Intelligent peak regulation optimization transformation method and system for thermal power generating unit

Abstract

The invention discloses an intelligent peak regulation optimization and deep transformation method and system for a thermal power generating unit, and the optimization and transformation method comprises the following steps: (1) model prediction control: developing a control strategy for improving the thermal power generating unit through model prediction control, building a mathematical model of the thermal power generating unit, collecting the historical operation data of the thermal power generating unit, and carrying out the preprocessing of the data, a model prediction controller is developed and optimized; (2) load prediction and renewable energy output prediction are carried out through an HCNN-LSTM model; (3) scheduling optimization model design by optimizing cost and constraint conditions; and (4) solving the scheduling optimization model by using a genetic algorithm, and determining the optimal output of each unit in different time periods. The control strategy of the thermal power generating unit is improved through model predictive control (MPC); furthermore, the accuracy of load prediction and renewable energy output prediction is improved through an HCNN-LSTM model; and finally, in combination with a prediction result, a scheduling plan is optimized by using a genetic algorithm, so that the peak regulation capability and economy of the thermal power generating unit are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system optimization, and in particular relates to a method and system for intelligent peak-shaving optimization transformation of a thermal power unit. Background Art

[0002] With the transformation of the global energy structure, the proportion of renewable energy in the power system continues to increase. However, due to the intermittent and volatile nature of renewable energy such as wind and solar energy, it poses a huge challenge to the stable operation of the power grid. In this context, thermal power units, as an important means of regulation in the power system, are particularly important in terms of peak load regulation capability and flexibility.

[0003] The existing technology has many shortcomings in the deep peak regulation of thermal power units. For example, the minimum stable output of traditional thermal power units is usually high, which limits their operating capacity at low loads, resulting in the inability to effectively regulate some power demand. The load response speed of thermal power units is slow, and it is difficult to quickly respond to the drastic fluctuations in the power grid load, affecting the stability and reliability of the power grid. With the increase in peak-shaving demand, the start-stop frequency of thermal power units also increases, which not only increases energy consumption, but also aggravates equipment wear and reduces the service life of the units. In the scheduling optimization process of multiple units and multiple time periods, various constraints and economics need to be considered comprehensively. The optimization problem is complex and it is difficult for traditional methods to solve it efficiently. Therefore, it is urgent to develop a method and system for intelligent peak-shaving optimization and transformation of thermal power units to effectively improve the peak-shaving capacity and economy of thermal power units. Summary of the invention

[0004] Purpose of the invention: To solve the above technical problems, the present invention provides a method and system for intelligent peak-shaving optimization and transformation of thermal power units, which improves the control strategy of thermal power units through model predictive control, improves the accuracy of load forecasting and renewable energy output forecasting; improves the flexibility of thermal power units so that they can operate stably under lower loads and quickly respond to changes in power grid demand.

[0005] Technical solution: The present invention provides a method for intelligent peak load optimization transformation of a thermal power unit, comprising the following steps:

[0006] (1) Model Predictive Control (MPC): Develop model predictive control to improve the control strategy of thermal power units, establish a mathematical model of thermal power units, collect historical operating data of thermal power units and pre-process the data, develop model predictive controllers and optimize them;

[0007] (2) Load forecasting and renewable energy output forecasting through HCNN-LSTM model;

[0008] (3) Design of scheduling optimization model by optimizing costs and constraints;

[0009] (4) Use genetic algorithm to solve the scheduling optimization model and determine the optimal output of each unit in different time periods.

[0010] Furthermore, the mathematical model of the thermal power unit in step (1) includes a boiler model, a steam turbine model, and a generator model. The boiler model mainly describes the relationship between fuel input and steam parameters (such as pressure and temperature); the steam turbine model is used to describe how steam parameters are converted into mechanical power output; and the generator model describes how mechanical power is converted into electrical power.

[0011] Furthermore, the historical operating data of the thermal power unit in step (1) includes load change data, fuel consumption rate, and key parameters of the boiler, steam turbine and generator.

[0012] After collecting historical operating data of thermal power units and preprocessing the data, a model predictive controller is developed and optimized. The workflow of the model predictive controller (MPC) is as follows: Use a nonlinear dynamic model to predict the system state changes in the future.

[0013]

[0014] The optimization technique is to solve the optimization problem using gradient descent or other nonlinear optimization algorithms:

[0015] Initialize control input u(t);

[0016] Calculate the prediction error and objective function value;

[0017] Compute the gradient and update the control input:

[0018]

[0019] Where: η is the learning rate.

[0020] Furthermore, step (2) performs load forecasting and renewable energy output forecasting through the HCNN-LSTM model, including the following steps:

[0021] Collect, organize and pre-process historical load data and meteorological data, build an HCNN-LSTM model, and use historical load data and meteorological data to train the HCNN-LSTM model to perform load forecasting and output forecasting.

[0022] Furthermore, the HCNN-LSTM model is designed and constructed by combining the advantages of convolutional neural network (CNN) and long short-term memory network (LSTM). The model structure includes convolution layer, pooling layer, LSTM layer and fully connected layer.

[0023] The convolution layer is used to extract the spatial features of time series data, the pooling layer is used to reduce the dimension, the LSTM layer is used to capture long-term dependencies, and the fully connected layer is used for the final prediction output.

[0024] Hyperparameters such as convolution kernel size, pooling type, number of LSTM units, etc. should be optimized through cross-validation.

[0025] Furthermore, the cost in step (3) mainly includes the following five cost factors: equipment investment cost, peak-shaving market transaction cost, incremental power generation cost, power sales revenue loss and system power abandonment cost. The optimization goal is to minimize the sum of all the above costs.

[0026] In order to ensure the safe and stable operation of the system, the following constraints must be met:

[0027] (1) The technical upper limit of the capacity of each unit after transformation. The output of the unit after transformation must be within its minimum and maximum output range:

[0028] P min ≤P i ≤P max

[0029] Where: P min and P max They are the minimum and maximum output of the unit before and after the transformation respectively.

[0030] (2) System power balance constraint: the total power generation of the system must meet the load demand:

[0031]

[0032] Where: P renew is the renewable energy output and D is the system load demand.

[0033] (3) Unit operation constraints. The operation of the unit must meet the following conditions:

[0034] P i，min ≤P i ≤P i，max

[0035] ΔP i ≤R i

[0036] Where: P i，min and P i，max are the minimum and maximum output of unit i, ΔP i is the output change of the unit, R i is the ramp rate of the unit.

[0037] By optimizing the above costs and constraints, the optimal solution for the transformation of thermal power units can be achieved. Specifically, the objective function F is optimized to minimize the total cost while ensuring that the constraints such as technology, operation and system power balance are met. In this way, the operating efficiency and economy of thermal power units can be effectively improved while ensuring the stability and reliability of the system.

[0038] Furthermore, step (4) uses a genetic algorithm to solve the scheduling optimization model, which mainly includes the following steps: generating an initial population including multiple individuals, calculating the fitness of each individual, selecting excellent individuals, crossing individuals, mutating and retaining elites, and iterating in a loop until a predetermined termination condition is reached.

[0039] The termination condition is that a predetermined number of iterations is reached, the fitness value reaches a certain threshold, or the fitness value no longer increases significantly in multiple generations.

[0040] The present invention also provides a thermal power unit intelligent peak-shaving optimization and transformation system, comprising a model prediction module, a cost calculation module, an HCNN-LSTM model for load prediction and renewable energy output prediction, and an optimization configuration module.

[0041] The HCNN-LSTM model includes a convolutional layer, a pooling layer, an LSTM layer and a fully connected layer.

[0042] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following significant advantages: improving the control strategy of thermal power units through model predictive control (MPC); and then improving the accuracy of load forecasting and renewable energy output forecasting through the HCNN-LSTM model; finally, combining the prediction results, using genetic algorithms to optimize the scheduling plan to improve the economy and stability of the system. The use of intelligent optimization algorithms can improve the system calculation and response speed, realize fast, automatic and accurate power system scheduling, and improve the absorption capacity and resource utilization of wind power and solar power. The flexibility of thermal power units is improved, enabling them to operate stably at lower loads and quickly respond to changes in power grid demand. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 The figure is a schematic diagram of the working process of the intelligent peak-shaving optimization transformation method of a thermal power unit according to the present invention. DETAILED DESCRIPTION

[0044] The technical solution of the present invention is further described in detail below through specific embodiments. However, it is necessary to point out that the following embodiments are only used to describe the content of the invention and do not constitute a limitation on the protection scope of the present invention.

[0045] A thermal power unit intelligent peak load optimization and transformation system includes a model prediction module, a cost calculation module, an HCNN-LSTM model for load prediction and renewable energy output prediction, and an optimization configuration module. The HCNN-LSTM model includes a convolution layer, a pooling layer, an LSTM layer, and a fully connected layer.

[0046] A method for intelligent peak load optimization transformation of a thermal power unit comprises the following steps:

[0047] Step S1: Model Predictive Control (MPC)

[0048] 1. Establish a mathematical model of thermal power units

[0049] Boiler model: The boiler model mainly describes the relationship between fuel input and steam parameters (such as pressure and temperature). It can be expressed by the following nonlinear equation:

[0050]

[0051] Where: x b is the boiler state variable, such as steam pressure, temperature, etc., u f For fuel input.

[0052] Steam Turbine Model: The steam turbine model describes how steam parameters are converted into mechanical power output:

[0053]

[0054] Where: x t is the turbine state variable, such as speed, torque, etc., x b It is the steam parameter output by the boiler.

[0055] Generator Model: The generator model describes how mechanical power is converted into electrical power:

[0056] P gen =f g (x t )

[0057] Where: P gen The electrical power output of the generator.

[0058] 2. Data Collection and Preprocessing

[0059] ①Data Collection

[0060] Collect the following historical operating data: load change data; fuel consumption rate; key parameters of boilers, turbines and generators (such as pressure, temperature, speed, etc.)

[0061] ②Data cleaning

[0062] In this example, interpolation method is used to fill missing data, remove outliers, and fill missing values.

[0063] x clean = interpolate(x raw )

[0064] ③Data standardization processing

[0065] Standardize the data:

[0066]

[0067] In the formula: x is the original data, μ is the mean, σ is the standard deviation

[0068] 3. Design of model predictive control

[0069] 1) Develop a model predictive controller (MPC), whose workflow is as follows:

[0070] Use nonlinear dynamic models to predict system state changes over a period of time in the future.

[0071]

[0072] 2) Optimize the objective function and design the optimized objective function J, for example:

[0073]

[0074] Where: P ref is the reference load, λ is the adjustment coefficient, and N is the prediction time domain length

[0075] 3) Constraints

[0076] l≤u(t)≤u max

[0077] x mi n≤x(t)≤x max

[0078] g(x(t),u(t))≤0

[0079] 4) Optimize technical details

[0080] Solve the optimization problem using gradient descent or other nonlinear optimization algorithms:

[0081] Initialize control input u(t);

[0082] Calculate the prediction error and objective function value;

[0083] Compute the gradient and update the control input:

[0084]

[0085] Where: η is the learning rate.

[0086] 4. Repeat steps 2 and 3 until convergence.

[0087] Through the above steps, a thermal power unit optimization control system based on model predictive control can be realized. This system can minimize fuel consumption and ensure the stability of output power while meeting various constraints.

[0088] Step S2: Load forecasting and renewable energy output forecasting

[0089] S2-1 Data Preprocessing

[0090] Collect and organize historical load data and meteorological data, and clean and standardize the data.

[0091] S2-2 Model construction

[0092] Combining the advantages of convolutional neural network (CNN) and long short-term memory network (LSTM), the HCNN-LSTM model is designed and constructed. The model structure includes convolution layer, pooling layer, LSTM layer and fully connected layer.

[0093] The convolution layer is used to extract the spatial features of time series data, the pooling layer is used to reduce the dimension, the LSTM layer is used to capture long-term dependencies, and the fully connected layer is used for the final prediction output.

[0094] Hyperparameters such as convolution kernel size, pooling type, number of LSTM units, etc. should be optimized through cross-validation.

[0095] S2-3 Model Training

[0096] The HCNN-LSTM model is trained using historical load data and meteorological data, and the model parameters θ are optimized to minimize the prediction error. The loss function is the mean square error (MSE):

[0097]

[0098] Where: is the predicted value, y i is the true value

[0099] Also, consider using other loss functions such as mean absolute error (MAE) or Huber loss to better handle outliers.

[0100] S2-4 Load Forecast and Output Forecast

[0101] The load forecast is as follows:

[0102] D(t)=f HCNN-LSTM(D(t-1), D(t-2),..., D(tn))

[0103] Where: is the forecast load at time t, f HCNN-LSTM It is a new variant algorithm, D(t-1), D(t-2), ..., D(tn) are historical load data.

[0104] The renewable energy output forecast is as follows:

[0105]

[0106] Where: Forecast renewable energy output at time t. Make sure the training data covers enough seasonal and cyclical variations to improve forecast accuracy.

[0107] Step S3: Scheduling Optimization Model Design

[0108] In the optimization model, the following five cost factors are considered:

[0109] 1. Equipment investment cost C inv

[0110] Equipment investment cost represents the investment required in the transformation of thermal power units:

[0111] C inv =N t ΔP

[0112] Where: N t is the transformation investment coefficient, which represents the investment cost per unit adjustable capacity; ΔP is the adjustable capacity after transformation, which represents the unit capacity added or reduced through the transformation.

[0113] 2. Peak load market transaction cost C market

[0114] The transaction costs of the peak-shaving market include the transaction costs incurred by the units participating in the peak-shaving market:

[0115]

[0116] Where: N b1 and N b2 is the compensation price of the first and second tiers, E b1，t and E b2，t It is the deep adjustment compensation power corresponding to the unit in time period t.

[0117] 3. Incremental power generation cost C gen

[0118] The incremental power generation cost refers to the additional cost caused by the increased coal consumption of the unit in the deep adjustment operation state:

[0119]

[0120] Where: k oc，i is the coal consumption cost coefficient of deep regulation operation of unit i, k i，t and k 0，i are the coal consumption rate of deep regulation operation of unit i in period t and the coal consumption rate in the original state, P i，t is the output of unit i in period t.

[0121] 4. Loss of revenue from electricity sales C loss

[0122] The loss of electricity sales revenue refers to the reduction in electricity sales revenue caused by the reduction in unit output after the transformation:

[0123]

[0124] Where: N d is the unit electricity price, P i，t，0 is the power generation of unit i in period t before transformation, P i，t is the power generation of unit i in period t after transformation.

[0125] 5. System power abandonment cost C curt

[0126] System curtailment costs are the costs caused by the excess electricity generated in the system that cannot be used:

[0127]

[0128] Where: N Curt is the unit power abandonment cost, E curt，t is the amount of power abandoned by the system in period t.

[0129] 6. Optimization goals

[0130] The optimization goal is to minimize the sum of all the above costs:

[0131] min F=C inv +C market +C gen +C loss +C curt

[0132] In order to ensure the safe and stable operation of the system, the following constraints must be met:

[0133] 1. Technical upper limit of the transformation capacity of each unit

[0134] The output of the modified unit must be within its minimum and maximum output range:

[0135] P min ≤P i ≤P max

[0136] Where: P min and P max They are the minimum and maximum output of the unit before and after the transformation respectively.

[0137] 2. System power balance constraints

[0138] The total power generation of the system must meet the load demand:

[0139]

[0140] Where: P renew is the renewable energy output and D is the system load demand.

[0141] 3. Unit operation constraints

[0142] The operation of the unit must meet the following conditions:

[0143] P i，min ≤P i ≤P i，max

[0144] ΔP i ≤R i

[0145] Where: P i，min and P i，max are the minimum and maximum output of unit i, ΔP i is the output change of the unit, R i is the ramp rate of the unit.

[0146] By optimizing the above costs and constraints, the optimal solution for the transformation of thermal power units can be achieved. Specifically, the objective function F is optimized to minimize the total cost while ensuring that the constraints such as technology, operation and system power balance are met. In this way, the operating efficiency and economy of thermal power units can be effectively improved while ensuring the stability and reliability of the system.

[0147] Step S4: Genetic algorithm solves the scheduling optimization model

[0148] Genetic algorithm is used to solve the scheduling optimization model and determine the optimal output of each unit in different time periods. Genetic algorithm is an optimization algorithm based on natural selection and genetic mutation. Its main steps include initializing the population, calculating fitness, selection, crossover, mutation and elite retention.

[0149] 1. Initialize the population

[0150] First, an initial population containing multiple individuals is generated. Each individual represents a possible solution, that is, the output of each unit in different time periods. These initial individuals are randomly generated to ensure the diversity of the search space. Assume that the population size is N, and each individual is represented by a matrix P, where P i，t Represents the output of unit i in time period t:

[0151]

[0152] 2. Calculate fitness

[0153] Calculate the fitness value of each individual. The fitness function is usually the negative of the optimization objective function. In this case, the fitness function considers the sum of the following costs:

[0154] F=C inv +C market +C gen +C loss +C curt

[0155] The fitness value is the negative of the total cost:

[0156] Fitness=-F

[0157] 3. Select

[0158] According to the fitness value of each individual, it enters the next generation. Using the roulette wheel selection method, the probability of selection is p i Determined by the fitness value:

[0159]

[0160] 4. Cross

[0161] Generate new offspring by exchanging some genes of the parents. Use the single-point crossover method and select a crossover point c:

[0162] For each parent individual, split it into two parts: the first c time steps and the remaining time steps.

[0163] The two parts are exchanged between the two parent individuals to produce two offspring individuals.

[0164] The offspring individuals are represented by

[0165] Child 1:

[0166]

[0167] Child 2:

[0168]

[0169] In the formula: N represents the number of units, T represents the total number of time steps, c represents the time step position of the intersection point, P i，t represents the active power output of the i-th unit at time t, Q i，t Represents the reactive power output of the i-th unit at time t.

[0170] 5. Mutation

[0171] Randomly change one or several genes of an individual to introduce new genetic diversity. Assume that the mutation probability is pm, for each gene P i，t If the random number is less than p m , then mutate it into a new random value:

[0172] 6. Elite Retention

[0173] Retain the individuals with the highest fitness in the previous generation to ensure that the optimal solution is not lost. Assume that the elite retention rate is r e , keep the first r e ×N individuals with the highest fitness.

[0174] 7. Loop Iteration

[0175] The selection, crossover, mutation, and elite retention steps are repeated until a predetermined termination condition is reached. The termination condition can be a predetermined number of iterations or no significant improvement in fitness.

[0176] 8. Termination Conditions

[0177] The genetic algorithm ends when a predetermined termination condition is met, which may be a predetermined number of iterations, a fitness value reaching a certain threshold, or a fitness value no longer significantly improving over multiple generations.

[0178] 9. Result Output

[0179] After the algorithm is finished, the individual with the highest fitness is output as the final optimal solution, that is, the optimal output of each unit in different time periods. This solution minimizes the total cost while satisfying the system power balance and unit operation constraints.

[0180] In summary, the present invention proposes a method and system for intelligent peak-shaving optimization and deep transformation of thermal power units. Based on the covariance matrix adaptive evolutionary algorithm, the matching degree of renewable energy power generation and load, as well as the output characteristics of renewable energy power generation are considered. The use of intelligent optimization algorithm can improve the system calculation and response speed, realize fast, automatic and accurate power system scheduling, and improve the renewable energy power generation and consumption capacity and resource utilization.

Claims

1. A method for optimizing and transforming intelligent peak load regulation of thermal power units, characterized in that: The following steps are involved: (1) Model Predictive Control: Develop model predictive control to improve the control strategy of thermal power units, establish a mathematical model of thermal power units, collect historical operating data of thermal power units and pre-process the data, develop model predictive controllers and optimize them; (2) Load forecasting and renewable energy output forecasting through HCNN-LSTM model; (3) Design of scheduling optimization model by optimizing costs and constraints; (4) Use genetic algorithms to solve the scheduling optimization model and determine the optimal output of each unit in different time periods.

2. The intelligent peak-shaving optimization and transformation method for thermal power units according to claim 1 is characterized in that: The mathematical model of the thermal power unit in step (1) includes a boiler model, a steam turbine model, and a generator model.

3. The intelligent peak-shaving optimization transformation method for thermal power units according to claim 1 is characterized in that: The historical operation data of the thermal power unit in step (1) includes load change data, fuel consumption rate, and key parameters of the boiler, steam turbine and generator.

4. The intelligent peak-shaving optimization and transformation method for thermal power units according to claim 1 is characterized in that: The step (2) of performing load forecasting and renewable energy output forecasting through the HCNN-LSTM model includes the following steps: collecting and organizing historical load data and meteorological data and preprocessing them, building an HCNN-LSTM model, and using the historical load data and meteorological data to train the HCNN-LSTM model to perform load forecasting and output forecasting.

5. The intelligent peak-shaving optimization and transformation method for thermal power units according to claim 4 is characterized in that: The HCNN-LSTM model is designed and constructed by combining the advantages of convolutional neural networks (CNN) and long short-term memory networks (LSTM).

6. The intelligent peak-shaving optimization and transformation method for thermal power units according to claim 1 is characterized in that: The costs in step (3) include equipment investment costs, peak-shaving market transaction costs, incremental power generation costs, power sales revenue losses, and system power abandonment costs.

7. The intelligent peak-shaving optimization and transformation method for thermal power units according to claim 1 is characterized in that: The step (4) of using a genetic algorithm to solve the scheduling optimization model mainly includes the following steps: Generate an initial population containing multiple individuals, calculate the fitness of each individual, select excellent individuals, cross individuals, mutate and retain the elite, and iterate until the predetermined termination condition is reached.

8. The intelligent peak-shaving optimization and transformation method for thermal power units according to claim 7 is characterized in that: The termination condition is that a predetermined number of iterations is reached, the fitness value reaches a certain threshold, or the fitness value no longer increases significantly in multiple generations.

9. An intelligent peak-shaving optimization and transformation system for thermal power units, characterized in that: It includes model prediction module, cost calculation module, HCNN-LSTM model for load forecasting and renewable energy output forecasting, and optimization configuration module.

10. The intelligent peak-shaving optimization and transformation method for thermal power units according to claim 1 is characterized in that: The HCNN-LSTM model includes a convolutional layer, a pooling layer, an LSTM layer and a fully connected layer.

Images