Smart platform-based environmental control unit management system for coal-fired power plant

Abstract

Provided is a smart platform-based environmental control unit management system for a coal-fired power plant, relating to the technical field of flue gas treatment. The system comprises a sensing layer, an application layer, and a decision-making layer, wherein the sensing layer comprises a plurality of data acquisition devices for collecting data of each piece of operating equipment within an environmental control unit; the application layer comprises a flue gas management module, wherein the flue gas management module creates a flue gas management prediction model on the basis of the data collected by the sensing layer and historical data records; and the decision-making layer determines an optimal coordinated control strategy for the environmental control unit on the basis of the model created by the application layer and the data. According to the platform-based environmental control unit management system, parameter information is acquired by means of a sensing layer, and emission prediction is achieved by means of modeling by an application layer; the application layer uploads a prediction model and sensing layer parameters to a decision-making layer; the decision-making layer applies a global particle swarm optimization algorithm to output parameters and the created models, and adjusts a parameter range and a pollutant emission threshold as constraint conditions with the aim of minimizing costs, so as to obtain an optimal system control strategy.

Description

A Smart Environmental Protection Island Platform Management System for Coal-fired Power Plants Technical Field

[0001] This application relates to the field of container terminal loading technology, specifically to a method and apparatus for intelligent loading within the hold of a container terminal vessel. Background Technology

[0002] Currently, most smart power plants focus primarily on the construction of boiler main units, coal yards, or the overall plant concept, lacking systematic research and engineering demonstrations of smart environmental protection operation and management platforms. There is a lack of intelligent analysis of massive amounts of data from environmental protection facilities, and a lack of the ability to propose early warnings and optimization suggestions based on the results of intelligent analysis and automated learning to guide the optimized operation of power plant environmental protection facilities and achieve energy conservation, emission reduction, and consumption reduction. Furthermore, there is a lack of a comprehensive analysis of facility and equipment characteristics, research and summary of operating conditions, and the construction of a smart operation optimization technology and management system that integrates monitoring, intelligent analysis, system diagnosis, refined adjustment, early warning, environmental evaluation, and supervision. Summary of the Invention

[0003] The purpose of this invention is to provide a smart environmental protection island platform management system for coal-fired power plants to solve the problems mentioned in the background art, including:

[0004] The sensing layer includes multiple data acquisition devices for collecting data from all working equipment within the environmental protection island;

[0005] The application layer includes a flue gas control module, which establishes a flue gas control prediction model based on data collected by the perception layer and historical data records.

[0006] The decision-making layer determines the optimal collaborative strategy for the environmental protection island based on the models and data established by the application layer.

[0007] Furthermore, the flue gas control prediction model includes a desulfurization prediction model, a denitrification prediction model, and a dust removal prediction model.

[0008] Furthermore, the data collected by the sensing layer includes desulfurization system data, which includes boiler load, flue gas flow rate, inlet flue gas SO2 concentration, outlet flue gas SO2 concentration, limestone slurry pH value, calcium flow ratio, absorption tower spray volume, and inlet flue gas PM concentration.

[0009] Furthermore, the process of establishing the desulfurization optimization model is as follows:

[0010] Based on the desulfurization system data acquired by the sensing layer, an MA-RELM prediction model is constructed. Specific steps include:

[0011] S31 sets the number of iterations n for the mayfly algorithm, the population size m, the range of individual population variation, and related parameters;

[0012] S32 initializes the population;

[0013] S33 updates the population individuals according to the algorithm principle and feeds the individuals into the RELM model, using the training set for training;

[0014] S34 feeds the validation set into the trained model for prediction and uses the prediction error as the fitness function.

[0015] S35 updates the mayfly algorithms pbest and gbest by mating male and female individuals to produce two offspring.

[0016] S36 calculates the fitness values ​​of the offspring and updates pbest and gbes;

[0017] S37 is the iteration termination condition. The best individual in the population is output, and the optimal value is used to train the RELM model.

[0018] S38 inputs the test set into the trained RELM prediction model to obtain the final prediction result.

[0019] Furthermore, the data collected by the sensing layer includes desulfurization system data, and the denitrification data includes boiler load, flue gas flow rate, inlet flue gas NOx concentration, outlet flue gas NOx concentration, reaction temperature, ammonia-nitrogen molar ratio, and inlet flue gas oxygen content.

[0020] Furthermore, the denitrification prediction model establishment process involves constructing an MA-RELM prediction model based on the denitrification system data obtained by the sensing layer.

[0021] Furthermore, the data collected by the sensing layer includes dust removal system data, which includes boiler load, flue gas flow rate, dust removal electric field voltage / current value, and dust removal chamber temperature.

[0022] Furthermore, the dust removal prediction model establishment process involves constructing an MA-RELM prediction model based on the dust removal system data obtained by the sensing layer.

[0023] Furthermore, the decision-making layer, based on the desulfurization prediction model, denitrification prediction model, and dust removal prediction model, determines the optimal control strategy with the goal of minimizing costs through a particle swarm optimization algorithm.

[0024] The constraints of the particle swarm optimization algorithm are adjustment range constraints and pollutant emission limit constraints.

[0025] The beneficial effects of this invention are as follows:

[0026] The environmental island platform management system provided by this invention obtains various parameter information through the perception layer, models it through the application layer, and realizes the prediction of emissions. The application layer uploads the prediction model and the parameters of the perception layer to the decision layer. The decision layer performs particle swarm optimization algorithm on the output parameters and each model, with the goal of minimizing cost, and adjusts the range of parameters and pollutant emission limits as constraints to obtain the optimal system control strategy. Attached Figure Description

[0027] Figure 1 is a flowchart of this prediction model;

[0028] Figure 2 is a flowchart of the decision-making algorithm. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] This invention provides a smart environmental protection island platform management system for coal-fired power plants, as shown in Figure 1. It includes a perception layer for acquiring data, an application layer for establishing predictive models, and a decision-making layer for determining the lowest cost. The perception layer acquires data by relying on existing industrial control systems such as DCS and SIS, and data acquisition equipment such as data lakes, to collect data on power plant production facilities and environmental protection facilities. The perception layer itself has a data anomaly alarm module. When one or more data items are abnormal, the perception layer reacts by issuing an on-site alarm and simultaneously feeding back to the application layer. The application layer records the abnormal data and predicts other out-of-control data based on its predictive models, reporting to the decision-making layer while issuing a warning to the perception layer. The predictive models established by the application layer include desulfurization prediction models, denitrification prediction models, and dust removal prediction models.

[0031] Example 1:

[0032] This embodiment provides a desulfurization prediction model. The desulfurization system in this embodiment adopts a limestone-ridge wet desulfurization system, with main components including an absorption tower, oxidation fan, slurry circulation pump, slurry agitator, and demister. The desulfurization reaction can be summarized into five processes: SO2 absorption, limestone dissolution, neutralization, oxidation, and crystallization. Sulfur dioxide removal involves multiple physicochemical changes; boiler load, operating parameters, and system structure all affect sulfur dioxide removal efficiency. This embodiment uses the following parameters as input parameters for the prediction model: load, flue gas flow rate, inlet flue gas SO2 concentration, outlet flue gas SO2 concentration, limestone slurry pH value, calcium flow ratio, absorption tower spray rate, and inlet flue gas PM concentration.

[0033] The prediction model used in this embodiment is the RELM model optimized by the floating optimization algorithm, and the MA-RELM prediction model. ELM is a single-layer feedforward neural network, but unlike traditional neural networks that use backpropagation to change the hidden layer weights, the hidden layer weights of the Extreme Learning Machine are random or manually given and do not need to be updated. During the neural network training process, only the weights of the output layer are calculated. Because ELM does not need to adjust the hidden layer nodes, but only needs to solve for the output weights, it has high training efficiency and generalization ability. The training process of ELM is as follows:

[0034] Let the number of neurons in the input layer of the ELM model be s, the number of neurons in the output layer be m, the number of neurons in the hidden layer be l, and the connection weights between the input layer and the hidden layer be w. We can obtain the weight matrix w as follows:

[0035] Let the bias on the hidden layer node be b, then we get the bias vector b:

[0036] Let the output of the hidden layer be H(x). Then the formula for calculating H(x) is as follows: H(x) = [h1(x), ..., h m (x)] h i (x)=g(W i x+b i )

[0037] In the formula, hi(x) is the output of the i-th hidden layer node; g is the activation function.

[0038] Let the connection weight between the hidden layer and the output layer be β:

[0039] ELM obtains the model output f(x) through the hidden layer connection weights β:

[0040] Because of w i b i Since β is unknown, the training process of ELM is actually a process of solving for these three unknowns, including the hidden layer node parameters (w... i b i The output layer weights β can be randomly generated based on any continuous probability distribution or selected manually. The weights β are primarily calculated by minimizing the training error, with the objective function being: min||Hβ-T|| 2

[0041] In the formula, T represents the target matrix of the training data.

[0042] RLEM introduces an L2 regularization term into the original ELM. By adding a regularization term, it prevents the model from overfitting, thereby improving the generalization ability of the prediction model. The improved objective function is:

[0043] In the formula, C is the regularization coefficient.

[0044] The mayfly algorithm is an intelligent algorithm with advantages in convergence accuracy and speed. This embodiment applies it to the aforementioned RLEM model to optimize the number of neurons in the hidden layer. During training, the data is divided into training, validation, and test sets in a 4:1:1 ratio. The training set is used to train the model and determine the output weights β of the RELM model. The validation set is used for internal evaluation of the model, optimizing and adjusting hyperparameters. The test set is used to test the model's generalization ability, i.e., the prediction error when applied to real-world scenarios. Both the validation and test sets are new datasets for training the model, but the validation set is used as a known dataset to evaluate the model's performance in actual modeling. By feeding the validation set into the trained model for prediction, and using the prediction error as the fitness function of the optimization algorithm, the model parameters are optimized and adjusted through continuous iterative updates. The trained model optimized through the validation set cannot reflect the model's generalization ability in real-world situations, but it can reflect the overall expressive power of the model, that is, the model's ability to fit complex controlled objects. The results prove the feasibility of the algorithm model.

[0045] In summary, the steps for constructing a MA-RELM prediction model include:

[0046] S31 sets the number of iterations n for the mayfly algorithm, the population size m, the range of individual population variation, and related parameters;

[0047] S32 initializes the population;

[0048] S33 updates the population individuals according to the algorithm principle and feeds the individuals into the RELM model, using the training set for training;

[0049] S34 feeds the validation set into the trained model for prediction and uses the prediction error as the fitness function.

[0050] S35 updates the mayfly algorithms pbest and gbest by mating male and female individuals to produce two offspring.

[0051] S36 calculates the fitness values ​​of the offspring and updates pbest and gbes;

[0052] S37 is the iteration termination condition. The best individual in the population is output, and the optimal value is used to train the RELM model.

[0053] S38 inputs the test set into the trained RELM prediction model to obtain the final prediction result.

[0054] Example 2:

[0055] The denitrification system used in this invention is an SCR catalytic reaction system. SCR, as a mature and efficient denitrification technology, has been applied in denitrification processes across multiple fields. Parameters affecting SCR denitrification efficiency include boiler load, flue gas flow rate, inlet flue gas NOx concentration, outlet flue gas NOx concentration, reaction temperature, ammonia-nitrogen molar ratio, and inlet flue gas oxygen content.

[0056] Based on the above parameters, this embodiment uses the method of constructing a desulfurization model in Example 1 to obtain a denitrification prediction model.

[0057] Example 3:

[0058] Dust removal systems typically employ either dry or wet electrostatic precipitators. This embodiment uses a dry electrostatic precipitator system as an example. The system's main components include a power supply, electrodes, a dust hopper, and a rapping device. The dust removal process is primarily divided into four stages: gas ionization, dust charging, dust collection, and dust removal via rapping. Factors affecting dust removal efficiency include boiler load, flue gas flow rate, dust removal electric field voltage / current, and dust collection chamber temperature.

[0059] Based on the above parameters, this embodiment uses the method of constructing a desulfurization model in Embodiment 1 to obtain a dust removal prediction model.

[0060] The application layer displays the prediction results in real time based on each prediction model and sends the prediction models and results to the decision layer.

[0061] Example 4:

[0062] Based on the models constructed in Examples 1-3 and the costs of each system, the decision-making layer uses the particle swarm optimization algorithm to obtain the lowest cost decision that meets the environmental protection requirements of each system.

[0063] The cost calculation methods for each system are as follows:

[0064] In desulfurization systems, operating costs are primarily comprised of energy and material consumption. Energy consumption is mainly generated by the motor equipment in the desulfurization system, including the power consumption of booster fans, oxidation fans, slurry circulation pumps, and slurry agitators. The cost formula is as follows:

[0065] Where g is the real-time load of the boiler, n bf n sa n scp n oabThese represent the number of operating units of the booster fan, oxidation fan, slurry circulation pump, and slurry agitator, respectively. i I i Let be the voltage and current of the i-th device, respectively. P is the power factor. E For electricity price, α WFGD This indicates the proportion of the desulfurization tower resistance to the total resistance in the latter half of the section, and its calculation method is as follows:

[0066] Where, p dt It is the pressure drop of the desulfurization tower, p WESP It is the resistance voltage drop of the wet electrostatic precipitator, p gd2 It is the pressure drop due to resistance in the flue section.

[0067] In selective catalytic reduction (SCR) denitrification systems, the main cost components are energy consumption and material consumption. Energy consumption primarily includes the power consumption of the induced draft fan, soot blowing fan, and dilution fan, which are calculated as follows:

[0068] Where, n idf ,n sb ,n adf These represent the number of operating induced draft fans, soot blowing motors, and dilution fans, respectively. SCR This indicates the proportion of the denitrification reactor resistance to the total resistance in the first half of the process. The calculation method is as follows:

[0069] The calculation method for the operating cost of a soot blowing system varies depending on the soot blowing method, where P... steam It is empirical steam energy consumption, CV s The first is the empirical reference catalyst dosage, while the second is the actual catalyst dosage.

[0070] The main material consumption of the denitrification system is the cost of liquid ammonia and catalyst. Based on material balance, the cost of liquid ammonia is calculated as follows:

[0071] The method for calculating catalyst loss cost is as follows:

[0072] In the formula, Pc is the catalyst price, Q is the unit capacity, and h is the annual operating hours of the unit.

[0073] In summary, the overall operating cost of the catalytic reduction denitrification system in this embodiment can be expressed as:

[0074] The operating cost of electrostatic precipitators is mainly due to power consumption. For dry electrostatic precipitators, the power consumption primarily consists of the induced draft fan and the power supply. The calculation formulas are as follows:

[0075] In the formula, ne represents the electric field quantity, α ESP The ratio of the electrostatic precipitator resistance to the total resistance in the first half of the stage is calculated as follows:

[0076] The operating cost of an electrostatic precipitator is: COST ESP =COST ESP_idf +COST ESP_e

[0077] The power consumption of the induced draft fan in a wet electrostatic precipitator is related to the proportion of its resistance to the first half of the resistance. The calculation method is as follows:

[0078] The power consumption of a wet electrostatic precipitator is:

[0079] In addition to the costs mentioned above, there are also fixed costs for denitrification, such as depreciation costs, maintenance costs, and labor costs.

[0080] In summary, the total cost of the environmental island is min COST = COST WFGD +COST WFGD_fix +COST SCR +COST SCR_fix +COST ESP +COST ESP_fix +COST WESP +COST WESP_fix

[0081] The constraints include emission limits for NOx, SO2, and PM.

[0082] Based on the above MA-RELM prediction models, the maximum number of iterations of the particle swarm optimization algorithm is set to 100, and the number of particles is set to 100. Limits are set for the working conditions under typical load and inlet concentration. The input parameters include the ammonia injection rate, the electric field voltage of the electrostatic precipitator, the pH value of the limestone slurry, and the number of circulating pumps in operation. Finally, the optimal solution for each parameter and the lowest cost optimization scheme for the environmental protection island can be obtained.

[0083] In addition to the above embodiments, this application also includes other implementation methods. All technical solutions formed by equivalent transformation or equivalent substitution should fall within the protection scope of the claims of this invention.

Claims

1. A smart environmental protection island platform management system for coal-fired power plants, characterized in that, include: The sensing layer includes multiple data acquisition devices for collecting data from all working equipment within the environmental protection island; The application layer includes a flue gas control module, which establishes a flue gas control prediction model based on data collected by the perception layer and historical data records. The decision-making layer determines the optimal collaborative strategy for the environmental protection island based on the models and data established by the application layer.

2. The intelligent environmental protection island platform management system for coal-fired power plants according to claim 1, characterized in that: The dust control prediction model includes a desulfurization prediction model, a denitrification prediction model, and a dust removal prediction model.

3. The intelligent environmental protection island platform management system for coal-fired power plants according to claim 2, characterized in that: The data collected by the sensing layer includes desulfurization system data, which includes boiler load, flue gas flow rate, inlet flue gas SO2 concentration, outlet flue gas SO2 concentration, limestone slurry pH value, calcium flow ratio, absorption tower spray volume, and inlet flue gas PM concentration.

4. The intelligent environmental protection island platform management system for coal-fired power plants according to claim 3, characterized in that: The process of establishing the desulfurization optimization model is as follows. Based on the desulfurization system data acquired by the sensing layer, an MA-RELM prediction model is constructed. Specific steps include: S31, set the number of iterations n for the mayfly algorithm, the population size m, the range of individual population variation, and related parameters; S32, Initialize the population; S33, update the population individuals according to the algorithm principle, and put the individuals into the RELM model, and train using the training set; S34, the validation set is fed into the trained model for prediction, and the prediction error is used as the fitness function; S35, update the mayfly algorithms pbest and gbest, and use male and female individuals to mate and produce two offspring; S36, calculate the fitness value of the offspring and update pbest and gbes; S37, the iteration termination condition, outputs the best individual in the population, and uses the best value to train the RELM model; S38. Input the test set into the trained RELM prediction model to obtain the final prediction result.

5. The intelligent environmental protection island platform management system for coal-fired power plants according to claim 2, characterized in that: The data collected by the sensing layer includes desulfurization system data, and the denitrification data includes boiler load, flue gas flow rate, and inlet flue gas NO. x Concentration, NO in flue gas x Concentration, reaction temperature, ammonia-nitrogen molar ratio, and oxygen content of inlet flue gas.

6. The intelligent environmental protection island platform management system for coal-fired power plants according to claim 5, characterized in that: The denitrification prediction model is established by constructing an MA-RELM prediction model based on the denitrification system data obtained by the sensing layer.

7. The intelligent environmental protection island platform management system for coal-fired power plants according to claim 2, characterized in that: The data collected by the sensing layer includes dust removal system data, which includes boiler load, flue gas flow rate, dust removal electric field voltage / current value, and dust removal chamber temperature.

8. The intelligent environmental protection island platform management system for coal-fired power plants according to claim 7, characterized in that: The dust removal prediction model establishment process involves constructing an MA-RELM prediction model based on the dust removal system data obtained by the sensing layer.

9. The intelligent environmental protection island platform management system for coal-fired power plants according to claim 2, characterized in that: The decision-making layer, based on desulfurization prediction models, denitrification prediction models, and dust removal prediction models, uses adjustment parameters as input parameters and the lowest cost as the objective, and determines the optimal control strategy through a particle swarm optimization algorithm under constraints.

10. The intelligent environmental protection island platform management system for coal-fired power plants according to claim 9, characterized in that: The controllable parameters include the ammonia injection rate, the electric field voltage of the electrostatic precipitator, the pH value of the limestone slurry, and the number of circulating pumps in operation. The constraints of the particle swarm optimization algorithm are the range constraints of the adjustment parameters and the pollutant emission limit constraints.

Images