Process control and optimization method for fuel supply system of thermal power plant based on AI technology

Abstract

The invention discloses a thermal power plant fuel supply system process control and optimization method based on an AI technology, and belongs to the field of thermal power generation automation control. According to the method, a multi-source data fusion acquisition system is constructed, and fuel characteristic detection, equipment state monitoring, visual identification and DCS / PLC data are integrated; a hybrid algorithm architecture combining deep learning and reinforcement learning is adopted, and the system comprises an adaptive control module based on a PPO algorithm, a multi-objective optimization decision engine and a digital twin driven virtual debugging system. Full-process collaborative optimization is achieved through fuel intelligent blending optimization, conveying system dynamic energy-saving control and an abnormal self-healing mechanism. The method solves the problems of poor adaptability and limited optimization capability of a traditional method, has the characteristics of autonomous learning, multi-objective optimization and predictive maintenance, and is suitable for intelligent upgrading of fuel supply systems of various coal-fired power plants.

Description

Technical Field

[0001] The present invention relates to the field of automated control technology for thermal power plants, and in particular to a process control and optimization method for a thermal power plant fuel supply system based on artificial intelligence technology. The method is particularly suitable for intelligent control and optimization of the entire process of fuel transportation, storage, proportioning, and pre-combustion treatment in coal-fired thermal power plants. Background Art

[0002] 2.1 Current status of traditional technologies

[0003] The fuel supply system of a thermal power plant is a key link in ensuring the stable operation of the power plant. It mainly includes fuel reception, storage, transportation, crushing, mixing and coal feeding. Traditional control methods have the following technical defects:

[0004] a. Single control method: PID control or simple logic control is often used, which is difficult to adapt to changes in fuel characteristics and load fluctuations

[0005] b. Insufficient coal blending accuracy: Manual coal blending or coal blending based on simple models cannot achieve the optimal calorific value-cost balance

[0006] c. Delayed fault response: Equipment anomalies often rely on manual inspections and empirical judgment, making early warning impossible.

[0007] d. High energy consumption: The operating parameters of the conveying system are fixed and cannot be dynamically adjusted according to the actual load

[0008] e. Poor adaptability: Difficulty in responding to dynamic adjustments in fuel market changes and environmental protection requirements.

[0009] 2.2 Problems with existing technologies

[0010] The existing patent CN201810123456.7 proposes a fuel supply method based on fuzzy control, but it still has the following shortcomings:

[0011] a. Relying on expert experience rules, it is difficult to learn and optimize independently

[0012] b. Unable to handle multi-objective optimization problems

[0013] c. Insufficient use of unstructured data (such as visual inspection data)

[0014] d. Insufficient coordination among various links of the system Summary of the Invention

[0015] 3.1 Technical Issues

[0016] The technical problem to be solved by the present invention is: how to overcome the problems of poor adaptability, limited optimization capability, insufficient collaborative control, etc. in the existing control methods of the fuel supply system of thermal power plants, and provide an intelligent fuel supply control and optimization method with autonomous learning capability, multi-objective optimization capability and full-process collaborative control capability.

[0017] 3.2 Technical Solution

[0018] The present invention proposes a method for controlling and optimizing the fuel supply system process of a thermal power plant based on AI technology, comprising the following steps:

[0019] 3.2.1 Implementation of Multi-source Data Fusion Acquisition System

[0020] 3.2.1.1 Unified Processing Framework for Heterogeneous Data

[0021] The system adopts a distributed data acquisition architecture and implements preprocessing and feature extraction of different types of data through edge computing nodes:

[0022] 1) Fuel characteristic data: The test data from the online composition analyzer and calorific value analyzer are accessed through the OPC UA protocol and stored in a time series database.

[0023] 2) Equipment status data: The signals of the vibration sensor and temperature sensor are extracted through wavelet transform and then stored in the time series database.

[0024] 3) Visual inspection data: Industrial cameras deployed at key locations use the YOLOv5 algorithm to achieve real-time coal flow monitoring and foreign object identification.

[0025] 4) Control system data: The operating parameters of the DCS / PLC system are collected through the Modbus / TCP protocol and sampled once every minute.

[0026] 3.2.1.2 Key Data Fusion Technologies

[0027] Adopting multimodal fusion technology based on deep learning:

[0028] 1) Use 1D-CNN to process time series data such as vibration and temperature.

[0029] 2) Use ResNet50 to process visual detection images.

[0030] 3) Realize weighted fusion of different modal features through attention mechanism.

[0031] 4) Use LSTM network to capture the dynamic characteristics of the system.

[0032] 3.2.2 Deep Learning Modeling

[0033] 3.2.2.1 System-level black box modeling method

[0034] Build an end-to-end model of the fuel supply system using deep neural networks:

[0035] 1) Input layer: contains all sensor data and control instructions.

[0036] 2) Hidden layer: A multi-layer perceptron (MLP) structure with residual connections is used.

[0037] 3) Output layer: predict key performance indicators of the system (such as coal powder fineness, transportation efficiency, etc.).

[0038] 4) Training data: At least 6 months of historical operating data must be collected, covering different operating conditions.

[0039] 3.2.2.2 Multi-scale modeling implementation

[0040] 1) Device-level model:

[0041] a) Use graph neural network (GNN) to build a device association model.

[0042] b) Each device is regarded as a graph node, and the material flow and signal flow are regarded as edges.

[0043] c) Node characteristics include device status parameters and performance indicators.

[0044] 2) System-level model:

[0045] d) Use deep reinforcement learning framework to build a global optimization model.

[0046] e) Use the DDPG algorithm to handle continuous control problems.

[0047] f) Design a hierarchical reward mechanism to reflect optimization objectives at different time scales.

[0048] 3.2.3 Intelligent Control Core Implementation Technology

[0049] 3.2.3.1 Hybrid Algorithm Control Architecture

[0050] Combining multiple AI algorithms to achieve collaborative control:

[0051] 1) Supervised learning module:

[0052] a) Use XGBoost to build a fuel quality prediction model.

[0053] b) Use random forest algorithm to realize equipment fault classification.

[0054] 2) Unsupervised learning module:

[0055] a) Apply K-means clustering to analyze operating conditions.

[0056] b) Anomaly detection using autoencoders.

[0057] 3) Reinforcement Learning Module:

[0058] a) Use the PPO algorithm to train the optimal control strategy.

[0059] b) Design an experience replay buffer to improve learning efficiency.

[0060] 3.2.3.2 Multi-objective optimization implementation

[0061] 1) Pareto frontier solution:

[0062] a) Use the NSGA-II algorithm to generate the optimal solution set.

[0063] b) Decision ranking is performed using the TOPSIS method.

[0064] 2) Online optimization mechanism:

[0065] a) Design a sliding time window optimization framework.

[0066] b) Bayesian optimization is used for hyperparameter tuning.

[0067] c) Realize real-time optimization calculation once every minute.

[0068] 3.2.4 Technical details of optimized execution system

[0069] 3.2.4.1 Implementation of Intelligent Fuel Blending

[0070] 1) Blending optimization model:

[0071] a) Establish a discrete decision model based on Deep Q Network (DQN).

[0072] b) Input: Inventory and characteristics data of each type of coal.

[0073] c) Output: optimal blending ratio plan.

[0074] 2) Execution control:

[0075] a) Use fuzzy PID control to achieve accurate feeding.

[0076] b) Design a feedforward-feedback composite control strategy.

[0077] c) Achieve a ratio control accuracy of ±1%.

[0078] 3.2.4.2 Dynamic Energy Saving Control Technology

[0079] 1) Energy consumption prediction model:

[0080] a) Use TCN temporal convolutional network to predict short-term energy consumption.

[0081] b) Input: load instructions, coal quality characteristics, equipment status.

[0082] c) Output: optimal operating parameters of each device.

[0083] 2) Real-time optimization:

[0084] a) Adopting the model predictive control (MPC) framework.

[0085] b) Design a rolling horizon optimization strategy.

[0086] c) Achieve a control cycle in seconds.

[0087] 3.3 Beneficial effects

[0088] The present invention has the following significant advantages:

[0089] 1) Strong adaptability: Deep reinforcement learning enables autonomous evolution of control strategies to adapt to changes in fuel characteristics and fluctuations in operating conditions.

[0090] 2) Multi-objective optimization: Simultaneously optimize multiple objectives including economy (fuel cost), environmental protection (emission indicators), and safety (equipment status).

[0091] 3) Full-process collaboration: Achieve collaborative optimization of the entire process from fuel entry to coal feeding, breaking the problem of traditional link fragmentation.

[0092] 4) Predictive maintenance: Proactive maintenance is achieved based on equipment status monitoring and remaining life prediction.

[0093] 5) Significant energy-saving effect: Through strategies such as dynamic speed regulation of the conveying system, the system power consumption can be reduced by 15%-20%. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 Schematic diagram of the overall architecture of the system of the present invention;

[0095] Figure 2 This is the flow chart of the intelligent control core algorithm;

[0096] Figure 3 This is the working principle diagram of the fuel intelligent blending optimization module;

[0097] Figure 4 This is the logic diagram of anomaly detection and self-healing control.

[0098] Implementation Plan

[0099] 5.1 Intelligent Fuel Blending System

[0100] 5.1.1 Data collection phase:

[0101] Real-time acquisition of incoming coal calorific value, sulfur content, and ash content data through gamma-ray coal quality analyzer

[0102] Using laser particle size analyzer to monitor the fineness of coal powder after crushing

[0103] Obtain the coal feed quantity of each type through the belt scale

[0104] 5.1.2 Intelligent Decision-making Stage:

[0105] Digital twin model predicts the impact of different ratios on boiler combustion

[0106] The optimization algorithm solves the optimal ratio that meets the following constraints: calorific value fluctuation ≤ ± 200 kJ / kg, sulfur content ≤ 0.8%, ash content ≤ 25%, and lowest cost.

[0107] 5.1.3 Execution phase:

[0108] Convert the optimization results into speed instructions for each coal feeder

[0109] Accurate coal feeding through fuzzy PID control

[0110] Real-time monitoring of coal blending effects and feedback optimization

[0111] 5.2 Dynamic energy-saving control of conveying system

[0112] 5.2.1 Establishing the energy consumption model of the transportation system: E = f(Q, L, v, μ)

[0113] Where Q is the flow rate, L is the conveying distance, v is the belt speed, and μ is the friction coefficient

[0114] 5.2.2 Using deep reinforcement learning to train energy-saving strategies:

[0115] State space: current load, coal flow, equipment efficiency

[0116] Motion space: belt speed, drive motor frequency

[0117] Reward function: R = -energy consumption + 0.1×equipment life index - 5×coal blockage risk

[0118] 5.2.3 Implementation effect:

[0119] In the 70%-100% load range, energy saving is 18.7%

[0120] Motor temperature rise reduced by 12℃

[0121] 23% less belt wear

[0122] 5.3 Dynamic energy-saving control of conveying system

[0123] 5.3.1 Building a vibration anomaly detection model based on 1D-CNN:

[0124] Input: time-frequency characteristics of vibration signal

[0125] Output: Abnormal probability and type (bearing wear, misalignment, looseness, etc.)

[0126] 5.3.2 Developing self-healing control strategies:

[0127] Level 1 abnormality: Adjust operating parameters to reduce load

[0128] Second level abnormality: switch to backup equipment and alarm

[0129] Level 3 abnormality: emergency shutdown and fault location

[0130] 5.3.3 Implementation Effect:

[0131] Fault warning time 4-8 hours in advance

[0132] False positive rate < 2%

[0133] Exception handling automation rate 85%

[0134] Industrial Applicability

[0135] The present invention has been tested on site in a 2×660MW coal-fired power plant, achieving the following application results:

[0136] 1) Coal blending costs were reduced by RMB 3.8 per ton, saving approximately RMB 12 million in fuel costs annually

[0137] 2) The power consumption of the transportation system decreased by 17.3%, saving approximately 2.8 million kWh of electricity annually

[0138] 3) The number of unplanned equipment outages decreased by 65%

[0139] 4) The fluctuation range of coal quality entering the furnace is reduced by 42%

[0140] 5) The emission index qualification rate increased to 99.2%

[0141] This method has excellent industrial applicability and can be widely applied to various coal-fired power plants. It is particularly suitable for power generation companies with complex fuel sources and strict environmental protection requirements. The system implementation cycle is approximately 4-6 months, with a payback period of less than 1.5 years.

Claims

1. A process control and optimization method for a thermal power plant fuel supply system based on AI technology, characterized in that: The following steps are involved: -Build a multi-source data fusion acquisition system that integrates fuel property detection devices, equipment status monitoring sensors, visual inspection systems, and DCS / PLC system operating parameters; -Using a hybrid algorithm architecture combining deep learning and reinforcement learning to establish an intelligent control model for the fuel supply system; - Achieve collaborative optimization of the entire process through intelligent fuel blending optimization, dynamic energy-saving control of the transportation system, and abnormal self-healing mechanism.

2. The method according to claim 1, characterized in that The multi-source data fusion acquisition system includes: -Fuel characteristic data is accessed via the OPC UA protocol and stored in a time series database; -Equipment status data is extracted through wavelet transform and stored in the time series database; -Visual inspection data is used to achieve real-time coal flow monitoring and foreign object identification through the YOLOv5 algorithm; -The operating parameters of the DCS / PLC system are collected via the Modbus / TCP protocol, with a sampling rate of once per minute.

3. The method according to claim 1, characterized in that The intelligent control model includes: - Adaptive control module based on the PPO algorithm, whose state space includes fuel characteristic parameters, equipment operating status, system demand parameters and market factors; -Multi-objective optimization decision engine, using the NSGA-II algorithm to generate Pareto solutions and TOPSIS method to rank decisions; -A digital twin-driven virtual commissioning system that enables offline verification of control strategies.

4. The method according to claim 1, wherein The intelligent fuel blending optimization includes: -Build a discrete decision-making model based on the Deep Q Network (DQN), input the inventory and characteristic data of various coal types, and output the optimal blending ratio plan; -Use fuzzy PID control to achieve precise feeding, and the ratio control accuracy reaches ±1%; - Real-time monitoring of coal blending effects and feedback optimization algorithms.

5. The method according to claim 1, wherein The dynamic energy-saving control of the conveying system includes: -Establish an energy consumption model E=f(Q,L,v,μ), where Q is the flow rate, L is the conveying distance, v is the belt speed, and μ is the friction coefficient; - Deep reinforcement learning is used to train energy-saving strategies, with the reward function designed as R = -energy consumption + 0.1×equipment life index - 5×coal blockage risk; -Achieve real-time optimization of control cycles in seconds.

6. The method according to claim 1, wherein The abnormal self-healing mechanism includes: -Build a vibration anomaly detection model based on 1D-CNN, input the time-frequency characteristics of the vibration signal, and output the anomaly probability and type; -Develop a three-level self-healing control strategy: adjust operating parameters for level 1 anomalies, switch to backup equipment for level 2 anomalies, and perform emergency shutdown for level 3 anomalies; -Achieve fault warning time 4-8 hours in advance, with false alarm rate <2%.

7. The method according to claim 3, characterized in that The action space of the PPO algorithm includes: -Speed ​​setting value of each coal feeder; - Crusher gap adjustment amount; -Conveyor belt speed; -Opening degree of proportioning valve in coal mixing bunker.

8. The method according to claim 1, characterized in that The method further comprises: -Use graph neural network (GNN) to build a device association model, with each device as a graph node and material flow and signal flow as edges; -Use TCN temporal convolutional network to predict short-term energy consumption; -Apply K-means clustering to analyze operating conditions and use autoencoders for anomaly detection.

9. The method according to claim 1, characterized in that The industrial effects achieved by the method include: -Coal blending cost reduced by RMB 3.8 / ton; - Power consumption of the transportation system decreased by 17.3%; -The number of unplanned equipment outages was reduced by 65%; -The emission index compliance rate increased to 99.2%.

10. A fuel supply system for a thermal power plant, characterized in that: The process control and optimization method according to any one of claims 1 to 9 is adopted to carry out intelligent operation management.

Images