Coal-fired unit multi-dimensional index optimization method and device based on neural network multi-model fusion prediction

Abstract

The invention discloses a coal-fired unit multi-dimensional index optimization method and device based on neural network multi-model fusion prediction, and the method comprises the steps: collecting the main steam temperature and target parameters meeting the multi-dimensional index prediction after a target unit operates under variable working conditions; calculating primary load damage and power generation coal consumption rate of key components of the coal-fired unit according to the collected target parameters, and performing multi-model fusion prediction on three key indexes including the main steam temperature, the primary load damage and the power generation coal consumption rate; and a multi-objective optimization model is constructed based on the prediction result, and the optimized main steam temperature gradient, the main steam pressure gradient and the power generation power change rate are obtained, so that the main steam temperature, the primary load damage and the power generation coal consumption rate are optimized, and the economical efficiency and the stability of the unit are both considered under the safe operation condition. A reliable technical scheme is provided for the coal-fired unit to realize safe, economical, stable and flexible peak regulation operation under the renewable energy source permeation background.

Description

Technical Field

[0001] This invention relates to the field of energy information technology, and in particular to a method and apparatus for optimizing multi-dimensional indicators of coal-fired power units based on neural network multi-model fusion prediction. Background Technology

[0002] With the deepening of energy structure transformation, the application of renewable energy sources such as wind power and photovoltaics in the power system continues to increase. Coal-fired units, as the main power source in the current power supply, need to undertake deep peak-shaving tasks to ensure the stable operation of the power grid. Under low-load conditions, the unit's output power and main steam parameters easily deviate from the design conditions, causing increasingly prominent issues related to economy, safety, and stability. Under low-load conditions, the main steam temperature drops significantly, deviating from the design value and affecting boiler thermal efficiency, leading to a significant increase in the unit's coal consumption rate for power generation, thus reducing the unit's operational economy. Frequent start-ups and shutdowns and drastic load fluctuations may cause critical components to bear more severe working loads, thereby affecting the unit's safety. Furthermore, frequent fluctuations in main steam temperature cause fluctuations in thermal parameters, increasing the difficulty of regulation and control, thus weakening the unit's operational stability. To ensure the stability of the unit's power generation, it is necessary to constantly monitor and maintain the main steam temperature within a reasonable range.

[0003] During the deep peak shaving process with frequent load changes in power units, the main steam temperature, primary load damage, and power generation coal consumption rate are interdependent and mutually inversely related, becoming the core challenge for the flexible operation of coal-fired power units. Therefore, real-time prediction of power generation coal consumption rate, primary load damage, and main steam temperature under the deep peak shaving and rapid load change operation conditions of coal-fired power units, and the realization of multi-objective collaborative optimization based on this prediction, are of great significance for the economic, safe, and stable operation of coal-fired power units.

[0004] Traditional methods for predicting main steam temperature and load damage include mechanistic models based on heat conduction equations and statistical models. However, these methods have poor adaptability and insufficient real-time performance under off-design conditions. In recent years, data-driven methods such as neural networks have been widely used for predicting key unit parameters due to their strong nonlinear fitting capabilities. For example, models such as LSTM have high prediction accuracy under normal operating conditions, but their generalization performance deteriorates under sudden load changes or abnormal conditions. Existing research has not yet fully combined the advantages of various neural network models to improve the robustness and accuracy of predictions under drastic load changes.

[0005] In terms of optimization algorithms, when the optimization objective is output by a deep learning prediction model, the objective function often exhibits highly non-convex and non-differentiable "black box" characteristics, making traditional gradient optimization methods difficult to apply. Intelligent optimization algorithms such as particle swarm optimization (PSO) can handle such multi-modal problems and have been applied in combustion and energy-saving optimization of coal-fired power units. However, their slow convergence speed, premature convergence, and sensitivity to parameter settings limit their optimization efficiency. Therefore, it is necessary to design a fast-converging and robust optimization algorithm to solve multi-objective optimization problems.

[0006] In summary, existing research has achieved significant results in the prediction and operational optimization of key indicators, but the following problems still need to be addressed. At the prediction level, both single-mechanism models and data-driven models have obvious limitations. How to effectively integrate mechanistic knowledge and data information to construct a hybrid prediction model adaptable to the diverse operating conditions of deep peak shaving remains a challenge. At the optimization level, traditional mathematical programming methods struggle to handle optimization problems with neural network embeddings. Designing efficient and reliable optimization algorithms to accelerate convergence and avoid local optima is a major challenge in current research. Therefore, there is an urgent need for a method that integrates multi-model fusion prediction and efficient intelligent optimization to improve the overall performance of coal-fired power units in peak shaving operation. Summary of the Invention

[0007] Purpose of the invention: This invention provides a method for optimizing multi-dimensional indicators of coal-fired power units based on neural network multi-model fusion prediction. It solves the problems of low prediction accuracy, poor real-time performance, and difficulty in optimization solutions of multi-dimensional indicators of coal-fired power units in related technologies. It realizes dynamic prediction of key indicators of the unit and multi-objective real-time optimization of operating parameters, thereby improving the safety, economy and stability of the unit's peak-shaving operation.

[0008] Technical solution: The method described in this invention includes the following steps:

[0009] After the target unit operates under varying conditions, the main steam temperature and target parameters that meet the multi-dimensional index prediction are collected.

[0010] Based on the collected target parameters, the primary load damage and power generation coal consumption rate of key components of the coal-fired unit are calculated. Parameters that have a significant impact on the main steam temperature, primary load damage, and power generation coal consumption rate are selected from the target parameters. The selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate are divided into training and testing sets, respectively. A prediction model for main steam temperature, a prediction model for primary load damage, and a prediction model for power generation coal consumption rate are constructed. The training set is used to perform multi-model fusion prediction on the three key indicators of main steam temperature, primary load damage, and power generation coal consumption rate.

[0011] Based on the prediction results, the main steam temperature, primary load damage, and power generation coal consumption rate are optimized as objectives. The main steam temperature gradient, main steam pressure gradient, and power generation change rate are used as decision variables. Constraints are set, and a multi-objective optimization model is established. The multi-objective optimization model is solved to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate, thereby optimizing the main steam temperature, primary load damage, and power generation coal consumption rate.

[0012] Furthermore, the target parameters that satisfy the multidimensional index prediction are the parameters that affect the main steam temperature, primary load damage and power generation coal consumption rate during the unit's variable operating condition operation.

[0013] Principal component analysis was used to denoise the collected main steam temperature and the target parameters that meet the multidimensional index prediction.

[0014] Furthermore, the Pearson correlation coefficient method was used to analyze the correlation between each target parameter and the main steam temperature, primary load damage, and power generation coal consumption rate. Parameters with absolute values ​​of correlation coefficients with the main steam temperature, primary load damage, and power generation coal consumption rate that are greater than the set threshold were selected. The selected parameters include: coal feed rate, water feed rate, main steam flow rate, power generation capacity, water spray desuperheater flow rate, and main steam pressure.

[0015] The selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate are divided into training set and test set, respectively, and the data in the training set are normalized.

[0016] Furthermore, the main steam temperature prediction model includes a CNN part and a BiGRU part. The CNN part extracts the parameters that affect the main steam temperature at each time step. That is, at each time step, the parameters that have a great impact on the main steam temperature are identified from the coal feed rate, water feed rate, power generation, water spray desuperheater flow rate and historical main steam temperature. Then, the input parameters at each time step are sent to the BiGRU part to capture the time relationship.

[0017] The primary load damage prediction model consists of a BiGRU part and an MHA part. The BiGRU part first obtains the temporal relationship between the primary load damage and the input quantities, such as coal feed rate, water feed rate, main steam flow rate, power generation, main steam pressure, and historical primary load damage. The output sequence is then fed into the MHA. Based on this, the MHA learns the global temporal dependency, that is, it assigns different weights to the input quantities at a certain time. The weighted sum of the inputs together constitutes the expression for the primary load damage.

[0018] The power generation coal consumption rate prediction model includes BiGRU, which captures long-term dependencies in historical operating data from both forward and reverse directions. It fully learns the cumulative and delayed effects of load changes, coal supply, and environmental factors on the power generation coal consumption rate for prediction.

[0019] Furthermore, after optimizing the main steam temperature, primary load damage, and power generation coal consumption rate, the following also applies:

[0020] A time sliding window is used to perform sliding prediction on the test set in the target parameters to obtain new multidimensional indicators and operating parameters of the coal-fired unit.

[0021] Based on the new multidimensional indicators and operating parameters, the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate are updated to continuously optimize the main steam temperature, primary load damage, and power generation coal consumption rate.

[0022] Furthermore, the multi-objective optimization model is expressed as:

[0023] ;

[0024] st ;

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] in, Let be the objective function. For time steps, Main steam temperature, Main steam pressure, For power generation capacity, It is the current moment. Is the future number At that moment, , , These are the prediction models. Standardized power generation coal consumption rate, main steam temperature, and primary load damage at specific times. , , These are the weighting coefficients for power generation coal consumption rate, main steam temperature, and primary load damage, respectively. , , These are the lower limit thresholds for main steam temperature, main steam pressure, and unit power generation, respectively. , , These are the upper limit thresholds for main steam temperature, main steam pressure, and unit power generation, respectively. , , These are the main steam temperature, main steam pressure, and unit power generation. The corresponding setting value at that time , , These are the main steam temperature gradient, the main steam pressure gradient, and the power change rate, respectively. , , These are the maximum allowable gradient of main steam temperature, the maximum allowable gradient of main steam pressure, and the maximum allowable rate of change of power, respectively.

[0031] The apparatus of the present invention comprises:

[0032] The data acquisition module is used to collect the main steam temperature and target parameters that meet the multi-dimensional index prediction after the target unit is operating under different conditions.

[0033] The indicator prediction module is used to calculate the primary load damage and power generation coal consumption rate of key components of coal-fired power units based on the collected target parameters. It also selects parameters that have a significant impact on the main steam temperature, primary load damage, and power generation coal consumption rate from the target parameters, and divides the selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate into training sets and test sets, respectively. It constructs a prediction model for main steam temperature, a prediction model for primary load damage, and a prediction model for power generation coal consumption rate, and uses the training set to perform multi-model fusion prediction on the three key indicators of main steam temperature, primary load damage, and power generation coal consumption rate.

[0034] The index optimization module is used to optimize the main steam temperature, primary load damage, and power generation coal consumption rate based on the prediction results. It sets the main steam temperature gradient, main steam pressure gradient, and power generation change rate as decision variables, sets constraints, and establishes a multi-objective optimization model. The multi-objective optimization model is solved to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate, thereby optimizing the main steam temperature, primary load damage, and power generation coal consumption rate.

[0035] The present invention also provides an electronic device, comprising: a memory, a processor, and a program stored in the memory that can run on the processor, the processor executing the program to implement the method.

[0036] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is used to implement the method described thereon.

[0037] The present invention also provides a computer program product comprising a computer program stored in a non-transitory computer-readable storage medium, which, when executed by a processor, is configured to implement the method described herein.

[0038] Beneficial Effects: Compared with existing technologies, the significant technical effects of this invention are as follows: After the target unit operates under varying conditions, it can collect target parameters that meet the prediction of multi-dimensional indicators. Through finite element modeling, thermodynamic calculations, and actual data collection, it obtains data on the main steam temperature, primary load damage, and power generation coal consumption rate of the coal-fired unit. It uses a neural network model to perform multi-model fusion prediction on the three key indicators of main steam temperature, primary load damage, and power generation coal consumption rate. Based on the prediction results, a multi-objective optimization model is constructed to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation power change rate, thereby optimizing the main steam temperature, primary load damage, and power generation coal consumption rate. This allows the unit to simultaneously consider economy and stability under safe operating conditions, solving the problems of low prediction accuracy, poor real-time performance, and difficult optimization solutions for multi-dimensional indicators of coal-fired units in related technologies. It realizes dynamic prediction of key unit indicators and multi-objective real-time optimization of operating parameters, thereby improving the safety, economy, and stability of the unit's peak-shaving operation. Attached Figure Description

[0039] Figure 1 This is a flowchart of the method of the present invention;

[0040] Figure 2 This is a schematic diagram of the system framework of the present invention;

[0041] Figure 3 This is a schematic diagram of a specific embodiment of the CNN (Convolutional Neural Network) of the present invention;

[0042] Figure 4 This is a schematic diagram of a BiGRU (Bidirectional Gated Recurrent Unit) according to a specific embodiment of the present invention;

[0043] Figure 5 This is a schematic diagram of the MHA (Multi-Head Attention) mechanism according to a specific embodiment of the present invention;

[0044] Figure 6 This is a comparison chart of multi-dimensional index data of a coal-fired power unit before and after optimization according to a specific embodiment of the present invention;

[0045] Figure 7 This is a schematic diagram of the structure of the device provided according to an embodiment of the present invention;

[0046] Figure 8This is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present invention. Detailed Implementation

[0047] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0048] The following describes, with reference to the accompanying drawings, an adaptive quantum particle swarm optimization method and apparatus for multi-dimensional indicators of coal-fired power units based on neural network multi-model fusion prediction of embodiments of the present invention. Addressing the significant limitations of single-mechanism models or data-driven models mentioned in the background section, and the difficulty of traditional mathematical programming methods in handling optimization problems involving neural network embeddings, the present invention provides a method for optimizing multi-dimensional indicators of coal-fired power units based on neural network multi-model fusion prediction. In this method, after the target unit operates under varying conditions, the main steam temperature and target parameters that meet the multi-dimensional indicator prediction are collected. Based on the collected target parameters, the primary load damage of key components of the coal-fired power unit is calculated using the elastic-ideal plastic finite element method, and the power generation coal consumption rate is calculated through thermodynamics. Multi-model fusion prediction is performed on the three key indicators—main steam temperature, primary load damage, and power generation coal consumption rate—separately. Based on the prediction results, a multi-objective optimization model is constructed to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation power change rate, thereby optimizing the main steam temperature, primary load damage, and power generation coal consumption rate, enabling the unit to simultaneously achieve economic efficiency and stability under safe operating conditions. Therefore, this method solves the problems of low prediction accuracy, poor real-time performance, and difficulty in optimization of multi-dimensional indicators of coal-fired power units in related technologies, and realizes dynamic prediction of key indicators of the unit and real-time optimization of multi-objective operating parameters, thereby improving the safety, economy and stability of the unit's peak-shaving operation.

[0049] like Figure 1 As shown, the method of the present invention includes the following steps:

[0050] S101. After the target unit operates under different conditions, collect the main steam temperature and target parameters that meet the multi-dimensional index prediction.

[0051] In this embodiment of the invention, the target unit is a coal-fired unit currently undergoing multi-dimensional index operation optimization; the target parameters for multi-dimensional index prediction are the parameters that affect the main steam temperature, primary load damage, and power generation coal consumption rate during the unit's variable operating condition operation.

[0052] It is understood that, in embodiments of the present invention, after the unit is operating under varying conditions, the main steam temperature and target parameters that meet the prediction of multi-dimensional indicators can be collected, for example, such as... Figure 2As shown, the embodiments of the present invention can collect parameters that affect the main steam temperature, primary load damage, and power generation coal consumption rate during the variable operating conditions of the unit, such as coal feed rate, feedwater flow rate, main steam flow rate, power generation, spray desuperheater flow rate, and main steam pressure, thereby improving the reliability of multi-dimensional index prediction for coal-fired units.

[0053] S102. Data denoising is performed on the collected main steam temperature and target parameters to ensure the data used for modeling is as clean as possible. Based on the collected target parameters, primary load damage data of key components of the coal-fired unit is calculated using the elastic-ideal plastic finite element method. Thermodynamic calculations are used to obtain the power generation coal consumption rate data, as well as the denoised main steam temperature data. To avoid data irrelevance and redundancy, feature filtering is performed on the denoised target parameters, i.e., parameters with significant impact on main steam temperature, primary load damage, and power generation coal consumption rate are selected from 20 target parameters. The selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate are divided into training and test sets, respectively. The data in the training set is normalized, and a main steam temperature prediction model CNN-BiGRU, a primary load damage prediction model BiGRU-MHA, and a power generation coal consumption rate prediction model BiGRU are constructed. U, where the inputs to the main steam temperature prediction model are coal feed rate, feedwater flow rate, power generation, spray desuperheater flow rate, and main steam temperature; the inputs to the primary load damage prediction model are coal feed rate, feedwater flow rate, main steam flow rate, power generation, main steam pressure, and calculated primary load damage; and the inputs to the power generation coal consumption rate prediction model are coal feed rate, main steam flow rate, power generation, and calculated power generation coal consumption rate. The training sets of the selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate are normalized respectively. Then, the main steam temperature prediction model, primary load damage prediction model, and power generation coal consumption rate prediction model are trained using the normalized training sets respectively. Finally, the trained main steam temperature prediction model, primary load damage prediction model, and power generation coal consumption rate prediction model are used to make predictions using the selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate test sets respectively.

[0054] In actual implementation, such as Figure 2 As shown, in this embodiment of the invention, the collected main steam temperature and target parameters are first denoised to calculate the primary load damage and power generation coal consumption rate. The input parameters of the prediction model are determined through feature filtering. Then, the filtered parameters, main steam temperature, primary load damage, and power generation coal consumption rate are divided into training and testing sets, respectively. The data in the training set is then normalized. Finally, the processed training data is input into the pre-constructed multi-model fusion task layer corresponding to the main steam temperature, primary load damage, and power generation coal consumption rate to obtain the multi-dimensional index prediction model. This effectively improves the accuracy and reliability of the prediction model. The specific steps are as follows:

[0055] (1) First, Principal Component Analysis (PCA) is used to reduce noise in the target parameters and main steam temperature. The original multidimensional variables are projected onto several mutually orthogonal principal component directions through orthogonal transformation, retaining most of the variance to achieve signal purification and dimensionality compression. The calculation formula is as follows:

[0056] ;

[0057] ;

[0058] ;

[0059] in, For target parameters or main steam temperature matrix; for transpose; The number of samples for the target parameter or main steam temperature; It is the covariance matrix; The eigenvector matrix of the covariance matrix; The corresponding eigenvalue diagonal matrix, For the first One eigenvalue; The number of principal components to be retained; For the reason before A matrix composed of eigenvectors; for transpose; Principal component score matrix; This represents the data after noise reduction of the target parameter or main steam temperature.

[0060] (2) Next, the Pearson correlation coefficient method (PCC) was used for correlation analysis. The input of the prediction model is required to contain a sufficiently rich set of relevant variables to fully learn the patterns in the historical data. At the same time, data irrelevance and redundancy should be avoided. Specifically, the parameters that have a greater impact on the main steam temperature, primary load damage, and power generation coal consumption rate are selected from 20 collected and noise-reduced target parameters (primary air flow, feedwater flow, secondary air flow, power generation, coal feed, furnace temperature, main steam pressure, flue gas temperature, excess air coefficient, spray desuperheater flow, reheat steam temperature, feedwater temperature, coal powder fineness, boiler load, main steam flow, condenser vacuum, reheat steam temperature, ambient temperature, atmospheric pressure, and feedwater pressure). The calculation formula is as follows:

[0061] ;

[0062] in, For the noise reduction of the first One target parameter, Main steam temperature or primary load damage or power generation coal consumption rate; The number of samples for the target parameter, main steam temperature, primary load damage, or power generation coal consumption rate; for The mean, for The mean; for and The Pearson correlation coefficient is used, with a value ranging from -1 to 1. A value close to 1 indicates a strong positive correlation, close to -1 indicates a strong negative correlation, and close to 0 indicates no correlation. Using this method, six parameters with relatively large absolute values ​​(less than 1 and closer to 1) of correlation coefficients with main steam temperature, primary load damage, and power generation coal consumption rate were selected from 20 collected and denoised target parameters: coal feed rate, feedwater flow rate, main steam flow rate, power generation, spray desuperheater flow rate, and main steam pressure. Three historical parameters were also selected: main steam temperature, calculated primary load damage, and calculated power generation coal consumption rate. These nine parameters constitute the input to the prediction model. Specifically, the inputs to the main steam temperature prediction model are coal feed rate, feedwater flow rate, power generation, spray desuperheater flow rate, and historical main steam temperature; the inputs to the primary load damage prediction model are coal feed rate, feedwater flow rate, main steam flow rate, power generation, main steam pressure, and historical primary load damage; and the inputs to the power generation coal consumption rate prediction model are coal feed rate, main steam flow rate, power generation, and historical power generation coal consumption rate.

[0063] (3) Secondly, the six selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate are normalized using the Min-max normalization method, as shown in the following formula:

[0064] ;

[0065] in, , These represent the minimum and maximum values ​​of six parameters in the training set, namely, main steam temperature, primary load damage, or power generation coal consumption rate. To train a set of six parameters, namely, main steam temperature, primary load damage, or power generation coal consumption rate, The normalized results of six parameters (main steam temperature, primary load damage, or power generation coal consumption rate) in the training set are in the range of [0, 1].

[0066] (4) Finally, the prediction models for main steam temperature, primary load damage and power generation coal consumption rate are set as CNN-BiGRU, BiGRU-MHA and BiGRU models respectively. Then, the normalized coal feed rate, water feed rate, main steam flow rate, power generation, water spray desuperheater flow rate, main steam pressure, historical main steam temperature, historical primary load damage and historical power generation coal consumption rate are fed into the prediction models for main steam temperature, primary load damage and power generation coal consumption rate constructed above to obtain the prediction results of main steam temperature, primary load damage and power generation coal consumption rate.

[0067] Among them, the CNN-BiGRU hybrid model, due to its cascaded architecture, achieves efficient linkage between local feature extraction and global time series modeling, enabling the model to focus on both detailed changes and overall trends. This synergistic effect makes it exhibit stronger predictive power than single-structure or other combined models when dealing with complex time series problems such as main steam temperature, which combines instantaneous changes with slow thermal inertia. Therefore, the CNN-BiGRU hybrid model is used for main steam temperature prediction. The CNN-BiGRU calculation formula is as follows:

[0068] CNN section:

[0069] ;

[0070] ;

[0071] ;

[0072] in, To output the feature vector, the eigenvalues ​​of the main steam temperature, The input data includes local sequence data of coal feed rate, water flow rate, power generation, water spray desuperheater flow rate, and historical main steam temperature. The weights of the convolution kernel, For bias terms, The output feature values ​​after pooling. For compensation of the pooled window, For the local region of the input feature vector, Main steam temperature output vector The input vectors are: coal feed rate, water feed rate, power generation, water spray desuperheater flow rate, and historical main steam temperature. This is the weight matrix. For non-linear activation functions, see the CNN schematic diagram. Figure 3 .

[0073] BiGRU section:

[0074] ;

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] ;

[0080] The CNN described above extracts the main steam temperature at each time step. The influencing parameters, i.e., at each time step Identify the parameters that significantly affect the main steam temperature from data on coal feed rate, water feed rate, power generation, spray desuperheater flow rate, and historical main steam temperature. Then, analyze the data at each time point. The input parameters are fed into the BiGRU to capture the timing relationship. Among them, For a moment The input vector of the main steam temperature prediction model after extraction by CNN The hidden state of the main steam temperature prediction model. Candidate states for the main steam temperature prediction model. This represents the hidden state of the main steam temperature prediction model from the previous time step. The update gate for the main steam temperature prediction model. Reset gate for the main steam temperature prediction model. This represents the sigmoid function. This indicates element-wise multiplication. , , , , , , , , For trainable parameters, and For forward GRU and backward GRU at various time steps The hidden state of the main steam temperature prediction model. Forward GRU, For backward GRU, Forward 1 to The input vector of the main steam temperature prediction model after being extracted by CNN at each time step. For backward The input vector of the main steam temperature prediction model after extraction by CNN at time 1. For at any time The hidden state of the main steam temperature prediction model after forward and backward fusion, see the BiGRU schematic diagram. Figure 4 .

[0081] The BiGRU-MHA hybrid model, due to its multi-head self-attention mechanism, enables the model to simultaneously focus on the importance of different time points in the input sequence, adaptively weighting the stress state characteristics of key time steps. This allows it to accurately capture key transient processes and abrupt changes in the primary load damage formation process, which is crucial for stress fluctuations and historical cumulative changes driven by the main steam temperature and pressure gradients. The combination of these two models allows the model to grasp both the overall trend of stress accumulation and evolution and to respond sensitively to local stress abrupt changes. Especially when model parameters are limited, it exhibits stronger predictive ability than single-structure or other combined models. Therefore, the BiGRU-MHA hybrid model is used for primary load damage prediction. The BiGRU-MHA calculation formula is as follows:

[0082] BiGRU section:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] in, For a moment The input vector of the single-load damage prediction model. This represents the hidden state of a single-load damage prediction model. These are candidate states for a single-load damage prediction model. This represents the hidden state of the load damage prediction model from the previous time step. Update the gate for the single-load damage prediction model. Reset the gate for a single load damage prediction model. This represents the sigmoid function. This indicates element-wise multiplication. , , , , , , , , For trainable parameters, and For forward GRU and backward GRU at various time steps The hidden state of a single-load damage prediction model Forward GRU, For backward GRU, Forward 1 to The input vector of the single-load damage prediction model at time step. For backward The input vector of the single-load damage prediction model at time 1. For at any time The hidden state of the load damage prediction model after forward and backward fusion.

[0090] MHA section:

[0091] ;

[0092] ;

[0093] ;

[0094] The aforementioned BiGRU first obtains the temporal relationship between a single load damage and the input quantities: coal feed rate, water flow rate, main steam flow rate, power generation, main steam pressure, and historical single load damage. The output sequence is then fed into the MHA. Based on this, the MHA learns the global temporal dependency, i.e., at a certain moment... Assign different weights to the inputs, for example, at time 1. Since the main steam pressure has the greatest impact on primary load damage, its weight is set to the maximum and weighted together with other inputs to form the expression for primary load damage. , , These are query, key, and value matrices, respectively. is the scaling factor for the key dimension. For sequence length, For attention mechanisms, For activation function, For the first One's attention, , The number of attention heads, multi-head attention in parallel Attention is calculated for each head, and then the outputs of each head are concatenated and linearly transformed. , , and For trainable projection matrices, For the attention of the bulls, For the assembly operation, the MHA schematic diagram is as follows: Figure 5 As shown.

[0095] In predicting power generation coal consumption rate, BiGRU can simultaneously capture long-term dependencies in historical operating data from both positive and negative directions, fully learning the cumulative and delayed effects of load changes, coal supply, and environmental factors on power generation coal consumption rate. In contrast, attention mechanisms or multi-layer convolutional hybrid models, when limited in parameter size, may learn noise in the training data, leading to overfitting and a decline in performance on the test set. BiGRU, with its moderate model complexity, can fully capture key features in the data while effectively suppressing the risk of overfitting. Therefore, BiGRU is used for predicting power generation coal consumption rate. The BiGRU calculation formula is as follows:

[0096] ;

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] ;

[0102] in, For a moment The input vector of the power generation coal consumption rate prediction model. This represents the hidden state of the power generation coal consumption rate prediction model. These are candidate states for the power generation coal consumption rate prediction model. This represents the hidden state of the power generation coal consumption rate prediction model for the previous time step. Update the gate for the power generation coal consumption rate prediction model. Reset the gate for the power generation coal consumption rate prediction model. This represents the sigmoid function. This indicates element-wise multiplication. , , , , , , , , For trainable parameters, and For forward GRU and backward GRU at various time steps The hidden state of the power generation coal consumption rate prediction model Forward GRU, For backward GRU, Forward 1 to The input vector of the time-varying coal consumption rate prediction model for power generation. For backward The input vector of the power generation coal consumption rate prediction model at time 1. For at any time The hidden state of the power generation coal consumption rate prediction model after the fusion of forward and backward approaches.

[0103] Therefore, the present invention can predict the main steam temperature, primary load damage, and power generation coal consumption rate through a hybrid neural network model, thereby improving prediction accuracy, significantly shortening calculation time, and improving prediction efficiency.

[0104] S103. Based on the prediction results, a multi-objective optimization model is constructed to obtain the optimized main steam temperature gradient, main steam pressure gradient and power generation change rate, thereby optimizing the main steam temperature, primary load damage and power generation coal consumption rate, so that the unit can simultaneously take into account economy and stability under safe operation conditions.

[0105] It is understood that, based on the predicted main steam temperature, primary load damage, and power generation coal consumption rate in the above steps, the embodiments of the present invention can use the Adaptive Quantum Particle Swarm Optimization (A-QPSO) algorithm to determine the main steam temperature, primary load damage, and power generation coal consumption rate as optimization objectives, and use the main steam temperature gradient, main steam pressure gradient, and power generation change rate as decision variables to establish a multi-objective optimization model. In addition, the embodiments of the present invention can optimize the main steam temperature gradient, main steam pressure gradient, and power generation change rate based on the multi-objective optimization model to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate, so that the unit can simultaneously take into account economy and stability under safe operating conditions.

[0106] Therefore, the embodiments of the present invention can simultaneously predict the main steam temperature, primary load damage, and power generation coal consumption rate, obtain multi-step prediction results of the above indicators, improve prediction accuracy, and then perform multi-objective optimization through intelligent optimization algorithm, reduce the main steam temperature, primary load damage, and power generation coal consumption rate by coordinating and optimizing decision variables, so as to enable the unit to operate safely, economically, and stably.

[0107] In some embodiments, such as Figure 2 As shown, the above steps use the main steam temperature, primary load damage, and power generation coal consumption rate as optimization objectives, and the main steam temperature gradient, main steam pressure gradient, and power generation change rate as decision variables. Constraints are set to establish a multi-objective optimization model, the mathematical form of which is:

[0108] ;

[0109] st ;

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] in, Let be the objective function. For time steps, Main steam temperature, Main steam pressure, For power generation capacity, It is the current moment. Is the future number At that moment, , , These are the prediction models. Standardized power generation coal consumption rate, main steam temperature, and primary load damage at specific times. , , These are the weighting coefficients for power generation coal consumption rate, main steam temperature, and primary load damage, respectively. , , These are the lower limit thresholds for main steam temperature, main steam pressure, and unit power generation, respectively. , , These are the upper limit thresholds for main steam temperature, main steam pressure, and unit power generation, respectively. , , These are the main steam temperature, main steam pressure, and unit power generation. The corresponding setting value at that time , , These are the main steam temperature gradient, the main steam pressure gradient, and the power change rate, respectively. , , These are the maximum allowable gradient of main steam temperature, the maximum allowable gradient of main steam pressure, and the maximum allowable rate of change of power, respectively.

[0116] Secondly, the above steps can be optimized using the A-QPSO algorithm. A-QPSO is an improved algorithm of the classical PSO algorithm. It is based on the stochastic behavior model of quantum mechanics and introduces a quantum potential well and a probabilistic position update mechanism. This allows particles to no longer depend on the velocity vector, but rather on global and local guiding terms to determine their position updates, thus possessing stronger global convergence capabilities. Each particle in the swarm updates its position around the average attractor with a certain probability distribution during iterations. The calculation formula is as follows:

[0117] ;

[0118] ;

[0119] ;

[0120] ;

[0121] ;

[0122] in, For the first The random mixing coefficients of individual particles are used to construct a local attractor between individual optima and global optima. Uniformly random numbers, To conform to a uniform distribution, the interval is between (0,1). For the first The particle in the first The position of the iteration, i.e., the relationship between the main steam temperature gradient, the main steam pressure gradient, and the rate of change of power generation at the [number]th iteration. The position of the next iteration. For the first The individual historical optimal position of each particle, i.e., the set of optimal values ​​of the decision variables (main steam temperature gradient, main steam pressure gradient, and rate of change of power generation) that minimizes the objective function (weighted sum of main steam temperature, primary load damage, and power generation coal consumption rate). The globally optimal position is the set of decision variables in the global history that minimizes the objective function. For the first The local attractors of each particle are weighted combinations of individual and globally optimal decision variables. The average attractor, i.e., the mean of all local attractors. For the number of particles, For the first Individual particles At the iteration position, i.e., the ratio of the main steam temperature gradient, the main steam pressure gradient, and the rate of change of power generation at the [number]th iteration, [the following is a partial equation]. The position of the next iteration, where ± is determined by random numbers. The contraction-expansion coefficient is used to control the search stride size. and These are the initial and final values, respectively. For group dispersion, For adjustment coefficients, The average position of the particles. When the population distribution is more dispersed through A-QPSO... Maintaining a larger size enhances exploration, while the population gradually converges. It automatically reduces, thereby inhibiting premature convergence and enhancing local refinement capabilities.

[0123] The A-QPSO algorithm parameters can be set as follows: particle number M is 30, and the upper and lower bounds of the contraction-expansion coefficient are... Take 1, Set to 0.1, adaptive coefficient Set the value to 2, and set the maximum number of iterations MaxDT to 50. When the number of iterations reaches 50, the optimization algorithm will also terminate.

[0124] Optionally, in one embodiment of the present invention, after optimizing the main steam temperature, primary load damage, and power generation coal consumption rate based on the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate, the method further includes: performing sliding prediction on the test set in the target parameters using a time sliding window to obtain new multidimensional indicators and operating parameters of the coal-fired unit; and updating the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate based on the new multidimensional indicators and operating parameters to continuously optimize the main steam temperature, primary load damage, and power generation coal consumption rate.

[0125] In some embodiments, such as Figure 2 As shown, embodiments of the present invention can use a time sliding window to perform sliding prediction on the test set in the target parameters to obtain new multidimensional indicators and operating parameters of the coal-fired unit. Thus, based on the new multidimensional indicators and operating parameters, the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate can be updated to continuously optimize the main steam temperature, primary load damage, and power generation coal consumption rate.

[0126] Therefore, the present invention can establish a multi-objective optimization model by setting the main steam temperature, primary load damage, and power generation coal consumption rate as optimization objectives, and using the main steam temperature gradient, main steam pressure gradient, and power generation change rate as decision variables, setting constraints, and establishing the model. This can achieve optimized control of key indicators and operating parameters of the unit, and improve the economy, safety, and stability of peak-shaving operation of coal-fired units.

[0127] Based on the aforementioned technical means, this method achieves accurate prediction and operational optimization of main steam temperature, primary load damage, and power generation coal consumption rate through multi-model fusion prediction and A-QPSO optimization. In the multi-objective optimization process, the synergistic relationship between safety, stability, and economy is fully considered, ensuring that the optimized scheme guarantees stable operation of main steam parameters while reducing coal consumption and fatigue damage, thereby improving the overall operating performance of the unit.

[0128] For example, such as Figure 6 Figures (a)-(c) show a comparison of the main steam temperature, primary load damage, and power generation coal consumption rate before and after optimization in this invention example. Figure 6 Figures (a), (b), and (c) show comparisons before and after optimization of power generation coal consumption rate, main steam temperature, and primary load damage, respectively. The experimental results demonstrate that this invention significantly reduces main steam temperature, primary load damage, and power generation coal consumption rate, improves unit operating economy, mitigates the abrupt changes and long-term accumulation of high values ​​in primary load damage, suppresses instantaneous peaks in main steam temperature, and makes the temperature sequence more stable, comprehensively enhancing the economy, safety, and stability of the unit's peak-shaving operation.

[0129] The multi-dimensional index optimization method for coal-fired power units based on neural network multi-model fusion prediction proposed in this invention can collect the main steam temperature and target parameters that meet the multi-dimensional index prediction after the target unit operates under varying conditions. Based on the collected target parameters, the primary load damage of key components of the coal-fired power unit is calculated using the elastic-ideal plastic finite element method, and the power generation coal consumption rate is calculated through thermodynamics. Multi-model fusion prediction is performed on the three key indicators of main steam temperature, primary load damage, and power generation coal consumption rate. Based on the prediction results, a multi-objective optimization model is constructed to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation power change rate, thereby optimizing the main steam temperature, primary load damage, and power generation coal consumption rate. This allows the unit to balance economy and stability under safe operating conditions, solving the problems of low prediction accuracy, poor real-time performance, and difficult optimization solutions for multi-dimensional indexes of coal-fired power units in related technologies. It realizes dynamic prediction of key unit indicators and multi-objective real-time optimization of operating parameters, thereby improving the safety, economy, and stability of the unit's peak-shaving operation.

[0130] Next, referring to the accompanying drawings, we describe the multi-dimensional index optimization device for coal-fired power units based on neural network multi-model fusion prediction proposed according to an embodiment of the present invention.

[0131] Figure 7 This is a schematic diagram of the structure of a multi-dimensional index optimization device for coal-fired power units based on neural network multi-model fusion prediction, according to an embodiment of the present invention.

[0132] like Figure 7 As shown, the multi-dimensional index optimization device 10 for coal-fired power units based on neural network multi-model fusion prediction includes: a data acquisition module 101, an index prediction module 102, and an index optimization module 103.

[0133] Specifically, the data acquisition module 101 is used to collect the main steam temperature and target parameters that meet the multi-dimensional index prediction after the target unit is operating under different conditions.

[0134] The indicator prediction module 102 is used to calculate the primary load damage and power generation coal consumption rate of key components of coal-fired power units based on the collected target parameters. It also selects parameters that have a significant impact on the main steam temperature, primary load damage, and power generation coal consumption rate from the target parameters, and divides the selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate into training sets and test sets, respectively. It constructs a prediction model for main steam temperature, a prediction model for primary load damage, and a prediction model for power generation coal consumption rate, and uses the training set to perform multi-model fusion prediction on the three key indicators of main steam temperature, primary load damage, and power generation coal consumption rate.

[0135] The index optimization module 103 is used to optimize the main steam temperature, primary load damage, and power generation coal consumption rate based on the prediction results. It sets the main steam temperature gradient, main steam pressure gradient, and power generation change rate as decision variables, sets constraints, and establishes a multi-objective optimization model. The multi-objective optimization model is solved to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate, thereby optimizing the main steam temperature, primary load damage, and power generation coal consumption rate, so that the unit can simultaneously take into account economy and stability under safe operation conditions.

[0136] Optionally, in one embodiment of the present invention, the indicator prediction module 102 includes a data processing unit and a neural network prediction unit.

[0137] The data processing unit is used to perform data noise reduction on the target parameters and main steam temperature, perform feature screening on the noise-reduced target parameters, and then divide the screened parameters, main steam temperature, primary load damage, and power generation coal consumption rate into training set and test set respectively, and normalize the data in the training set.

[0138] The neural network prediction unit is used to input the processed data into the pre-built multi-model fusion corresponding task layer of main steam temperature, primary load damage and power generation coal consumption rate, respectively, to obtain the multi-dimensional index prediction model.

[0139] Optionally, in one embodiment of the present invention, the apparatus 10 of the present invention further includes a scrolling module and an updating module.

[0140] The rolling module is used to perform sliding prediction on the test set of the target parameters based on the optimized main steam temperature gradient, main steam pressure gradient and power generation change rate, after optimizing the main steam temperature, primary load damage and power generation coal consumption rate, so as to obtain new multidimensional indicators and operating parameters of the coal-fired unit.

[0141] The update module updates the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate based on the new multidimensional indicators and operating parameters, so as to continuously optimize the main steam temperature, primary load damage, and power generation coal consumption rate.

[0142] By introducing a time-sliding window to continuously predict the test set and dynamically updating and optimizing parameters based on the prediction results, the main steam temperature, primary load damage and power generation coal consumption rate can be optimized in real time, thereby improving the economy, safety and stability of coal-fired units during peak shaving.

[0143] Optionally, in one embodiment of the present invention, the optimization module 103 includes a modeling module and a solution module.

[0144] The modeling module, based on the prediction results of multi-dimensional indicators, determines the main steam temperature, primary load damage, and power generation coal consumption rate as optimization objectives, and establishes a multi-objective optimization model by taking the main steam temperature gradient, main steam pressure gradient, and power generation change rate as decision variables.

[0145] The solution module optimizes the main steam temperature gradient, main steam pressure gradient, and power generation rate of change based on the multi-objective optimization model to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation rate of change.

[0146] It should be noted that the explanation of the above-mentioned embodiment of the multi-dimensional index optimization method for coal-fired power units based on neural network multi-model fusion prediction also applies to the multi-dimensional index optimization device for coal-fired power units based on neural network multi-model fusion prediction in this embodiment, and will not be repeated here.

[0147] The multi-dimensional index optimization device for coal-fired power units based on neural network multi-model fusion prediction proposed in this invention can collect the main steam temperature and target parameters that meet the multi-dimensional index prediction after the target unit is operating under varying conditions. Based on the collected target parameters, the primary load damage of key components of the coal-fired power unit is calculated using the elastic-ideal plastic finite element method, and the power generation coal consumption rate is calculated through thermodynamics. Multi-model fusion prediction is performed on the three key indicators of main steam temperature, primary load damage, and power generation coal consumption rate. Based on the prediction results, a multi-objective optimization model is constructed to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation power change rate, thereby optimizing the main steam temperature, primary load damage, and power generation coal consumption rate. This allows the unit to balance economy and stability under safe operating conditions, solving the problems of low prediction accuracy, poor real-time performance, and difficult optimization solutions for multi-dimensional indexes of coal-fired power units in related technologies. It realizes dynamic prediction of key unit indicators and multi-objective real-time optimization of operating parameters, thereby improving the safety, economy, and stability of the unit's peak-shaving operation.

[0148] Figure 8 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. The electronic device may include:

[0149] The memory 801, the processor 802, and the computer program stored in the memory 801 and executable by the processor 802.

[0150] When the processor 802 runs the program, it can implement the adaptive quantum particle swarm optimization method for multi-dimensional indicators of coal-fired power units based on neural network multi-model fusion prediction as described in the various embodiments of this specification.

[0151] Furthermore, the electronic device may also include:

[0152] The communication interface 803 is used to realize information exchange between the memory 801 and the processor 802.

[0153] The memory 801 is used to store computer programs that can be executed on the processor 802. The memory 801 may include high-speed random access memory (RAM) or one or more non-volatile memories, such as at least one disk storage device.

[0154] If the memory 801, processor 802, and communication interface 803 are implemented as independent components, these components can be interconnected via a bus to complete data transmission and control signal exchange. The bus can be implemented using industry-standard bus architectures (such as ISA), peripheral interconnect bus (PCI), or extended industry standard bus (EISA), and can typically be further subdivided into address buses, data buses, and control buses. (See attached figures for example...) Figure 8For ease of illustration, a thick line is usually used to represent bus connections, but this representation does not mean that the system contains only a single bus or a single type of bus.

[0155] Alternatively, if the memory 801, processor 802, and communication interface 803 are integrated on a single chip, they can communicate with each other through the interface inside the chip, thereby achieving higher integration and a more compact hardware implementation.

[0156] The processor 802 may be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the functions of the embodiments of the present invention.

[0157] This embodiment also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can realize the above-mentioned method for optimizing multi-dimensional indicators of coal-fired power units based on neural network multi-model fusion prediction.

[0158] In addition, this embodiment also provides a computer program product, which exists in an executable form and can implement the above-mentioned method for optimizing multi-dimensional indicators of coal-fired power units based on neural network multi-model fusion prediction when executed.

[0159] A flowchart or other form of logic or steps can be considered as a sequence of executable instructions arranged to achieve a corresponding function. These instructions can be stored on any type of computer-readable medium for reading and execution by an instruction execution system or apparatus (such as a processor-based system). The term "computer-readable medium" as used in this specification broadly includes any medium that can be used to store, transmit, or transport a program for use by an instruction execution system, apparatus, or device. Examples (but not limited to) include electrical connection components in electronic devices (electronic devices with wiring), portable disk enclosures (magnetic media), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), optical media (such as CD-ROM), and fiber optic devices. Furthermore, a computer-readable medium can even be a program carrier printed on paper, from which the corresponding program can be electronically recovered and stored in computer memory through optical scanning and subsequent processing of the paper medium.

[0160] It should be understood that the functional modules of the present invention can be implemented by hardware, software, firmware, or a combination thereof. Several steps or methods described in the above embodiments can be implemented by software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, techniques known in the art can be used, such as combinational or sequential logic circuits composed of discrete logic gates, application-specific integrated circuits (ASICs), programmable gate arrays (PGAs), or field-programmable gate arrays (FPGAs) to accomplish the corresponding functions.

[0161] Those skilled in the art will understand that all or part of the steps of the above methods can be implemented by program instructions driving related hardware. The program can be stored in one or more computer-readable media, and when executed, it includes one or more of the method steps described in this invention.

[0162] Furthermore, in various embodiments of the present invention, each functional unit can be integrated into a single processing module, implemented as an independent physical unit, or implemented by a combination of two or more units. The integrated module can be implemented as a hardware device or as a software module and deployed or sold as an independent product. Such software modules can be stored in a computer-readable storage medium and loaded and executed by a corresponding processor.

[0163] The storage media types listed above, such as read-only memory, disks, or optical discs, are merely examples and not an exhaustive list. Although exemplary embodiments of the present invention have been described, those skilled in the art should understand that various modifications, substitutions, or equivalent variations can be made to the above embodiments without departing from the technical solutions and protection scope of the present invention.

Claims

1. A method for optimizing multi-dimensional indicators of coal-fired power units based on neural network multi-model fusion prediction, characterized in that, Includes the following steps: After the target unit operates under varying conditions, the main steam temperature and target parameters that meet the multi-dimensional index prediction are collected. Based on the collected target parameters, the primary load damage and power generation coal consumption rate of key components of the coal-fired unit are calculated. Parameters that have a significant impact on the main steam temperature, primary load damage, and power generation coal consumption rate are selected from the target parameters. The selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate are divided into training and testing sets, respectively. A prediction model for main steam temperature, a prediction model for primary load damage, and a prediction model for power generation coal consumption rate are constructed. The training set is used to perform multi-model fusion prediction on the three key indicators of main steam temperature, primary load damage, and power generation coal consumption rate. Based on the prediction results, the main steam temperature, primary load damage, and power generation coal consumption rate are optimized as objectives. The main steam temperature gradient, main steam pressure gradient, and power generation change rate are used as decision variables. Constraints are set, and a multi-objective optimization model is established. The multi-objective optimization model is solved to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate, thereby optimizing the main steam temperature, primary load damage, and power generation coal consumption rate.

2. The method according to claim 1, characterized in that, The target parameters that meet the multidimensional index prediction are the parameters that affect the main steam temperature, primary load damage and power generation coal consumption rate during the unit's variable operating condition operation. Principal component analysis was used to denoise the collected main steam temperature and the target parameters that meet the multidimensional index prediction.

3. The method according to claim 1, characterized in that, The Pearson correlation coefficient method was used to analyze the correlation between each target parameter and the main steam temperature, primary load damage, and power generation coal consumption rate. Parameters with absolute values ​​of correlation coefficients with the main steam temperature, primary load damage, and power generation coal consumption rate that are greater than the set threshold were selected. The selected parameters include: coal feed rate, water feed rate, main steam flow rate, power generation, water spray desuperheater flow rate, and main steam pressure. The selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate are divided into training set and test set, respectively, and the data in the training set are normalized.

4. The method according to claim 1, characterized in that, The main steam temperature prediction model includes a CNN part and a BiGRU part. The CNN part extracts the parameters that affect the main steam temperature at each time step. That is, at each time step, the parameters that have a great impact on the main steam temperature are found from the coal feed rate, water feed rate, power generation, water spray desuperheater flow rate and historical main steam temperature. Then, the input parameters at each time step are sent to the BiGRU part to capture the time relationship. The primary load damage prediction model consists of a BiGRU part and an MHA part. The BiGRU part first obtains the temporal relationship between the primary load damage and the input quantities, such as coal feed rate, water feed rate, main steam flow rate, power generation, main steam pressure, and historical primary load damage. The output sequence is then fed into the MHA. Based on this, the MHA learns the global temporal dependency, that is, it assigns different weights to the input quantities at a certain time. The weighted sum of the inputs together constitutes the expression for the primary load damage. The power generation coal consumption rate prediction model includes BiGRU, which captures long-term dependencies in historical operating data from both forward and reverse directions. It fully learns the cumulative and delayed effects of load changes, coal supply, and environmental factors on the power generation coal consumption rate for prediction.

5. The method according to claim 1, characterized in that, After optimizing the main steam temperature, primary load damage, and power generation coal consumption rate, the following is also included: A time sliding window is used to perform sliding prediction on the test set in the target parameters to obtain new multidimensional indicators and operating parameters of the coal-fired unit. Based on the new multidimensional indicators and operating parameters, the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate are updated to continuously optimize the main steam temperature, primary load damage, and power generation coal consumption rate.

6. The method according to claim 1, characterized in that, The multi-objective optimization model is expressed as follows: ； s.t. ; ； ； ； ； ； in, Let be the objective function. For time steps, Main steam temperature, Main steam pressure, For power generation, It is the current moment. Is the future number At that moment, , , These are the prediction models. Standardized power generation coal consumption rate, main steam temperature, and primary load damage at specific times. , , These are the weighting coefficients for power generation coal consumption rate, main steam temperature, and primary load damage, respectively. , , These are the lower limit thresholds for main steam temperature, main steam pressure, and unit power generation, respectively. , , These are the upper limit thresholds for main steam temperature, main steam pressure, and unit power generation, respectively. , , These are the main steam temperature, main steam pressure, and unit power generation. The corresponding setting value at that time , , These are the main steam temperature gradient, the main steam pressure gradient, and the power change rate, respectively. , , These are the maximum allowable gradient of main steam temperature, the maximum allowable gradient of main steam pressure, and the maximum allowable rate of change of power, respectively.

7. A device for optimizing multi-dimensional indicators of coal-fired power units based on neural network multi-model fusion prediction, characterized in that, include: The data acquisition module is used to collect the main steam temperature and target parameters that meet the multi-dimensional index prediction after the target unit is operating under different conditions. The indicator prediction module is used to calculate the primary load damage and power generation coal consumption rate of key components of coal-fired power units based on the collected target parameters. It also selects parameters that have a significant impact on the main steam temperature, primary load damage, and power generation coal consumption rate from the target parameters, and divides the selected parameters, main steam temperature, primary load damage, and power generation coal consumption rate into training sets and test sets, respectively. It constructs a prediction model for main steam temperature, a prediction model for primary load damage, and a prediction model for power generation coal consumption rate, and uses the training set to perform multi-model fusion prediction on the three key indicators of main steam temperature, primary load damage, and power generation coal consumption rate. The index optimization module is used to optimize the main steam temperature, primary load damage, and power generation coal consumption rate based on the prediction results. It sets the main steam temperature gradient, main steam pressure gradient, and power generation change rate as decision variables, sets constraints, and establishes a multi-objective optimization model. The multi-objective optimization model is solved to obtain the optimized main steam temperature gradient, main steam pressure gradient, and power generation change rate, thereby optimizing the main steam temperature, primary load damage, and power generation coal consumption rate.

8. An electronic device, characterized in that, include: A memory, a processor, and a program stored in the memory that can run on the processor, the processor executing the program to implement the method of any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program is used to implement any of the methods described in claims 1-6 above.

10. A computer program product comprising a computer program stored in a non-transitory computer-readable storage medium, characterized in that, When the program is executed by a processor, it is configured to implement the method described in any one of claims 1-6 above.

Images