A method, system and device for dynamic optimization of working conditions of a thermal power unit based on a neural network and a genetic algorithm, and a storage medium

Abstract

The application discloses a kind of based on neural network and genetic algorithm's thermal power unit working condition dynamic optimization method, system, equipment and storage medium, it is related to thermal power unit intelligent dynamic optimization technical field, including determining fitness function parameter, constructs BP neural network model;Establish the boundary line equation of parameter to be optimized, the upper and lower boundary lines of parameter to be optimized under different loads are fitted out;Using genetic algorithm receives BP neural network model and boundary line equation, publishes working condition optimization AI model, configures non-optimized parameter;Boiler efficiency working condition optimization real-time model is constructed, and based on non-optimized parameter real-time optimization is carried out to parameter to be optimized, obtains efficiency deviation and corresponding parameter to be optimized deviation;The method described in the application realizes thermal power unit working condition dynamic optimization, parameter to be optimized reasonable boundary constraint and efficiency deviation and parameter deviation linkage output, improves the pertinence, coherence and executability of optimization result under complex load condition.

Description

Technical Field

[0001] This invention relates to the field of intelligent dynamic optimization technology for thermal power units, specifically to a method, system, device, and storage medium for dynamic optimization of operating conditions of thermal power units based on neural networks and genetic algorithms. Background Technology

[0002] Thermal power unit operating condition optimization technology mainly focuses on the coordinated control of operating parameters of boilers, turbines, and main and auxiliary equipment under multivariate coupling conditions. Research emphasizes the correlation between variables such as main steam pressure, main steam temperature, reheat steam parameters, feedwater parameters, and flue gas parameters, and boiler efficiency, heat consumption, and plant power consumption. Early operating condition analyses primarily employed heat balance calculations, inverse balance methods, design value conversions, and empirical comparisons, adjusting based on differences between historical optimal operating conditions and current operating data. With increasing unit capacity, more frequent load fluctuations, and expanded operating boundaries, operating condition optimization has gradually shifted from manual experience-based judgment to data-driven calculations. Online monitoring, time-series databases, neural network modeling, and swarm iterative search methods have begun to be used to establish mapping relationships between operating parameters and performance indicators, supporting parameter combination analysis under dynamic operating conditions.

[0003] Existing technologies still have significant limitations in engineering applications. Methods based on benchmark operating conditions or empirical rules usually rely on limited samples and are difficult to cover continuously changing load ranges and complex operating scenarios. Real-time calculation paths based on mechanism formulas are long and involve many intermediate variables, making them difficult to use directly as evaluation criteria for high-frequency iterative searches. Although some intelligent algorithms can fit the target parameters for optimization, they do not adequately consider the stability of input parameters under operating conditions, the coupling relationship between parameters, and the adjustable range under load changes. This can easily lead to mathematically feasible but operationally mismatched parameter combinations. In particular, during multi-parameter joint search, if there are no parameter boundary constraints corresponding to the power generation load, the candidate solutions generated by cross-variation may deviate from the actual adjustment range, resulting in a disconnect between the calculation results and the on-site operating conditions. Many schemes can only output the trend of the target parameter changes and cannot further provide the current efficiency deviation and the corresponding deviation of the parameters to be adjusted, making it difficult to form a foundation for refined operating condition analysis for real-time operation. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing methods for optimizing the operating conditions of thermal power units have insufficient global representativeness of benchmark operating conditions, difficulty in constructing fitness functions suitable for genetic algorithms for complex performance indicators, low constraint ability of reasonable variation range of parameters to be optimized under different loads, insufficient ability to output efficiency deviation and corresponding parameter deviation in real time, and the problem of how to achieve dynamic optimization of parameters to be optimized under current non-optimized parameter conditions.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms, comprising: determining fitness function parameters; fitting upper and lower boundary lines of the parameters to be optimized under different loads; using a genetic algorithm to receive the fitness function parameters and upper and lower boundary lines, configuring non-optimized parameters in real time, and combining them with candidate solutions of the parameters to be optimized to form an input parameter vector; optimizing the parameters to be optimized in real time based on the non-optimized parameters, and obtaining the efficiency deviation and the corresponding deviation of the parameters to be optimized.

[0007] As a preferred embodiment of the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms described in this invention, the fitness function parameters include input parameters and output parameters; the input parameters include parameters to be optimized and non-optimized parameters; the output parameters include optimization target parameters; the parameters to be optimized are specified as parameters that can be adjusted or modified during boiler operation to enable the boiler to operate at maximum efficiency; the genetic algorithm is used to optimize the target parameters and, based on the actual values ​​of the non-optimized parameters, adjusts the parameters to be optimized to lock in the parameter combination that can achieve the optimal boiler efficiency under the current fixed conditions, thus constructing a BP neural network model.

[0008] As a preferred embodiment of the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms described in this invention, the following steps are taken: A BP neural network model is constructed as the fitness function of the genetic algorithm, and the input parameters are used as feature parameters and the output parameters as label parameters for training to obtain the boiler efficiency under the current operating conditions. The BP neural network model includes: an input layer configured to receive an input parameter vector representing the current operating conditions of the boiler; at least six hidden layers for nonlinear mapping and feature extraction of the input parameter vector; and an output layer that outputs the optimization target parameter representing the power generation performance of the unit. The optimization target parameter is the boiler efficiency. Using the mean square error between the predicted boiler efficiency value and the normalized historical boiler efficiency value as the loss function, a gradient descent algorithm is employed to adjust the connection weights and bias terms between neurons in each layer from the output layer to the input layer, so that the loss function converges or reaches a preset number of training rounds.

[0009] As a preferred embodiment of the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms described in this invention, wherein: in the at least six hidden layers, each layer uses a modified linear function as the activation function to perform nonlinear transformation; the number of neurons in the first to fifth hidden layers is set to be greater than the number of neurons in the input layer, and the input parameter vector is subjected to dimensionality expansion feature extraction; the number of neurons in the sixth hidden layer is set to be greater than the number of neurons in the output layer, forming a convergent feature expression before the output layer.

[0010] As a preferred embodiment of the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms described in this invention, the step of establishing the boundary line equations of the parameters to be optimized includes: constructing a two-dimensional scatter plot of the parameters to be optimized with respect to the power generation load; setting the X-axis as the power generation load and the Y-axis as the parameters to be optimized; dividing the power generation load into different load intervals at equal intervals according to a predetermined step size; within each load interval, sorting the corresponding parameters to be optimized according to their numerical values; selecting the point corresponding to a predetermined low quantile as the lower boundary point and the point corresponding to a predetermined high quantile as the upper boundary point; performing curve fitting on the lower boundary points of all load intervals to form the lower boundary line equations of the parameters to be optimized; performing curve fitting on the upper boundary points of all load intervals to form the upper boundary line equations of the parameters to be optimized; and passing the upper and lower boundary line equations of the parameters to be optimized to the genetic algorithm.

[0011] As a preferred embodiment of the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms described in this invention, the operating condition optimization AI model includes: setting the encoding of the parameters to be optimized as DNA in a genetic algorithm, with any DNA forming a chromosome, initializing a population containing N individuals of the parameters to be optimized, and each individual representing a set of candidate solutions for the parameters to be optimized.

[0012] As a preferred embodiment of the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms described in this invention, the real-time model for boiler efficiency optimization includes: fitting the trained BP neural network model and the boundary line equations of the parameters to be optimized, then passing them to the genetic algorithm for configuring the input parameters and attribute parameters of the genetic algorithm; combining the candidate solutions with the non-optimized parameters under the current operating condition to form a complete input parameter vector, inputting it into the trained BP neural network model to obtain the corresponding boiler efficiency, and setting it as the fitness value of the current individual; and selecting a suitable individual from the current population using roulette wheel selection or tournament selection based on the individual's fitness value. Individuals with high fitness are selected as parents. For the selected parents, crossover and mutation operations are performed according to preset crossover and mutation probabilities to generate a new generation population. During the crossover and mutation operations, the newly generated values ​​of the parameters to be optimized are constrained within a reasonable variation range under the current load conditions by the parameter boundary line equation. The deviation between the current boiler efficiency and the optimal efficiency is identified, and the corresponding deviation of the parameters to be optimized under the efficiency deviation is obtained. The process of repetitively executing and iterating the parameters to be optimized continues until a preset number of iterations is reached or the fitness value of individuals in the population tends to stabilize. The individual with the highest fitness is then decoded to obtain the optimal combination of parameters that maximizes boiler efficiency under the current non-optimized parameter conditions.

[0013] Another objective of this invention is to provide a dynamic optimization system for thermal power unit operating conditions based on neural networks and genetic algorithms. This system can solve the problems of difficulty in constructing fitness functions, insufficient constraints on the reasonable variation range of the parameters to be optimized, and difficulty in linking efficiency deviation with the deviation of the parameters to be optimized in current thermal power unit operating condition optimization technologies through parameter modeling, boundary fitting, and optimization execution.

[0014] As a preferred embodiment of the dynamic optimization system for thermal power unit operating conditions based on neural networks and genetic algorithms described in this invention, the system includes: a parameter modeling module, a boundary fitting module, and an optimization execution module. The parameter modeling module determines the fitness function parameters, including the parameters to be optimized, non-optimized parameters, and optimization target parameters. It constructs and trains a BP neural network model using input parameters as feature parameters and output parameters as label parameters. The boundary fitting module constructs a two-dimensional scatter plot of the parameters to be optimized relative to the power generation load. It divides the power generation load into equal-interval operating conditions and selects the lower boundary point corresponding to a predetermined low quantile and the upper boundary point corresponding to a predetermined high quantile, respectively fitting these points to form the lower boundary line equation and the upper boundary line equation of the parameters to be optimized. The optimization execution module encodes the parameters to be optimized as DNA in the genetic algorithm and forms chromosomes. It then passes the BP neural network model and the boundary line equation of the parameters to be optimized to the genetic algorithm to generate an operating condition optimization AI model. It configures non-optimized parameters, constructs a real-time boiler efficiency operating condition optimization model, and optimizes the parameters to be optimized based on the non-optimized parameters, obtaining the efficiency deviation and the corresponding deviation of the parameters to be optimized.

[0015] Another objective of this invention is to provide a dynamic optimization device for thermal power unit operating conditions based on neural networks and genetic algorithms, comprising a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement a dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms.

[0016] Another object of the present invention is to provide a storage medium for dynamic optimization of thermal power unit operating conditions based on neural networks and genetic algorithms, wherein a computer program is stored thereon, and when the computer program is executed by a processor, the steps of the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms are implemented.

[0017] The beneficial effects of this invention are as follows: The dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms provided by this invention provides synergistic support in terms of fitness function construction capability, load correlation parameter constraint capability, and real-time operating condition optimization capability. This enables the operating condition optimization process to simultaneously possess the technical characteristics of continuous calculation, boundary limitation, and online execution, achieving better results in complex operating condition expression, candidate solution rationality maintenance, parameter deviation positioning, and dynamic operating condition adaptation. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 The above is an overall flowchart of a dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms, provided in Embodiment 1 of the present invention.

[0020] Figure 2 The diagram shows a BP neural network model of a dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms, as provided in Embodiment 1 of the present invention.

[0021] Figure 3 This is a boundary line equation fitting diagram for a dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms, provided in Embodiment 1 of the present invention.

[0022] Figure 4 This is a comparison diagram of three boiler efficiency calculation methods for a dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms, as provided in Embodiment 2 of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0024] Example 1, referring to Figure 1-3 As an embodiment of the present invention, a dynamic optimization method for the operating conditions of thermal power units based on neural networks and genetic algorithms is provided, comprising: S1: Determine the fitness function parameters.

[0025] It should be noted that the fitness function parameters include input parameters and output parameters; the input parameters include the parameters to be optimized and the non-optimized parameters; the output parameters include the target parameters for optimization; the parameters to be optimized are specified as those that can be adjusted or modified during boiler operation to enable the boiler to perform at maximum efficiency. The genetic algorithm (GA) is used to optimize the target parameters and adjusts the parameters to be optimized based on the actual values ​​of the non-optimized parameters, thereby locking in the parameter combination that can achieve the optimal boiler efficiency under the current fixed conditions.

[0026] It should also be noted that, based on the business object of boiler efficiency optimization, the fitness function parameters are determined, and the fitness function parameters are divided into input parameters and output parameters. The parameters to be optimized are selected as those that can be adjusted or modified during boiler operation, while the non-optimized parameters are selected as those that cannot be adjusted during boiler operation. The output parameters are determined as the optimization target parameters. Sample records corresponding to the parameters to be optimized, non-optimized parameters, and optimization target parameters at the same time are extracted from historical data. The samples are then screened, deduplicated, and unstable operating conditions are removed. The retained samples are assembled into an input parameter vector according to a unified parameter order.

[0027] A BP neural network model 100 is constructed as the fitness function of the genetic algorithm. The BP neural network model 100 includes an input layer C1, at least six hidden layers C2, and an output layer C3. The input layer C1 is used to receive the input parameter vector representing the current operating condition of the boiler. The at least six hidden layers C2 are used to perform nonlinear mapping and feature extraction on the input parameter vector. Each layer uses a modified linear function as the activation function. The number of neurons in the first to fifth hidden layers C2 is set to 2 to 4 times the number of neurons in the input layer C1. The number of neurons in the sixth hidden layer C2 is set to 1.5 to 2 times the number of neurons in the output layer C3. The output layer C3 is used to output the optimization target parameters representing the power generation performance of the unit.

[0028] The input parameters received by the input layer C1 include main steam flow rate, main steam pressure, main steam temperature, reheat steam pressure, reheat steam temperature, ambient temperature, average moisture content of coal fed into the furnace, total coal quantity in the boiler, boiler air volume, flue gas oxygen content, and flue gas temperature. These parameters together form the input parameter vector, which serves as the input basis for the BP neural network model 100.

[0029] like Figure 2 As shown, the input parameter vector is passed to the hidden layers 1, 2, 3, 4, 5, and 6 in sequence. For example, the number of neurons in each layer is set to 60, 50, 40, 30, 20, and 10 in sequence. This is used to map and extract the multi-parameter coupling relationship in the boiler operation condition layer by layer, so that the nonlinear influence relationship between different parameters on boiler efficiency is expressed step by step in the network. The result after processing by the six hidden layers C2 is passed to the output layer C3, which outputs the "final boiler efficiency". The "final boiler efficiency" is the output parameter of the BP neural network model 100, and it is also the result of the fitness function called by the subsequent genetic algorithm.

[0030] Using the mean square error between the predicted boiler efficiency and the normalized historical boiler efficiency as the loss function, the gradient descent algorithm is used to adjust the connection weights and bias terms between neurons in each layer from the output layer C3 to the input layer C1 until the loss function converges or reaches the preset number of training rounds, thus obtaining a BP neural network model 100 that can output the boiler efficiency under the current operating conditions.

[0031] The trained BP neural network model 100 is passed to the genetic algorithm so that after receiving the boundary equation 200 of the parameters to be optimized, the genetic algorithm can calculate the boiler efficiency corresponding to the candidate solution based on the BP neural network model 100, providing a fitness function basis for the subsequent release of the operating condition optimization AI model 300 and the construction of the boiler efficiency operating condition optimization real-time model 400.

[0032] S2: Fit the upper and lower boundary lines of the parameter to be optimized under different loads.

[0033] It should be noted that establishing the boundary equation 200 for the parameters to be optimized includes: constructing a two-dimensional scatter plot of the parameters to be optimized with respect to the power generation load; setting the X-axis as the power generation load and the Y-axis as the parameters to be optimized; dividing the power generation load into different load intervals at predetermined step sizes; within each load interval, sorting the corresponding parameters to be optimized according to their numerical values; selecting the point corresponding to a predetermined low quantile as the lower boundary point and the point corresponding to a predetermined high quantile as the upper boundary point; performing curve fitting on the lower boundary points of all load intervals to form the lower boundary equation for the parameters to be optimized; performing curve fitting on the upper boundary points of all load intervals to form the upper boundary equation for the parameters to be optimized; and passing the upper and lower boundary equations of the parameters to be optimized to the genetic algorithm.

[0034] It should also be noted that, firstly, a corresponding sample sequence of "power generation load - parameter to be optimized" is established for each parameter to be optimized, and the sample sequences are sorted in ascending order of power generation load to form a parameter sample matrix for boundary fitting.

[0035] The power generation load is divided into equal-interval operating conditions according to a predetermined step size. The number of samples in each load interval is counted. When the number of samples in a certain load interval is lower than the preset lower limit, the current load interval is merged with the adjacent load interval before boundary extraction. In each load interval, the corresponding parameter sample values ​​to be optimized are reordered according to their numerical values, and the parameter values ​​at the predetermined low quantile position and the predetermined high quantile position are extracted as the lower boundary candidate point and the upper boundary candidate point of the current load interval, respectively.

[0036] If multiple identical quantile values ​​exist within the same load interval, the sample point closest to the central load of that load interval is preferentially retained as the boundary point. The lower boundary points corresponding to all load intervals are combined into a lower boundary point sequence, and the upper boundary points corresponding to all load intervals are combined into an upper boundary point sequence. The continuity of the two sequences is checked. When the boundary point jump variable of adjacent load intervals exceeds a preset threshold, the boundary points of the adjacent intervals are used to perform local smoothing correction on the abnormal jump point. Then, curve fitting is performed to form the lower boundary line equation and upper boundary line equation of the parameter to be optimized.

[0037] For each parameter to be optimized, the above process of working condition segmentation, quantile extraction, boundary point verification and curve fitting is repeated to obtain the upper and lower boundary line equations corresponding to the main steam pressure, main steam temperature, hot reheat pressure, hot reheat temperature, feedwater flow rate, feedwater temperature, feedwater pressure, flue gas oxygen content and flue gas temperature. After establishing a parameter index relationship between each set of boundary line equations 200 and the corresponding parameter to be optimized, they are uniformly written into the boundary line equation set.

[0038] During the genetic algorithm invocation phase, the current power generation load is determined based on the real-time input non-optimized parameters. The upper and lower boundary values ​​of each parameter to be optimized under the current power generation load are read from the boundary line equation set. These upper and lower boundary values ​​are used as parameter constraint boundaries for DNA crossover mutation and gene mutation in the genetic algorithm, so that the newly generated candidate values ​​of the parameters to be optimized are always within the reasonable variation range corresponding to the current load condition.

[0039] S3: Use a genetic algorithm to receive fitness function parameters and upper and lower boundary lines, configure non-optimized parameters in real time, and combine them with candidate solutions of the parameters to be optimized to form an input parameter vector.

[0040] It should be noted that the working condition optimization AI model 300 includes setting the encoding of the parameters to be optimized as DNA in a genetic algorithm, with any DNA forming a chromosome, initializing a population containing N individuals of the parameters to be optimized, and each individual representing a set of candidate solutions for the parameters to be optimized.

[0041] It should also be noted that the input parameters of the genetic algorithm are first configured to be the same as the input parameter set of the BP neural network model 100, keeping the parameter order of the parameters to be optimized and the non-optimized parameters consistent. Each parameter to be optimized is assigned a DNA, and any DNA is spliced ​​together to form a chromosome according to the preset parameter order, so that each chromosome uniquely corresponds to a set of candidate solutions for the parameters to be optimized. According to the boundary equation 200 of the parameters to be optimized, the values ​​of each DNA are initialized according to the upper and lower boundary values ​​corresponding to the current power generation load, forming a population containing N individuals of the parameters to be optimized, where each individual corresponds to a chromosome and represents a set of candidate solutions for the parameters to be optimized that satisfy the boundary constraints of the current load condition.

[0042] After initializing the population, the attribute parameters of the genetic algorithm are set, including DNA length, population size, target value optimization direction, crossover mutation probability, gene mutation probability, and number of iterations. In the implementation of boiler efficiency condition optimization, the DNA length is set to 8, the population size is set to 50, the target value optimization direction is set to 1, the crossover mutation probability is set to 0.2, the gene mutation probability is set to 0.05, and the number of iterations is set to 100. Then, the trained BP neural network model 100 is used as the fitness function of the genetic algorithm, and the boundary equation 200 of the parameters to be optimized is used as the reasonable boundary for DNA crossover mutation and gene mutation. The fitness function, boundary equation 200, genetic algorithm input parameters, and attribute parameters are uniformly encapsulated to generate the condition optimization AI model 300.

[0043] When the working condition optimization AI model 300 was released, an independent parameter configuration interface was established for non-optimized parameters. This allows non-optimized parameters to participate in the candidate solution calculation as fixed input conditions during the operation of the working condition optimization AI model 300, and together with the candidate solutions of the parameters to be optimized, they form a complete input parameter vector.

[0044] S4: Optimize the parameters to be optimized in real time based on non-optimized parameters, and obtain the efficiency deviation and the corresponding deviation of the parameters to be optimized.

[0045] After the AI ​​model 300 for optimizing operating conditions is released, the non-optimized parameters collected in real time are first written into the real-time model 400 for optimizing boiler efficiency operating conditions. Based on the power generation load in the non-optimized parameters, the current load condition is determined. Then, the boundary line equations 200 corresponding to each parameter to be optimized are called to calculate the upper and lower boundary values ​​of the main steam pressure, main steam temperature, heat reheat pressure, heat reheat temperature, feedwater flow rate, feedwater temperature, feedwater pressure, flue gas oxygen content, and flue gas temperature under the current load condition, forming the parameter constraint interval corresponding to the current optimization round.

[0046] Based on the real-time values ​​of each parameter to be optimized under the current operating conditions, initial candidate solutions are generated within the corresponding parameter constraint interval. Each set of candidate solutions is concatenated with the non-optimized parameters at the current time in a predetermined input order to form a complete input parameter vector, which is then input into the BP neural network model 100 in batches to obtain the boiler efficiency prediction value corresponding to each candidate solution. The boiler efficiency prediction value is then set as the fitness value of the corresponding individual.

[0047] After the fitness values ​​are formed, individuals with higher fitness are selected from the current population as parents using roulette wheel selection or tournament selection. Crossover operations are performed on the parent individuals according to the preset crossover probability, and mutation operations are performed according to the preset mutation probability. After the crossover and mutation are completed, the newly generated parameters to be optimized are checked one by one. When a newly generated value exceeds the upper or lower boundary value of the corresponding parameter under the current load condition, the newly generated value is reset to the corresponding boundary value, or regenerated within the boundary interval.

[0048] The process of "input parameter vector construction—boiler efficiency calculation—fitness value assignment—parent selection—crossover mutation—boundary check" is repeatedly executed on the new generation population. After each iteration, the individual with the highest fitness and its corresponding combination of parameters to be optimized are recorded. When the preset number of iterations is reached or the individual with the highest fitness remains unchanged for multiple consecutive iterations, the iteration stops. The individual with the highest fitness is decoded as the optimal combination of parameters to be optimized under the current non-optimized parameter conditions. Simultaneously, the predicted boiler efficiency value corresponding to this optimal combination is compared with the input parameter vector composed of the real-time parameters to be optimized and the non-optimized parameters under the current operating condition. The calculated boiler efficiency is compared to obtain the efficiency deviation between the current boiler efficiency and the optimal efficiency. The optimal combination of parameters to be optimized is compared with the real-time value of the current parameters to be optimized item by item to obtain the deviation of main steam pressure, main steam temperature, hot reheat pressure, hot reheat temperature, feedwater flow rate, feedwater temperature, feedwater pressure, flue gas oxygen content, and flue gas temperature. The deviation results of the parameters to be optimized are formed in a one-to-one correspondence with the efficiency deviation. The individual with the highest fitness obtained in the previous iteration is used as one of the candidate individuals for the population initialization in the next moment, and participates in the optimization calculation again together with the current non-optimized parameters.

[0049] Example 2, refer to Figure 4 As an embodiment of the present invention, a dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms is provided. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0050] First, 10,000 sets of working condition optimization data were selected, including traditional benchmark working condition optimization data and working condition optimization data based on neural networks and genetic algorithms, for verification of correctness and consistency.

[0051] The optimization data for operating conditions includes traditional benchmark optimization data and optimization data based on neural networks and genetic algorithms. Using boiler efficiency as the optimization target, in each sample, the current boiler efficiency, the boiler efficiency optimized from the benchmark operating condition, and the boiler efficiency optimized based on neural networks and genetic algorithms are recorded. The boiler efficiency optimized from the benchmark operating condition is compared with the current boiler efficiency; if it is larger, it is recorded as a positive direction; if it is smaller, it is recorded as a negative direction. Similarly, the boiler efficiency optimized based on neural networks and genetic algorithms is compared with the current boiler efficiency; if it is larger, it is recorded as a positive direction; if it is smaller, it is recorded as a negative direction. The discrepancies are statistically analyzed to determine if the directions of deviation are consistent, and the number of discrepancies, N, is recorded.

[0052] like Figure 4 As shown, the black curve represents the boiler efficiency using the reverse balancing method, the red curve represents the boiler efficiency using big data AI, and the blue curve represents the historical best boiler efficiency. At each sampling time, the red curve is generally above the black curve and its trend is consistent with the black curve. The blue curve, as the historical best boiler efficiency benchmark, is located above it. Thus, at each sampling time, the efficiency deviation between the current boiler efficiency and the optimized boiler efficiency can be obtained, and the corresponding deviation of the parameter to be optimized can be further deduced. Based on the above implementation process, the real-time model for optimizing boiler efficiency can continuously optimize the parameter to be optimized under the current non-optimized parameter conditions and output the deviation result of the parameter to be optimized corresponding to the efficiency deviation, which can be used to form the basis for parameter adjustment under boiler operating conditions.

[0053] Table 1. Verification of the correctness and consistency of the optimization under operating conditions.

[0054] The accuracy of the optimization target direction is expressed as: , in, This indicates the accuracy of the optimization target direction. This indicates the number of times the deviation direction is inconsistent.

[0055] The deviation rate obtained from benchmarking is expressed as: , in, This indicates the deviation rate in benchmark condition optimization. This indicates the current boiler efficiency. This indicates the boiler efficiency obtained by optimizing the benchmark operating conditions.

[0056] The deviation rate obtained by neural networks and genetic algorithms is expressed as: , in, This represents the deviation rate of the optimization search in neural networks and genetic algorithms. This represents the boiler efficiency optimized by neural networks and genetic algorithms.

[0057] The degree of agreement between the deviation of the optimization target parameters and the target parameters is expressed as: , in, This indicates the degree of agreement between the deviation of the target parameters and the target parameters.

[0058] Experimental results show that the accuracy of the optimization target direction is 99%, and the consistency of the deviation of the optimization target parameters is 95%.

[0059] Example 3, an embodiment of the present invention, provides a dynamic optimization system for thermal power unit operating conditions based on neural networks and genetic algorithms, including a parameter modeling module, a boundary fitting module, and an optimization execution module.

[0060] The parameter modeling module is used to determine the fitness function parameters, including the parameters to be optimized, the non-optimized parameters, and the optimization target parameters. It constructs and trains a BP neural network model 100 by using the input parameters as feature parameters and the output parameters as label parameters.

[0061] The boundary fitting module is used to construct a two-dimensional scatter plot of the parameters to be optimized with respect to the power generation load. It divides the power generation load into equal-interval operating conditions and selects the lower boundary point corresponding to the predetermined low quantile and the upper boundary point corresponding to the predetermined high quantile. The lower boundary line equation and the upper boundary line equation of the parameters to be optimized are then fitted respectively.

[0062] The optimization execution module is used to encode the parameters to be optimized as DNA in the genetic algorithm and form chromosomes. It passes the BP neural network model 100 and the boundary line equation 200 of the parameters to be optimized to the genetic algorithm to generate the operating condition optimization AI model 300, configure the non-optimization parameters, construct the boiler efficiency operating condition optimization real-time model 400, and optimize the parameters to be optimized based on the non-optimization parameters to obtain the efficiency deviation and the corresponding deviation of the parameters to be optimized.

[0063] This embodiment also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements a dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as proposed in the above embodiment.

[0064] This embodiment also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements a dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as proposed in the above embodiment.

[0065] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0066] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0067] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0068] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms, characterized in that, include: Determine the fitness function parameters; Fit the upper and lower boundary lines of the parameter to be optimized under different loads; The genetic algorithm is used to receive fitness function parameters and upper and lower boundary lines, configure non-optimized parameters in real time, and combine them with candidate solutions of parameters to be optimized to form an input parameter vector. Based on the non-optimized parameters, the optimization of the parameters to be optimized is performed in real time to obtain the efficiency deviation and the corresponding deviation of the parameters to be optimized.

2. The dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as described in claim 1, characterized in that: The fitness function parameters include input parameters and output parameters; The input parameters include parameters to be optimized and non-optimized parameters; The output parameters include the optimization target parameters; The parameters to be optimized are specified as those that can be adjusted or modified during boiler operation, so that the boiler can perform at its maximum efficiency. The genetic algorithm is used to find the target parameters and adjust the parameters to be optimized based on the actual values ​​of the non-optimized parameters. The parameter combination that can achieve the optimal boiler efficiency under the current fixed conditions is locked, and a BP neural network model (100) is constructed.

3. The dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as described in claim 2, characterized in that: The constructed BP neural network model (100) is used as the fitness function of the genetic algorithm. The input parameters are used as feature parameters and the output parameters are used as label parameters for training to obtain the boiler efficiency under the current working conditions. The BP neural network model (100) includes, The input layer (C1) is configured to receive an input parameter vector that characterizes the current operating condition of the boiler. At least six hidden layers (C2) are used to perform nonlinear mapping and feature extraction on the input parameter vector; The output layer (C3) outputs the optimization target parameters characterizing the power generation performance of the unit; The optimization target parameter is boiler efficiency; Using the mean square error between the predicted boiler efficiency and the normalized historical boiler efficiency as the loss function, the gradient descent algorithm is used to adjust the connection weights and bias terms between neurons in each layer from the output layer (C3) to the input layer (C1) layer by layer, so that the loss function converges or reaches the preset training rounds.

4. The dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as described in claim 3, characterized in that: In the at least six hidden layers (C2), each layer uses a modified linear function as the activation function to perform a nonlinear transformation; The number of neurons in the first to fifth hidden layers (C2) is set to be greater than the number of neurons in the input layer (C1). The input parameter vector is expanded to extract features. The number of neurons in the sixth hidden layer (C2) is set to be greater than the number of neurons in the output layer (C3), forming a convergent feature expression before the output layer.

5. The dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as described in any one of claims 1, 2, and 4, characterized in that: The fitted upper and lower boundary lines of the parameters to be optimized under different loads include: Establish the boundary line equation (200) for the parameters to be optimized, construct a two-dimensional scatter plot of the parameters to be optimized with respect to the power generation load, set the X-axis as the power generation load and the Y-axis as the parameters to be optimized, divide the power generation load into equal intervals according to a predetermined step size to form different load intervals, sort the corresponding parameters to be optimized in each load interval according to their numerical values, select the point corresponding to the predetermined low quantile as the lower boundary point, and select the point corresponding to the predetermined high quantile as the upper boundary point; Curve fitting is performed on the lower boundary points of all load intervals to form the lower boundary line equation of the parameter to be optimized; Curve fitting is performed on the upper boundary points of all load intervals to form the upper boundary line equation of the parameter to be optimized; The equations of the upper and lower boundary lines of the parameters to be optimized are passed to the genetic algorithm.

6. The dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as described in claim 5, characterized in that: The method of using a genetic algorithm to receive fitness function parameters and upper and lower boundary lines includes, Release the working condition optimization AI model (300), set the encoding of the parameters to be optimized as DNA in the genetic algorithm, any DNA makes up a chromosome, initialize a population containing N individuals of the parameters to be optimized, and each individual represents a set of candidate solutions of the parameters to be optimized.

7. The dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as described in any one of claims 1, 2, 4, and 6, characterized in that: The method of optimizing parameters in real time based on non-optimized parameters includes... A real-time model for optimizing boiler efficiency (400) is constructed. The trained BP neural network model (100) and the boundary line equation (200) of the parameters to be optimized are fitted and then passed to the genetic algorithm to configure the input parameters and attribute parameters of the genetic algorithm. The candidate solution is combined with the non-optimal parameters under the current operating condition to form a complete input parameter vector, which is then input into the trained BP neural network model (100) to obtain the corresponding boiler efficiency and set as the fitness value of the current individual. Based on the individual's fitness value, roulette wheel selection or tournament selection is used to select individuals with high fitness from the current population as parents. For the selected parent individuals, crossover and mutation operations are performed according to the preset crossover and mutation probabilities to generate a new generation population; In the crossover and mutation operations, the newly generated values ​​of the parameters to be optimized are constrained within a reasonable range of variation under the current load condition by the parameter boundary line equation; Find the deviation between the current boiler efficiency and the optimal efficiency, and obtain the corresponding deviation of the parameters to be optimized under the effect of efficiency deviation; Repeatedly execute and iterate the parameters to be optimized until the preset number of iterations is reached or the fitness value of an individual in the population tends to stabilize. Decode the individual with the highest fitness to obtain the optimal combination of parameters that maximizes boiler efficiency under the current non-optimized parameter conditions.

8. A dynamic optimization system for thermal power unit operating conditions based on neural networks and genetic algorithms, employing the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as described in any one of claims 1 to 7, characterized in that: Parametric modeling module, boundary fitting module, optimization execution module; The parameter modeling module is used to determine the fitness function parameters, including the parameters to be optimized, the non-optimized parameters, and the optimization target parameters, and to construct and train a BP neural network model (100) using the input parameters as feature parameters and the output parameters as label parameters. The boundary fitting module is used to construct a two-dimensional scatter plot of the parameters to be optimized with respect to the power generation load. It divides the power generation load into equal-interval operating conditions and selects the lower boundary point corresponding to the predetermined low quantile and the upper boundary point corresponding to the predetermined high quantile, respectively fitting the lower boundary line equation and the upper boundary line equation of the parameters to be optimized. The optimization execution module is used to encode the parameters to be optimized as DNA in the genetic algorithm and form chromosomes, pass the BP neural network model (100) and the boundary line equation (200) of the parameters to be optimized to the genetic algorithm, generate the working condition optimization AI model (300), configure the non-optimized parameters, construct the boiler efficiency working condition optimization real-time model (400), and optimize the parameters to be optimized based on the non-optimized parameters to obtain the efficiency deviation and the corresponding deviation of the parameters to be optimized.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the dynamic optimization method for thermal power unit operating conditions based on neural networks and genetic algorithms as described in any one of claims 1 to 7.

Images