanism; Step S3: in the prediction optimization correction mechanism, filtering out the current power PID controller parameters without tampering from the standardized time domain data set, inputting the current power PID controller parameters into a GPR model for processing, and outputting a predicted mean value, a predicted standard deviation and a confidence interval; comparing the current power PID controller parameters with the confidence interval, and determining that the current power PID controller parameters are illegally tampered with when the current power PID controller parameters fall outside the confidence interval range, and issuing a correction instruction; Step S4: after the correction instruction is issued, first calculating an uncertainty coefficient based on the predicted mean value and the predicted standard deviation, comparing the uncertainty coefficient with a preset coefficient threshold, and when the uncertainty coefficient is greater than the coefficient threshold, performing global search path optimization on the predicted mean value and the predicted standard deviation using a particle swarm algorithm, and when the uncertainty coefficient is less than or equal to the coefficient threshold, performing local optimization search on the predicted mean value and the predicted standard deviation using the particle swarm algorithm, obtaining a current optimal solution, and updating the standardized time domain data set based on the current optimal solution, and fine-tuning and correcting the illegally tampered current power PID controller parameters based on the updated standardized time domain data set; Step S5: periodically comparing the actual measured values with the predicted mean value and the predicted standard deviation and the confidence interval output by the GPR model to determine whether to release the warning or keep the warning triggered.

2. The method of claim 1, wherein the method further comprises: The specific process of step S1 is: The operation data of the primary frequency modulation of the thermal power unit based on the sensor includes main steam pressure , rotating speed , unit load , valve opening , current power PID controller parameter and power target value key parameter; the current power PID controller parameter includes proportional gain , integral gain , and differential gain ; Based on the operation data of the thermal power unit primary frequency modulation, a time domain data set is constructed; By normalizing, sliding window slicing and cleaning the time domain data set, a standardized time series sample is obtained; The operation data of the thermal power unit under different time points and load conditions are collected, and the operation data includes operation states covering high load, medium load, low load, frequency disturbance and load step; Reasonable time tolerance window is adopted for data points with different collection frequencies in operation data of thermal power generating units The data points with different collection frequencies are regarded as effective data at the same time by processing The data points with different collection frequencies are regarded as effective data at the same time by processing Uniformly sort all valid data's time stamp, set a reasonable time tolerance window Size, along the sorted time axis in turn; in the process of traversal, the time stamp falls into the current uniform reasonable time tolerance window Data points within the range are grouped; in the reasonable time tolerance window In the grouping process, synchronous identification and elimination of time stamp abnormal drift and beyond the reasonable time tolerance window Threshold discrete data, obtain the grouped data, and fill in the reasonable time tolerance window by using linear interpolation function Missing values, map the data with supplemented missing values to the [0, 1] interval through Min-Max normalization to obtain multi-source data; wherein the linear interpolation function and the normalization are represented as: ； ； In the formula, represents the estimated value of the missing position in the time tolerance window after processing by the linear interpolation function, that is, the missing operation data value; and respectively represent the values of the adjacent two known data points in the operation data of the thermal power unit; and respectively represent the timestamps of the adjacent two known data points in the operation data of the thermal power unit; and respectively represent the timestamps of the adjacent two known data points in the operation data of the thermal power unit; is the original operation data of the thermal power unit to be normalized; is the minimum value of the operation data of the thermal power unit; is the maximum value of the operation data of the thermal power unit; is the normalized multi-source data; The final standardized time series samples and multi-source data form a standardized time domain data set containing load, main steam pressure, valve opening, and rotating speed ; is the standardized time domain data set at time t; is the main steam pressure at time t; is the rotating speed at time t; is the unit load at time t; is the valve opening at time t; is the proportional gain at time t; is the integral gain at time t; is the differential gain at time t; is the power target value at time t.

3. The method of claim 2, wherein the method further comprises: The processing process of the PyCT-PsLSTM prediction network model is: The PyCT-PsLSTM prediction network model comprises a channel independence unfolding module, a patching module, a feature extraction module, a channel-time collaborative attention module and a long short-term memory network module; The standardized time domain data set is input into the channel independence unfolding module for processing to obtain the output of the channel independence unfolding module; The output of the channel independence unfolding module is input into the patching module, the patching module divides the time series in each channel of the output of the channel independence unfolding module into fixed-length time segments Patch, and each fixed-length time segment Patch is mapped to a unified feature dimension through a linear embedding layer to capture local dependent information and cross-time segment association in the time series of the feature dimension, and the output of the patching module is obtained; The output of the patching module is input into a double-path stacked feature extraction module, the feature extraction module comprising two structurally parallel sub-paths: a first structurally parallel sub-path is a local path, the local path being composed of two series one-dimensional convolution-maximum pooling units, the output of the first sub-path being obtained by extracting short-term local features in the time sequence in the output of the patching module; a second structurally parallel sub-path is a global path, the global path being composed of a down-sampling-large convolution kernel-up-sampling three units, the output of the second sub-path being obtained by extracting global trend information in the time span in the output of the patching module; the output of the first sub-path and the output of the second sub-path are fused by element-by-element addition to form a fusion feature tensor ​ Channel-temporal collaborative attention module for fusion feature tensors Global average pooling is performed on both the time and spatial dimensions, followed by linear transformation and the sigmoid function to finally generate the channel attention weights. ; The channel dimension and the spatial dimension of the fusion feature tensor are maximum-pooled to obtain a time attention weight , wherein ⊙ represents element-wise multiplication.​​​​​ The After being flattened, the input vector is input into the long short-term memory network module for processing to obtain the main steam pressure, valve opening and load as the predicted values.

4. The method of claim 3, wherein the method further comprises: The specific process of difference calculation between the output main steam pressure, valve opening and unit load in step S2 and the actual measured values at the current time is as follows: When the absolute error between the predicted value and the actual measured value exceeds the set error threshold , the inference of the GPR model is triggered and a warning is given, and the PID controller parameter prediction optimization correction process is started. not exceeding the error threshold the safety is displayed; the average absolute error of the network, the redundancy error threshold, the actual measurement value being the main steam pressure , the unit load , the valve opening .

5. The method of claim 4, wherein the method further comprises: The specific process of outputting the predicted mean and the predicted standard deviation and the confidence interval in step S3 is as follows: When the absolute error between the predicted value and the actual measured value exceeds a set error threshold When the absolute error between the predicted value and the actual measured value exceeds a set error threshold When the absolute error between the predicted value and the actual measured value exceeds a set error threshold Real-time process variable, i.e. valve opening at each sampling instant , main steam pressure p and unit load ; The extended statistical features include the unit load moving average mean( ), the main steam pressure standard deviation std(p), and the valve opening change trend( ) calculated by linear fitting. The standardized time domain data set is proportionally divided into a training data set and a test data set in time sequence; The training data set is wherein represents the i-th feature vector, is the true value of the current power PID controller parameter at the i-th corresponding moment; is the sample number; Capturing valve opening using radial basis kernel Nonlinear coupling with main steam pressure p, characterized by linear kernel Steady-state correlation with valve opening Introducing periodic kernel to describe unit load Diurnal regularity of fluctuations, respectively Simultaneously Satisfying normalization constraints And adjusting through adaptive optimization Specifically: ； ； ； ； ； wherein is a covariance matrix; is a weight coefficient corresponding to a radial basis kernel; is a radial basis kernel function; is a weight coefficient corresponding to a linear kernel; is a linear kernel function; is a weight coefficient corresponding to a periodic kernel; is a periodic kernel function; is a variance of an observation noise; is an identity matrix; is a weight coefficient of an i-th kernel function; ; is a transpose of an i-th weight vector; is an eigenvector; is a transpose of a j-th weight vector; is a log probability; is a log probability is a partial derivative of ; is a log probability is a partial derivative of ; is a is a partial derivative of ; is a weight vector at a t+1-th iteration; is a weight vector at a t-th iteration; is a gradient of a log probability with respect to a weight vector w; is a weight change amount at a t-th iteration; is a learning rate, is a momentum coefficient; Displaying the addition of a noise variance term in the covariance matrix K , constituting the complete covariance matrix definition as , where is the N x 6 dimensional training data matrix; Based on a Bayesian framework, the set of hyperparameters is estimated by maximizing the marginal likelihood function , and the set of hyperparameters is obtained after optimization , and the complete covariance matrix is constructed, wherein is the length scale of the radial basis kernel, is the period parameter of the periodic kernel, is the smoothing coefficient of the periodic kernel, is the amplitude coefficient of the periodic kernel, is the noise standard deviation, is the standard deviation of the radial basis kernel; The GPR model with the optimized hyperparameter set is used to calculate the test data set, and the predicted mean and the observation noise variance of the current power PID controller parameters are obtained, which are represented as follows: ； ； wherein, is the predicted mean of the test dataset using the GPR model with the optimized hyperparameters set; k is the covariance function; is the predicted mean of the test dataset using the GPR model with the optimized hyperparameters set; k is the covariance function; is the test dataset, is the true value of the current power PID controller parameters at the corresponding time instant; is the predicted standard deviation of the test dataset; Final output prediction mean Including parameter values And confidence region , Prediction mean for proportional term; Prediction mean for integral term; Prediction mean for derivative term.

6. The method of claim 5, wherein the method further comprises: The specific process of obtaining the current optimal solution in step S4 is as follows: After issuing the correction command, it is first based on the predicted mean. and the predicted standard deviation Calculate the uncertainty coefficient. The coefficient threshold is 0.15, when When, execute the global search path; when At that time, local optimization exploration is carried out; When , a global search path is executed; then in the global parameter space , a certain number of particles are randomly generated as corresponding to candidate parameters; is a variable defined in the global parameter space ; The sum of the error of the predicted value and the actual measured value and the predicted mean value The sum of the error of the predicted value and the actual measured value and the predicted mean value The sum of the error of the predicted value and the actual measured value and the predicted mean value ； ； In the formula, For the first Fitness function for each particle; for Weighting coefficients; For the first The error between the predicted value and the actual measured value of each particle; for Weighting coefficients; For the first Each particle predicts the error between the current power PID controller parameters and the candidate parameters; For the first During the nth iteration The particle in the first The amount of change in position across each parameter dimension; Inertial weights; For the first During the nth iteration The particle in the first Speed ​​across all parameter dimensions; For individual learning factors; To generate a random number within the range [0,1], adjust... The randomness of learning; For the first The particle in the first The optimal position of a particle in each parameter dimension; For the first During the nth iteration The particle in the first Current position in each parameter dimension; As a social learning factor; To generate a random number within the range [0,1], adjust... The randomness of learning; For the entire particle swarm in the th The global optimal position in each parameter dimension; For the first During the nth iteration The particle in the first Current position in each parameter dimension; when Local optimization is explored; in the global parameter space... In this process, the particle swarm optimization algorithm is used to iteratively optimize the predicted value by considering the error between the predicted value and the actual measured value, and the predicted mean. The sum of errors with the candidate parameters forms the fitness function. The particle's velocity and position are continuously updated until the number of iterations is satisfied or the fitness converges. Finally, the particle's optimal position is taken as the current optimal solution. ; wherein the current optimal solution the change amount denotes: 。 7. The method of claim 6, wherein the method further comprises: The specific process of fine tuning and correcting the illegally tampered current power PID controller parameters in step S4 is as follows: The set confidence bias threshold is 0.1, and the current optimal solution In the neighborhood of the current optimal solution The current optimal solution is stored together with the feature vector as a new training sample, and is expanded to the standardized time domain data set , the hyperparameter set of the illegal tampered current power PID controller parameter is fine-tuned and corrected by using the expanded standardized time domain data set through the maximum marginal likelihood function; A new training sample is added to the standardized time domain data set; The feature vector composed of real-time process variables and extended statistical features; current optimal solution amount of change when the set confidence bias threshold is exceeded the current optimal solution amount of change the PID controller parameters are flagged as abnormal; For the PID controller parameters in the anomaly, the candidate parameters with the minimum error are selected from the global parameter space as the optimal control parameters The feature vectors and the optimal control parameters are combined into a sample combination, and the sample combination is added to the standardized time domain data set The feature vectors and the optimal control parameters are combined into a sample combination, and the sample combination is added to the standardized time domain data set The feature vectors and the optimal control parameters are combined into a sample combination, and the sample combination is added to the standardized time domain data set The feature vectors and the optimal control parameters are combined into a sample combination, and the sample combination is added to the standardized time domain data set The feature vectors and the optimal control parameters are combined into a sample combination, and the sample combination is added to the standardized time domain data set 8. The method of claim 7, wherein the method further comprises: The specific process of step S5 is as follows: After the current power PID controller parameter correction instruction is executed, the current power PID controller parameter optimization correction mechanism is used to continuously supervise and dynamically judge the current power PID controller parameters marked as illegally tampered, and the specific process is as follows: Current optimal solution After the confidence interval is deviated, the sustained observation stage is entered: when the absolute error between the predicted value and the actual measured value is less than the set error threshold , the current power PID controller parameter adjustment effective indicating illegal tampering is released, the early warning is triggered, and when the absolute error between the predicted value and the actual measured value is greater than the set error threshold , the next cycle judgment process is entered, and the warning is maintained; Based on the real-time operation data of the primary frequency modulation of the thermal power unit obtained by the sensor, the GPR model is trained through the real-time operation data, the real-time fitting update is carried out on the corrected current power PID controller parameter operation data and the real-time operation data based on the trained GPR model, and the uncertainty coefficient of the current time step is calculated according to the real-time fitting update , the uncertainty coefficient of the current time step , which indicates the judgment of supporting fine tuning and correction of the current power PID controller parameters subject to illegal tampering.

Images