Power plant heat tracing system intelligent optimization method based on ambient temperature fluctuation

Abstract

The application provides a power plant heat tracing system intelligent optimization method based on environmental temperature fluctuation, relates to the temperature control technical field of the power plant heat tracing system and comprises the following steps: S1, data acquisition and preprocessing: deploying a sensor array and collecting operation data signals of the power plant heat tracing pipeline, preprocessing the operation data signals and outputting a standardized feature vector; S2, model establishment and parameter optimization: constructing a physical-data double-driven heat transfer model and inputting the standardized feature vector to optimize initial parameters.Compared with the prior art, the application has the following beneficial effects: 1. The application can automatically adjust control parameters according to factors such as heat load change, external environmental temperature fluctuation and equipment aging; 2. The application can predict temperature trends in the future short time by constructing a heat transfer dynamic model, adjust the control strategy in advance, avoid temperature out of control, increase heat dissipation or adjust heat source input in advance, and effectively guarantee safe and efficient operation of the system.

Description

Technical Field

[0001] This invention relates to the field of temperature control technology for heat tracing systems in power plants, and in particular to an intelligent optimization method for heat tracing systems in power plants based on ambient temperature fluctuations. Background Technology

[0002] In existing technologies, power plant heat tracing systems are mainly used to prevent media such as water, steam, or other liquids in pipelines from freezing or solidifying in low-temperature environments, ensuring the normal operation of the system. Electric heat tracing temperature control typically employs fixed-parameter control algorithms. While this method can meet basic temperature control requirements to a certain extent, it often exhibits insufficient adaptability when facing complex and variable operating conditions.

[0003] When heat load changes, ambient temperature fluctuates, or equipment aging causes changes in heat transfer efficiency, the fixed-parameter control algorithm cannot adjust the control parameters in time, which may lead to unstable temperature control. Throughout the year, ambient temperature varies greatly. Traditional heat tracing systems still control according to fixed parameters, which may result in excessively high temperatures in summer and excessively low temperatures in winter, failing to meet the system's requirements for accurate and stable temperature control, and thus affecting the normal operation of the system. At the same time, traditional methods only react when the temperature is about to run out of control, making it difficult to predict and adjust the control strategy in advance. This poses a potential threat to the safe and efficient operation of the system. Therefore, the temperature control technology of existing power plant heat tracing systems needs to be improved to adapt to complex operating conditions and improve the system's adaptability, stability, and safety. Summary of the Invention

[0004] To address the shortcomings of the existing technologies, the technical problem to be solved by this invention is to provide an intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations, which can automatically adjust control parameters and predict temperature trends in advance based on real-time operating conditions, thereby reducing temperature fluctuations.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: The present invention provides an intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations, comprising the following steps:

[0006] S1. Data Acquisition and Preprocessing: Deploy sensor arrays and acquire operational data signals from the power plant's heat tracing pipelines. For running data signals Perform preprocessing to output standardized feature vectors;

[0007] S2. Model Building and Parameter Optimization: Construct a physical-data dual-driven heat transfer model and optimize the initial parameters by inputting standardized feature vectors;

[0008] S3. Seasonal Intelligent Control: Construct a differentiated parameter library based on seasonal characteristics, and execute intelligent switching of transitional seasons based on the optimized heat transfer model;

[0009] S4. Fault self-healing and model updating: Establish a fault graded response mechanism and construct an online evolution framework for the heat transfer model.

[0010] In the preferred embodiment, the running data signal is processed in step S1. The specific steps for preprocessing are as follows: A 4-layer Coiflet wavelet decomposition and reconstruction algorithm is used to process the running data signal. Denoising is performed, and the denoised signal is obtained. As input data matrix RobustPCA is used to separate low-rank normal data from sparse outlier data, and spatiotemporal K-nearest neighbor interpolation is used for data repair. The repaired data is then input into PCA to calculate and select principal components Z. Finally, PCA and mutual information are used for weighted fusion to construct a fused feature vector. ;

[0011] The sensor array includes temperature sensors, current sensors, voltage sensors, and flow sensors.

[0012] Running data signals Including pipe temperature Heat tracing power Ambient temperature Medium flow rate .

[0013] In the preferred embodiment, step S1 further includes the following step:

[0014] The rate of temperature change r was calculated by fitting a cubic spline curve using a 10-second sliding window.

[0015] set up , , These are three rate-of-change thresholds, and ( The sampling frequency is dynamically switched based on the relationship between the temperature change rate r and the threshold. The formula is as follows:

[0016] ;

[0017] in, This refers to the sampling frequency.

[0018] In the preferred embodiment, the specific steps in step S2 are as follows:

[0019] S21. Construction of a Physical-Data Dual-Driven Heat Transfer Model: A steady-state heat transfer model is constructed based on the Fourier equation, and the nonlinearity of the medium flow rate is introduced to correct the thermal conductivity. The steady-state temperature distribution is calculated, and the pipe foundation temperature distribution is obtained. Based on the time-series dynamic characteristics captured by LSTM, the normalized historical temperature is input. Heat tracing power Ambient temperature Medium flow rate aging factors And update the hidden state at the current moment through the internal control mechanism. The output is mapped from the hidden state to the temperature fluctuation component via a fully connected layer, generating a dynamic correction term. To correct the steady-state temperature of the steady-state heat transfer model;

[0020] Fourier tube base temperature distribution With LSTM dynamic correction term Overlaying can be used for temperature prediction;

[0021] S22. Bayesian Hyperparameter Optimization: Adjusting the number of nodes in the LSTM hidden layer using Bayesian optimization. Learning rate Regularization coefficient Based on historical evaluation results, a hyperparameter distribution is constructed, and the hyperparameter space is sampled using the Monte Carlo sampling method. The expected improvement value of each sampling point is calculated, and the hyperparameter combination with the largest expected improvement value is selected as the hyperparameter for the next iteration using the expected improvement criterion.

[0022] The optimal combination of hyperparameters is: , , ;

[0023] Add Gaussian noise to LSTM weights Generate 100 sets of prediction results, calculate the mean and standard deviation based on the prediction results to obtain the 95% confidence interval, and evaluate the uncertainty of the model;

[0024] S23. Dynamically update aging compensation factors: Fit historical aging data using linear regression and incrementally update aging factors online. aging factors The embedded temperature prediction formula compensates for the decrease in thermal efficiency caused by equipment aging.

[0025] In the preferred embodiment, in step S21, the hidden state is mapped through a fully connected layer. Dynamic correction term for temperature fluctuation component The calculation formula is as follows:

[0026] ;

[0027] Fourier pipe foundation temperature distribution With LSTM dynamic correction term Temperature prediction output after superposition The formula is expressed as follows:

[0028] ;

[0029] in, It is the weight matrix of the fully connected layer. It is a bias vector. The hidden state output by the LSTM;

[0030] In step S22, the calculation formula for maximizing the selection of hyperparameter combinations by expectation is as follows:

[0031] ;

[0032] The formula for outputting the confidence interval based on the mean and standard deviation of the prediction results is as follows:

[0033] ;

[0034] Among them, the mean The average level of the predicted results is represented by the formula: Standard deviation The formula for measuring the dispersion of prediction results is: ;

[0035] In step S23, the formula for adjusting the attenuation rate in real time based on the new data is as follows:

[0036] ;

[0037] in, It is the learning rate, used to control the step size of updates. These are the currently observed aging factors. It is based on aging factors predicted by previous models;

[0038] The aging factors are embedded in the prediction formula as follows:

[0039] .

[0040] In the preferred embodiment, the specific steps in step S3 are as follows:

[0041] S31. Construction of Seasonal Feature Parameter Database: Collection of Historical Environmental Temperature Data Based on historical ambient temperature data mean and standard deviation Define reference temperature Quantifying seasonal characteristic indices through formulas Based on the quantification results, establish a control parameter library corresponding to different seasons;

[0042] S32. Intelligent Seasonal Switching: Using a trained LSTM model and inputting current and recent ambient temperature data, predict the temperature sequence for the next 72 time steps. Calculate the mean of the time series and the predicted temperature series, and calculate the slope of the temperature trend using a specific formula. A threshold is set, and when a specific condition is met for 72 consecutive time steps, the slope of the temperature trend is considered. Positive and negative triggers switch seasonal modes;

[0043] S33. Dynamic weight allocation: based on seasonal characteristic indices. and real-time temperature deviation Dynamically allocate winter weights and summer weight By combining winter and summer control parameters, the final adjustment step size is obtained. With response time ;

[0044] S34. Predictive Reactive Control: Input variable definition: real-time temperature deviation and rate of temperature change The input variables are fuzzified using Gaussian and triangular membership functions respectively, and the fuzzified input variables are mapped to the corresponding duty cycle adjustment values. The duty cycle adjustment amount is obtained by deblurring the output using the center of gravity method. Combined with seasonal sensitivity coefficient aging factors Taking into account time factors, a heat balance equation is established to dynamically adjust the power of the heat tracing cable, and the adjustment amount is based on the duty cycle of the fuzzy output. Seasonal weighting and response time Dynamically optimize PWM duty cycle;

[0045] aging factors Calculating the dynamic attenuation factor in conjunction with seasonal parameters Comprehensive thermal balance power Dynamic decay factor and real-time temperature deviation Make a final correction to the power.

[0046] In the preferred embodiment, in step S31, the seasonal characteristic index is quantified. The formula is as follows:

[0047] ;

[0048] in, This indicates the relationship between the dispersion of temperature data and the average level. This indicates a non-linear adjustment based on the ratio of the average temperature to the reference temperature.

[0049] The formula for storing parameter settings for different seasons in the parameter library is:

[0050] ;

[0051] Among them, when At that time, the mode is set to winter mode, and the adjustment step size is set. Response time Second;

[0052] when At that time, the summer mode was selected, and the adjustment step size was set. Response time Second;

[0053] In step S32, the temperature trend slope The calculation formula is:

[0054] ;

[0055] The numerator represents the degree of coordinated change of time and temperature from their respective mean values, while the denominator is used for normalization.

[0056] Based on the slope of the temperature trend The formula for determining the switching of positive and negative seasonal patterns is:

[0057] ;

[0058] Among them, the specific condition is when This indicates that the temperature is trending downwards, and the system is switching to winter mode.

[0059] when This indicates that the temperature is trending upwards, and the system is switching to summer mode.

[0060] In step S33, the winter weight is calculated first: Then calculate the summer weights: ;

[0061] Among them, seasonal characteristic index If it rises, then the winter weight will increase. It is on the rise. Represents the absolute value of the temperature deviation, with winter weighting. Follow Increase and grow;

[0062] Substitute the parameters from the parameter library to adjust the step size. and response time Then the step size will be adjusted in the end. With response time The formula is expressed as follows:

[0063] ;

[0064] ;

[0065] In step S34, the duty cycle is adjusted according to the fuzzy output. Seasonal weighting and response time The formula for dynamically optimizing the PWM duty cycle is expressed as follows:

[0066] ;

[0067] in, The current duty cycle, Weighting for winter This refers to the final response time;

[0068] Power after comprehensive thermal balance adjustment Dynamic decay factor and real-time temperature deviation The power is finally corrected using the following formula:

[0069] .

[0070] In the preferred embodiment, the specific steps in step S4 are as follows:

[0071] S41. Sensor Fault Detection and Compensation: Acquiring Historical Temperature Data from Sensors And calculate the historical temperature standard deviation. By triggering Bayesian anomaly probability calculation through redundancy consistency score, faulty sensors are located, and the faulty sensor data is repaired according to spatiotemporal kriging interpolation.

[0072] S42. Heat tracing cable damage detection and switching: Drive backup system power adjustment by determining abnormal harmonic energy ratio;

[0073] Among them, the abnormal harmonic energy ratio includes the abnormal energy ratio of the characteristic frequency band of the current signal detected by frequency domain analysis and the abnormal correlation between power input and temperature response detected by covariance analysis.

[0074] S43. Incremental training for model update: Assign data weights based on prediction error and adjust LSTM weights using an online Bayesian update formula;

[0075] S44, Q-learning reinforcement learning optimization: Constructing a state-action space and reward function, and updating the policy network using the Actor-Critic framework. parameter;

[0076] S45. Dynamic calibration of aging factors: Establishing an exponential decay model for aging factors. By defining state variables, deriving state transition equations, and constructing observation equations, linearized observation equations are obtained based on these equations. Extended Kalman filtering is then used to estimate the aging rate in real time. The estimated aging factors Embedded heat transfer model to correct thermal conductivity .

[0077] In the preferred embodiment, in step S41, the formula for the redundancy consistency score within the sensor array is:

[0078] ;

[0079] in, The standard deviation of historical temperatures. This represents the maximum temperature difference between any two sensors within the sensor array. It is represented as a reference value for normalization.

[0080] If the consistency score is <0.8, an abnormal sensor is identified.

[0081] When an abnormal sensor is detected, a Bayesian anomaly probability calculation is triggered, and expressed by the following formula:

[0082] ;

[0083] in, It is the first The weights of a Gaussian distribution The mean, For variance;

[0084] like Then mark the sensor. This is a fault;

[0085] The formula for repairing faulty sensor data using spatiotemporal kriging interpolation is as follows:

[0086] ;

[0087] in, Location of the current faulty sensor With neighboring sensor locations Euclidean distance, Current time Time measurement with neighboring sensors The difference, The weights are calculated based on the spatiotemporal distance between nearby sensors and faulty sensors. and These are parameters based on spatial and temporal scales;

[0088] In step S42, the specific calculation steps for the abnormal energy proportion of the characteristic frequency band of the current signal detected by frequency domain analysis are as follows:

[0089] For the heat tracing current signal Perform a Fast Fourier Transform to obtain the frequency domain signal. The formula for extracting energy from the 1kHz frequency band is: The total energy is calculated as follows: ;

[0090] in, The frequency range of the energy signal is 1kHz ± 50Hz. Represented as the total energy across the entire frequency domain;

[0091] like If so, the abnormal energy percentage indicates an open circuit;

[0092] The specific calculation steps for detecting anomalies in the correlation between power input and temperature response using covariance analysis are as follows:

[0093] ;

[0094] in, The change in temperature The change in power Temperature change The average value, Power change The average value;

[0095] like If the energy ratio is abnormal, it indicates that the tracing cable has failed.

[0096] When an abnormal energy ratio is determined, the formula for adjusting the power of the standby heating cable according to the power compensation formula is as follows:

[0097] ;

[0098] in, This is the nominal power of the backup heat tracing cable. To compensate for the gain, This is a reference temperature.

[0099] In the preferred embodiment, in step S43, the model-predicted temperature is obtained. and actual measured temperature Temperature predicted based on the model and actual measured temperature Calculate prediction error And based on the prediction error Calculate the variance estimate of the error Through prediction error and variance estimates Define data weights The formula is as follows:

[0100] ;

[0101] Set model parameters It follows a Gaussian distribution, and is then updated based on the verified Bayesian update formula:

[0102] ;

[0103] in, It is the variance of observation noise, the numerator part The denominator represents the correction of the parameter mean by the new data. Used to balance the impact of new data and prior information;

[0104] In step S44, the state space is defined as follows: ;

[0105] in, It's a temperature deviation. It is the rate of temperature change. It is an aging factor. It is a seasonal characteristic index;

[0106] The action space is defined as follows: ;

[0107] in, It is the power adjustment amount of the heat tracing cable. It is the PWM duty cycle adjustment amount;

[0108] Calculating the total power, the reward function is:

[0109] ;

[0110] in, , , These are weighting coefficients. To measure the magnitude of temperature deviation, This represents the total power consumption of the system. Used to control energy consumption Measuring the range of motion To avoid frequent system adjustments;

[0111] Update the policy network using the Actor-Critic framework. Policy Network Based on the current state Output Action The probability distribution is determined, and the policy network is updated according to the following policy gradient formula. parameter:

[0112] ;

[0113] in, Let the action value function be... The state value function;

[0114] In step S45, an exponential decay model of the aging factor is established. The formula is:

[0115] ;

[0116] in, This is the initial value of the aging factor. It is the aging rate. It is a noise term;

[0117] Extended Kalman filter for real-time estimation of aging rate The formula is:

[0118] ;

[0119] in, The Kalman gain is dynamically adjusted based on the current state and observation data.

[0120] Real-time estimated aging factors Embedded heat transfer model to correct thermal conductivity The formula is:

[0121] .

[0122] This invention provides an intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations, which has the following advantages compared with existing technologies:

[0123] First, the present invention can automatically adjust the control parameters according to factors such as changes in heat load, fluctuations in ambient temperature and equipment aging. Compared with the fixed parameter control algorithm in the prior art, it has higher adaptability. In different seasons, it can adjust parameters such as cooling medium flow rate and heating power in a timely manner according to changes in ambient temperature to ensure that the temperature is accurately and stably within the set range.

[0124] Secondly, this invention can predict temperature trends in the near future by constructing a dynamic heat transfer model, allowing for advance adjustment of control strategies and prevention of temperature runaway. Existing technologies typically only react when the temperature is about to run away with the heat, failing to prevent problems in advance. During peak electricity consumption periods, this invention can increase heat dissipation or adjust heat source input in advance, effectively ensuring the safe and efficient operation of the system. Attached Figure Description

[0125] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0126] Figure 1 This is the main flowchart of the present invention;

[0127] Figure 2 This is the present invention. Figure 1 Data preprocessing flowchart;

[0128] Figure 3 This is the present invention. Figure 2 The flowchart of dynamic sampling adjustment in the middle;

[0129] Figure 4 This is the present invention. Figure 1 Flowchart of the heat transfer model construction process;

[0130] Figure 5 This is the present invention. Figure 1 The flowchart of seasonal adaptive control in the text;

[0131] Figure 6 This is the present invention. Figure 1 The flowchart of fault self-healing and update in the process. Detailed Implementation

[0132] To better understand the purpose, structure, and function of this invention, the embodiments and features described herein can be combined with each other without conflict. The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0133] Example 1

[0134] like Figures 1 to 6 As shown in the figure, this embodiment provides an intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations. The specific steps are as follows:

[0135] S1. The specific steps for data acquisition and preprocessing are as follows:

[0136] S11. Sensor Deployment and Data Acquisition: Deploy redundant sensor arrays in critical areas of the pipeline to acquire and output measured operational data signals in real time. Including pipe temperature Heat tracing power Ambient temperature Medium flow rate ;

[0137] The sensor array consists of five sensors in each group, including temperature sensors, current and voltage sensors, and flow sensors.

[0138] For the data collected by each sensor array, taking temperature data as an example, other types of data are processed in the same way.

[0139] The temperature range covered by the temperature sensor is:

[0140] First, calculate the measured values ​​from multiple temperature sensors. The median M, The median avoids the influence of individual extreme values ​​on the overall data. Then, each measurement is calculated. Absolute deviation from the median M: This formula can be used to determine the degree to which each data point deviates from the average level, and then a deviation threshold can be set. ,like Then determine If it is an outlier, remove it.

[0141] For non-outliers, a weighted average method based on standard deviation is used for fusion, and the standard deviation of the non-outliers is calculated, followed by weighting. The fused temperature value was calculated based on the historical stability data of the sensor. ;

[0142] S12. Data Preprocessing: Utilizing Coiflet wavelets to process the acquired running data signals. The data signals are decomposed into four levels. By decomposing at different scales and locations, coefficients of different frequency components are obtained, thereby achieving the goal of extracting low-frequency (approximate) and high-frequency (detail) components.

[0143] By estimating the noise variance to determine a threshold, the detail coefficients are processed to effectively suppress high-frequency noise while preserving the effective signal. The thresholded detail coefficients are then reconstructed to generate the denoised signal. This enables signal restoration and noise reduction.

[0144] S122. Outlier Removal: Robust Principal Component Analysis (RobustPCA) algorithm is used to process the data, and the denoised signal is analyzed. Arranged into a data matrix according to time series , data matrix Decompose it into a low-rank matrix L and a sparse matrix S, and solve for... Under constraints Next, separate normal data from abnormal data;

[0145] The Alternating Direction Multiplier Method (ADMM) is used for iterative solution. and By using an alternating optimization method, the optimal value can be gradually approached. and Matrix, improving the accuracy of outlier separation, especially for sparse matrices. Non-zero elements (i.e., outlier locations) in the data are used... Nearest neighbor interpolation ( Repair is performed by calculating the repair value of outliers using a weighted interpolation method. The corrected outlier values ​​are obtained.

[0146] Calculate the covariance matrix of the denoised and repaired data. The projection matrix is ​​formed by selecting the eigenvectors corresponding to the three largest eigenvalues. Through the principal component projection formula Reduce the data to a 3D principal component space This achieves data dimensionality reduction, reduces data redundancy, and calculates principal components. ( ) and target temperature mutual information We employ a mutual information calculation method based on kernel density estimation to first estimate the joint probability density function. and marginal probability density function , .

[0147] The correlation between principal components and target temperature is measured by calculating mutual information. The principal components that contribute most to temperature prediction are then selected. These selected principal components are then fused with dynamic changes in medium flow rate data, and the medium flow rate data is incorporated into the calculation. Then the final feature vector is: ;

[0148] The above fusion formula combines the selected principal components and flow data into a feature vector containing multiple key information, providing more comprehensive and representative input data for subsequent model training.

[0149] S13. Dynamic sampling frequency setting based on sensor array

[0150] The sensor array calculates the rate of temperature change based on a sliding window.

[0151] Where window size = 10 seconds;

[0152] Let the temperature data within the window be: ;

[0153] in, This represents the number of data points within the window.

[0154] By fitting a cubic spline curve To approximate the temperature change over time, a cubic spline curve is assumed to exist in each subinterval. The expression above is: ;

[0155] The coefficients are determined by satisfying certain boundary conditions and continuity conditions. Then calculate the window point time. Rate of temperature change at: ,Right now: ;

[0156] in, It includes The starting point of the subinterval;

[0157] By fitting cubic spline curves to calculate the rate of temperature change, the trend of temperature change can be captured more accurately, especially when the temperature change is more complex.

[0158] Set three rate of change thresholds , , ( );

[0159] like Then the sampling frequency ;like Then the sampling frequency ;like Then the sampling frequency ;like Then the sampling frequency .

[0160] in, , , , , , , .

[0161] S2. Selection and Optimization of Heat Transfer Dynamic Model

[0162] S21, Construction of a Dual-Driven Model Based on Physical Data

[0163] S211, Physically driven layer (Fourier heat conduction equation)

[0164] Calculating the pipe foundation temperature distribution based on the Fourier equation while neglecting the time term. The specific formula is as follows: ;

[0165] in, For Hamiltonian operators, in Cartesian coordinates This equation represents the conduction of heat in a medium under steady state, where the divergence of heat flux density is 0, meaning that heat does not accumulate in space.

[0166] In practical applications, the finite element method is used for complex pipe geometries.

[0167] Let the pipeline region be discretized as Each unit, for each unit According to Galerkin's method, the Fourier equation is: Multiply by the test function Integrating over the unit, we obtain the following: ;

[0168] Assemble all the equations from each unit to form a system of linear equations: ;

[0169] in, It is the stiffness matrix. It is a node temperature vector. It is the load vector.

[0170] Solving the above system of linear equations yields the temperature distribution of the pipe foundation. .

[0171] S212, Data-Driven Layer (LSTM Dynamic Correction)

[0172] The input sequence contains historical temperatures Heat tracing power Ambient temperature Medium flow rate aging factors To eliminate dimensional differences and improve model convergence speed, the input features are normalized using the following formula: ;

[0173] in, Representing different input features (e.g.) , wait), It is a feature The mean, It is a feature The standard deviation.

[0174] Next, calculate the features. The formula for the mean is as follows: ;

[0175] in, It is the sample size;

[0176] Calculation features Standard deviation: ;

[0177] The LSTM hidden state update formula is expressed as follows: ;

[0178] in, It is the normalized input sequence. It is the hidden state from the previous moment. It is the parameter set of LSTM, including the weight matrix and the bias vector.

[0179] The LSTM cell internally uses input gates Forgotten Gate Output gate and memory unit To control the flow of information, by calculating the input gate Forgotten Gate Output gate and memory unit The final hidden state is: ;

[0180] in, It is the sigmoid function. It is element-wise multiplication.

[0181] Using the above formula, LSTM can capture the dynamic characteristics of time series and output the hidden state at the current time step. It effectively handles long-term dependencies in time series.

[0182] The hidden state is mapped to the temperature fluctuation component through a fully connected layer, as shown in the following formula: ;

[0183] in, It is the weight matrix of the fully connected layer. It is the bias vector.

[0184] The above formula will output the hidden state of the LSTM. Mapped to temperature fluctuation space, generating dynamic correction terms. It is used to correct the steady-state temperature of the physical model in order to reflect the dynamic changes in the actual heat transfer process.

[0185] S22, Model Training and Optimization

[0186] S221. Hybrid Model Fusion: Temperature prediction is performed by combining the steady-state solution of the physical model with the dynamic correction term of LSTM, as shown in the following formula: ;

[0187] Among them, the steady-state temperature calculated by the physical model Provides a basic temperature distribution, LSTM dynamic correction term This takes into account the dynamic factors in the heat transfer process, and the combination of the two forms a complete temperature prediction. This dual-drive approach using physical data enhances the model's generalization ability, enabling it to better adapt to different working conditions.

[0188] S222, Bayesian Hyperparameter Optimization (Treestructured Parzen Estimator, TPE)

[0189] To improve efficiency, a portion of the hyperparameter space is selected as the search space;

[0190] The search space includes the number of LSTM hidden layer nodes. Learning rate Regularization coefficient .

[0191] Among them, the number of hidden layer nodes The learning rate has a value range of [64, 256]. The range of values ​​is Regularization coefficient The range of values ​​is Its function is to control model complexity and prevent overfitting;

[0192] TPE probabilistic modeling: Based on historical evaluation results, a hyperparameter distribution is constructed, with the following formula: ;

[0193] in, Represents the set of hyperparameters. This represents historical assessment data. It is the likelihood function, representing the likelihood of hyperparameters. The following observed data The probability, It is the prior distribution, representing the distribution with respect to hyperparameters. The initial understanding.

[0194] By using Bayesian inference, the prior distribution is updated with historical data to obtain the posterior distribution. Therefore, a high-potential combination of hyperparameters is selected. In the above calculations, a Gaussian process is used to estimate the likelihood function: ;

[0195] Historical evaluation data for , ;

[0196] in, It is a combination of hyperparameters. These are the corresponding model evaluation metrics.

[0197] Gaussian processes assume the data follows a Gaussian distribution. A kernel function is used to measure the similarity between hyperparameters; a radial basis function kernel is employed. ;

[0198] in, It is the signal variance. It is the length scale parameter. Through calculation and optimization of the Gaussian process, the estimated value of the likelihood function is obtained, and then the posterior distribution is calculated. In this process, the values ​​of the hyperparameters all fall within... , , Within the range.

[0199] Maximizing Expected Improvement (EI): Select the hyperparameters that maximize the expected improvement, as shown in the formula: ;

[0200] in, In hyperparameters The following model evaluation metrics It is currently the optimal model evaluation metric. This formula balances exploration and exploitation, searching for hyperparameter combinations in the hyperparameter space that may bring greater improvements and accelerating model convergence.

[0201] In actual calculations, the Monte Carlo sampling method is used to approximate the expected improvement. This involves sampling the hyperparameter space, calculating the expected improvement value for each sampling point, and then selecting the hyperparameter with the largest expected improvement value as the hyperparameter for the next iteration.

[0202] S223, Uncertainty Assessment (Monte Carlo Sampling)

[0203] Weight perturbation sampling: Gaussian noise is added to the LSTM weights. Generate 100 sets of prediction results using the following formula: ;

[0204] in, It is the first noise vector of the next sample This indicates that with the weights after adding noise, the LSTM performs better on the input. The output.

[0205] The above formula simulates the uncertainty of model parameters. Considering that the model weights have a certain degree of randomness in actual training, different weights may lead to different prediction results.

[0206] The formulas for the mean and standard deviation of statistical prediction results are:

[0207] ; ;

[0208] Then output the 95% confidence interval: ;

[0209] Among them, the mean The average value and standard deviation of the predicted results represent the average value and standard deviation of the predicted results. The degree of dispersion of the prediction results is measured. Similarly, the optimized values ​​of hyperparameters will indirectly affect the degree of dispersion and mean of the prediction results, thereby changing the range of the confidence interval and making the uncertainty assessment results more consistent with the actual performance of the model under specific hyperparameter settings.

[0210] S224, Dynamic Update of Aging Compensation Factor

[0211] Linear regression fitting of aging history data: Let aging factor be... It decays linearly over time, as shown in the formula: ;

[0212] Fitting by least squares method (Annual decay rate). Collect historical data on the changes of aging factors over time. , Define the objective function: ;

[0213] right about Take the derivative and set it to 0, that is: ;

[0214] Solving the above equation yields the following results: The estimated value is used to determine the decay relationship of the aging factor over time. The overall performance of the model is affected by the hyperparameter optimization results, and the calculation of the aging compensation factor depends on the model's ability to learn and fit the data. Therefore, changes in the hyperparameter values ​​within the specified range will indirectly affect the model's performance. The fitting results make the calculation of the aging factor more closely match the actual performance of the model.

[0215] Online incremental update: The decay rate is adjusted daily based on new data, using the following formula: ;

[0216] in, It is the learning rate, used to control the step size of updates. These are the currently observed aging factors. It is based on the aging factor predicted by previous models.

[0217] Temperature prediction correction: The aging factor is embedded in the prediction formula, which is as follows: .

[0218] In practical applications, thermal efficiency decreases as equipment ages. This formula is used to correct temperature predictions, improving their accuracy. This is due to aging factors. The calculation is indirectly affected by hyperparameter optimization, so the final temperature prediction correction result will also change accordingly, making the predicted value more consistent with the actual situation.

[0219] S3, Seasonal Intelligent Control Strategy

[0220] S31. Construction of Differentiated Parameter Library

[0221] S311, Quantification of Seasonal Characteristics

[0222] First, collect historical ambient temperature data. And calculate historical ambient temperature mean of the data Next, the standard deviation of historical ambient temperatures is calculated as follows: ;

[0223] The above formula is used to measure the dispersion of historical ambient temperature data.

[0224] Then, define the reference temperature. Through the formula: Quantify seasonal characteristics.

[0225] in, This reflects the relationship between the dispersion of temperature data and the average level. The nonlinear adjustment is then made based on the ratio of the average temperature to the reference temperature.

[0226] In winter, ambient temperatures are low and fluctuate significantly. Smaller Relatively large, making ;

[0227] Summer temperatures are relatively high and fluctuate little. Larger Relatively small, resulting in This allows for the quantitative differentiation of seasonal characteristics.

[0228] S312, Parameter Pre-storage Rules

[0229] Based on the quantified seasonal characteristic index To pre-store control parameters for different seasons in order to build a parameter library.

[0230] when When the system is in winter mode, adjust the step size accordingly. Response time Seconds. In winter, when temperatures are low, a larger adjustment step size can respond more quickly to temperature changes, shortening the response time and compensating for heat loss in a timely manner, thus ensuring stable system operation.

[0231] when At that time, it was determined to be summer mode. , Seconds. In summer, when temperatures are high, smaller adjustment steps can prevent over-adjustment, while longer response times help prevent frequent system adjustments and improve system stability.

[0232] The formula for storing parameter settings for different seasons is: ;

[0233] The corresponding parameters can be quickly called up based on seasonal characteristics during the control process.

[0234] S32, Intelligent Seasonal Switching

[0235] S321, Temperature Trend Forecast

[0236] Using a trained LSTM model, input current and recent ambient temperature data, and output the temperature sequence for the next 3 days (72 time steps, with each hour as one time step). .

[0237] Calculate the mean of a time series and the mean of the temperature series These two formulas are used to obtain the average level of the time series and the predicted temperature series, respectively.

[0238] Temperature trend slope The calculation formula is: ;

[0239] The numerator reflects the degree of coordinated change between time and temperature deviating from their respective mean values, while the denominator is used for normalization. By calculating this slope, the future temperature trend can be accurately predicted.

[0240] S322, Switching Conditions

[0241] Set a threshold, and when the condition is met for 3 consecutive days (i.e., 72 time steps) When a seasonal mode switch is triggered, the seasonal mode is determined based on the sign of the temperature trend slope, using the following formula:

[0242] ;

[0243] when This indicates a significant downward trend in temperature, signaling a switch to winter mode; when This indicates a significant upward trend in temperature, prompting a switch to summer mode to achieve intelligent seasonal transition.

[0244] S33, Dynamic Weight Allocation

[0245] S331, Weight Calculation: Based on seasonal characteristic index and real-time temperature deviation Calculate winter weights and summer weight First, calculate the winter weight: ;

[0246] in, The larger the value, the closer the season is to winter, and the greater its impact on the weight. This represents the absolute value of the temperature deviation. The larger the temperature deviation, the greater the weight of winter.

[0247] Then, using the formula for calculating the summer weight: .

[0248] S332. Control Parameter Fusion: Using the calculated weights, the control parameters for winter and summer are fused to obtain the final adjustment step size and response time. The final adjustment step size formula is: ;

[0249] Among them, 2 is the step size adjustment in winter, and 1 is the step size adjustment in summer.

[0250] Based on the weighted allocation, as it gets closer to winter... The step size is relatively large, and the final adjustment step size is closer to that used in winter to achieve a rapid response; as summer approaches, The adjustment step size is relatively large, closer to the summer adjustment step size, to avoid over-adjustment. The final response time formula is: ;

[0251] A balance between fast response time in winter and stable response time in summer is achieved through weight allocation.

[0252] S34, Predictive Reactive Control

[0253] S341. Fuzzy Logic Controller Design: Input Variable Definition, Real-time Temperature Deviation : ;

[0254] in, It is the set target temperature. This is the actual measured temperature. This formula is used to calculate the difference between the actual temperature and the target temperature, providing a basis for subsequent control.

[0255] Rate of temperature change for: ;

[0256] in, It is the temperature at the current moment. It was the temperature a moment ago. It is a time interval; this formula is used to measure how quickly temperature changes over time.

[0257] For actual temperature deviation Gaussian membership functions are used: , ;

[0258] in, Represents actual temperature deviation The value of , It is the center value of the membership function. It's the standard deviation. Different The values ​​correspond to different fuzzy sets, such as "negative large" (NB), "negative small" (NS), "zero" (ZO), "positive small" (PS), and "positive large" (PB).

[0259] By adjusting and The value can be defined according to the actual situation, specifying the coverage and membership degree of each fuzzy set for temperature deviation, thereby achieving fuzzification processing of temperature deviation, and addressing the temperature change rate. Using triangular membership functions: , ;

[0260] in, Represents the rate of temperature change The value of , It is the vertex value of the membership function. It refers to the bottom width. Different... The value corresponds to three fuzzy sets: "negative" (Neg), "zero" (Zero), and "positive" (Pos). This can be achieved by adjusting... and The division and membership degree of each fuzzy set for the rate of temperature change can be determined according to actual needs.

[0261] A fuzzy rule base was developed based on actual operational experience and control objectives.

[0262] like and Then the duty cycle adjustment amount This means that when the temperature deviation is large and the rate of temperature change is positive, the duty cycle needs to be increased significantly.

[0263] like and Then the duty cycle adjustment amount That is, when the temperature deviation is small and the rate of temperature change is negative, the duty cycle should be appropriately reduced.

[0264] Deblurring (center-of-gravity method) converts the blurred output into a precise duty cycle adjustment using a formula, as shown below: ;

[0265] in, It is the first The membership degree of a rule. It is the first The duty cycle adjustment amount corresponding to each rule. This represents the total number of rules. Based on the center-of-gravity principle, this formula comprehensively considers the influence of all rules to calculate a precise duty cycle adjustment amount, which is used for subsequent adjustments to the PWM duty cycle.

[0266] S342. Energy Balance Formula and Power Compensation: Based on the law of conservation of energy, considering seasonal sensitivity coefficients. aging factors The power of the heat tracing cable is dynamically adjusted based on time factors using a formula, as shown below: ;

[0267] in, It is the base power and the seasonal sensitivity coefficient. The value is 0.15 in winter and 0.05 in summer, reflecting the sensitivity to temperature changes in different seasons; It is the aging effect coefficient. It is the current time. It is the initial time. Used to describe the cumulative effects of equipment aging over time; as time goes on, the impact of aging on power gradually increases.

[0268] Duty cycle adjustment in combination with fuzzy output With seasonal weights, the PWM duty cycle is optimized using a formula, expressed as follows: ;

[0269] in, This is the current duty cycle. It is the winter weight. This is the final response time. In winter, Larger Smaller A larger value allows for a greater range of duty cycle adjustments, enabling a faster response; in summer, Smaller The duty cycle is relatively small, which ensures system stability.

[0270] S343, Aging and Seasonal Coupling Compensation

[0271] Dynamic decay factor: aging factor Coupled with seasonal parameters, the dynamic attenuation factor is calculated using the following formula: ;

[0272] In winter, , The effects of aging factors are amplified;

[0273] In summer, , The impact of aging factors is reduced, enabling dynamic adjustment of aging effects with the seasons. The final power correction considers the power after comprehensive thermal balance adjustment. Dynamic decay factor and real-time temperature deviation The power is finally corrected using the following formula: ;

[0274] S4, Self-healing

[0275] S41. Sensor Fault Detection and Compensation: First, collect historical temperature data of the sensor over a period of time (e.g., the past 24 hours). Calculate its historical temperature standard deviation: ;

[0276] in, It is the amount of historical temperature data. It is the average value of historical temperature data, and this formula is used to measure the dispersion of historical temperature data.

[0277] The temperature data collected by the sensor array at the current moment ( (Indicating different sensor numbers), calculate the consistency index within the sensor array: ;

[0278] in, This represents the maximum temperature difference between any two sensors within the sensor array. It is a reference value used for normalization. The formula evaluates the consistency of the sensor array by comparing the temperature differences between sensors with historical temperature fluctuations.

[0279] If the consistency score is less than 0.8, it indicates that the temperature difference between the sensors is too large and exceeds the normal fluctuation range, indicating that there is an abnormal sensor.

[0280] Assume the sensor data follows a Gaussian mixture model (GMM), i.e., the formula is: ;

[0281] in, It represents the number of Gaussian distributions in the Gaussian mixture model. It is the first The weights are distributed according to a Gaussian distribution, satisfying... , The mean is variance is The Gaussian probability density function.

[0282] Model parameters are estimated using the Expectation-Maximization (EM) algorithm. , and .

[0283] In the E-step of the EM algorithm, each data point is calculated. Belongs to the The probabilities of a Gaussian distribution are: ;

[0284] In step M, the updated parameters are: ; ; ;

[0285] Repeat the E-step and M-step until the parameters converge.

[0286] The probability of an anomaly is calculated as follows: ;

[0287] This formula represents a sensor measurement value. The maximum probability that it does not belong to any Gaussian distribution, if This indicates that the sensor's measured values ​​differ significantly from the normal data distribution, thus marking the sensor. This is a malfunction.

[0288] Repairing faulty sensor data based on spatiotemporal kriging.

[0289] First, determine the set of nearby sensors. The proximity sensors are selected based on spatial distance and time sequence;

[0290] Assume the sensor is located in two-dimensional space. This indicates that, for spatial distance, the current location of the faulty sensor is calculated. With neighboring sensor locations The Euclidean distance is: ;

[0291] For time distance, calculate the current time. Time measurement with neighboring sensors The difference .

[0292] Repair using the following formula: ;

[0293] in, The weights are calculated based on the spatiotemporal distance between nearby sensors and faulty sensors. and These are spatial and temporal scale parameters used to adjust the scope of influence of spatiotemporal correlation.

[0294] Weight It can be calculated using the inverse distance weighting method, i.e., the formula is: ;

[0295] The formula utilizes the spatiotemporal correlation of neighboring sensors to repair data, making the repaired data more consistent with the actual situation.

[0296] S42. Detection and switching of heat tracing cable damage:

[0297] Harmonic energy ratio analysis: analysis of the current signal of the heat tracing cable. Perform a Fast Fourier Transform (FFT) to obtain the frequency domain signal. .

[0298] According to the Fourier transform formula: ;

[0299] Implemented on a computer using the Discrete Fourier Transform (DFT), namely: , ;

[0300] in, It represents the number of sampling points.

[0301] Extracting energy from the 1kHz frequency band: The total energy is calculated as follows: ;

[0302] in, The frequency range of the energy signal is 1kHz ± 50Hz. It represents the total energy across the entire frequency domain.

[0303] like This indicates that the energy proportion in the 1kHz frequency band is relatively large, exceeding the normal range, and is determined to be an open circuit.

[0304] Power-temperature dynamic correlation test:

[0305] The formula for calculating the change in temperature is: The formula for the change in power is: ;

[0306] in, This is a time interval (e.g., 1 minute). Calculate the average value of the temperature change. and the average value of power change .

[0307] The dynamic correlation coefficient is calculated using the following formula: ;

[0308] like This indicates a strong negative correlation between power and temperature. If the power is normal but the temperature drops abnormally, the heat tracing cable is confirmed to be faulty. Once the heat tracing cable is confirmed to be faulty, the backup heat tracing cable is activated.

[0309] The temperature loss is calculated as follows: ;

[0310] in, It is the set target temperature. This is the actual temperature measured at present.

[0311] Adjust the standby heating cable power according to the following power compensation formula: ;

[0312] in, This is the nominal power of the backup heat tracing cable. To compensate for the gain, it is generally set to 1.2. This is a reference temperature, including the average temperature during normal operation. The formula compensates for the standby heating cable power based on temperature loss, ensuring the system can maintain normal operation.

[0313] S5, Closed-Loop Model Update and Optimization

[0314] S51, Incremental Training (Online Learning): For model prediction of temperature and actual measured temperature The prediction error is calculated as follows: The variance estimate of the calculation error is: ;

[0315] in, It represents the number of recent prediction error data. It is the average value of the prediction error.

[0316] Define data weights: ;

[0317] Among them, prediction error Larger samples, their weights The smaller the noise level, the less the impact of noise samples on the model during model updates, ensuring more accurate model updates.

[0318] Set model parameters Following a Gaussian distribution, the initial prior distribution of the parameters is set as follows: ;

[0319] in, It is the prior mean. It is the a priori covariance.

[0320] For new sample data ( It is the input feature matrix. (This is the corresponding output vector), and it is updated according to the following online Bayesian update formula: ;

[0321] in, This is the variance of the observation noise. In this formula, the numerator part... The denominator represents the correction of the parameter mean by the new data. This is used to balance the influence of new data and prior information. By continuously updating with new data, the LSTM weights are dynamically adjusted, allowing the model to adapt to shifts in data distribution.

[0322] S52, Q-learning reinforcement learning optimization

[0323] The state space defines the state as follows: ;

[0324] in, It's a temperature deviation. It is the rate of temperature change. It is an aging factor. These are seasonal characteristic indices. These state variables comprehensively reflect the current operating state of the system.

[0325] The action space is defined as follows: ;

[0326] in, It is the power adjustment amount of the heat tracing cable. It is the PWM duty cycle adjustment amount, and the action space represents the control operations that the system can take.

[0327] The total power is calculated as follows: ;

[0328] in, This is the current power of the heat tracing cable.

[0329] The reward function is then: ;

[0330] in, , , These are weighting coefficients used to balance different optimization objectives. To measure the magnitude of temperature deviation, The larger the value, the greater the importance placed on temperature stability; This represents the total power consumption of the system. Used to control energy consumption; Measuring the range of motion This is to avoid frequent system adjustments.

[0331] Update the policy network using the Actor-Critic framework. First, policy network Based on the current state Output Action The probability distribution is implemented using a neural network, with the following parameters: .

[0332] The action value function is calculated as follows: The state-value function is: ;

[0333] Indicates the state Take action below The expected cumulative reward can be calculated iteratively using the following Bellman equation: ;

[0334] in, It is a discount factor, usually set to 0.9, used to balance current rewards and future rewards.

[0335] Indicates the state The value can be obtained through all possible actions. The weighted average is obtained as follows: ;

[0336] Update the policy network parameters according to the following policy gradient formula: ;

[0337] This formula calculates the gradient of the policy network parameters, causing the policy network to update in the direction of maximizing cumulative reward, thus avoiding local optima.

[0338] S53, Dynamic calibration of aging factors

[0339] Assume that the aging factor follows an exponential decay model: ;

[0340] in, This is the initial value of the aging factor. It is the aging rate. It is a noise term, with a mean of 0 and a variance of . The Gaussian distribution.

[0341] Aging rate is estimated in real time using an extended Kalman filter (EKF). The specific steps are as follows:

[0342] First, the aging factor is linearized according to an exponential decay model, and the state variables are set. Then the state transition equation is: The observation equation is: ;

[0343] For the observation equation at the current estimate Performing a Taylor expansion at the given point yields the following linearized observation equation: ;

[0344] in, yes The estimated value.

[0345] Calculate the Kalman gain: ;

[0346] in, It is the state estimation error covariance. It is the Jacobian matrix of the linearized observation equation. If the observation noise covariance is given, then: ; ;

[0347] The aging rate is updated using the following formula. : ;

[0348] in, It dynamically adjusts based on the current status and observation data to achieve real-time tracking of the aging rate.

[0349] Real-time estimated aging factors Embed the heat transfer model and correct the thermal conductivity.

[0350] Assume the original thermal conductivity is The corrected effective thermal conductivity is obtained using the following formula. : ;

[0351] Example 2

[0352] In conjunction with Embodiment 1, this application provides an electronic device. This electronic device may include:

[0353] Memory, processor, and computer programs stored in memory and capable of running on the processor.

[0354] When the processor executes the program, it implements the intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations provided in the above embodiments.

[0355] Furthermore, electronic devices also include:

[0356] A communication interface used for communication between the memory and the processor.

[0357] Memory is used to store computer programs that can run on the processor.

[0358] The memory may include high-speed RAM, and may also include non-volatile memory, such as at least one disk drive.

[0359] If the memory, processor, and communication interface are implemented independently, they can be interconnected via a bus to communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0360] A processor may include one or more processing units, such as an application processor (AP), an application-specific integrated circuit (ASIC), a modem processor, a central processing unit (CPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors. The controller may serve as a central nervous system and command center. The controller generates operation control signals based on instruction opcodes and timing signals to control instruction fetching and execution. The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has recently used or that is used repeatedly. If the processor needs to reuse the instruction or data, it can directly retrieve it from the memory. This avoids repeated access, reduces processor waiting time, and thus improves system efficiency.

[0361] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations.

[0362] This application also provides a computer program product that can run computer instructions. When these computer instructions are executed by a processor, they implement the above-mentioned intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations.

[0363] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for intelligent optimization of power plant heat tracing systems based on ambient temperature fluctuations, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Deploy sensor arrays and acquire operational data signals from the power plant's heat tracing pipelines. For running data signals Perform preprocessing to output standardized feature vectors; S2. Model Building and Parameter Optimization: Construct a physical-data dual-driven heat transfer model and optimize the initial parameters by inputting standardized feature vectors; S3. Seasonal Intelligent Control: Construct a differentiated parameter library based on seasonal characteristics, and execute intelligent switching of transitional seasons based on the optimized heat transfer model; The specific steps in step S3 are as follows: S31. Construction of Seasonal Feature Parameter Database: Collection of Historical Environmental Temperature Data Based on historical ambient temperature data mean and standard deviation Define reference temperature Quantifying seasonal characteristic indices through formulas Based on the quantification results, establish a control parameter library corresponding to different seasons; S32. Intelligent Seasonal Switching: Using a trained LSTM model and inputting current and recent ambient temperature data, predict the temperature sequence for the next 72 time steps. Calculate the mean of the time series and the predicted temperature series, and calculate the slope of the temperature trend using a specific formula. A threshold is set, and when a specific condition is met for 72 consecutive time steps, the slope of the temperature trend is considered. Positive and negative triggers switch between seasonal modes; S33. Dynamic weight allocation: based on seasonal characteristic indices. and real-time temperature deviation Dynamically allocate winter weights and summer weight By combining winter and summer control parameters, the final adjustment step size is obtained. With response time ; S34. Predictive Reactive Control: Input variable definition: real-time temperature deviation and rate of temperature change The input variables are fuzzified using Gaussian and triangular membership functions respectively, and the fuzzified input variables are mapped to the corresponding duty cycle adjustment values. The duty cycle adjustment amount is obtained by deblurring the output using the center of gravity method. Combined with seasonal sensitivity coefficient aging factors Taking into account time factors, a heat balance equation is established to dynamically adjust the power of the heat tracing cable, and the adjustment amount is based on the duty cycle of the fuzzy output. Seasonal weighting and response time Dynamically optimize PWM duty cycle; aging factors Calculating the dynamic attenuation factor in conjunction with seasonal parameters Comprehensive thermal balance power Dynamic decay factor and real-time temperature deviation Make a final correction to the power; S4. Fault self-healing and model updating: Establish a fault graded response mechanism and construct an online evolution framework for the heat transfer model.

2. The intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations according to claim 1, characterized in that, In step S1, the running data signal The specific steps for preprocessing are as follows: A 4-layer Coiflet wavelet decomposition and reconstruction algorithm is used to process the running data signal. Denoising is performed, and the denoised signal is obtained. As input data matrix RobustPCA is used to separate low-rank normal data from sparse outlier data, and spatiotemporal K-nearest neighbor interpolation is used for data repair. The repaired data is then input into PCA to calculate and select principal components Z. Finally, PCA and mutual information are used for weighted fusion to construct a fused feature vector. ; The sensor array includes temperature sensors, current sensors, voltage sensors, and flow sensors. Running data signals Including pipe temperature Heat tracing power Ambient temperature Medium flow rate .

3. The intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations according to claim 2, characterized in that, Step S1 also includes the following steps: The rate of temperature change r was calculated by fitting a cubic spline curve using a 10-second sliding window. set up , , These are three threshold rates of change, and The sampling frequency is dynamically switched based on the relationship between the temperature change rate r and the threshold. The formula is as follows: ； in, This refers to the sampling frequency.

4. The intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations according to claim 1, characterized in that, The specific steps in step S2 are as follows: S21. Construction of a Physical-Data Dual-Driven Heat Transfer Model: A steady-state heat transfer model is constructed based on the Fourier equation, and the nonlinearity of the medium flow rate is introduced to correct the thermal conductivity. The steady-state temperature distribution is calculated, and the pipe foundation temperature distribution is obtained. Based on the time-series dynamic characteristics captured by LSTM, the normalized historical temperature is input. Heat tracing power Ambient temperature Medium flow rate aging factors And update the hidden state at the current moment through the internal control mechanism. The output is mapped from the hidden state to the temperature fluctuation component via a fully connected layer, generating a dynamic correction term. To correct the steady-state temperature of the steady-state heat transfer model; Fourier tube base temperature distribution With LSTM dynamic correction term Overlaying can be used for temperature prediction; S22. Bayesian Hyperparameter Optimization: Adjusting the number of nodes in the LSTM hidden layer using Bayesian optimization. Learning rate Regularization coefficient Based on historical evaluation results, a hyperparameter distribution is constructed, and the hyperparameter space is sampled using the Monte Carlo sampling method. The expected improvement value of each sampling point is calculated, and the hyperparameter combination with the largest expected improvement value is selected as the hyperparameter for the next iteration using the expected improvement criterion. The optimal combination of hyperparameters is: , , ; Add Gaussian noise to LSTM weights Generate 100 sets of prediction results, calculate the mean and standard deviation based on the prediction results to obtain the 95% confidence interval, and evaluate the uncertainty of the model; S23. Dynamically update aging compensation factors: Fit historical aging data using linear regression and incrementally update aging factors online. aging factors The embedded temperature prediction formula compensates for the decrease in thermal efficiency caused by equipment aging.

5. The intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations according to claim 4, characterized in that, In step S21, the hidden state is mapped through the fully connected layer. Dynamic correction term for temperature fluctuation component The calculation formula is as follows: ； Fourier pipe foundation temperature distribution With LSTM dynamic correction term Temperature prediction output after superposition The formula is expressed as follows: ； in, It is the weight matrix of the fully connected layer. It is a bias vector. The hidden state output by the LSTM; In step S22, the calculation formula for maximizing the selection of hyperparameter combinations by expectation is as follows: ； The formula for outputting the confidence interval based on the mean and standard deviation of the prediction results is as follows: ； Among them, the mean The average level of the predicted results is represented by the formula: Standard deviation The formula for measuring the dispersion of prediction results is: ; In step S23, the formula for adjusting the attenuation rate in real time based on the new data is as follows: ； in, It is the learning rate, used to control the step size of updates. These are the currently observed aging factors. It is based on aging factors predicted by previous models; The aging factors are embedded in the prediction formula as follows: 。 6. The intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations according to claim 1, characterized in that, In step S31, the seasonal characteristic index is quantified. The formula is as follows: ； in, This indicates the relationship between the dispersion of temperature data and the average level. This indicates a non-linear adjustment based on the ratio of the average temperature to the reference temperature. The formula for storing parameter settings for different seasons in the parameter library is as follows: ； Among them, when At that time, the mode is set to winter mode, and the adjustment step size is set. Response time Second; when At that time, the summer mode was selected, and the adjustment step size was set. Response time Second; In step S32, the temperature trend slope The calculation formula is: ； The numerator represents the degree of coordinated change of time and temperature from their respective mean values, while the denominator is used for normalization. Based on the slope of the temperature trend The formula for determining the switching of positive and negative seasonal patterns is: ； Among them, the specific condition is when This indicates that the temperature is trending downwards, and the system is switching to winter mode. when This indicates that the temperature is trending upwards, and the system is switching to summer mode. In step S33, the winter weight is calculated first: Then calculate the summer weights: ; Among them, seasonal characteristic index If it rises, then the winter weight will increase. It is on the rise. Represents the absolute value of the temperature deviation, with winter weighting. Follow Increase and grow; Substitute the parameters from the parameter library to adjust the step size. and response time Then the step size will be adjusted in the end. With response time The formula is expressed as follows: ； ； In step S34, the duty cycle is adjusted according to the fuzzy output. Seasonal weighting and response time The formula for dynamically optimizing the PWM duty cycle is expressed as follows: ； in, The current duty cycle, Weighting for winter This refers to the final response time; Power after comprehensive thermal balance adjustment Dynamic decay factor and real-time temperature deviation The power is finally corrected using the following formula: 。 7. The intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations according to claim 1, characterized in that, The specific steps in step S4 are as follows: S41. Sensor Fault Detection and Compensation: Acquiring Historical Temperature Data from Sensors And calculate the historical temperature standard deviation. By triggering Bayesian anomaly probability calculation through redundancy consistency score, faulty sensors are located, and the faulty sensor data is repaired according to spatiotemporal kriging interpolation. S42. Heat tracing cable damage detection and switching: Drive backup system power adjustment by determining abnormal harmonic energy ratio; Among them, the abnormal harmonic energy ratio includes the abnormal energy ratio of the characteristic frequency band of the current signal detected by frequency domain analysis and the abnormal correlation between power input and temperature response detected by covariance analysis. S43. Incremental training for model update: Assign data weights based on prediction error and adjust LSTM weights using an online Bayesian update formula; S44, Q-learning reinforcement learning optimization: Constructing a state-action space and reward function, and updating the policy network using the Actor-Critic framework. parameter; S45. Dynamic calibration of aging factors: Establishing an exponential decay model for aging factors. By defining state variables, deriving state transition equations, and constructing observation equations, linearized observation equations are obtained based on these equations. Extended Kalman filtering is then used to estimate the aging rate in real time. The estimated aging factors Embedded heat transfer model to correct thermal conductivity .

8. The intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations according to claim 7, characterized in that, In step S41, the formula for the redundancy consistency score within the sensor array is: ； in, The standard deviation of historical temperatures. This represents the maximum temperature difference between any two sensors within the sensor array. It is represented as a reference value for normalization. If the consistency score is <0.8, an abnormal sensor is identified. When an abnormal sensor is detected, a Bayesian anomaly probability calculation is triggered, and expressed by the following formula: ； in, It is the first The weights of a Gaussian distribution The mean, For variance; like Then mark the sensor. This is a fault; The formula for repairing faulty sensor data using spatiotemporal kriging interpolation is as follows: ； in, Location of the current faulty sensor With neighboring sensor locations Euclidean distance, Current time Time measurement with neighboring sensors The difference, The weights are calculated based on the spatiotemporal distance between nearby sensors and faulty sensors. and These are parameters based on spatial and temporal scales; In step S42, the specific calculation steps for the abnormal energy proportion of the characteristic frequency band of the current signal detected by frequency domain analysis are as follows: For the heat tracing current signal Perform a Fast Fourier Transform to obtain the frequency domain signal. The formula for extracting energy from the 1kHz frequency band is: The total energy is calculated as follows: ; in, The frequency range of the energy signal is 1kHz ± 50Hz. Represented as the total energy across the entire frequency domain; like If so, the abnormal energy percentage indicates an open circuit; The specific calculation steps for detecting anomalies in the correlation between power input and temperature response using covariance analysis are as follows: ； in, The change in temperature The change in power Temperature change The average value, Power change The average value; like If the energy ratio is abnormal, it indicates that the tracing cable has failed. When an abnormal energy ratio is determined, the formula for adjusting the power of the standby heating cable according to the power compensation formula is as follows: ； in, This is the nominal power of the backup heat tracing cable. To compensate for the gain, This is a reference temperature.

9. The intelligent optimization method for power plant heat tracing systems based on ambient temperature fluctuations according to claim 7, characterized in that, In step S43, the model-predicted temperature is obtained. and actual measured temperature Temperature prediction based on model and actual measured temperature Calculate prediction error And based on the prediction error Calculate the variance estimate of the error Through prediction error and variance estimates Define data weights The formula is as follows: ； Set model parameters It follows a Gaussian distribution, and is then updated based on the verified Bayesian update formula: ； in, It is the variance of observation noise, the numerator part The denominator represents the correction of the parameter mean by the new data. Used to balance the impact of new data and prior information; In step S44, the state space is defined as follows: ; in, It's a temperature deviation. It is the rate of temperature change. It is an aging factor. It is a seasonal characteristic index; The action space is defined as follows: ; in, It is the power adjustment amount of the heat tracing cable. It is the PWM duty cycle adjustment amount; Calculating the total power, the reward function is: ； in, , , These are weighting coefficients. To measure the magnitude of temperature deviation, This represents the total power consumption of the system. Used to control energy consumption Measuring the range of motion To avoid frequent system adjustments; Update the policy network using the Actor-Critic framework. Policy Network Based on the current state Output Action The probability distribution is determined, and the policy network is updated according to the following policy gradient formula. parameter: ； in, Let the action value function be... The state value function; In step S45, an exponential decay model of the aging factor is established. The formula is: ； in, This is the initial value of the aging factor. It is the aging rate. It is a noise term; Extended Kalman filter for real-time estimation of aging rate The formula is: ； in, The Kalman gain is dynamically adjusted based on the current state and observation data. Real-time estimated aging factors Embedded heat transfer model to correct thermal conductivity The formula is: 。

Images