Boiler thermal efficiency soft measurement method, system and electronic equipment based on LTC

Abstract

The present invention provides a boiler thermal efficiency soft-measurement method, system, and electronic device based on LTC, belonging to the technical field of boiler thermal efficiency soft-measurement technology. The method comprises the following steps: collecting and preprocessing historical operating data from a unit's distributed control system; constructing an online boiler thermal efficiency soft-measurement model based on a four-layer liquid time constant network (LTC); training the model using training data and validating it with a validation set to determine the optimal network structure and hyperparameters; inputting the optimized model into a test set to output predicted values, which are then denormalized to evaluate performance; and inputting real-time operating data into the trained model to output predicted thermal efficiency values. By leveraging the LTC model's ability to adjust weights in real time using dynamic differential equations, the method overcomes the fixed parameter limitations of traditional neural network models, adapts to the time-varying nature of the measurement object, and achieves high-precision dynamic soft-measurement of boiler thermal efficiency, improving measurement accuracy and real-time response.

Description

Technical Field

[0001] The present invention belongs to the technical field of boiler thermal efficiency soft measurement, and in particular relates to a boiler thermal efficiency soft measurement method, system and electronic equipment based on LTC. Background Art

[0002] Over long-term operation, coal-fired boilers experience a gradual decline in energy conversion performance due to the variability of incoming coal quality (e.g., changes in ash, sulfur, and volatile matter) and the slow, dynamic degradation of the heating surfaces. High-ash flue gas causes dust accumulation on the radiation and convection heating surfaces, increasing heat transfer resistance and leading to higher exhaust temperatures. The coupled deposition of dust and scale reduces the heat transfer coefficient. These factors collectively lead to a gradual decline in boiler thermal efficiency with increasing operating time. Coal-fired units were initially designed as a stable power source for the power system. However, with the rapid development of renewable energy, coal-fired units are being forced to participate in deep peak regulation, shifting from a primary source to a regulating source. Under this deep peak regulation scenario, coal-fired units are forced to operate at low loads, deviating from their design conditions, for extended periods. This results in a significant and time-varying decline in boiler thermal efficiency. Therefore, developing an accurate boiler thermal efficiency prediction model is crucial for improving boiler thermal efficiency under complex operating conditions such as deep peak regulation and transient load fluctuations.

[0003] The research on boiler combustion process modeling is currently divided into two major technical branches: mechanism modeling method and data-driven modeling method. The mechanism modeling method is based on the basic theories such as the kinetic model of combustion and related physical and chemical processes, the law of conservation of energy, etc., to construct a mathematical model that reflects the internal mechanism of the process. In existing studies, some researchers have established a method to describe the relationship between boiler thermal efficiency and NO based on the distribution characteristics of coal in the furnace and the combustion mechanism of over-fire air (OFA). X Mathematical models of emission characteristics are developed. By rationally simplifying the combustion system, a discrete-time linear state-space model is constructed. Using computational fluid dynamics (CFD) technology to model the boiler combustion process is also a common technique, for example, establishing a coupled CFD model of the boiler's air, smoke, and steam / water sides.

[0004] However, using a 3 GHz Dell 3600 workstation with four threads running in parallel, the solution for a single operating condition takes approximately one week, making it difficult to meet the engineering requirements for real-time operation and control of coal-fired power plants. In the field of boiler thermal efficiency prediction, factors such as coking, wear, and component replacement on the heating surface affect the characteristics of the thermodynamic process. As operating time increases, the thermal efficiency of the boiler slowly decreases, leading to a gradual decrease in the performance match between the mechanistic model and the actual boiler operating conditions. Furthermore, the combustion process of coal-fired boilers involves complex multidisciplinary physical processes such as aerodynamics, two-phase flow, and heat and mass transfer. Online thermal efficiency prediction methods based on mechanistic models lack the accuracy and real-time performance required for engineering applications. In summary, the engineering applicability of mechanistic modeling methods is significantly limited.

[0005] Compared to mechanism-based modeling, data-driven modeling leverages machine learning algorithms to directly mine mapping relationships between system parameters from operational data. This offers the technical advantages of faster modeling and higher prediction accuracy. For industrial processes whose mechanisms are not yet fully understood, data-driven modeling approaches using machine learning algorithms, such as neural networks, are more appropriate.

[0006] Existing modeling methods do not consider the time scale factor in the input dimension design, and all belong to the category of static models, which are only applicable to the steady-state operating conditions of the boiler. Specifically, when the boiler is in the cold start process or the load switching stage during the deep peak regulation process, the historical operating conditions will have a significant impact on the current thermal efficiency. This is because the heat storage of the working fluid will be released or rebalanced during the change of boiler load, and under stable operating conditions, the heat storage effect of the working fluid can usually effectively improve the combustion efficiency and thermal efficiency, and accelerate the load response of the unit. Therefore, in view of the time-series dynamic characteristics of the boiler thermal efficiency, it is necessary to construct a dynamic prediction model and incorporate the historical operating condition parameters into the model input system. Summary of the Invention

[0007] To address the above-mentioned technical problems, the present invention provides a method, system, and electronic device for soft-sensing boiler thermal efficiency based on LTC. Liquid Time-Constant Networks (LTC) are inspired by the nervous system of the nematode Caenorhabditis elegans (C. elegans). The present invention constructs an LTC-based soft-sensing model for boiler thermal efficiency. This model employs a Neural Circuit Policies (NCP) architecture and consists of four layers of neurons: sensory neurons, interneurons, command neurons, and motor neurons. Each neuron in each layer is constructed using an LTC model. The LTC model uses dynamic differential equations to adjust weights in real time, effectively overcoming the limitation of previous soft-sensing models, which often retain fixed parameters after training. This model enables high-precision dynamic soft-sensing of boiler thermal efficiency, improving the accuracy and real-time performance of boiler thermal efficiency measurement.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] A soft measurement method for boiler thermal efficiency based on LTC includes the following steps:

[0010] Step 1: Collect historical operating data of the unit's distributed control system;

[0011] Step 2: Preprocess the historical operation data obtained in step 1, including missing data processing, data dimensionality reduction, and data reconstruction, to form a data set, which is divided into a training set, a validation set, and a test set;

[0012] Step 3: Construct a soft-sensing model of boiler thermal efficiency based on LTC, which includes four layers of neurons connected in sequence: a sensory neuron layer as an input layer, an intermediate neuron layer, a command neuron layer, and a motor neuron layer as an output layer; LTC stands for liquid time constant network;

[0013] Step 4: Use the training set to train the LTC-based boiler thermal efficiency soft measurement model constructed in step 3, optimize the model hyperparameters, and use the validation set to determine the optimal hyperparameter settings to obtain the optimized LTC-based boiler thermal efficiency soft measurement model;

[0014] Step 5: Input the test set data into the optimized LTC-based boiler thermal efficiency soft measurement model obtained in step 4 to obtain the prediction results. After denormalization, the prediction performance of the LTC-based boiler thermal efficiency soft measurement model under different operating conditions is evaluated using the evaluation function.

[0015] Step 6: Input the real-time collected boiler operation data into the optimized LTC-based boiler thermal efficiency soft measurement model obtained in step 4, and output the predicted value of boiler thermal efficiency.

[0016] Furthermore, in step 1, the historical operating data includes boiler thermal efficiency and auxiliary variables; wherein, boiler thermal efficiency is the percentage of boiler effective heat utilization to fuel input heat; auxiliary variables include fuel parameters, operating condition parameters, flue gas parameters and environmental parameters.

[0017] Furthermore, in step 2, the missing data processing includes: detecting abnormal data using the 3σ-rule, and replacing the abnormal data based on the rate of change of the auxiliary variables at adjacent moments.

[0018] Furthermore, in step 2, the data dimensionality reduction includes: using a random forest algorithm to sort the auxiliary variables by importance, and selecting the top 10 auxiliary variables as input variables.

[0019] Furthermore, in step 2, data reconstruction includes: using a sliding window method to convert the time series data into a two-dimensional tensor, where the time dimension includes 20 consecutive sampling moments, and the feature dimension is the top 10 auxiliary variables of importance screened.

[0020] Furthermore, in step 3, each neuron of the four neuron layers of the LTC-based boiler thermal efficiency soft-sensing model is constructed using an LTC model.

[0021] The present invention also provides a boiler thermal efficiency soft measurement system for implementing the above-mentioned boiler thermal efficiency soft measurement method based on LTC, comprising:

[0022] Data acquisition module, used to obtain historical operating data of the unit's distributed control system;

[0023] A preprocessing module, configured to perform missing data processing, data dimensionality reduction, and data reconstruction on the historical operation data;

[0024] The model operation module is used to deploy the LTC-based boiler thermal efficiency soft measurement model and perform prediction calculations to obtain the boiler thermal efficiency prediction results;

[0025] The visualization terminal is used to display the boiler thermal efficiency prediction result online.

[0026] The present invention also provides an electronic device, comprising a processor and a memory, wherein the memory stores a computer program, and when the processor executes the program, the steps of the above-mentioned LTC-based boiler thermal efficiency soft measurement method are implemented.

[0027] Furthermore, the electronic device is integrated into the unit distributed control system or edge computing terminal to receive boiler operation data in real time and output the boiler thermal efficiency prediction result.

[0028] Beneficial effects:

[0029] The LTC-based soft-sensing model for boiler thermal efficiency constructed by the present invention uses an LTC model for each neuron in each layer. The LTC model uses dynamic differential equations to adjust weights in real time. This effectively overcomes the limitation of previous soft-sensing models, which often retain fixed parameters after training, and enables high-precision dynamic soft-sensing of boiler thermal efficiency, improving both measurement accuracy and real-time performance. Furthermore, the present invention designs a dynamic soft-sensing method applicable to boiler operating data under different operating conditions. This method can adapt to the multivariate and nonlinear characteristics of complex operating conditions, exhibits strong robustness, interpretability, generalization capabilities, and is computationally efficient. Specifically, the present invention utilizes a random forest algorithm to preprocess boiler operating data, particularly for dimensionality reduction. Subsequently, the reduced eigenvalues ​​are input into the LTC-based soft-sensing model for rapid modeling and prediction, thereby achieving real-time, dynamic measurement of boiler thermal efficiency. Finally, the LTC-based soft-sensing model is evaluated using real data from a coal-fired boiler under variable load conditions. The electronic device provided by the present invention can automate the aforementioned method, further improving the efficiency and practicality of soft-sensing. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Flowchart of the soft measurement method of boiler thermal efficiency based on LTC of the present invention;

[0031] Figure 2 Schematic diagram of a single burner in the embodiment: (a) is a schematic diagram of the rich fuel nozzle and the exhaust gas nozzle; (b) is a schematic diagram of the arrangement of the secondary air nozzles and air ducts at different levels in the boiler furnace;

[0032] Figure 3 Schematic diagram of data reconstruction in the embodiment: (a) time series variables, (b) tensor graph obtained by sliding window;

[0033] Figure 4 Schematic diagram of data set segmentation in the embodiment;

[0034] Figure 5 Schematic diagram of a closed-form continuous-time (CFC) neuron;

[0035] Figure 6 This is a structural diagram of a soft-sensing model for boiler thermal efficiency based on LTC in an embodiment;

[0036] Figure 7Graphs of iterative loss values ​​of the boiler thermal efficiency soft-sensing model for different activation functions in the embodiment; (a) is the tanh hyperbolic tangent activation function, (b) is the sigmoid activation function, and (c) is the rectified linear unit (ReLU) activation function;

[0037] Figure 8 The following are the prediction results of LSTM, GRU and LTC on the test set in the embodiment; (a) is the load reduction interval, (b) is the load increase interval, and (c) is the steady-state interval;

[0038] Figure 9 Graph showing the correlation between unit load and boiler thermal efficiency in the embodiment. DETAILED DESCRIPTION

[0039] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the following embodiments may be combined with each other without conflict.

[0040] like Figure 1 As shown, the present invention provides a dynamic soft measurement method for boiler thermal efficiency based on LTC, comprising the following steps:

[0041] Step 1: Collect historical operation data;

[0042] Step 2: Preprocess the historical operation data, including missing data processing, data dimensionality reduction, data reconstruction, and divide the preprocessed data into training set, validation set and test set;

[0043] Step 3: construct a boiler thermal efficiency soft measurement model based on LTC, which includes four neuron layers connected in sequence: sensory neuron layer (input layer), intermediate neuron layer, command neuron layer and motor neuron layer (output layer); that is, Figure 1 The four-layer network described in step 3: sensory, intermediate, command, and motor neurons are connected in sequence;

[0044] Step 4: Use the training set to train the LTC-based boiler thermal efficiency soft measurement model, and use the validation set to determine the optimal hyperparameter settings to obtain the optimized LTC-based boiler thermal efficiency soft measurement model;

[0045] Step 5: Input the test set into the LTC-based boiler thermal efficiency soft measurement model to obtain the prediction results, perform denormalization, and perform evaluation and analysis;

[0046] Step 6: Real-time data collection is input into the LTC-based boiler thermal efficiency soft measurement model to output the dynamic thermal efficiency prediction value.

[0047] Furthermore, in step 1, the historical operating data includes thermal efficiency and auxiliary variables related thereto (as shown in Table 2 below).

[0048] Furthermore, in step 2, the missing data processing step is: using the 3σ-rule to detect abnormal data and replacing the abnormal data based on the following formula:

[0049] ;

[0050] Among them, i represents the auxiliary variable index value (i-th variable), t represents the current time point, and k is the time window length. The index of the historical sampling time in the window, ranging from [2, t] or [tk, t-1], represents the value of the i-th auxiliary variable at time t, represents the value of the i-th auxiliary variable at time t-1, Represents auxiliary variables The rate of change at adjacent moments t and t-1 is, express The average rate of change before time t, Represents standard deviation.

[0051] Furthermore, in step 2, the step of data dimensionality reduction is: using the random forest algorithm (RF) to sort the importance of auxiliary variables, and selecting the top 10 important variables as input variables.

[0052] Furthermore, in step 2, the data reconstruction step is: using a sliding window method to convert the time series data into a two-dimensional tensor, where the time dimension contains 20 consecutive sampling moments and the feature dimension is the 10 variables after screening.

[0053] Furthermore, in step 3, the LTC-based soft-sensing model for boiler thermal efficiency includes four neuron layers, and each neuron in each layer is constructed using the LTC model.

[0054] Furthermore, in step 4, the hyperparameters of the LTC-based boiler thermal efficiency soft measurement model are set as follows: the learning rate is set to 0.005, the activation function adopts ReLU (Rectified Linear Unit), the number of neurons in each layer is 128, the number of iterations is 200, and the optimizer adopts Adam.

[0055] The present invention will be further explained below with reference to the accompanying drawings and embodiments.

[0056] Example

[0057] A boiler thermal efficiency soft measurement method based on LTC in this embodiment includes the following steps:

[0058] Step 1: Collect historical operating data from the unit's distributed control system (DCS).

[0059] This example studies a W-flame boiler in a 600MW supercritical thermal power plant. The boiler features supercritical parameters, W-flame combustion, and variable-pressure operation. The boiler's main design parameters are shown in Table 1. The power plant is located in the heart of southwestern China, where the climate is humid. The plant burns low-quality anthracite with a volatile content below 12%, resulting in significant coking within the furnace. This coking phenomenon severely impacts the combustion atmosphere and heat transfer within the furnace, making it difficult to predict the boiler's thermal efficiency using mechanistic models. Therefore, data-driven modeling is more suitable for this type of industrial scenario.

[0060] Table 1 Main design parameters of boiler

[0061]

[0062] The target boiler is equipped with 6 double-inlet and double-outlet coal mills and 24 double-cyclone direct current pulverized coal burners specially designed for burning low-volatile coal. Figure 2 The diagram shows a single burner. Figure 2 (a) shows the structural appearance of the fuel-rich nozzle and the exhaust gas nozzle in the combustor; Figure 2 (b) presents the position distribution of the secondary air dampers (A-layer, B-layer, C-layer) and secondary air dampers (D-layer, F-layer) at different layers on the burner as well as OFA (overburn air). Figure 2 In this model, OFA (Over-Fire Air) refers to over-fire air, which is injected in the late stages of combustion to provide additional air for the fuel burnout phase, promoting further reaction of incompletely burned substances, improving combustion efficiency, and reducing incomplete combustion losses. Fuel-rich nozzles are specifically designed to inject a fuel-rich mixture, providing the primary fuel source for combustion. Exhaust gas nozzles are used to inject exhaust gas, generated during processes such as pulverizing, which carries a certain amount of energy and substances. This exhaust gas participates in the combustion reaction and also helps organize the airflow. The secondary air for each burner is individually controlled. The air distribution unit consists of an upper and lower air box. The F damper provides the maximum airflow and controls the primary secondary air volume required for combustion. The opening degrees of the F-layer secondary air dampers of the 24 burners are included as auxiliary variables required for modeling.

[0063] The thermal efficiency of coal-fired power generation boilers is mainly determined by the counter-balance method, that is, the various heat losses of the boiler are measured through experiments, and then the thermal efficiency of the boiler is calculated according to the following formula :

[0064] ;

[0065] Where, It represents the percentage of each heat loss to the input heat. , m=2,3,…,6, is the input heat, where is the percentage of exhaust heat loss to input heat (%); is the percentage of heat loss from incomplete combustion of gas to input heat (%); is the percentage of heat loss due to incomplete combustion of solids to input heat (%); The percentage of heat loss from the boiler to the input heat (%); is the percentage (%) of other heat losses to the input heat. is the heat loss of the boiler, m=2,3,…,6, where is the heat loss from exhaust (kJ / kg); is the heat loss due to incomplete combustion of gas (kJ / kg); is the heat loss from incomplete combustion of solid (kJ / kg); Heat loss due to heat dissipation by the boiler (kJ / kg); is the heat loss of other heat (kJ / kg). By calculating each heat loss, we can understand the boiler operation status and find optimization measures to improve the boiler thermal efficiency. Analyze the variables that affect each heat loss and then compile a list of variables that affect thermal efficiency. It is the largest heat loss in the boiler, accounting for 5% to 6% in general. The increase will be about 1%. Therefore, the exhaust gas temperature is selected as the input variable for predicting the thermal efficiency of the boiler. In the coal-fired boiler, the only product of incomplete combustion of gas is CO. Since the target boiler is not equipped with a CO sensor, the CO content is not considered in the auxiliary variable. Second only to flue gas heat loss, this heat loss is primarily influenced by factors such as coal quality, combustion method, flue gas oxygen content, coal blending, and air distribution. Therefore, auxiliary variables include flue gas oxygen content, mill capacity airflow (coal blending), and secondary air damper opening (air distribution).

[0066] The present invention integrates the above variables to form an auxiliary variable table based on the combustion mechanism, expert experience and the advice of operation engineers. The auxiliary variable table includes 53 parameters related to air, coal and water. The detailed information of all auxiliary variables is shown in Table 2. The selection of auxiliary variables comprehensively considers the influencing factors such as the aerodynamic field in the furnace, the air-coal two-phase flow field and the heat absorption of the working fluid. Among them, the secondary air controls the main air required for combustion in the furnace, the pulverizer inlet capacity air flow controls the distribution of the air-coal two-phase flow field in the furnace, and variables such as the main steam pressure and main steam temperature reflect the heat absorption level of the working fluid in the furnace.

[0067] Table 2 Auxiliary variable table

[0068]

[0069] Step 2: Preprocess the collected historical operation data.

[0070] Approximately 10 days of continuous operation data, totaling 14,419 samples, were extracted from the target boiler's DCS. The raw data was preprocessed to meet the model input requirements. This preprocessing process included missing data handling, dimensionality reduction, and data reconstruction. The dataset was then divided into training, validation, and test sets.

[0071] Step 2.1 Missing data processing:

[0072] Due to the complex environment of boiler production sites and the presence of noise in the signals, the raw data stored in the DCS contains abnormal values ​​to varying degrees. This embodiment uses the 3σ-rule to detect abnormal data, as shown in the formula above.

[0073] Step 2.2 Data dimensionality reduction:

[0074] As research into boiler combustion process modeling deepens, researchers have discovered that reducing the dimensionality of input variables not only improves model learning efficiency but also enhances generalization performance. Yang et al. used principal component analysis to reduce the dimensionality of 35 original boiler variables to six principal components for predicting NOx emissions. Tang et al. used mutual information to identify 13 variables highly correlated with NOx emissions. They used the random forest (RF) algorithm to rank the auxiliary variables by importance, selecting the top 10 as input variables.

[0075] Random Forest is an extended variant of Bagging, which is particularly suitable for evaluating the importance of high-dimensional nonlinear variables. RF uses several independent decision trees for parallel prediction. Its random sampling with replacement makes the out-of-bag data (OOB) naturally have the function of evaluating the model prediction error. The degree of deterioration of this error indirectly reflects the importance of the variable. For each decision tree, the in-bag data is used to train the tree, and the out-of-bag data (OOB) is used to evaluate the tree. The deterioration of the prediction accuracy will be achieved by applying noise interference to the important variables. Before and after the interference, the importance of the input variable j is The measurement is calculated using the following formula:

[0076] ;

[0077] in, Quantifies the influence of input variable j on the prediction accuracy of the model. The larger the value, the more important the variable is to the model. and are the prediction errors of out-of-bag data on decision tree b before and after interference, is the number of decision trees, b is the order of the decision tree. The sum of the importance of all input variables is 1, that is , j=1,2,…,n represents the auxiliary variable sequence number, and n represents the total number of auxiliary variables.

[0078] This embodiment uses a random forest algorithm to reduce dimensionality based on the high-dimensional nonlinearity and large inertia of the boiler combustion system.

[0079] The RF algorithm was used to evaluate the feature importance of 53 comprehensive auxiliary variables that affect thermal efficiency, and the decision tree was set to 500. The sum of the importance of the top 6 variables exceeded 0.9. In this example, combined with the knowledge of boiler combustion mechanism, the F-layer secondary air damper, which mainly affects the combustion in the furnace, was also included as an input variable. The load has a significant impact on the overall unit. Finally, 10 variables (x1~x2) were selected. 10 The total weight of the top 10 variables in terms of importance reached 0.933, and the re-ranking is shown in Table 3.

[0080] Table 3 Importance of thermal efficiency-related variables (mean ± standard value)

[0081]

[0082] According to the evaluation results of the importance of each variable, its necessity in thermal efficiency modeling is analyzed. From Table 3, it can be seen that in the importance ranking of variables in RF calculation, exhaust temperature has the highest importance, and the importance of a single variable occupies an absolute advantage, which is related to the exhaust heat loss. This is consistent with the fact that it is the maximum heat loss of the boiler. In addition, the excess air coefficient at the furnace outlet of a coal-fired boiler has a great influence on the thermal efficiency. The excess air coefficient at the furnace outlet is generally characterized by the oxygen content of the flue gas, which refers to the ratio of the actual amount of air required for coal supply to the theoretical amount of air. The larger the excess air coefficient, the more actual air volume exceeds the theoretical required amount of air, that is, there is an excess of air supply during the boiler combustion process. The influence trend of the excess air coefficient on the thermal efficiency of a coal-fired boiler is relatively complex. When the value of the excess air coefficient is relatively small, so that it cannot guarantee the air volume requirement for fuel combustion, it will obviously cause mechanical incomplete combustion heat loss. Increasing the excess air coefficient will reduce heat loss. However, if the excess air coefficient can already meet the needs of combustion in the furnace, increasing the excess air coefficient will increase the flow rate of the flue gas and reduce the heat loss of the exhaust gas. The fuel will also increase, while the residence time of the fuel in the furnace will be shortened, and the flue gas's ability to carry large particle fuel will be enhanced, which will cause the incomplete combustion of solids to lose heat. Therefore, there is an optimal excess air coefficient that can guarantee both the air volume required for fuel combustion and the combustion time. At this time, the heat loss from incomplete combustion of solids is The minimum value will be reached, and the boiler thermal efficiency will also remain high. Therefore, in actual operation, it is necessary to find a suitable balance by adjusting the excess air coefficient to achieve the optimal thermal efficiency of coal-fired boilers. The above analysis shows that the variable importance calculated by RF conforms to the changing pattern of boiler thermal efficiency. Based on this importance ranking and boiler combustion knowledge, this example selects the top 10 variables as the model input variables.

[0083] Step 2.3 Data reconstruction:

[0084] To construct an input structure suitable for time series modeling, this paper uses a sliding window method to convert data into a two-dimensional tensor (time dimension x feature dimension). The data is segmented according to the feature dimension and time dimension. The time dimension contains 20 sampling moments, i.e., H = 20; the feature dimension contains 10 variable types, i.e., W = 10. The window slides rightward along the time dimension, with a step size of 1. The label corresponding to each two-dimensional tensor is the boiler thermal efficiency of the last time dimension of the tensor. Figure 3 The figure shows a schematic diagram of data reconstruction, where: Figure 3 (a) represents the time series variable, where variable-1 represents the load in Table 3, variable-2 represents the Mill-E-left in Table 3, and so on. In addition, output-y represents the predicted target variable, corresponding to thermal efficiency, and label-1 represents the first label; Figure 3(b) represents the tensor graph obtained by sliding the window, in which x represents the variable, the superscript of x represents the sampling time of the variable, and the subscript of x represents the variable type 1, 2, ... W; y H Represents the thermal efficiency at the Hth sampling moment.

[0085] Step 2.4 Dataset Segmentation:

[0086] like Figure 4 The figure shows a dataset segmentation diagram. The feature dimensions are 10 variables (feature #1-feature #10) with high correlation with thermal efficiency. t represents the sampling time. The original data from 14,419 sampling times is segmented into 14,400 standard input matrices. The labels represent the predicted target variable y, i.e., thermal efficiency. The original sampling sequence of the labels corresponding to each matrix is ​​20-14,419. In this invention, these are renumbered from 1 to 14,400 to align with the matrix sequence. That is, the sequence numbers of the input matrix and labels all start at 1. The segmented data is divided into training, validation, and test sets in a 6:2:2 ratio. The training set consists of 8,640 sampling points, the validation set consists of 2,880 sampling points, and the test set consists of 2,880 sampling points.

[0087] Step 3: Construct a soft-sensing model for boiler thermal efficiency based on LTC. Its network structure consists of four sequentially connected neuron layers: the sensory neuron layer (input layer), the intermediate neuron layer, the command neuron layer, and the motor neuron layer (output layer). These layers are connected by feedforward connections, while the command neurons have recurrent connections. Each neuron in each layer uses the liquid time constant (LTC) model.

[0088] The Liquid Time Constant (LTC) model is based on ordinary differential equations (ODEs), whose temporal behavior adaptively adjusts weights in real time based on input, making it a universal approximator for modeling system dynamics. The LTC model and its closed-form approximation, the CFC model, are inspired by the hierarchical information processing mechanism of biological nervous systems (C. elegans), consisting of a four-layer topology consisting of sensory neurons, interneurons, command neurons, and motor neurons. The LTC model is represented by the following ordinary differential equation:

[0089] ;

[0090] ;

[0091] Where: represents a nonlinear synapse, ; represents nonlinear synaptic release, A and τ represent synaptic reversal potential and time constant, respectively. θ is a set of parameters used to adjust the behavior of nonlinear synaptic release function. Its specific value and meaning depend on the adopted Function form, t is time.

[0092] Solving the ordinary differential equations for LTC typically involves computationally intensive iterative methods. A closed-form expression has been introduced to approximate the solution with lower complexity. The closed-form continuous-time (CFC) model significantly reduces the computational complexity through approximate solution. Figure 5 The structural diagram of the CFC model is shown. This model achieves a closed-form approximate solution to the ordinary differential equation of the LTC model through a combination of specific neural network branches (f, g, h), activation functions (such as Sigmoid), and operations (such as Hadamard product and addition). Figure 6 This is the structure diagram of the soft measurement model of boiler thermal efficiency based on LTC. The input data passes through sensory neurons (input), intermediate neurons, command neurons, motor neurons (output), and finally output.

[0093] Compared to the LTC model that requires a numerical solver, CFC significantly improves training and inference efficiency. This layer efficiently extracts time-related dynamic features, and its core operation can be expressed as:

[0094] x ( t ) = σ ( − f ( x ( t ), i ( t ); θ f ) t ) ⊙ g( x ( t ), i ( t ); θ g ) + [ 1- σ (- f ( x ( t ), i ( t ); θ f ) t ) ] ⊙ h( x ( t ), i ( t ); θ h ) ;

[0095] Where f, g and h are trainable neural layers (parameters are θ f ,θ g and θ h ), represents input, σ represents sigmoid function, ⊙ represents Hadamard product (element-wise product), is the hidden state, and t represents the time.

[0096] Step 4: Use the training set to train the LTC-based boiler thermal efficiency soft measurement model, optimize the network hyperparameters, and use the validation set to determine the optimal hyperparameter settings to obtain the optimized LTC-based boiler thermal efficiency soft measurement model.

[0097] The LTC model uses a four-layer network structure based on the NCP architecture. The hyperparameters of the LTC model include: a learning rate of 0.005, a ReLU activation function, 128 neurons in each layer, 200 iterations, and an Adam optimizer.

[0098] Figure 7 The training loss values ​​under different activation functions (tanh, Sigmoid, ReLU) are compared, and the results show that the loss value is the lowest when the ReLU activation function is used. Figure 7 (a) is the tanh hyperbolic tangent activation function, Figure 7 (b) is the Sigmoid activation function, Figure 7 (c) is the rectified linear unit ReLU activation function.

[0099] Step 5: Input the test set data into the optimized boiler thermal efficiency soft measurement model to obtain the prediction results. After denormalization, the prediction performance of the model under different working conditions is analyzed and evaluated using the evaluation function.

[0100] To evaluate model performance, the LTC model was compared with LSTM and GRU. The mean (Mean) and standard deviation (Std) of the model prediction error were obtained from twenty independent experiments.

[0101] The LTC network was compared with the LSTM and GRU models using the test set. Both the LSTM and GRU models use a single hidden layer structure, with 128 neurons in the LSTM and 128 in the GRU. The output of the last neuron in the LSTM and GRU is used as the overall prediction result of the model.

[0102] The mean and standard deviation of the prediction errors of LSTM, GRU, and LTC networks are shown in Table 4.

[0103] Table 4 Prediction performance of LSTM, GRU and LTC on the test set

[0104]

[0105] The RMSE of the LTC network is 0.091 ± 0.005%, the MAE is 0.051 ± 0.005%, and the R 2 The root mean square error RMSE, mean absolute error MAE and determination coefficient R of the LTC network are 0.986 ± 0.002. 2 The indicators are better than LSTM and GRU networks.

[0106] Randomly select one experiment from the twenty experiments. Figure 8 The boiler thermal efficiency prediction curve and load curve of a random experiment are shown. Figure 8 (a) Figure 8 (b) Figure 8 (c) corresponds to load reduction, load increase, and steady-state conditions, respectively. In the load reduction range, the deviation between the boiler thermal efficiency predicted by the LTC network and the true value is minimal. In the load increase range, the LTC model has the smallest prediction deviation. In the steady-state range, the prediction performance of the three models is similar. Figure 9 The Pearson's coefficient r between load and thermal efficiency is 0.826, confirming a strong positive correlation between the two. In summary, the LTC-based boiler thermal efficiency prediction model exhibits low prediction errors under both transient (load ramping) and steady-state conditions.

[0107] Step 6: Input the real-time collected boiler operation data into the trained boiler thermal efficiency soft measurement model to output the boiler thermal efficiency prediction value.

[0108] The present invention also provides a system for implementing the LTC-based dynamic soft measurement method for boiler thermal efficiency, comprising:

[0109] Data acquisition module, collecting historical operation data;

[0110] Preprocessing module, used to perform missing data processing, data dimensionality reduction, and data reconstruction on historical operation data;

[0111] The model operation module is used to deploy the LTC-based boiler thermal efficiency soft measurement model and perform prediction calculations to obtain the boiler thermal efficiency prediction results;

[0112] Visualization terminal, used to display boiler thermal efficiency prediction results online.

[0113] The present invention also provides an electronic device, comprising a processor and a memory, wherein the memory stores a computer program, and when the processor executes the program, the steps of the above-mentioned LTC-based boiler thermal efficiency soft measurement method are implemented.

[0114] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned LTC-based boiler thermal efficiency soft measurement method are implemented.

[0115] Furthermore, the electronic device is integrated into the DCS system or edge computing terminal of the coal-fired power plant, receives boiler operation data in real time and outputs boiler thermal efficiency prediction results.

[0116] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk drives, CD-ROMs, optical storage devices, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0117] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0118] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0119] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0120] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0121] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A soft measurement method for boiler thermal efficiency based on LTC, characterized in that: The following steps are involved: Step 1: Collect historical operating data of the unit's distributed control system; the historical operating data includes boiler thermal efficiency and auxiliary variables; boiler thermal efficiency is the percentage of boiler effective heat utilization to fuel input heat; auxiliary variables include fuel parameters, operating condition parameters, flue gas parameters, and environmental parameters; Step 2: Preprocess the historical operating data obtained in step 1, including: missing data processing, data dimensionality reduction, and data reconstruction, to form a data set, and divide the data set into a training set, a validation set, and a test set. Missing data processing includes: using the 3σ-rule to detect abnormal data and replacing abnormal data based on the rate of change of auxiliary variables at adjacent moments. Data dimensionality reduction includes: using the random forest algorithm to sort the auxiliary variables by importance and screening the top 10 auxiliary variables as input variables. Data reconstruction includes: using the sliding window method to convert the time series data into a two-dimensional tensor, where the time dimension contains 20 consecutive sampling moments and the feature dimension is the top 10 auxiliary variables screened by importance. Step 3: Construct a boiler thermal efficiency soft-sensing model based on LTC, which includes four sequentially connected neuron layers: a sensory neuron layer as an input layer, an intermediate neuron layer, a command neuron layer, and a motor neuron layer as an output layer. LTC stands for liquid time constant network. Each neuron in the four neuron layers of the LTC-based boiler thermal efficiency soft-sensing model is constructed using an LTC model. There are feedforward connections between the sensory neuron layer, the intermediate neuron layer, the command neuron layer, and the motor neuron layer as the output layer, and there are recurrent connections between the command neurons. Step 4: Use the training set to train the LTC-based boiler thermal efficiency soft measurement model constructed in step 3, optimize the model hyperparameters, and use the validation set to determine the optimal hyperparameter settings to obtain the optimized LTC-based boiler thermal efficiency soft measurement model; Step 5: Input the test set data into the optimized LTC-based boiler thermal efficiency soft measurement model obtained in step 4 to obtain the prediction results. After denormalization, the prediction performance of the LTC-based boiler thermal efficiency soft measurement model under different operating conditions is evaluated using the evaluation function. Step 6: Input the real-time collected boiler operation data into the optimized LTC-based boiler thermal efficiency soft measurement model obtained in step 4, and output the predicted value of boiler thermal efficiency.

2. A boiler thermal efficiency soft measurement system for implementing the boiler thermal efficiency soft measurement method based on LTC according to claim 1, characterized in that: include: Data acquisition module, used to obtain historical operating data of the unit's distributed control system; A preprocessing module, configured to perform missing data processing, data dimensionality reduction, and data reconstruction on the historical operation data; The model operation module is used to deploy the LTC-based boiler thermal efficiency soft measurement model and perform prediction calculations to obtain the boiler thermal efficiency prediction results; The visualization terminal is used to display the boiler thermal efficiency prediction result online.

3. An electronic device comprising a processor and a memory, characterized in that: The memory stores a computer program, and when the processor executes the program, the steps of the soft measurement method of boiler thermal efficiency based on LTC according to claim 1 are implemented.

4. The electronic device according to claim 3, wherein: The electronic device is integrated into the unit's distributed control system or edge computing terminal, receives boiler operation data in real time, and outputs boiler thermal efficiency prediction results.

Images