Intelligent scheduling and optimizing method for thermal power plant

Abstract

The invention discloses an intelligent scheduling and optimization method for a thermal power plant, and the method comprises the steps: data collection: monitoring the temperature, pressure, current, voltage, flow and other data of a boiler in real time through a sensor, and carrying out the abnormal value removal and missing value filling of the transmission data; data storage: providing historical data and real-time data support for an optimization algorithm by using a time sequence database and a stream processing framework; a scheduling and optimization algorithm: adopting a mixed integer linear programming model and a deep reinforcement learning algorithm to realize power generation cost minimization and efficiency maximization; prediction and intelligent decision making: using load prediction, climate prediction and fuel price prediction models to improve the perspectiveness of scheduling; control and execution: based on a reinforcement learning control strategy, unit operation parameters are adjusted in real time; and monitoring and feedback: displaying a real-time operation state through a graphical interface, detecting abnormity by using a data mining algorithm, and dynamically adjusting a strategy. And the economical efficiency and the environmental protection property of thermal power generation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal power generation, and in particular to an intelligent scheduling and optimization method for a thermal power plant. Background Art

[0002] The scheduling and optimization of traditional thermal power plants mainly rely on manual experience and preset rules, which are difficult to dynamically adapt to real-time data changes, resulting in low power generation efficiency, fuel waste, and excessive pollutant emissions. With the increasing volatility of energy demand and increasingly stringent environmental protection requirements, the existing scheduling methods can no longer meet the dual needs of economy and environmental protection. In recent years, the development of artificial intelligence, optimization algorithms, and real-time monitoring technology has provided new ways for the intelligent scheduling of thermal power plants, but the existing solutions still have shortcomings in data collection, scheduling optimization, and feedback control. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent scheduling and optimization method for a thermal power plant.

[0004] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0005] The present invention comprises the following steps:

[0006] S1: Data collection: Real-time monitoring of the operating status of thermal power plants and collection of various data, including boiler temperature, pressure, current, voltage, and flow; preliminary processing of transmitted data, removal of outliers, and filling in missing data;

[0007] S2: Data storage: provides historical data storage and real-time data stream processing for optimization algorithms;

[0008] S3: Scheduling and optimization algorithm: Optimize power generation scheduling based on real-time data and historical data to minimize the cost of energy production and maximize power generation efficiency. Considering the requirements of environmental protection, a mixed integer linear programming model is used for power generation scheduling. The constraints include: equipment operation restrictions, economic constraints, environmental emission constraints, and load requirements. The optimization algorithm uses a deep reinforcement learning model to optimize the power generation scheduling strategy.

[0009] S4: Forecasting and intelligent decision-making: Use machine learning models to predict future loads, consider the impact of climate and temperature factors on electricity demand, use meteorological data forecasting models to jointly forecast weather and load, and use time series forecasting models to predict future fuel price fluctuations;

[0010] S5: Control and execution: According to the scheduling plan output by the optimization algorithm, the actual control operation is performed, and the unit operating parameters, including load, temperature, and flow, are adjusted in real time based on the reinforcement learning control strategy;

[0011] S6: Monitoring and feedback: The current load of the power plant, the operating status of each unit, energy consumption, and emission data are displayed through a graphical interface. Abnormal data is detected through data mining algorithms to promptly discover system failures or operating abnormalities, trigger alarms, monitor the operating status of the power plant in real time, and make dynamic adjustments based on feedback information.

[0012] The beneficial effects of the present invention are:

[0013] The present invention is an intelligent scheduling and optimization method for a thermal power plant. Compared with the prior art, the present invention improves power generation efficiency and economy, reduces fuel consumption costs, and dynamically adapts to changes in load demand to ensure the stability of power supply. It significantly reduces environmental pollution by optimizing and controlling emissions. It provides a real-time monitoring and feedback mechanism that can quickly respond to anomalies and make dynamic adjustments. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0016] like Figure 1 As shown: The present invention comprises the following steps:

[0017] S1: Data collection (real-time monitoring of the power plant operation status, data collection, and preliminary processing): real-time monitoring of the operation status of the thermal power plant, and collection of various data, including boiler temperature, pressure, current, voltage, flow; preliminary processing of the transmitted data, removal of outliers, and filling of missing data; the removal of outliers is detected by statistical methods. Suppose the data set is {x 1 , x 2 , ..., x n}, calculate the mean μ and standard deviation σ, and the outlier is determined as: |x i -μ|>kσ, where k is the outlier detection threshold; the missing data is filled using the following formula:

[0018]

[0019] The missing data is x m . Ensure that the input data is complete and unbiased, providing a reliable basis for subsequent analysis.

[0020] Data acquisition obtains the operating data of the power plant in real time through sensors and SCADA systems.

[0021] S2: Data storage (providing historical data storage and real-time data stream processing for optimization algorithms): providing historical data storage and real-time data stream processing for optimization algorithms; the historical data storage uses a time series database (such as InfluxDB) to store operation data, and the historical data D hist for:

[0022]

[0023] Where: T b is the boiler temperature, P b is the boiler pressure, I is the current, V is the voltage, and Q is the flow rate;

[0024] The real-time data stream processing uses a stream processing framework (such as Apache Kafka) to collect and process data streams in real time. i Stream data at time D stream for:

[0025] D stream (t i )=(T b (t i ), P b (t i ), I(t i ), V(t i ), Q(t i ))

[0026] Aggregate and analyze real-time data to provide input for optimization models. Achieve efficient data management and provide data support for optimization algorithms.

[0027] S3: Scheduling and optimization algorithm (optimizing power generation scheduling based on real-time data and historical data to minimize costs, maximize efficiency, and meet environmental protection constraints): Optimize power generation scheduling based on real-time data and historical data to minimize the cost of energy production and maximize power generation efficiency, and consider environmental protection requirements. A mixed integer linear programming model is used for power generation scheduling. Constraints include: equipment operation restrictions, economic constraints, environmental emission constraints, and load requirements. The optimization algorithm uses a deep reinforcement learning model to optimize the power generation scheduling strategy.

[0028] The mixed integer linear programming model is:

[0029]

[0030] Where: min C f To minimize fuel cost, N is the number of units, c iis the unit fuel cost of unit i, P i is the power generation power of unit i;

[0031] The constraints are: Unit start and stop constraints: The switch state of unit i at time t is u i (t),u i (t)∈{0,1}, where:

[0032]

[0033] Generation load balancing constraints:

[0034]

[0035] Where: P load (t) is the load demand at time t;

[0036] Emission constraint: Emission of the unit i (t) Meet environmental regulations:

[0037] E i (t) = α i ·P i (t)≤E max，i

[0038] Where: α i is the emission factor of unit i, E max，i is the emission limit of unit i;

[0039] The deep reinforcement learning model is:

[0040] State space S t : Contains information such as current load, unit status, temperature, pressure, etc.; Action space A t :Each action corresponds to adjusting the power generation or start / stop status of each unit; the reward function R t : The reward function is calculated comprehensively based on indicators such as fuel cost, emissions, and efficiency.

[0041] R t =-(C f (t)+λ·E total (t)+γ,ΔP load (t))

[0042] Among them, R t is the reward function, C f (t) is the fuel cost at time t; E total (t) is the total emission at time t; λ and γ are the weights of emission and load fluctuation, ΔP load (t) is the difference between time t and the predicted load;

[0043] Optimization update:

[0044]

[0045] Where: Q(S t , A t ) is the state-action value function; α is the learning rate, and A′ is the possible action at the next moment.

[0046] S4: Forecasting and intelligent decision-making (forecasting future load, climate change, fuel prices, etc., to provide support for scheduling decisions): Use machine learning models to predict future loads, consider the impact of climate and temperature factors on power demand, use meteorological data prediction models to jointly predict weather and load, and use time series prediction models to predict future fuel price fluctuations; the load prediction model is: let the load time series data be P load (t), the load forecasting model is trained through historical load data to predict future load P load (t+1), the calculation formula is:

[0047] P load (t+1)=f LSTM (P load (t), P load (t-1), ..., P load (tk))

[0048] Among them, f LTSM is the function of the load forecasting model, k is the time step number of the input historical data;

[0049] Climate and temperature forecasts predict future temperature T(t+1) through meteorological data, and use the ARIMA time series model for modeling:

[0050] T(t+1)=φ 0 +φ 1 T(t)+∈ t

[0051] Among them, φ 0 and φ 1 is the parameter of the ARIMA time series model, ∈ t is the noise term;

[0052] Fuel price forecasting uses the Prophet time series model to predict fuel prices:

[0053] P fuel (t+l)=f Prophet (P fuel (t), holiday, trend)

[0054] Among them, f ProphetIt is a Prophet time series model that combines holiday and trend information to predict fuel prices.

[0055] S5: Control and execution (according to the scheduling scheme output by the optimization algorithm, control operations are performed in real time to adjust the unit operating parameters): According to the scheduling scheme output by the optimization algorithm, actual control operations are performed, and based on the reinforcement learning control strategy, the unit operating parameters, including load, temperature, and flow, are adjusted in real time; the PID controller fine-tunes the operation of the unit:

[0056]

[0057] Among them, e(t) is the error, K p , K i , K d are the parameters of the PID controller.

[0058] S6: Monitoring and feedback (real-time monitoring of operating status, fault detection and alarm): Display the current load of the power plant, the operating status of each unit, energy consumption, and emission data information through a graphical interface, detect abnormal data through data mining algorithms (such as isolation forest, K-means), calculate abnormal value scores, and promptly discover system failures or operating abnormalities, trigger alarms, monitor the operating status of the power plant in real time, and make dynamic adjustments based on feedback information. The data mining algorithm is:

[0059]

[0060] Where: d i is the deviation between the data point and the model; σ i is the standard deviation of this point.

[0061] Real-time monitoring interface: Use visualization tools to display real-time data, such as load, temperature, emissions and other key indicators to ensure transparency of operating status. Based on feedback information, dynamically adjust the scheduling strategy to deal with emergencies.

[0062] Example:

[0063] The following is a practical application example of the intelligent dispatching and optimization method for a thermal power plant, which illustrates the functions and effects of the system in combination with specific data;

[0064] 1. Data collection and preprocessing: Real-time monitoring of multiple key parameters such as boiler temperature and pressure during operation of thermal power plants.

[0065] Operation: Collect operational data using sensor networks:

[0066] T b =520℃

[0067] P b=18MPa

[0068] I=300A

[0069] V=10kV

[0070] Q=120m 3 / s

[0071] Detect outliers: If the boiler temperature collected at a certain time is 650°C, which is far higher than the mean value of 520°C and the standard deviation σ=15°C, and satisfies the condition |650-520|>3·15, it will be marked as an anomaly.

[0072] Use linear interpolation to complete the missing flow value Q:

[0073]

[0074] 2. Data storage and management: Ensure that the dispatching system has sufficient historical and real-time data support.

[0075] Operation: Historical data is stored in a time series database. The example data structure is:

[0076] D hist ={(t 1 ,520,18,300,10,120),(t 2 , 518, 18.1, 305, 10.1, 119), ...}

[0077] Real-time data is stream processed through Apache Kafka. Example of structure:

[0078] D stream (t)=(T b =520,P b =18, I=300, V=10, Q=120)

[0079] 3. Dispatching and Optimization: There are N = 4 generating units in the thermal power plant, which need to meet P load (t) = 500MW demand, operation: The optimization goal is to minimize the fuel cost, assuming that the unit fuel cost of each unit is c 1 =50 yuan / MWh, c 2 =60 yuan / MWh, c 3 =55 yuan / MWh, C 4 =65 yuan / MWh.

[0080] constraint:

[0081] Load balance: P 1 +P 2 +P 3 +P 4=500MW

[0082] Emission Limitation: Assumption α i =0.9kg / MWh, the maximum emission limit is 450kg: 0.9(P 1 +P 2 +P 3 +P 4 )≤450

[0083] Calculation results:

[0084] The optimal allocation is P 1 =120MW, P 2 =130MW, P 3 =140MW, P 4 =110MW.

[0085] Total fuel cost: C f =50.120+60.130+55.140+65.110=30600 yuan

[0086] 4. Forecasting and intelligent decision-making: Predict future load demand and avoid problems of untimely scheduling.

[0087] Operation: Use the LSTM model to predict the load demand for the next 24 hours. The example prediction curve is:

[0088] P load (t+1)=f LSTMM (P load (t), ..., P load (tk))

[0089] Prediction results:

[0090] 08:00: P load =520MW

[0091] 09:00: P load =530MW

[0092] The Prophet model is used to predict fuel price fluctuations, and the results show that:

[0093] Fuel costs are expected to rise by 5% tomorrow.

[0094] 5. Control and execution: Adjust the unit operating status according to the optimization plan.

[0095] Operation: Adjust the unit load:

[0096] Unit 1: Increased from 110MW to 120MW.

[0097] Unit 2: Remains unchanged at 130MW.

[0098] PID controller fine-tuning parameters:

[0099]

[0100] Assume e(t) = 2MW, K p =1.5, the error is reduced to 0.2MW after adjustment.

[0101] 6. Monitoring and feedback: Monitor system status in real time and respond to exceptions promptly.

[0102] Operation: Graphical interface displays real-time operation data:

[0103] Current load P = 500MW

[0104] Current emission E = 450 kg

[0105] Abnormal detection: If the temperature of a unit rises abnormally, an alarm is triggered.

[0106] Data mining algorithm to identify outliers T b =650℃, the system automatically reduces the unit load to 90MW.

[0107] The technical solution of the present invention is not limited to the above-mentioned specific embodiments. All technical variations made according to the technical solution of the present invention fall within the protection scope of the present invention.

Claims

1. An intelligent dispatching and optimization method for a thermal power plant, characterized in that: The following steps are involved: S1: Data collection: Real-time monitoring of the operating status of thermal power plants and collection of various data, including boiler temperature, pressure, current, voltage, and flow; preliminary processing of transmitted data, removal of outliers, and filling in missing data; S2: Data storage: provides historical data storage and real-time data stream processing for optimization algorithms; S3: Scheduling and optimization algorithm: Optimize power generation scheduling based on real-time data and historical data to minimize the cost of energy production and maximize power generation efficiency. Considering the requirements of environmental protection, a mixed integer linear programming model is used for power generation scheduling. The constraints include: equipment operation restrictions, economic constraints, environmental emission constraints, and load requirements. The optimization algorithm uses a deep reinforcement learning model to optimize the power generation dispatch strategy; S4: Forecasting and intelligent decision-making: Use machine learning models to predict future loads, consider the impact of climate and temperature factors on electricity demand, use meteorological data forecasting models to jointly forecast weather and load, and use time series forecasting models to predict future fuel price fluctuations; S5: Control and execution: According to the scheduling plan output by the optimization algorithm, the actual control operation is performed, and the unit operating parameters, including load, temperature, and flow, are adjusted in real time based on the reinforcement learning control strategy; S6: Monitoring and feedback: The current load of the power plant, the operating status of each unit, energy consumption, and emission data are displayed through a graphical interface. Abnormal data is detected through data mining algorithms to promptly discover system failures or operating abnormalities, trigger alarms, monitor the operating status of the power plant in real time, and make dynamic adjustments based on feedback information.

2. The intelligent dispatching and optimization method of a thermal power plant according to claim 1, characterized in that: In step S1, outliers are removed by detecting outliers using statistical methods. Suppose the data set is {x1, x2, ..., x n }, calculate the mean μ and standard deviation σ, and the outlier is determined as: |x i -μ|>κσ, where κ is the outlier detection threshold; the following formula is used to fill in missing data: The missing data is x m .

3. The intelligent dispatching and optimization method of a thermal power plant according to claim 2, characterized in that: In step S2, the historical data storage uses a time series database to store the operation data. hist for: Where: T b is the boiler temperature, P b is the boiler pressure, I is the current, V is the voltage, and Q is the flow rate; The real-time data stream processing uses a stream processing framework to collect and process data streams in real time. i Stream data at time D stream for: D stream (t i )=(T b (t i ),P b (t i ),I(t i ),V(t i ),Q(t i )) Aggregate and analyze real-time data to provide input for optimization models.

4. The intelligent dispatching and optimization method of a thermal power plant according to claim 3 is characterized in that: The mixed integer linear programming model in step S3 is: Where: min C f To minimize fuel cost, N is the number of units, c i is the unit fuel cost of unit i, P i is the power generation power of unit i; The constraints are: Unit start and stop constraints: The switch state of unit i at time t is u i (t),u i (t)∈{0,1}, where: Generation load balancing constraints: Where: P load (t) is the load demand at time t; Emission constraint: Emission of the unit i (t) Meet environmental regulations: E i (t)=α i ·P i (t)≤E max，i Where: α i is the emission factor of unit i, E max，i is the emission limit of unit i; The deep reinforcement learning model is: R t =-(C f (t)+λ·E total (t)+γ·ΔP load (t)) Among them, R t is the reward function, C f (t) is the fuel cost at time t; E total (t) is the total emission at time t; λ and γ are the weights of emission and load fluctuation, ΔP load (t) is the difference between time t and the predicted load; Optimization update: Where: Q(S t , A t ) is the state-action value function; α is the learning rate, and A′ is the possible action at the next moment.

5. The intelligent dispatching and optimization method of a thermal power plant according to claim 4, characterized in that: The load forecasting model in step S4 is: Assume that the load time series data is P load (t), predict future load P load (t+1), the calculation formula is: P load (t+1)=f LSTM (P load (t),P load (t-1),...,P load (t-k)) Among them, f LSTM is the function of the load forecasting model, k is the time step number of the input historical data; Climate and temperature forecasts predict future temperature T(t+1) through meteorological data, and use ARIMA time series model for modeling: f(t+1)=φ0+φ1T(t)+∈ t Among them, φ0 and φ1 are the parameters of the ARIMA time series model, ∈ t is the noise term; Fuel price forecasting uses the Prophet time series model to predict fuel prices: P fuel (t+1)=f Prophet (P fuel (t),holiday,trend) Among them, f Prphet It is a Prophet time series model that combines holiday and trend information to predict fuel prices.

6. The intelligent dispatching and optimization method of a thermal power plant according to claim 5, characterized in that: In step S5, a PID controller is used to fine-tune the operation of the unit: Among them, e(t) is the error, K p , K i , K d are the parameters of the PID controller.

7. The intelligent dispatching and optimization method of a thermal power plant according to claim 6, characterized in that: The data mining algorithm in step S6 is: Where: d i is the deviation between the data point and the model; σ i is the standard deviation of this point.

Images