Thermal energy storage control method and control system based on flue gas parameter fluctuations

Abstract

This invention belongs to the field of thermal energy storage control technology, specifically involving a thermal energy storage control method and control system based on flue gas parameter fluctuations. First, flue gas parameters from the boiler tail end are collected and preprocessed. Then, the parameters are decomposed and fluctuation characteristics are extracted using the EEMD algorithm. An LSTM-BP fusion model is used to predict the fluctuation range and trend. Next, a multi-objective optimization function is constructed based on relevant parameters, and the dynamic operating condition baseline value is obtained by solving it. Then, based on a fuzzy adaptive PID algorithm, the thermal energy storage device is adjusted by combining the deviation and fluctuation frequency. Finally, the actual fluctuations are monitored, and the model and optimization function are corrected to form a closed-loop control. This invention can accurately capture the fluctuation characteristics of flue gas parameters, achieve accurate prediction of fluctuation trends and dynamic adaptation to the operating condition baseline value, improve the accuracy and stability of thermal energy storage control, solve the pain points of existing methods such as adjustment lag and insufficient accuracy, and improve the efficiency of flue gas waste heat recovery.

Description

Technical Field

[0001] This invention belongs to the field of thermal energy storage control technology, and particularly relates to a thermal energy storage control method and control system based on flue gas parameter fluctuations. Background Technology

[0002] In the power industry, boilers, as core thermal equipment, contain a large amount of recoverable waste heat in their flue gas. Flue gas waste heat recovery and storage systems have become key equipment for improving unit energy efficiency and reducing energy consumption, and are widely used in thermal power, combined heat and power (CHP) and other scenarios. However, during boiler operation, due to various factors such as power load scheduling and adjustment, fuel combustion stability, and furnace pressure fluctuations, parameters such as temperature, flow rate, component concentration, and pressure of the flue gas are constantly in a state of dynamic fluctuation. Moreover, the fluctuations exhibit nonlinearity, non-stationarity, and a coexistence of randomness and periodicity. Among them, random fluctuations mainly originate from flue gas turbulence and instantaneous sensor disturbances, while periodic fluctuations are directly related to the boiler combustion cycle and load adjustment cycle. The superposition of these two types of fluctuations leads to significant instability in the heat supply of flue gas waste heat.

[0003] Existing thermal energy storage control methods mostly employ open-loop or simple closed-loop control modes with fixed operating condition baselines, lacking the ability to accurately capture and predict flue gas parameter fluctuations. These methods suffer from the following main technical challenges:

[0004] 1. The thermal storage control does not fully consider the fluctuation characteristics of flue gas parameters, and only makes passive adjustments based on real-time collected flue gas parameters. The adjustment has a strong lag and is difficult to adapt to the rapid fluctuations of flue gas parameters. This can easily lead to a mismatch between the thermal storage system and the waste heat supply, resulting in insufficient waste heat recovery or overload of the thermal storage device.

[0005] Second, the traditional time-series decomposition method is used for fluctuation feature extraction, which is prone to mode aliasing and cannot accurately separate the steady-state components, periodic fluctuation components and random fluctuation components of flue gas parameters. This results in inaccurate fluctuation feature description, which in turn affects the accuracy of subsequent prediction and control.

[0006] Third, the prediction models mostly use a single model, which makes it difficult to take into account the time-dependent characteristics of flue gas parameter fluctuations and the nonlinear mapping requirements. The accuracy and reliability of the prediction results are insufficient, and they cannot provide accurate support for the calibration of the benchmark value of thermal storage conditions.

[0007] Fourth, the calibration of the operating condition reference value did not incorporate the flue gas parameter fluctuation coefficient as a constraint condition. The reference value is fixed and cannot be dynamically adjusted according to the flue gas fluctuation trend, which further aggravates the operational instability of the thermal storage system.

[0008] Therefore, given the inherent characteristics of flue gas parameter fluctuations in the power industry, how to accurately extract the fluctuation characteristics of flue gas parameters, accurately predict the fluctuation trend, construct a dynamic thermal energy storage control system adapted to the fluctuation characteristics, solve the technical problems of existing control methods such as adjustment lag, insufficient accuracy, and poor stability, and improve the efficiency of flue gas waste heat recovery and the operational reliability of thermal energy storage systems have become key technical issues that urgently need to be addressed in the field of flue gas waste heat recovery and thermal energy storage in the power industry. Summary of the Invention

[0009] The purpose of this invention is to provide a thermal energy storage control method and control system based on flue gas parameter fluctuations, in order to solve the technical problems of existing control methods such as adjustment lag, insufficient accuracy and poor stability, and improve the efficiency of flue gas waste heat recovery and the operational reliability of thermal energy storage systems.

[0010] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0011] Firstly, a thermal energy storage control method based on flue gas parameter fluctuations is provided, including the following steps:

[0012] S1: Real-time temperature, flow rate, component concentration and flue gas pressure of the boiler tail gas are collected synchronously to obtain the original flue gas parameter sequence. The original flue gas parameter sequence is preprocessed to obtain a standardized flue gas parameter dataset.

[0013] S2: Decompose each parameter into steady-state components, periodic fluctuation components and random fluctuation components, calculate the fluctuation amplitude, frequency and duration of each component, obtain the flue gas parameter fluctuation feature set, predict the fluctuation range and change trend of flue gas parameters within a preset time period, and output the parameter prediction results.

[0014] S3: Based on the parameter prediction results, combined with the thermodynamic characteristics of the heat storage medium, the rated capacity of the heat storage device and the waste heat recovery efficiency threshold, a multi-objective optimization function is established to solve for the target heat storage temperature, medium circulation flow rate and heat exchange area matching coefficient of the heat storage system, which serve as the operating condition benchmark value for heat storage control.

[0015] S4: Real-time acquisition of specified parameters of the thermal storage device, comparison with parameter prediction results and operating condition baseline values, calculation of parameter deviation, and dynamic adjustment of the PID controller correlation coefficient to control the thermal storage device based on the deviation and the fluctuation frequency of the flue gas parameter fluctuation characteristics.

[0016] S5: Monitor the actual fluctuation of flue gas parameters in real time, compare the actual fluctuation with the prediction results, and calculate the prediction error; if the prediction error exceeds the preset threshold, correct the weight parameters of the LSTM-BP fusion prediction model based on the error feedback, adjust the constraints of the multi-objective optimization function, and update the benchmark value of the thermal storage system.

[0017] Preferably, the specific process of extracting flue gas parameter fluctuation features from the standardized flue gas parameter dataset in step S2 to obtain the flue gas parameter fluctuation feature set is as follows:

[0018] S21: Time series decomposition based on EEMD algorithm to achieve three-component extraction including steady-state component, periodic fluctuation component and random fluctuation component;

[0019] S22: Perform fluctuation characteristic calculations, including amplitude, frequency, and duration;

[0020] S23: Summarize the fluctuation characteristics and combine the coupling relationships between various parameters to form a flue gas parameter fluctuation characteristic set.

[0021] Preferably, the specific process of step S21 is as follows:

[0022] S211: Adding Noise: Using the standardized flue gas parameter sequence x(t) as input, add Gaussian white noise n with an amplitude of 0.1 to 0.4 times its standard deviation to x(t). i (t), to obtain the noisy sequence x i (t)=x(t)+n i (t); i = 1, 2, ..., M, M = 200 ~ 500 is the number of Ensemble iterations;

[0023] S212: Noisy decomposition: for each x i (t) Perform EMD decomposition to obtain the intrinsic mode functions c i,j (t), j=1,2,...,K are the IMF order and the residual component r i (t);

[0024] S213: Average Denoising: Take the arithmetic mean of IMFs and residual components of the same order to eliminate the influence of noise;

[0025] S214: Component Classification:

[0026] Random fluctuation component C r (t): The first 1 to 3 high-frequency IMFs correspond to flue gas turbulence and instantaneous sensor disturbances;

[0027] Periodic fluctuation component C p (t): Intermediate-stage IMF, corresponding to the boiler combustion / load adjustment cycle;

[0028] steady-state component C s (t): Residual component This characterizes the baseline steady-state level of the parameters.

[0029] Preferably, the specific process of step S22 is as follows:

[0030] S221: Calculation of fluctuation range, the formula is as follows:

[0031] ;

[0032] Where k∈{r,p}, r corresponds to the random fluctuation component, p corresponds to the periodic fluctuation component, and C k (t) is the time series sequence of a certain type of fluctuation component after the time series decomposition of flue gas parameters, and the peak-to-peak value of the two types of components is A. r A p max(⋅) and min(⋅) are the maximum and minimum values ​​within the analysis period;

[0033] S222: Fluctuation frequency calculation, the formula is as follows:

[0034] Periodic component frequency f p First, calculate the autocorrelation function:

[0035] ;

[0036] The delay step is taken as the first peak value. ,but f s =1~5Hz, which is the sampling frequency;

[0037] random component frequency f r : FFT transformation to the frequency domain, taking the frequency corresponding to the peak value of the power spectrum as the random dominant frequency;

[0038] S223: Duration calculation, the formula is as follows:

[0039] T k = N / f s ·N k / N;

[0040] Among them, T k The effective duration of the k-th type of fluctuation component represents the actual duration of this type of fluctuation within the analysis period, where k∈{r,p}, r represents the random fluctuation component, p represents the periodic fluctuation component, and N k The effective fluctuation sample number for the k-th fluctuation component refers to the number of sample points whose fluctuation amplitude exceeds twice the standard deviation of its own mean. It is used to define the range of effective fluctuation. N / f s N represents the total analysis time. k / N represents the percentage of valid fluctuation samples.

[0041] Preferably, the specific process of using the LSTM-BP fusion prediction model to predict the output parameters in step S2 is as follows:

[0042] S24: Using the extracted flue gas parameter fluctuation feature set as the core input, the power load dispatch plan data P(t) and boiler combustion condition historical data B(t) are integrated. The three types of data are aligned based on the timestamp, and load and combustion condition parameters that are strongly correlated with flue gas parameter fluctuations are filtered. The filtered data are normalized to obtain the integrated input dataset.

[0043] S25: Construct a cascaded structure of LSTM feature extraction + BP regression prediction, use LSTM to capture the long-term and short-term dependencies of time series data, and optimize the nonlinear mapping capability through BP neural network.

[0044] S26: Prediction and output of flue gas parameter fluctuations.

[0045] Preferably, the specific process of step S26 is as follows:

[0046] S261: Input the fused input data into the trained LSTM-BP model. The LSTM layer extracts temporal correlation features, and the BP layer outputs the future T. pred Upper and lower limits of flue gas parameter fluctuations at each sampling point;

[0047] S262: Based on the output fluctuation interval sequence, calculate the rate of change of the interval mean of adjacent sampling points, and make trend judgment based on the rate of change;

[0048] S263: Outputs the fluctuation range, trend and confidence level of flue gas parameters within a preset time period in the future, forming the final parameter prediction result.

[0049] Preferably, the specific process of step S3 is as follows:

[0050] S31: Taking the parameter prediction results as the core, three types of basic parameters are collected and standardized. The three types of basic parameters include the thermodynamic property parameters of the thermal storage medium, the rated parameters of the thermal storage device, and the constraint correlation parameters.

[0051] S32: Construction of multi-objective optimization function: Construct a constrained multi-objective optimization function with the objectives of maximizing flue gas waste heat recovery, minimizing energy consumption of the thermal storage system, and optimizing subsequent heat exchange stability;

[0052] S33: Set constraints, including: waste heat recovery efficiency constraints, temperature constraints, flow rate constraints, fluctuation adaptation constraints, and matching coefficient constraints.

[0053] S34: Solve and output the baseline value using a genetic algorithm.

[0054] Preferably, the specific process of step S4 is as follows:

[0055] S41: Real-time data acquisition and parameter deviation calculation:

[0056] Real-time data acquisition targets: Synchronously acquire three types of core parameters during the operation of the thermal storage system, including parameters on the thermal storage device side, flue gas side, and control feedback side; compare with the benchmark to calculate the deviation, including temperature deviation, flow rate deviation, and flue gas parameter prediction deviation, and finally obtain the comprehensive deviation and deviation change rate;

[0057] S42: Set the fuzzy adaptive PID control algorithm, including dynamic tuning of PID parameters, fuzzification processing, and fuzzy rules;

[0058] S43: Perform PID parameter tuning and declarative analysis: First, set the initial reference parameters for the PID, and then obtain the real-time tuning parameters through the correction amount of the fuzzy output; the declarative analysis uses the centroid method to calculate the precise value of the correction amount;

[0059] S44: PID control output calculation: The incremental PID algorithm is used to calculate the control increment and control output.

[0060] S45: Control signal output and actuator regulation.

[0061] Preferably, the specific process of step S45 is as follows:

[0062] The control output calculated by PID is converted into specific execution signals according to the regulation characteristics of the controlled object, and the speed of the heat storage medium circulation pump, the opening degree of the heat exchange valve, and the switching state of the heat storage unit are adjusted respectively. The three are linked to achieve closed-loop control.

[0063] S451: Speed ​​control of heat storage medium circulation pump: Pump speed n and medium circulation flow rate q V Proportional, the control signal is the speed regulation duty cycle γ n :

[0064] γ n =γ n0 +k n ⋅u(t);

[0065] Wherein: γ n0 The pump reference speed duty cycle, k n This is the speed adjustment coefficient;

[0066] S452: Heat exchange valve opening control: Matching coefficient k between heat exchange valve opening θ and actual heat exchange area A Positive correlation, the control signal is the opening adjustment amount Δθ:

[0067] Δθ=k θ ⋅u(t);

[0068] Where: k θ This is the opening adjustment coefficient;

[0069] S453: Thermal storage unit switching status control: The switching threshold is set according to the absolute value of the comprehensive deviation |e| and the flue gas fluctuation frequency f. The parallel switching / series combination of thermal storage units is realized by the switch signal output by PID control.

[0070] Secondly, a thermal energy storage control system based on flue gas parameter fluctuations is provided to implement the thermal energy storage control method based on flue gas parameter fluctuations described in any one of the above embodiments, including:

[0071] Distributed flue gas parameter acquisition module: Composed of several temperature sensors, flow rate sensors, gas composition sensors and pressure sensors, it is deployed at different cross sections and key nodes of the flue gas duct at the tail of the boiler to synchronously acquire the original flue gas parameter sequence.

[0072] Data preprocessing and feature extraction unit: connected to the distributed flue gas parameter acquisition module, used to perform outlier removal, data smoothing and standardization on the original flue gas parameter sequence to generate a standardized flue gas parameter dataset;

[0073] LSTM-BP fusion prediction unit: It is connected to the data preprocessing and feature extraction unit by signal, and has a built-in trained LSTM-BP fusion prediction model. It takes as input flue gas parameter fluctuation feature set, power load scheduling plan and historical data of boiler combustion conditions, and predicts the fluctuation range and trend of flue gas parameters within a preset time period in the future.

[0074] Operating condition reference value calibration unit: It is connected to the LSTM-BP fusion prediction unit and has a built-in multi-objective optimization model and genetic algorithm solver. It combines the thermodynamic characteristic parameters of the heat storage medium and the rated parameters of the heat storage device to solve for the dynamic operating condition reference value with the goal of maximizing the recovery of waste heat from the flue gas, and outputs it to the control unit.

[0075] Fuzzy adaptive PID control unit: It is connected to the operating condition reference value calibration unit, the distributed flue gas parameter acquisition module and the thermal storage actuator respectively. It is used to collect the operating parameters of the thermal storage system and the actual parameters of the flue gas outlet, calculate the deviation from the operating condition reference value, and generate control signals through the fuzzy adaptive PID algorithm.

[0076] Thermal storage actuators include a thermal storage medium circulation pump, a heat exchange valve group, a thermal storage unit switching device, and a medium temperature regulation module. They are connected to a fuzzy adaptive PID control unit and adjust their operating status in response to control signals.

[0077] Feedback correction unit: It is connected to the fuzzy adaptive PID control unit, the LSTM-BP fusion prediction unit and the operating condition reference value calibration unit respectively. It is used to calculate the prediction error of flue gas parameters, correct the weights of the prediction model and the constraints of the optimization function, update the operating condition reference value and construct the closed-loop control link.

[0078] The beneficial effects of this invention include:

[0079] 1. Accurately capture flue gas parameter fluctuation characteristics, providing reliable support for subsequent prediction and control: The EEMD algorithm achieves time-series decomposition of flue gas parameters. Compared to traditional time-series decomposition methods, it effectively avoids mode aliasing problems and accurately separates the steady-state components, periodic fluctuation components, and random fluctuation components of flue gas parameters. Simultaneously, through fluctuation feature calculation, it accurately obtains core features such as amplitude, frequency, and duration of the two types of fluctuation components, and combines the coupling relationships between various parameters to form a complete fluctuation feature set, comprehensively and accurately characterizing the nonlinear and non-stationary characteristics of flue gas parameter fluctuations. This completely solves the problem of inaccurate fluctuation feature description in existing technologies, providing high-quality core input for subsequent fluctuation trend prediction and dynamic control.

[0080] 2. Improve the accuracy of flue gas parameter fluctuation trend prediction and achieve advance adaptation of thermal storage control: The LSTM-BP fusion prediction model is adopted. The LSTM layer is used to accurately capture the long-term and short-term time-series dependencies of flue gas parameter fluctuations. Combined with the optimization nonlinear mapping capability of the BP layer, and the integration of power load scheduling plan and historical data of boiler combustion conditions, the accuracy of prediction is further improved. In the prediction process, not only is the fluctuation range of flue gas parameters within the preset time period output, but the fluctuation trend is also determined by the rate of change of the mean of the interval, and the trend confidence is calculated, realizing the dual accurate prediction of fluctuation range and change trend.

[0081] 3. Achieve dynamic calibration of operating condition baseline values ​​to adapt to the dynamic characteristics of flue gas parameter fluctuations: During the calibration of operating condition baseline values, a multi-objective optimization function is constructed based on the parameter prediction results, combined with the thermodynamic characteristics of the heat storage medium, the rated capacity of the heat storage device, and the waste heat recovery efficiency threshold. Innovatively, a flue gas parameter fluctuation coefficient is introduced as a constraint condition. A genetic algorithm is used to solve for three baseline values: the target heat storage temperature, the medium circulation flow rate, and the heat exchange area matching coefficient. The operating condition baseline values ​​can be dynamically adjusted according to the fluctuation trend of flue gas parameters. The stronger the fluctuation, the stronger the adaptability of the baseline values, ensuring that the heat storage system is always matched with the waste heat supply from the flue gas. This avoids the problem of insufficient waste heat recovery and prevents overload of the heat storage device, significantly improving the operational rationality and adaptability of the heat storage system.

[0082] 4. Closed-loop feedback control adapts to fluctuation characteristics, improving the accuracy and stability of thermal storage control: A fuzzy adaptive PID control algorithm is adopted, using the comprehensive deviation, deviation change rate, and fluctuation frequency of flue gas parameter fluctuation characteristics as inputs. The proportional, integral, and derivative coefficients of the PID controller are dynamically adjusted, and an incremental PID algorithm is used to calculate the control output, which separately regulates the speed of the thermal storage medium circulation pump, the opening of the heat exchange valve, and the switching state of the thermal storage unit. These three factors work together to achieve closed-loop feedback control. Differentiated PID parameter adjustment strategies are adopted according to different flue gas fluctuation frequencies, significantly improving control accuracy and response speed. Simultaneously, real-time acquisition of thermal storage system operating parameters and flue gas outlet parameters, continuous comparison and correction, effectively suppresses the impact of flue gas parameter fluctuations on the operational stability of the thermal storage system, ensuring that the thermal storage system is always in a stable operating state, further improving waste heat recovery efficiency.

[0083] 5. Constructing a complete closed-loop correction link to ensure long-term operational accuracy and reliability: By monitoring the actual fluctuations of flue gas parameters in real time, comparing the actual fluctuations with the predicted results, and calculating the prediction error; if the error exceeds a preset threshold, the weight parameters of the LSTM-BP fusion prediction model are corrected in a timely manner, and the constraints of the multi-objective optimization function are adjusted to update the baseline values ​​of the thermal storage system, forming a complete closed-loop link of acquisition-prediction-control-correction. This effectively compensates for the decline in prediction error and control accuracy caused by prediction model aging and changes in operating parameters during long-term operation, ensuring that the prediction model and optimization function always maintain a high degree of adaptability, guaranteeing the long-term stable operation of the thermal storage system, continuously maintaining a high waste heat recovery efficiency, and reducing the operation and maintenance costs of the thermal storage system. Attached Figure Description

[0084] Figure 1 This is a schematic flowchart of the thermal storage control method based on flue gas parameter fluctuations of the present invention.

[0085] Figure 2 This is a schematic diagram of the flue gas parameter fluctuation feature extraction process of the present invention.

[0086] Figure 3 This is a schematic diagram of the architecture of the LSTM-BP fusion prediction model of the present invention. Detailed Implementation

[0087] The following is in conjunction with the appendix Figures 1-3 The present invention will be further described in detail below:

[0088] Example 1

[0089] See appendix Figure 1 As shown, the thermal energy storage control method based on flue gas parameter fluctuations includes the following steps:

[0090] S1: Multi-dimensional acquisition and preprocessing of flue gas parameters: The real-time temperature, flow rate, component concentration and flue gas pressure of the boiler tail flue gas are synchronously acquired through the distributed flue gas parameter acquisition module to obtain the original flue gas parameter sequence; the original flue gas parameter sequence is subjected to outlier removal, data smoothing and standardization processing to generate a standardized flue gas parameter dataset.

[0091] The distributed flue gas parameter acquisition module includes temperature sensors, flow rate sensors, gas composition sensors, and pressure sensors, which are deployed at different cross-sections and key nodes of the flue gas duct at the boiler tail end for synchronously acquiring raw flue gas parameter sequences. The sensors employ a high-temperature resistant and corrosion-resistant packaging structure, adaptable to the high-temperature flue gas conditions in the power industry, and possess a self-calibration function to periodically correct the acquisition accuracy.

[0092] Outlier removal: An algorithm combining the 3σ criterion and gradient mutation detection is used. First, the 3σ criterion is used to screen out suspected outliers that deviate from the parameter mean by more than three times the standard deviation. Then, gradient mutation detection is used to determine whether the value is a sudden abnormality caused by flue gas turbulence or instantaneous sensor failure (not a true fluctuation of the operating condition). After double verification, confirmed outliers are removed to avoid misjudgment by a single algorithm.

[0093] Data smoothing: A fusion strategy of moving average and wavelet denoising is adopted. First, moving average is used to eliminate high-frequency random disturbances in the parameter sequence, and then wavelet denoising is used to decompose the signal and filter out noise components. The two methods complement each other, effectively suppressing parameter fluctuations caused by flue gas turbulence and sensor errors, and outputting a smoothed parameter sequence.

[0094] Standardization processing: The min-max normalization algorithm is used to map the smoothed parameters of each dimension to the [0,1] interval, eliminate the differences in the dimensions of different parameters, ensure the comparability of multi-dimensional flue gas parameters, and provide a dataset of a unified scale for subsequent fluctuation feature extraction and model prediction.

[0095] S2: Flue gas parameter fluctuation feature extraction and trend prediction: Based on the standardized flue gas parameter dataset generated in step S1, the time series decomposition algorithm is used to decompose each parameter into steady-state components, periodic fluctuation components, and random fluctuation components. The fluctuation amplitude, frequency, and duration of each component are calculated to obtain the flue gas parameter fluctuation feature set. Combining the power load dispatch plan and historical data of boiler combustion conditions, an LSTM-BP fusion prediction model is constructed. The fluctuation feature set is input to predict the fluctuation range and trend of flue gas parameters within a preset time period in the future, and the parameter prediction results are output.

[0096] S3: Dynamic calibration of the operating condition benchmark value of the thermal storage system: Based on the parameter prediction results of step S2, combined with the thermodynamic characteristics of the thermal storage medium (heat transfer oil or molten salt), the rated capacity of the thermal storage device, and the waste heat recovery efficiency threshold, a multi-objective optimization function is established. The objectives are to maximize the waste heat recovery of flue gas, minimize the energy consumption of the thermal storage system, and optimize the subsequent heat exchange stability. The target thermal storage temperature, medium circulation flow rate, and heat exchange area matching coefficient of the thermal storage system are obtained by solving the problem through a genetic algorithm, which serve as the operating condition benchmark value for thermal storage control. The multi-objective optimization function introduces the flue gas parameter fluctuation coefficient as a constraint condition, so that the benchmark value is dynamically adjusted according to the predicted fluctuation trend.

[0097] S4: Closed-loop feedback control of the thermal storage process: Based on the operating condition baseline values ​​calibrated in step S3, the thermal storage system is started, and the inlet and outlet medium temperatures, pressures, and flue gas outlet parameters of the thermal storage device are collected in real time. These are compared with the parameter prediction results and operating condition baseline values ​​from step S2 to calculate the parameter deviation. A fuzzy adaptive PID control algorithm is adopted, and the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller are dynamically adjusted according to the deviation and the fluctuation frequency of the flue gas parameter fluctuation characteristics. The output control signal is used to adjust the speed of the thermal storage medium circulation pump, the opening degree of the heat exchange valve, and the switching status of the thermal storage unit.

[0098] S5: Dynamic Correction and Optimization for Flue Gas Adaptability: Real-time monitoring of actual fluctuations in flue gas parameters during the control process in step S4, comparing the actual fluctuations with the prediction results in step S2, and calculating the prediction error. If the prediction error exceeds a preset threshold, the weight parameters of the LSTM-BP fusion prediction model are corrected based on the error feedback, and the constraints of the multi-objective optimization function are adjusted. The baseline values ​​of the thermal storage system are updated, forming a closed-loop control link of acquisition-prediction-control-correction, ensuring that the thermal storage system adapts to the dynamic fluctuations of flue gas parameters and maintains stable waste heat recovery efficiency.

[0099] Example 2

[0100] Based on Example 1, see Figure 2 The specific process of extracting flue gas parameter fluctuation features from the standardized flue gas parameter dataset in step S2 is as follows:

[0101] S21: Time series decomposition is performed based on the EEMD algorithm to extract three components: steady-state component, periodic fluctuation component, and random fluctuation component. The core of the EEMD algorithm is to add Gaussian white noise to assist decomposition, eliminate the mode aliasing problem of single empirical mode decomposition (EMD), and adapt to the nonlinear and non-stationary fluctuation characteristics of flue gas parameters. The specific process is as follows:

[0102] S211: Adding Noise: Using the standardized flue gas parameter sequence x(t), t=1,2,...,N (where N is the data length, normalized to the [0,1] interval) as input, add Gaussian white noise n with an amplitude of 0.1~0.4 times its standard deviation to x(t). i (t), to obtain the noisy sequence x i (t)=x(t)+n i (t). i=1,2,...,M, M=200~500 is the number of Ensemble iterations, used to balance accuracy and efficiency.

[0103] S212: Noisy decomposition: for each x i (t) Perform EMD decomposition to obtain the intrinsic mode functions c i,j (t), j=1,2,...,K are the IMF order and the residual component r i (t), satisfying the following formula:

[0104] .

[0105] S213: Average Denoising: Take the arithmetic mean of the IMF and residual components of the same order to eliminate the influence of noise. The final decomposition is as follows:

[0106] ;

[0107] in, For the j-th order average IMF component, This represents the average residual component.

[0108] S214: Component Classification:

[0109] Random fluctuation component C r (t): The first 1-3 high-frequency IMFs (f>0.1Hz) correspond to flue gas turbulence and instantaneous sensor disturbances. j1 is a high-frequency threshold, determined by mutual information entropy.

[0110] Periodic fluctuation component C p (t): Intermediate-order IMF (0.001Hz≤f≤0.1Hz), corresponding to the boiler combustion / load adjustment cycle. j2 is the low-frequency threshold, which is determined empirically.

[0111] steady-state component C s (t): Residual component This characterizes the baseline steady-state level of the parameters.

[0112] S22: Perform fluctuation characteristic calculations, including amplitude, frequency, and duration. The specific process is as follows:

[0113] S221: Fluctuation Amplitude (Peak-to-Peak Value Method, characterizing fluctuation intensity):

[0114] ;

[0115] Where k∈{r,p} (r is random, p is periodic), A r A p These are the peak values ​​of the two types of components, and max(⋅) and min(⋅) are the maximum and minimum values ​​within the analysis period.

[0116] S222: Fluctuation frequency (characterizing the speed of fluctuation):

[0117] Periodic component frequency f p (Autocorrelation function method): First, calculate the autocorrelation function:

[0118] ;

[0119] The delay step is taken as the first peak value. ,but f s =1~5Hz, step S1 collects the frequency.

[0120] random component frequency f r (Power Spectral Density Method): FFT is used to transform the frequency domain, and the frequency corresponding to the peak value of the power spectrum is taken as the random main frequency.

[0121] S223: Duration (characterizing the duration of fluctuation):

[0122] T k = N / f s ·N k / N;

[0123] Among them, T k The effective duration of the k-th type of fluctuation component represents the actual duration of this type of fluctuation within the analysis period, where k∈{r,p}, r represents the random fluctuation component, p represents the periodic fluctuation component, and N k The effective fluctuation sample number for the k-th fluctuation component refers to the number of sample points whose fluctuation amplitude exceeds twice the standard deviation of its own mean. It is used to define the range of effective fluctuation. N / f s N represents the total analysis time. k / N represents the percentage of valid fluctuation samples.

[0124] S23: Perform feature set integration: summarize A r A p f r f p T r T p and the mean of steady-state components By combining the coupling relationships between various parameters (Pearson correlation coefficient), a flue gas parameter fluctuation feature set is formed, which provides input for subsequent LSTM-BP model prediction.

[0125] See the LSTM-BP fusion prediction model architecture. Figure 3 As shown, the specific process of using the LSTM-BP fusion prediction model to output the prediction results in step S2 is as follows:

[0126] S24: Data preprocessing and input set construction: Using the flue gas parameter fluctuation feature set extracted in step S2 as the core input, the power load dispatch plan data P(t) (unit: MW) is integrated, including the planned load curve and load adjustment nodes within the future preset time period, as well as the historical boiler combustion condition data B(t), including key operating condition parameters such as coal feed rate, air supply rate, and furnace pressure, with units of t / h, m³ / h, and kPa, respectively.

[0127] Data Alignment and Filtering: The three types of data are aligned based on timestamps. Pearson correlation coefficients are used to filter load and combustion condition parameters that are strongly correlated with flue gas parameter fluctuations (|ρ|>0.6), eliminating redundant data. The filtered data is then normalized to obtain the fused input dataset X=[X feat ,X P ,X B ], where X feat Let X be the wave characteristic set matrix. P For the load data matrix, X B This is a combustion condition data matrix.

[0128] Dataset partitioning: The dataset is divided into training, validation, and test sets in a 7:2:1 ratio. A sliding window method is used to construct time-series prediction samples. The window length L is determined based on the duration of periodic fluctuations in flue gas parameters, typically using 30-60 sampling points at a sampling frequency of 1-5Hz. The window duration is 6-60 seconds. The output is the future preset duration T. pred Flue gas parameter fluctuation range and trend characteristics within the range, T pred Usually, it takes 5 to 30 minutes.

[0129] S25: Employs a concatenated structure of LSTM feature extraction and BP regression prediction. LSTM captures the long-term and short-term dependencies in time-series data, adapting to the temporal characteristics of flue gas parameter fluctuations. BP neural network optimizes nonlinear mapping capabilities, improving the accuracy of fluctuation range prediction.

[0130] LSTM layer: Input dimension is the number of features D of the fused input dataset, and the number of neurons in the hidden layer is N. LSTM =2 k k is to satisfy 2 kThe smallest integer ≥D is usually 32~128. The activation function is tanh. The activation functions for the forget gate, input gate, and output gate are sigmoid. The dropout coefficient is set to 0.2~0.3 to suppress overfitting.

[0131] BP layer: The input is the temporal feature vector output by the LSTM layer. The dimension is the same as the number of neurons in the LSTM hidden layer. There are 1 to 2 hidden layers, and the number of neurons decreases layer by layer, from 128 to 64. The activation function is ReLU. The output layer dimension is 2×T. pred / Δt, where Δt is the sampling interval, outputs the upper and lower limits of flue gas parameters for each sampling point in the future, forming the fluctuation range.

[0132] Model training and optimization:

[0133] Loss function: The root mean square error (RMSE) is combined with the interval coverage loss to balance prediction accuracy and interval reliability. The formula is as follows:

[0134] ;

[0135] Among them, y i These are the actual flue gas parameter values; To predict the mean; , , respectively, represent the upper and lower limits of the prediction interval; I(⋅) is the indicator function, which takes the value 1 if the condition is met, and 0 otherwise; λ is the weight coefficient, which takes the value 0.1~0.3, and N is the number of samples.

[0136] Optimizer: The Adam optimizer is used, with the initial learning rate set to 0.001 and adjusted by a learning rate decay strategy, decreasing to 0.9 every 100 rounds. The number of training iterations is set to 300-500 rounds, and the termination condition is to minimize the validation set loss.

[0137] S26: Prediction and Result Output of Flue Gas Parameter Fluctuations:

[0138] S261: Model Inference: The preprocessed fused input data is input into the trained LSTM-BP model. The LSTM layer extracts temporal correlation features, and the BP layer outputs the future T. pred The upper and lower limits of flue gas parameter fluctuations at each sampling point.

[0139] S262: Trend Determination: Based on the output fluctuation interval sequence, calculate the rate of change of the interval mean of adjacent sampling points:

[0140] ;

[0141] ;

[0142] Set a threshold ±α, where α = 0.02 ~ 0.05:

[0143] like This is determined to be an upward trend;

[0144] like This is determined to be a downward trend;

[0145] Otherwise, the trend is stable.

[0146] S263: Results Integration: Output the fluctuation range of flue gas parameters (including the upper and lower limits of each sampling point), the trend of change (rising, falling or remaining stable) and the trend confidence level within a preset time period in the future. The trend confidence level is calculated based on the model prediction error.

[0147] Confidence level = 1 - test set RMSE / flue gas parameter fluctuation range, forming the final parameter prediction result.

[0148] Example 3

[0149] Based on Example 1 or Example 2, the specific process of step S3 is as follows:

[0150] S31: Based on the parameter prediction results output in step S2, including the fluctuation range, trend, and confidence level of flue gas parameters within a preset time period, three types of basic parameters are collected simultaneously and standardized to ensure input consistency.

[0151] Thermodynamic properties of the heat storage medium:

[0152] If it is a heat transfer oil, take the kinematic viscosity ν, in mm² / s, with a typical value of 20~40 at 40℃; specific heat capacity c. p , unit: kJ / (kg・℃), conventional value 2.0~2.5; thermal conductivity λ, unit: W / (m・℃), conventional value 0.12~0.18.

[0153] If it is a molten salt (sodium nitrate-potassium nitrate mixture), take the melting point T. m Unit: °C, typically 220~260; specific heat capacity c p , unit: kJ / (kg・℃), typically 1.5~1.8 in liquid state; thermal conductivity λ, unit: W / (m・℃), typically 0.5~0.8 in liquid state.

[0154] Rated parameters of thermal storage device: Rated capacity Q rat Unit: MJ, take 10 according to unit size 4 ~10 6 Rated operating temperature T rat Unit: °C; Thermal oil type: 280~320; Molten salt type: 350~400; Maximum allowable medium flow rate q V,maxUnit: m³ / h; Rated heat exchange surface area A rat , Unit: m².

[0155] Constraint correlation parameter: Waste heat recovery efficiency threshold η thr Unit: %, industry standard ≥85; flue gas parameter fluctuation coefficient β, calculated based on the fluctuation feature set in step S2, β=(A p +A r ) / C s A p A r These are the periodic fluctuation amplitude and the random fluctuation amplitude, respectively. The mean of the steady-state components represents the intensity of the fluctuation.

[0156] S32: Construction of Multi-Objective Optimization Function:

[0157] With the objectives of maximizing flue gas waste heat recovery, minimizing energy consumption of the thermal storage system, and optimizing subsequent heat exchange stability, a constrained multi-objective optimization function is constructed, with the variables being the baseline values ​​to be calibrated:

[0158] Target thermal storage temperature T tar Unit: °C; Medium circulation flow rate q V Unit: m³ / h; heat exchange area matching coefficient k A k A =A use / A rat A use For the actual heat exchange area used, 0.5 ≤ k A ≤1.

[0159] Objective function 1: Maximize flue gas waste heat recovery (core objective):

[0160] maxQ rec =k A ⋅A rat ⋅K⋅ΔT lm ;

[0161] Where K is the overall heat transfer coefficient (unit: W / (m²・℃), calculated using the following formula:

[0162] ;

[0163] α g α is the side convective heat transfer coefficient of the flue gas; f λ is the convective heat transfer coefficient on the medium side; δ is the heat exchange wall thickness; w ΔT is the thermal conductivity of the wall surface. lm The logarithmic mean temperature difference (unit: °C).

[0164] ;

[0165] , The flue gas inlet and outlet temperatures are taken from the prediction result in step S2, T in This refers to the inlet temperature of the medium (operating condition setpoint).

[0166] Objective function 2: Minimize the energy consumption of the thermal storage system (operating cost objective):

[0167] minP cons =P pump +P ins ;

[0168] Among them, P pump Power consumption of the circulating pump, unit: kW.

[0169] P pump =ρ⋅g⋅H⋅q V / 3600⋅η pump ;

[0170] ρ is the density of the medium; g = 9.81 m / s²; H is the pump head; η pump For pump efficiency, typically 0.7~0.85; P ins Energy consumption for insulation, unit: kW.

[0171] P ins =k ins ⋅A tank ⋅(T tar -T env );

[0172] k ins A is the heat transfer coefficient of the insulation layer. tank T represents the surface area of ​​the thermal storage tank. env The ambient temperature.

[0173] Objective function 3: Optimal subsequent heat transfer stability (operating condition adaptation objective):

[0174] ;

[0175] Among them, T tar,t The dynamic target temperature for each sampling point in the future is adjusted based on the trend prediction in step 2. The target average temperature; σ ΔT This is the standard deviation of the target thermal storage temperature, representing the degree of fluctuation in the dynamic target thermal storage temperature over a preset period. The smaller the value, the more stable the heat exchange temperature difference of the thermal storage system, and the less impact it has on subsequent equipment.

[0176] S33: Set constraints, including:

[0177] Waste heat recovery efficiency constraints: ;

[0178] Q g The total waste heat of the flue gas is taken from the prediction result in step 2, and η is the actual waste heat recovery efficiency of the heat storage system; Q rec η represents the actual waste heat recovered from the flue gas by the thermal storage system, measured in MJ or kJ. It is the net heat absorbed by the thermal storage medium from the flue gas. thr The waste heat recovery efficiency threshold, usually set at 85% or higher, is the minimum efficiency requirement set by the power industry to ensure the economic viability of thermal storage systems.

[0179] Temperature constraint (to prevent medium solidification / overheating):

[0180] ;

[0181] Flow constraints:

[0182] ;

[0183] Fluctuation adaptation constraints:

[0184] ;

[0185] ;

[0186] Matching coefficient constraints: .

[0187] S34: Genetic Algorithm Solution and Benchmark Value Output:

[0188] Algorithm parameter settings: Population size 50-100, number of iterations 100-200, crossover probability P c =0.7~0.8, mutation probability P m =0.01~0.05, the fitness function uses a weighted summation method, and the target weights are allocated according to industry priority: Q rec Weight 0.5, P cons Weight 0.3, σ ΔT Weight 0.2.

[0189] Solution process:

[0190] Initialize the population and randomly generate T that meet the constraints. tar q v k A combination;

[0191] Calculate the fitness value for each individual;

[0192] The population is updated through selection, crossover, and mutation operations;

[0193] Iterate until convergence (fitness value fluctuation ≤ 1%), and output the compromise solution in the Pareto optimal solution set.

[0194] Baseline value determination: Select the solution that satisfies both fluctuation coefficient adaptation and equipment safety from the optimal solutions as the final operating condition baseline value, including the target thermal storage temperature T. tar Medium circulation flow rate q V Heat exchange area matching coefficient k A The curve of the reference value being adjusted with the flue gas fluctuation coefficient β is output synchronously.

[0195] Example 4

[0196] Based on Example 1, Example 2, or Example 3, the specific process of step S4 is as follows:

[0197] S41: Real-time data acquisition and parameter deviation calculation:

[0198] Real-time data acquisition targets: Synchronously acquire three types of core parameters during the operation of the thermal storage system, including:

[0199] Thermal storage device side: Medium inlet temperature T f,in Outlet temperature T f,out Medium operating pressure P f ;

[0200] Flue gas side: Flue gas outlet temperature T g,out Outlet flow velocity v g,out , that is, the actual parameters of the flue gas after heat exchange;

[0201] Control feedback side: Actual circulating flow rate of the medium, q V,act Actual heat exchange area matching coefficient k A,act ;

[0202] Comparison with reference point and deviation calculation: The reference value of the operating condition calibrated in step 3, i.e., the target thermal storage temperature T, is used. tar Medium circulation flow rate reference q V,ref Heat exchange area matching coefficient benchmark k A,ref The core target value is determined by combining the flue gas parameter prediction results from step 2. Calculate the core deviation:

[0203] Temperature deviation: ;

[0204] Flow deviation: ;

[0205] Flue gas parameter prediction deviation: .

[0206] The combined deviation, after normalization, is used as the PID input:

[0207] ;

[0208] in: , , The weighting coefficient is ω, which is the industry standard. T =0.6、ω q =0.25、ω g =0.15, with temperature as the core control indicator; T rat The rated temperature of the thermal storage device, T m ΔT is the melting point of the heat storage medium. g,max The maximum allowable fluctuation value for flue gas outlet temperature is ±10~20℃, which is the standard range for the power industry. All deviations are normalized to eliminate dimensional differences and adapt to fuzzy control input requirements.

[0209] Deviation change rate: ;

[0210] Where: e(t) is the current time-to-time comprehensive deviation; e(t−1) is the previous time-to-time comprehensive deviation; Δt is the sampling interval.

[0211] S42: Fuzzy Adaptive PID Control Algorithm

[0212] Dynamic tuning of PID parameters:

[0213] Using the overall deviation e and the rate of change of deviation The flue gas parameter fluctuation frequency f extracted in step S2 is used as a fuzzy input, with the proportional coefficient K of the PID controller as the input. p Integral coefficient K i Differential coefficient K d Correction amount , , To achieve fuzzy output, and in conjunction with flue gas fluctuation frequency adaptation tuning rules, dynamic adjustment of PID parameters is realized:

[0214] Blur processing:

[0215] input domain: e, The universe of discourse is set to {−3,−2,−1,0,1,2,3}. The flue gas fluctuation frequency f is divided into the universe of discourse {0,1,2} according to the flue gas fluctuation characteristics of the industry, corresponding to micro fluctuations f<0.01Hz, medium fluctuations 0.01≤f≤0.1Hz, and large fluctuations f>0.1Hz, which is consistent with the fluctuation characteristic classification in step 2.

[0216] Output domain: , , The universe of discourse is set as {−3,−2,−1,0,1,2,3}.

[0217] Membership function: Both input and output adopt the triangular membership function of PID fuzzy control in the power industry, and the formula is as follows:

[0218] For any variable x in the universe of discourse, the trigonometric membership function is:

[0219] ;

[0220] Where: a, b, c are the inflection points of the membership function, which take values ​​uniformly according to the universe of discourse {−3,−2,−1,0,1,2,3}. For example, “negative large” corresponds to a=−4, b=−3, c=−2, and “zero” corresponds to a=−1, b=0, c=1.

[0221] Set fuzzy rules:

[0222] Based on the control experience of thermal storage systems in the power industry, a 7×7×3 fuzzy rule table (e7 level, ...) was developed. (f3 gear), the core rule logic is:

[0223] Significant fluctuations in flue gas (f>0.1Hz): Increase K p Decrease K i (To avoid integral saturation) Appropriately increase K d (Accelerate deviation response and suppress large fluctuations).

[0224] Flue gas amplitude fluctuation (0.01≤f≤0.1Hz): K p K i K d Choose a moderate value to balance response speed and control stability.

[0225] Slight fluctuations in flue gas (f < 0.01 Hz): Decrease K p Increase K i (Eliminate steady-state error), reduce K d， Avoid amplifying high-frequency disturbances.

[0226] Example: If e = positive, =positive small, f=large large, then ΔK p =Zhengda, ΔK i =Negative small, ΔK d = Center.

[0227] S43: PID parameter tuning and declarative analysis:

[0228] First, set the initial baseline parameter for the PID controller: K p0 K i0 K d0 Then, the real-time tuning parameters are obtained through the correction amount of the fuzzy output:

[0229] ;

[0230] in: , , This is a parameter correction factor, typically set to 0.1 to 0.3 in the industry, with minor adjustments made based on the scale of the thermal storage device.

[0231] The sharpening process uses the centroid method to calculate the precise value of the correction amount, as shown in the following formula:

[0232] ;

[0233] Where: μ(x) i ) represents the membership degree of the i-th domain point, x i Let be the actual value of the i-th universe point, and n be the number of universe points.

[0234] S44: PID control output calculation:

[0235] An incremental PID algorithm is used to calculate the control increment Δu(t), which serves as the basis for the control signal of the actuator. The formula is as follows:

[0236] Δu(t)=K p [e(t)−e(t−1)]+K i e(t)+K d [e(t)−2e(t−1)+e(t−2)];

[0237] Where: e(t−2) is the combined deviation between the first two time points;

[0238] The control output is u(t) = u(t−1) + Δu(t), where u(t−1) is the control output at the previous moment.

[0239] S45: Control signal output and actuator regulation, the specific process is as follows:

[0240] The control output u(t) calculated by PID is converted into specific execution signals according to the regulation characteristics of the controlled object, and the speed of the heat storage medium circulation pump, the opening degree of the heat exchange valve, and the switching state of the heat storage unit are adjusted respectively. The three are linked to achieve closed-loop control:

[0241] Heat storage medium circulation pump speed control: Pump speed n and medium circulation flow rate q V Proportional, the control signal is the speed regulation duty cycle γ n :

[0242] γ n =γ n0 +k n ⋅u(t);

[0243] Wherein: γ n0The duty cycle is the pump's reference speed; k n This is the speed adjustment coefficient, typically taken as 0.02~0.05 / ; speed adjustment directly changes the medium circulation flow rate, matching the heat transfer requirements of flue gas waste heat fluctuations.

[0244] Heat exchange valve opening control: Matching coefficient k between heat exchange valve opening θ and actual heat exchange area A Positive correlation, the control signal is the opening adjustment amount Δθ:

[0245] Δθ=k θ ⋅u(t) where: k θ The valve opening adjustment coefficient is typically 0.5~1℃ / ° in the industry. Adjusting the valve opening enables dynamic matching of the heat exchange area and controls the waste heat recovery rate.

[0246] Thermal storage unit switching status control: The switching threshold is set according to the absolute value of the comprehensive deviation |e| and the flue gas fluctuation frequency f. The industry standard is: when |e|>20% and f>0.1Hz, multiple units are started in parallel. The parallel switching / series combination of thermal storage units is realized by the switch signal output by PID control, which can adapt to the thermal storage capacity requirements when flue gas parameters fluctuate greatly and ensure the continuity of waste heat recovery.

[0247] In all the formulas, g = 9.81 m / s², and the pump efficiency η pump =0.7~0.85.

[0248] A thermal energy storage control system based on flue gas parameter fluctuations includes:

[0249] The distributed flue gas parameter acquisition module consists of several temperature sensors, flow rate sensors, gas composition sensors, and pressure sensors. These sensors are deployed at different cross-sections and key nodes of the boiler's tail flue gas duct to synchronously acquire raw flue gas parameter sequences. The sensors employ a high-temperature resistant and corrosion-resistant packaging structure, making them suitable for the high-temperature flue gas conditions in the power industry. They also feature self-calibration capabilities, periodically correcting the acquisition accuracy.

[0250] Data preprocessing and feature extraction unit: connected to the distributed flue gas parameter acquisition module, used to perform outlier removal, data smoothing and standardization on the original flue gas parameter sequence to generate a standardized flue gas parameter dataset; and extract flue gas parameter fluctuation feature set through time series decomposition algorithm and feature calculation model, and output it to the prediction unit.

[0251] LSTM-BP fusion prediction unit: It is connected to the data preprocessing and feature extraction unit by signal, and has a built-in trained LSTM-BP fusion prediction model. It takes into account the flue gas parameter fluctuation feature set, power load scheduling plan and historical data of boiler combustion conditions, predicts the fluctuation range and trend of flue gas parameters within a preset time period, and outputs the parameter prediction result.

[0252] Operating condition reference value calibration unit: It is connected to the LSTM-BP fusion prediction unit by signal, and has a built-in multi-objective optimization model and genetic algorithm solver. Combining the thermodynamic characteristic parameters of the heat storage medium and the rated parameters of the heat storage device, it solves the dynamic operating condition reference value with the goal of maximizing the recovery of waste heat from flue gas, and outputs it to the control unit.

[0253] Fuzzy adaptive PID control unit: It is connected to the operating condition reference value calibration unit, the distributed flue gas parameter acquisition module and the thermal storage actuator respectively. It is used to collect the operating parameters of the thermal storage system and the actual parameters of the flue gas outlet, calculate the deviation from the operating condition reference value, and generate control signals through the fuzzy adaptive PID algorithm.

[0254] Thermal storage actuator: includes thermal storage medium circulation pump, heat exchange valve group, thermal storage unit switching device and medium temperature adjustment module, which is connected to the fuzzy adaptive PID control unit and adjusts the operating state in response to control signals.

[0255] Feedback correction unit: It is connected to the fuzzy adaptive PID control unit, LSTM-BP fusion prediction unit and operating condition reference value calibration unit respectively. It is used to calculate the prediction error of flue gas parameters, correct the weight of prediction model and the constraint of optimization function, update the operating condition reference value and construct closed-loop control link.

Claims

1. A thermal energy storage control method based on flue gas parameter fluctuations, characterized in that, Includes the following steps: S1: Real-time temperature, flow rate, component concentration and flue gas pressure of the boiler tail gas are collected synchronously to obtain the original flue gas parameter sequence. The original flue gas parameter sequence is preprocessed to obtain a standardized flue gas parameter dataset. S2: Decompose each parameter in the standardized flue gas parameter dataset into steady-state components, periodic fluctuation components, and random fluctuation components. Calculate the fluctuation amplitude, frequency, and duration of the periodic and random fluctuation components to obtain the flue gas parameter fluctuation feature set. Based on the flue gas parameter fluctuation feature set, use the LSTM-BP fusion prediction model to predict the fluctuation range and trend of flue gas parameters within a preset time period in the future, and output the parameter prediction results. S3: Based on the parameter prediction results, combined with the thermodynamic characteristics of the heat storage medium, the rated capacity of the heat storage device and the waste heat recovery efficiency threshold, a multi-objective optimization function is established to solve for the target heat storage temperature, medium circulation flow rate and heat exchange area matching coefficient of the heat storage system, which serve as the operating condition benchmark value for heat storage control. S4: Real-time acquisition of specified parameters of the thermal storage device, comparison with parameter prediction results and operating condition baseline values, calculation of parameter deviation, and dynamic adjustment of the PID controller correlation coefficient to control the thermal storage device based on the deviation and the fluctuation frequency of the flue gas parameter fluctuation characteristics. S5: Monitor the actual fluctuation of flue gas parameters in real time, compare the actual fluctuation with the prediction results, and calculate the prediction error; if the prediction error exceeds the preset threshold, correct the weight parameters of the LSTM-BP fusion prediction model based on the error feedback, adjust the constraints of the multi-objective optimization function, and update the benchmark value of the thermal storage system.

2. The thermal energy storage control method based on flue gas parameter fluctuations according to claim 1, characterized in that, In step S2, each parameter in the standardized flue gas parameter dataset is decomposed into steady-state components, periodic fluctuation components, and random fluctuation components. The fluctuation amplitude, frequency, and duration of the periodic and random fluctuation components are calculated to obtain the flue gas parameter fluctuation feature set. The specific process is as follows: S21: Time series decomposition based on EEMD algorithm to achieve three-component extraction including steady-state component, periodic fluctuation component and random fluctuation component; S22: Perform fluctuation characteristic calculations, including amplitude, frequency, and duration; S23: Summarize the fluctuation characteristics and combine the coupling relationships between various parameters to form a flue gas parameter fluctuation characteristic set.

3. The thermal energy storage control method based on flue gas parameter fluctuations according to claim 2, characterized in that, The specific process of step S21 is as follows: S211: Adding Noise: Using the standardized flue gas parameter sequence x(t) as input, add Gaussian white noise n with an amplitude of 0.1 to 0.4 times its standard deviation to x(t). i (t), to obtain the noisy sequence x i (t)=x(t)+n i (t); i = 1, 2, ..., M, M = 200 ~ 500 is the number of Ensemble iterations; S212: Noisy decomposition: for each x i (t) Perform EMD decomposition to obtain the intrinsic mode functions c i,j (t), j=1,2,...,K are the IMF order and the residual component r i (t); S213: Average Denoising: Take the arithmetic mean of IMFs and residual components of the same order to eliminate the influence of noise; S214: Component Classification: Random fluctuation component C r (t): The first 1 to 3 high-frequency IMFs correspond to flue gas turbulence and instantaneous sensor disturbances; Periodic fluctuation component C p (t): Intermediate-stage IMF, corresponding to the boiler combustion / load adjustment cycle; steady-state component C s (t): Residual component This characterizes the baseline steady-state level of the parameters.

4. The thermal storage control method based on flue gas parameter fluctuations according to claim 2, characterized in that, The specific process of step S22 is as follows: S221: Calculation of fluctuation range, the formula is as follows: ； Where k∈{r,p}, r corresponds to the random fluctuation component, p corresponds to the periodic fluctuation component, and C k (t) is the time series sequence of a certain type of fluctuation component after the time series decomposition of flue gas parameters, and the peak-to-peak value of the two types of components is A. r A p max(⋅) and min(⋅) are the maximum and minimum values ​​within the analysis period; S222: Fluctuation frequency calculation, the formula is as follows: Periodic component frequency f p First, calculate the autocorrelation function: ； The delay step is determined by taking the first peak value. ,but f s =1~5Hz, which is the sampling frequency; random component frequency f r : FFT transformation to the frequency domain, taking the frequency corresponding to the peak value of the power spectrum as the random dominant frequency; S223: Duration calculation, the formula is as follows: T k = N / f s ·N k / N; Among them, T k Let N be the effective duration of the k-th type of fluctuation component, representing the actual duration of this type of fluctuation within the analysis period, where k∈{r,p}, r represents the random fluctuation component, p represents the periodic fluctuation component, and N k The effective fluctuation sample number for the k-th fluctuation component refers to the number of sample points whose fluctuation amplitude exceeds twice the standard deviation of its own mean. It is used to define the range of effective fluctuation. N / f s N represents the total analysis time. k / N represents the percentage of valid fluctuation samples.

5. The thermal energy storage control method based on flue gas parameter fluctuations according to claim 1, characterized in that, In step S2, the LSTM-BP fusion prediction model is used to predict the fluctuation range and trend of flue gas parameters within a preset time period based on the flue gas parameter fluctuation feature set, and the specific process of outputting the parameter prediction results is as follows: S24: Using the extracted flue gas parameter fluctuation feature set as the core input, the power load dispatch plan data P(t) and boiler combustion condition historical data B(t) are integrated. The three types of data are aligned based on the timestamp, and load and combustion condition parameters that are strongly correlated with flue gas parameter fluctuations are filtered. The filtered data are normalized to obtain the integrated input dataset. S25: Construct a cascaded structure of LSTM feature extraction + BP regression prediction, use LSTM to capture the long-term and short-term dependencies of time series data, and optimize the nonlinear mapping capability through BP neural network. S26: Prediction and output of flue gas parameter fluctuations.

6. The thermal energy storage control method based on flue gas parameter fluctuations according to claim 5, characterized in that, The specific process of step S26 is as follows: S261: Input the fused input data into the trained LSTM-BP fusion prediction model. The LSTM layer extracts temporal correlation features, and the BP layer outputs the future T. pred Upper and lower limits of flue gas parameter fluctuations at each sampling point; S262: Based on the output fluctuation interval sequence, calculate the rate of change of the interval mean of adjacent sampling points, and make trend judgment based on the rate of change; S263: Outputs the fluctuation range, trend and confidence level of flue gas parameters within a preset time period in the future, forming the final parameter prediction result.

7. The thermal energy storage control method based on flue gas parameter fluctuations according to claim 6, characterized in that, The specific process of step S3 is as follows: S31: Taking the parameter prediction results as the core, three types of basic parameters are collected and standardized. The three types of basic parameters include the thermodynamic property parameters of the thermal storage medium, the rated parameters of the thermal storage device, and the constraint correlation parameters. S32: Construction of multi-objective optimization function: Construct a constrained multi-objective optimization function with the objectives of maximizing flue gas waste heat recovery, minimizing energy consumption of the thermal storage system, and optimizing subsequent heat exchange stability; S33: Set constraints, including: waste heat recovery efficiency constraints, temperature constraints, flow rate constraints, fluctuation adaptation constraints, and matching coefficient constraints. S34: Solve and output the baseline value using a genetic algorithm.

8. The thermal storage control method based on flue gas parameter fluctuations according to claim 7, characterized in that, The specific process of step S4 is as follows: S41: Real-time data acquisition and parameter deviation calculation: Real-time data acquisition targets: Synchronously acquire three types of core parameters during the operation of the thermal storage system, including parameters on the thermal storage device side, parameters on the flue gas side, and parameters on the control feedback side; The deviation is calculated by comparing with the benchmark, including temperature deviation, flow rate deviation, and flue gas parameter prediction deviation, and finally the comprehensive deviation and deviation change rate are obtained. S42: Set the fuzzy adaptive PID control algorithm, including dynamic tuning of PID parameters, fuzzification processing, and fuzzy rules; S43: Perform PID parameter tuning and declarative analysis: First, set the initial reference parameters for the PID, and then obtain the real-time tuning parameters through the correction amount of the fuzzy output; the declarative analysis uses the centroid method to calculate the precise value of the correction amount; S44: PID control output calculation: The incremental PID algorithm is used to calculate the control increment and control output. S45: Control signal output and actuator regulation.

9. The thermal energy storage control method based on flue gas parameter fluctuations according to claim 8, characterized in that, The specific process of step S45 is as follows: The control output calculated by PID is converted into specific execution signals according to the regulation characteristics of the controlled object, and the speed of the heat storage medium circulation pump, the opening degree of the heat exchange valve, and the switching state of the heat storage unit are adjusted respectively. The three are linked to achieve closed-loop control. S451: Speed ​​control of heat storage medium circulation pump: Pump speed n and medium circulation flow rate q V Proportional, the control signal is the speed regulation duty cycle γ n : c n =c n0 +k n ⋅u(t); Wherein: γ n0 The pump reference speed duty cycle, k n This is the speed adjustment coefficient; S452: Heat exchange valve opening control: Matching coefficient k between heat exchange valve opening θ and actual heat exchange area A Positive correlation, the control signal is the opening adjustment amount Δθ: Δθ=k θ ⋅u(t); Where: k θ This is the opening adjustment coefficient; S453: Thermal storage unit switching status control: The switching threshold is set according to the absolute value of the comprehensive deviation |e| and the flue gas fluctuation frequency f. The parallel switching / series combination of thermal storage units is realized by the switch signal output by PID control.

10. A thermal storage control system based on flue gas parameter fluctuations, used to implement the thermal storage control method based on flue gas parameter fluctuations as described in any one of claims 1-9, characterized in that, include: Distributed flue gas parameter acquisition module: Composed of several temperature sensors, flow rate sensors, gas composition sensors and pressure sensors, it is deployed at different cross sections and key nodes of the flue gas duct at the tail of the boiler to synchronously acquire the original flue gas parameter sequence. Data preprocessing and feature extraction unit: connected to the distributed flue gas parameter acquisition module, used to perform outlier removal, data smoothing and standardization on the original flue gas parameter sequence to generate a standardized flue gas parameter dataset; LSTM-BP fusion prediction unit: It is connected to the data preprocessing and feature extraction unit by signal, and has a built-in trained LSTM-BP fusion prediction model. It takes as input flue gas parameter fluctuation feature set, power load scheduling plan and historical data of boiler combustion conditions, and predicts the fluctuation range and trend of flue gas parameters within a preset time period in the future. Operating condition reference value calibration unit: It is connected to the LSTM-BP fusion prediction unit and has a built-in multi-objective optimization model and genetic algorithm solver. It combines the thermodynamic characteristic parameters of the heat storage medium and the rated parameters of the heat storage device to solve for the dynamic operating condition reference value with the goal of maximizing the recovery of waste heat from the flue gas, and outputs it to the control unit. Fuzzy adaptive PID control unit: It is connected to the operating condition reference value calibration unit, the distributed flue gas parameter acquisition module and the thermal storage actuator respectively. It is used to collect the operating parameters of the thermal storage system and the actual parameters of the flue gas outlet, calculate the deviation from the operating condition reference value, and generate control signals through the fuzzy adaptive PID algorithm. Thermal storage actuators include a thermal storage medium circulation pump, a heat exchange valve group, a thermal storage unit switching device, and a medium temperature regulation module. They are connected to a fuzzy adaptive PID control unit and adjust their operating status in response to control signals. Feedback correction unit: It is connected to the fuzzy adaptive PID control unit, the LSTM-BP fusion prediction unit and the operating condition reference value calibration unit respectively. It is used to calculate the prediction error of flue gas parameters, correct the weights of the prediction model and the constraints of the optimization function, update the operating condition reference value and construct the closed-loop control link.

Images