Intelligent cruise control method for state transition peak regulation of thermal power generating unit based on intelligent control algorithm

Abstract

The invention discloses a thermal power generating unit transition peak regulation intelligent cruise control method based on an intelligent control algorithm, and relates to the technical field of intelligent control algorithms, and the method comprises the steps: constructing a distributed sensor network through deploying distributed sensors at all key positions of a thermal power generating unit, and obtaining the operation parameters of the thermal power generating unit in real time; external data are introduced and integrated, and short-term load demand prediction is generated according to an intelligent control algorithm; constructing a dynamic optimization function by combining a power grid dispatching instruction and a load prediction result and by taking the minimum coal consumption, the shortest peak regulation response time and the minimum equipment life loss as targets; adopting a hybrid algorithm of model predictive control and reinforcement learning to obtain an optimal control instruction; the control instruction is transmitted to a boiler combustion system and a steam turbine speed regulating system, and the fuel amount, the water supply amount and the valve opening degree are regulated; and dynamically adjusting model parameters and fuzzy rule weights of model prediction control through reinforcement learning to realize adaptive control of the thermal power generating unit.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control algorithms, and in particular to an intelligent cruise control method for peak-shaving of thermal power units based on an intelligent control algorithm. Background Art

[0002] As the proportion of renewable energy generation in power systems continues to increase, load fluctuations in power systems are becoming increasingly frequent and severe. As a crucial component of power systems, thermal power units must possess rapid and flexible peak-shaving capabilities to adapt to these load fluctuations and ensure their safe and stable operation. However, traditional peak-shaving control methods for thermal power units suffer from slow response, low control accuracy, and high energy consumption, making them unable to meet the peak-shaving requirements of modern power systems.

[0003] Traditional peak-shaving control for thermal power plants typically employs control strategies based on experience and fixed parameters. These strategies are unable to adapt to the unit's real-time operating status and changes in external load demand. When the unit's operating conditions change, traditional control methods are prone to control lag and overshoot, resulting in poor load tracking performance and impacting the unit's stability and economic efficiency during peak-shaving. Furthermore, traditional control methods often struggle to achieve ideal control results when dealing with multivariable, strongly coupled, and nonlinear thermal power plant systems. Summary of the Invention

[0004] In order to solve the above technical problems, an intelligent cruise control method for peak-shaving of thermal power units based on an intelligent control algorithm is provided. This technical solution solves the above-mentioned limitations of current real-time data acquisition. The control strategy based on experience and fixed parameters is unable to make adaptive adjustments according to the real-time operating status of the unit and changes in external load demand. When the operating conditions of the unit change, traditional control methods are prone to control lag and overshoot, resulting in poor load tracking performance of the unit, and affecting the stability and economy of the unit during the peak-shaving process.

[0005] In order to achieve the above objects, the technical solution adopted by the present invention is:

[0006] The intelligent cruise control method for peak load regulation of thermal power units based on intelligent control algorithms includes:

[0007] By deploying various types of distributed sensors at key locations of thermal power units, a distributed sensor network is formed to obtain the operating parameters of thermal power units in real time;

[0008] Based on the collected operating parameters, external data is introduced and integrated to generate short-term load demand forecasts according to intelligent control algorithms;

[0009] Combining grid dispatch instructions and load forecast results, a dynamic optimization function is constructed with the goal of minimizing coal consumption, shortest peak-shaving response time, and minimum equipment life loss.

[0010] Based on the dynamic optimization function, a hybrid algorithm of model predictive control and reinforcement learning is used to obtain the optimal control instructions;

[0011] Transmit control instructions to the boiler combustion system and turbine speed control system to adjust the fuel quantity, water supply and valve opening;

[0012] Based on the adjustment of control instructions, the control effect is monitored in real time, and the model parameters and fuzzy rule weights of model predictive control are dynamically adjusted through reinforcement learning to achieve adaptive control of thermal power units.

[0013] Preferably, the sensor nodes and the data acquisition unit are connected via a communication network;

[0014] Sensor nodes: They are the basic units of distributed sensor networks. Various types of distributed sensor nodes are deployed at key locations in thermal power units.

[0015] Data acquisition unit: collects data from each sensor node and performs preliminary processing and caching.

[0016] Preferably, the external data mainly includes meteorological data, date type and seasonal factors;

[0017] Select features that have a significant impact on load demand forecasting from the above operating parameters and external data, and use correlation analysis methods to screen out features with high correlation with load demand;

[0018] According to the above operating parameters, external data and load demand characteristics, a long short-term memory network neural network model is constructed;

[0019] The long short-term memory neural network model is trained using historical data, and the real-time collected operating parameters and external data are input into the trained long short-term memory neural network model. The model predicts the grid load demand in the short term based on the learned patterns and rules.

[0020] Preferably, the calculation formula of the Pearson correlation coefficient is:

[0021]

[0022] ρzc is the linear correlation between the characteristic and the load demand, n is the number of samples, z v and c v are the vth observation values ​​of feature z and load demand c, respectively, and are the sample means of characteristic z and load demand c, respectively. The value of ρzc ranges from -1 to 1. The closer the absolute value is to 1, the stronger the linear correlation between the two variables; the closer it is to 0, the weaker the linear correlation.

[0023] Preferably, in order to balance the three objectives of minimum coal consumption, shortest peak load response time, and minimum equipment life loss, a weighted summation method is used to construct a dynamic optimization function, and the calculation formula is:

[0024]

[0025] Where J is the value of the dynamic optimization function, w1, w2, and w3 are the weight coefficients of the three objectives of coal consumption, peak load response time, and equipment life loss, respectively, and w1+w2+w3=1, w1≥0, w2≥0, w3≥0; p d is the target load required by the grid dispatch instruction, p(t) is the actual load of the unit at time t, B is the coal consumption, L is the equipment life loss index, and T is the set time period.

[0026] Preferably, according to the operating characteristics of the thermal power unit, a state space model is established to describe the dynamic behavior of the unit, and the state equation of the system can be expressed as:

[0027] x(k+1)=f(x(k),u(k)

[0028] Where f is a nonlinear function that can be estimated by system identification method based on the historical operating data of the unit, x(k) is the state vector of the system at time k, and u(k) is the control vector at time k;

[0029] Based on the state space model, predict the future N p System status at a moment:

[0030] x(k+i|k)=f(x(k+i-1|k),u(x(k+i-1|k),i=1,2,…,N p

[0031] where x(k+i|k) represents the system state at time k+i predicted at time k, and u(x(k+i|k) is the predicted control input;

[0032] The previously constructed dynamic optimization function J is used as the objective function of the model predictive control, considering the future N p Cumulative optimization effect at each moment:

[0033]

[0034] F is expressed as the dynamic optimization function, J is the objective function of model predictive control, B is the coal consumption, p dis the target load required by the grid dispatch instruction, p is the actual power generation power of the unit, and L is the equipment life loss index.

[0035] Preferably, the calculation formula of the deep Q-network of the reinforcement learning algorithm is:

[0036]

[0037] Where y is the target Q value, s(h) represents the state of the agent at time h; a′ is any action in the action space at time h+1; r(h) refers to the reward value obtained by the agent after performing the action at time h; γ is the discount factor; θ - are the parameters of the target network, Q(s(h+1),a′;θ - ) is the Q-value estimate of the target network for action a′ in state s(h+1);

[0038] Use mean squared error as the loss function to update the parameters of the Q-network:

[0039]

[0040] L(θ) maintains the loss function, which is used to measure the difference between the Q value currently estimated by the Q-network and the target Q value, s j is the state in the jth sample, a j is the state s of the agent in the jth sample j The action taken, M is the number of samples sampled from the experience replay buffer, (s j ,a j ,r j ,s j+1 ) is the jth sample.

[0041] Preferably, the control instructions are transmitted from the control system to the boiler combustion system and the steam turbine speed control system via a communication network;

[0042] Feedback regulation: During the regulation process, the operating parameters of the boiler combustion system and the steam turbine speed control system must be monitored in real time to provide accurate feedback information to the control system;

[0043] The control system adjusts the control instructions according to the feedback information to form a closed-loop control, continuously optimize the control effect, and improve the stability and reliability of the system.

[0044] Preferably, the real-time monitoring of the control effect based on the adjustment of the control instruction specifically includes:

[0045] In each control cycle, the monitoring system collects the operating data of the thermal power unit and calculates the monitoring indicators of the control effect. The monitoring indicators include unit performance indicators, equipment operating status indicators, and environmental indicators.

[0046] Preferably, the agent selects an action based on the current state and adjusts the model parameters and fuzzy rule weights of the model predictive control;

[0047] Model predictive control generates new control instructions based on the adjusted model parameters and fuzzy rules, and sends them to the boiler combustion system and turbine speed control system to adjust the fuel quantity, feed water quantity and valve opening;

[0048] Repeating the above steps forms a closed-loop adaptive control process, which enables the thermal power unit to maintain good performance and stability under different operating conditions.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention proposes an intelligent cruise control method for peak-shaving of thermal power units based on an intelligent control algorithm. Through a distributed sensor network and multivariate data integration, it realizes all-round collection of operating parameters and accurate load forecasting, and constructs an optimization function with coal consumption, response time and equipment life as comprehensive objectives. It breaks through the limitations of traditional focus on a single objective, taking into account economy, stability and sustainability. The hybrid algorithm of model prediction and reinforcement learning allows the control strategy to be adjusted in real time according to the complex operating conditions of the unit, monitors the control effect in real time, and dynamically adjusts the model parameters and fuzzy rule weights through reinforcement learning, realizing adaptive control and timely optimization control. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of the intelligent cruise control method for peak load regulation of thermal power units based on intelligent control algorithms;

[0052] Figure 2 A flow chart of a method for obtaining operating parameters of a thermal power unit;

[0053] Figure 3 A flow chart of a method for generating short-term load demand forecasts;

[0054] Figure 4 A flow chart of a control instruction transmission adjustment method;

[0055] Figure 5 Flowchart of the adaptive control method for thermal power units. DETAILED DESCRIPTION

[0056] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0057] Reference Figure 1 As shown in FIG, a method for intelligent cruise control of thermal power unit transition peak load regulation based on an intelligent control algorithm includes:

[0058] By deploying various types of distributed sensors at key locations of thermal power units, a distributed sensor network is formed to obtain the operating parameters of thermal power units in real time;

[0059] Based on the collected operating parameters, external data is introduced and integrated to generate short-term load demand forecasts according to intelligent control algorithms;

[0060] Combining grid dispatch instructions and load forecast results, a dynamic optimization function is constructed with the goal of minimizing coal consumption, shortest peak-shaving response time, and minimum equipment life loss.

[0061] Based on the dynamic optimization function, a hybrid algorithm of model predictive control and reinforcement learning is used to obtain the optimal control instructions;

[0062] Transmit control instructions to the boiler combustion system and turbine speed control system to adjust the fuel quantity, water supply and valve opening;

[0063] Based on the adjustment of control instructions, the control effect is monitored in real time, and the model parameters and fuzzy rule weights of model predictive control are dynamically adjusted through reinforcement learning to achieve adaptive control of thermal power units.

[0064] This solution utilizes a distributed sensor network to obtain the operating parameters of thermal power units in real time, introduces and integrates external data, comprehensively considers the operating parameters, external data and load demand characteristics, generates short-term load demand forecasts based on intelligent control algorithms, and adaptively adjusts the real-time operating status and changes in external load demand. By monitoring the control effect in real time, reinforcement learning is used to dynamically adjust the model parameters and fuzzy rule weights of model predictive control to achieve adaptive control of thermal power units, promote the load tracking performance of the units, and improve the stability and economy of the units during peak regulation.

[0065] Reference Figure 2 As shown in the figure, various types of distributed sensors are deployed at key locations of thermal power units to form a distributed sensor network. The real-time operation parameters of the thermal power units are obtained, including:

[0066] The sensor nodes and data acquisition units are connected via a communication network;

[0067] Sensor nodes: They are the basic units of distributed sensor networks. Various types of distributed sensor nodes are deployed at key locations in thermal power units.

[0068] Data acquisition unit: collects data from each sensor node and performs preliminary processing and caching.

[0069] It should be noted that various types of distributed sensor nodes are deployed at various key locations of the thermal power unit, including but not limited to temperature sensors, pressure sensors, speed sensors, vibration sensors, flow sensors, voltage sensors, liquid level sensors, gas composition sensors, and displacement sensors.

[0070] Reference Figure 3 As shown in the figure, based on the collected operating parameters, external data is introduced and integrated, and a short-term load demand forecast is generated according to the intelligent control algorithm. Specifically, the following are included:

[0071] External data mainly include meteorological data, date types and seasonal factors;

[0072] Select features that have a significant impact on load demand forecasting from the above operating parameters and external data, and use correlation analysis methods to screen out features with high correlation with load demand;

[0073] According to the above operating parameters, external data and load demand characteristics, a long short-term memory network neural network model is constructed;

[0074] The long short-term memory neural network model is trained using historical data, and the real-time collected operating parameters and external data are input into the trained long short-term memory neural network model. The model predicts the grid load demand in the short term based on the learned patterns and rules.

[0075] This solution introduces and integrates external data. Based on the above operating parameters, external data and load demand characteristics, a long short-term memory neural network model is constructed to adaptively adjust the real-time operating status and changes in external load demand, thereby improving the accuracy of future power grid load demand forecasts in the short term.

[0076] Reference Figure 4 As shown, the control instructions are transmitted to the boiler combustion system and the turbine speed control system to adjust the fuel quantity, water supply quantity and valve opening. Specifically,

[0077] The control instructions are transmitted from the control system to the boiler combustion system and the turbine speed control system through the communication network;

[0078] Feedback regulation: During the regulation process, the operating parameters of the boiler combustion system and the steam turbine speed control system must be monitored in real time to provide accurate feedback information to the control system;

[0079] The control system adjusts the control instructions according to the feedback information to form a closed-loop control, continuously optimize the control effect, and improve the stability and reliability of the system.

[0080] This solution uses feedback regulation to provide accurate feedback information to the control system. The control system adjusts the control instructions based on the feedback information. The two form a closed-loop control to continuously optimize the control effect, greatly improve the stability and reliability of the control system, and promote intelligent cruise control for transition and peak regulation of thermal power units.

[0081] Reference Figure 5 As shown in the figure, based on the adjustment of control instructions, the control effect is monitored in real time, and the model parameters and fuzzy rule weights of the model predictive control are dynamically adjusted through reinforcement learning to achieve adaptive control of the thermal power unit. Specifically, the following are included:

[0082] In each control cycle, the monitoring system collects the operating data of the thermal power unit and calculates the monitoring indicators of the control effect. The monitoring indicators include unit performance indicators, equipment operating status indicators, and environmental indicators.

[0083] The agent selects actions based on the current state and adjusts the model parameters and fuzzy rule weights of the model predictive control;

[0084] Model predictive control generates new control instructions based on the adjusted model parameters and fuzzy rules, and sends them to the boiler combustion system and turbine speed control system to adjust the fuel quantity, feed water quantity and valve opening;

[0085] Repeating the above steps forms a closed-loop adaptive control process, which enables the thermal power unit to maintain good performance and stability under different operating conditions.

[0086] It should be noted that the monitoring system monitors the operating data of the thermal power unit in real time and determines the operating status of the thermal power unit by calculating the monitoring indicators of the control effect; it uses the intelligent agent to select actions according to the current status to complete the reinforcement learning dynamic adjustment of the model parameters and fuzzy rule weights of the model predictive control.

[0087] In summary, the advantages of the present invention are: through the integration of distributed sensor networks and multivariate data, all-round collection of operating parameters and accurate load forecasting are achieved, and an optimization function is constructed with coal consumption, response time and equipment life as comprehensive objectives, breaking through the limitations of traditional focus on a single objective, taking into account economy, stability and sustainability, and a hybrid algorithm of model prediction and reinforcement learning, so that the control strategy can be adjusted in real time according to the complex operating conditions of the unit, the control effect can be monitored in real time, and the model parameters and fuzzy rule weights can be dynamically adjusted through reinforcement learning to achieve adaptive control and timely optimization control.

[0088] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent cruise control of thermal power units for peak load regulation based on intelligent control algorithms, characterized in that: include: By deploying various types of distributed sensors at key locations of thermal power units, a distributed sensor network is formed to obtain the operating parameters of thermal power units in real time; Based on the collected operating parameters, external data is introduced and integrated to generate short-term load demand forecasts according to intelligent control algorithms; Combining grid dispatch instructions and load forecast results, a dynamic optimization function is constructed with the goal of minimizing coal consumption, shortest peak-shaving response time, and minimum equipment life loss. Based on the dynamic optimization function, a hybrid algorithm of model predictive control and reinforcement learning is used to obtain the optimal control instructions; Transmit control instructions to the boiler combustion system and turbine speed control system to adjust the fuel quantity, water supply and valve opening; Based on the adjustment of control instructions, the control effect is monitored in real time, and the model parameters and fuzzy rule weights of model predictive control are dynamically adjusted through reinforcement learning to achieve adaptive control of thermal power units.

2. The method for intelligent cruise control of thermal power unit transition peak load regulation based on intelligent control algorithm according to claim 1 is characterized in that: The distributed sensor network is formed by deploying various types of distributed sensors at various key locations of the thermal power unit to obtain the operating parameters of the thermal power unit in real time. Specifically, the following are performed: The sensor nodes and data acquisition units are connected via a communication network; Sensor nodes: They are the basic units of distributed sensor networks. Various types of distributed sensor nodes are deployed at key locations in thermal power units. Data acquisition unit: collects data from each sensor node and performs preliminary processing and caching.

3. The method for intelligent cruise control of thermal power unit transition peak load regulation based on intelligent control algorithm according to claim 2 is characterized in that: The process of introducing and integrating external data based on the collected operating parameters and generating a short-term load demand forecast according to the intelligent control algorithm specifically includes: External data mainly include meteorological data, date types and seasonal factors; Select features that have a significant impact on load demand forecasting from the above operating parameters and external data, and use correlation analysis methods to screen out features with high correlation with load demand; According to the above operating parameters, external data and load demand characteristics, a long short-term memory network neural network model is constructed; The long short-term memory neural network model is trained using historical data, and the real-time collected operating parameters and external data are input into the trained long short-term memory neural network model. The model predicts the grid load demand in the short term based on the learned patterns and rules.

4. The method for intelligent cruise control of thermal power unit transition peak load regulation based on intelligent control algorithm according to claim 3 is characterized in that: The features that have a significant impact on load demand prediction are selected from the above operating parameters and external data, and the features with a high correlation with load demand are screened using a correlation analysis method, specifically including: The calculation formula of Pearson correlation coefficient is: ρzc is the linear correlation between the characteristic and the load demand, n is the number of samples, z v and c v are the vth observation values ​​of feature z and load demand c, respectively, and are the sample means of characteristic z and load demand c, respectively. The value of ρzc ranges from -1 to 1. The closer the absolute value is to 1, the stronger the linear correlation between the two variables; the closer it is to 0, the weaker the linear correlation.

5. The method for intelligent cruise control of thermal power unit transition peak load regulation based on intelligent control algorithm according to claim 4 is characterized in that: The above-mentioned construction of a dynamic optimization function by combining the grid dispatching instructions and the load forecast results with the goal of minimizing coal consumption, minimizing peak load response time, and minimizing equipment life loss specifically includes: In order to balance the three goals of minimum coal consumption, shortest peak load response time, and minimum equipment life loss, a weighted summation method is used to construct a dynamic optimization function. The calculation formula is: Where J is the value of the dynamic optimization function, w1, w2, and w3 are the weight coefficients of the three objectives of coal consumption, peak load response time, and equipment life loss, respectively, and w1+w2+w3=1, w1≥0, w2≥0, w3≥0; p d is the target load required by the grid dispatch instruction, p(t) is the actual load of the unit at time t, B is the coal consumption, L is the equipment life loss index, and T is the set time period.

6. The method for intelligent cruise control of thermal power unit transition peak load regulation based on intelligent control algorithm according to claim 5 is characterized in that: The method of obtaining the optimal control instructions based on the dynamic optimization function and adopting a hybrid algorithm of model predictive control and reinforcement learning specifically includes: According to the operating characteristics of the thermal power unit, a state space model is established to describe the dynamic behavior of the unit. The state equation of the system can be expressed as: x(k+1)=f(x(k),u(k) Where f is a nonlinear function that can be estimated by system identification method based on the historical operating data of the unit, x(k) is the state vector of the system at time k, and u(k) is the control vector at time k; Based on the state space model, predict the future N p System status at a moment: x(k+i|k)=f(x(k+i-1|k),u(x(k+i-1|k),i=1,2,…,N p where x(k+i|k) represents the system state at time k+i predicted at time k, and u(x(k+i|k) is the predicted control input; The previously constructed dynamic optimization function J is used as the objective function of the model predictive control, considering the future N p Cumulative optimization effect at each moment: F is expressed as the dynamic optimization function, J is the objective function of model predictive control, B is the coal consumption, p d is the target load required by the grid dispatch instruction, p is the actual power generation power of the unit, and L is the equipment life loss index.

7. The method for intelligent cruise control of thermal power unit transition peak load regulation based on intelligent control algorithm according to claim 5 is characterized in that: The method of obtaining the optimal control instructions based on the dynamic optimization function and adopting a hybrid algorithm of model predictive control and reinforcement learning specifically includes: The calculation formula of the deep Q-network of the reinforcement learning algorithm is: Among them, y is the target Q value, s(h) represents the state of the agent at time h. The agent is based on the current state; a ′ is any action in the action space at time h+1; r(h) refers to the reward value obtained by the agent after performing the action at time h; γ is the discount factor; θ - are the parameters of the target network, Q(s(h+1),a ′ θ - ) is the target network in state s(h+1), for action a ′ Q value estimation; Use mean squared error as the loss function to update the parameters of the Q-network: L(θ) maintains the loss function, which is used to measure the difference between the Q value currently estimated by the Q-network and the target Q value, s j is the state in the jth sample, a j is the state s of the agent in the jth sample j The action taken, M is the number of samples sampled from the experience replay buffer, (s j ,a j ,r j ,s j+1 ) is the jth sample.

8. The method for intelligent cruise control of thermal power unit transition peak load regulation based on intelligent control algorithm according to claim 7 is characterized in that: The transmission of control instructions to the boiler combustion system and the steam turbine speed control system to adjust the fuel quantity, water supply quantity and valve opening specifically includes: The control instructions are transmitted from the control system to the boiler combustion system and the turbine speed control system through the communication network; Feedback regulation: During the regulation process, the operating parameters of the boiler combustion system and the steam turbine speed control system must be monitored in real time to provide accurate feedback information to the control system; The control system adjusts the control instructions according to the feedback information to form a closed-loop control, continuously optimize the control effect, and improve the stability and reliability of the system.

9. The method for intelligent cruise control of thermal power unit transition peak load regulation based on intelligent control algorithm according to claim 8, characterized in that: The adjustment based on the control instruction and the real-time monitoring of the control effect specifically include: In each control cycle, the monitoring system collects the operating data of the thermal power unit and calculates the monitoring indicators of the control effect. The monitoring indicators include unit performance indicators, equipment operating status indicators, and environmental indicators.

10. The method for intelligent cruise control of thermal power unit transition peak load regulation based on intelligent control algorithm according to claim 9, characterized in that: Dynamically adjusting the model parameters and fuzzy rule weights of the model predictive control through reinforcement learning to achieve adaptive control of the thermal power unit specifically includes: The agent selects actions based on the current state and adjusts the model parameters and fuzzy rule weights of the model predictive control; Model predictive control generates new control instructions based on the adjusted model parameters and fuzzy rules, and sends them to the boiler combustion system and turbine speed control system to adjust the fuel quantity, feed water quantity and valve opening; Repeating the above steps forms a closed-loop adaptive control process, which enables the thermal power unit to maintain good performance and stability under different operating conditions.

Images