Power plant optimal scheduling method and virtual control system based on machine learning model

Through the machine learning model-based power plant optimization scheduling method, combined with memory network model and fruit fly algorithm, the global considerations in power grid scheduling are solved, the comprehensive optimization of grid loss risk and operating costs are achieved, and the efficiency of grid operation is improved and stability.

CN118469257BActive Publication Date: 2024-11-05中煤新集利辛发电有限公司 +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410919464.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-10
Publication Date
2024-11-05
Estimated Expiration
2044-07-10

AI Technical Summary

Technical Problem

The prior art only considers the power interaction of the internal nodes of the power grid in power grid scheduling, and cannot consider the scheduling scheme from the global power grid, resulting in the inability to effectively solve the comprehensive optimization of the grid loss risk and operating costs.

Method used

The optimization scheduling method of power plant based on machine learning model is adopted, and by obtaining historical grid load requirements, establishing a memory network model, combining Monte Carlo simulation and fruit fly algorithm, an optimized scheduling solution is generated to minimize the grid loss risk and operating costs.

Benefits of technology

It improves the accuracy and reliability of grid load demand forecasting, comprehensively considers the diversity and uncertainty of grid loss risk, generates high-quality scheduling plans, reduces power plant operation risks, and improves grid operation efficiency and stability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118469257B_ABST
    Figure CN118469257B_ABST
Patent Text Reader

Abstract

The present invention provides a power plant optimization scheduling method and a virtual control system based on a machine learning model, which relate to the technical field of power grids, and include: obtaining historical power grid load demands, generating a data set, dividing the data set into a training set and a validation set, establishing a memory network model, determining the structure of the memory network model and defining a hyperparameter grid, adjusting the hyperparameter value range of the memory network model, cross-validating and determining the hyperparameter, and deploying the memory network model; determining input parameters in the memory network model, setting a probability distribution, generating random samples, for each input parameter, randomly extracting samples according to the probability distribution and solving the model to obtain uncertainty, evaluating the risk of power grid load loss, and constructing an objective function; generating a fruit fly population and setting the number of individuals, repeatedly iterating based on a fruit fly algorithm to obtain an optimal individual, determining whether the current optimal individual is accepted by a simulated annealing algorithm, and if so, determining a scheduling optimization plan according to the objective function.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of power grid technology, and in particular to a power plant optimization scheduling method and a virtual control system based on a machine learning model. Background Art

[0002] In the prior art, CN105391060A discloses a method for optimizing and dispatching a distribution network containing a microgrid, the method comprising calculating the unbalanced electricity of each of the microgrids; calculating the maximum output power of the corresponding microgrid and the maximum input power of the corresponding microgrid; establishing an optimization model with the output power of each controllable distributed power source of the distribution network, the interactive power of each microgrid entrance and the power purchased by the distribution network from the upper power grid as control variables, and with minimizing the power supply cost of the distribution network as the optimization target; calculating the optimal solution of the control variables, and dispatching the distribution network according to the optimal solution.

[0003] In summary, although the existing technology can achieve grid dispatch optimization, it only involves the power interaction of nodes within the grid and cannot consider the dispatch plan from the overall perspective of the grid. Therefore, a solution is needed to solve the problems existing in the existing technology. Summary of the invention

[0004] The embodiments of the present invention provide a power plant optimization scheduling method and a virtual control system based on a machine learning model, which can at least solve some of the problems existing in the prior art.

[0005] A first aspect of an embodiment of the present invention provides a method for optimizing power plant scheduling based on a machine learning model, comprising:

[0006] Obtain historical power grid load demand, generate a data set based on the historical power grid load demand, divide the data set into a training set and a validation set, establish a memory network model based on deep learning, determine the structure of the memory network model and define a hyperparameter grid, adjust the hyperparameter value range of the memory network model based on the hyperparameter grid, cross-validate the memory network model based on the data set and determine the hyperparameters, and deploy the memory network model;

[0007] Based on the memory network model, input parameters in the memory network model are determined, probability distribution is set, random samples are generated through Monte Carlo simulation, for each input parameter, samples are randomly drawn according to the set probability distribution and the model is solved to obtain uncertainty, the grid load loss risk is evaluated based on the uncertainty, and an objective function is constructed with minimization of the grid load loss risk and minimization of operating costs as goals;

[0008] Based on the objective function, a fruit fly population is generated and the number of individuals is set, and the initial temperature and temperature attenuation coefficient of simulated annealing are set, wherein each individual in the fruit fly population represents a solution of the objective function, and the optimal individual is obtained by repeated iteration based on the trained fruit fly algorithm. In combination with the initial temperature and the temperature attenuation coefficient, a simulated annealing algorithm is used to determine whether the current optimal individual is accepted. If so, a scheduling optimization plan is determined according to the objective function.

[0009] In an optional embodiment,

[0010] The method of obtaining historical power grid load demand, generating a data set based on the historical power grid load demand, dividing the data set into a training set and a validation set, establishing a memory network model based on deep learning, determining the structure of the memory network model and defining a hyperparameter grid, adjusting the hyperparameter value range of the memory network model based on the hyperparameter grid, cross-validating the memory network model based on the data set and determining the hyperparameter, and deploying the memory network model includes:

[0011] Obtain historical power grid load demand, save it in the form of timestamp and specific value and generate a data set, and divide the data set into a training set and a validation set, wherein the training set accounts for 80% of the data set and the validation set accounts for 20%;

[0012] Constructing a memory network model, determining the number of gated recurrent unit layers in the memory network model, determining the corresponding number of neurons for each gated recurrent unit layer according to computing requirements, and setting the values ​​of initial hyperparameters;

[0013] Based on the initial hyperparameters, a parameter grid is defined, and each hyperparameter combination in the parameter grid is traversed. Based on the structure and task requirements of the memory network model, a value range of each hyperparameter is determined. Based on the value range, the performance of the memory network model is evaluated through cross-validation. The validation is repeated until all hyperparameter combinations are validated. The hyperparameter combination that enables the memory network model to have the highest performance is selected and recorded as the optimal hyperparameter combination. The memory network model is deployed based on the optimal hyperparameter combination.

[0014] In an optional embodiment,

[0015] Based on the memory network model, the input parameters in the memory network model are determined, the probability distribution is set, random samples are generated through Monte Carlo simulation, for each input parameter, samples are randomly drawn according to the set probability distribution and the model is solved to obtain uncertainty, the grid load loss risk is evaluated based on the uncertainty, and the objective function is constructed with the minimization of the grid load loss risk and the minimization of the operating cost as the goal, including:

[0016] For the deployed memory network model, obtaining input parameters of the memory network model, namely, weather conditions and economic indicators that affect power grid load demand;

[0017] For the input parameters, determine the probability distribution corresponding to each parameter through discrete probability distribution or continuous probability distribution according to the data type, and set the corresponding normal distribution for the model error;

[0018] For each input parameter, random sampling is performed based on a probability density function according to a set probability distribution to generate a random sample, the random sample is added to the memory network model for solving, a first prediction result is obtained, a first standard deviation corresponding to the first prediction result is calculated, and the first standard deviation is used as the uncertainty;

[0019] Based on the uncertainty, combined with the predicted value of the grid load demand and the rated capacity, the overload time is calculated, wherein the longer the overload time, the greater the risk of grid unload. The objective function is constructed with the goal of minimizing the risk of grid unload and minimizing the operating cost.

[0020] In an optional embodiment,

[0021] The objective function is constructed with the goal of minimizing the risk of grid loss and the operating cost as shown in the following formula:

[0022] ;

[0023] Where F represents the objective function value, E[] + represents the expected value operation, D(t) represents the predicted load demand in time period t, and P gen,i (t) represents the power generation of the ith power generation unit in time period t, T represents the total number of time periods, C i represents the unit power generation cost of the i-th power generation unit, ∑ t It means to sum the expected values ​​over all time periods t, ∑ i It indicates summing the power generated by the power generation units.

[0024] In an optional embodiment,

[0025] Based on the objective function, a fruit fly population is generated and the number of individuals is set, and the initial temperature and temperature attenuation coefficient of simulated annealing are set, wherein each individual in the fruit fly population represents a solution of the objective function, and the optimal individuals are obtained by repeated iteration based on the trained fruit fly algorithm, including:

[0026] Generate a fruit fly population and set the number of individuals, the initial temperature and temperature attenuation coefficient corresponding to the simulated annealing algorithm, and represent each individual in the fruit fly population as a binary vector through binary coding;

[0027] For each individual in the fruit fly population, a corresponding first fitness value is calculated, an individual with the highest fitness value is selected as the optimal individual, and offspring individuals are randomly generated in the neighborhood by adding random perturbations in each dimension of the binary vector;

[0028] For the offspring individuals, according to a preset dynamic step size, the offspring individuals are moved toward the optimal individual, and the positions of the offspring individuals are repeatedly updated until the position of each offspring individual is updated, and the second fitness value corresponding to each offspring individual after the update is calculated, and the minimum second fitness value corresponding to the individual in the offspring individuals is compared with the first fitness value corresponding to the optimal individual. If the second fitness value is greater than the first fitness value, the current offspring individual is used to replace the optimal individual.

[0029] In an optional embodiment,

[0030] Combining the initial temperature and the temperature attenuation coefficient, determining whether the current optimal individual is accepted by a simulated annealing algorithm, and if so, determining a scheduling optimization solution according to the objective function includes:

[0031] For the optimal individual generated in each iteration in the fruit fly population, calculating the corresponding objective function value according to the objective function, and applying the simulated annealing algorithm;

[0032] For each objective function value, starting from the initial temperature, an alternative result is generated through neighborhood search and an alternative objective function value corresponding to the alternative result is calculated. If the alternative objective function value is better than the objective function value corresponding to the optimal individual, the alternative result is used to replace the current optimal individual, and the update is repeated by lowering the temperature until the preset minimum temperature is reached to obtain the scheduling optimization plan;

[0033] If the alternative objective function value is not better than the objective function value corresponding to the optimal individual, the alternative result is deleted. If the alternative results obtained by lowering the temperature search three times in a row are not selected, the temperature attenuation coefficient is doubled and repeatedly updated until the preset minimum temperature is reached to obtain the scheduling optimization plan.

[0034] In an optional embodiment,

[0035] According to the preset dynamic step length, the offspring individuals are moved toward the optimal individuals as shown in the following formula:

[0036] ;

[0037] in, represents the position of individual n in the x direction at time step t+1, x nrepresents the position of individual n in the x direction at time step t, x best Indicates the position of the current optimal individual in the x direction, S x represents the dynamic step size in the x direction, δ x represents the random perturbation parameter in the x direction, rand x represents a random number in the x direction, represents the position of individual n in the y direction at the t+1 time step, y n represents the position of individual n in the x direction at time step t, S y Indicates the dynamic step size in the y direction, y best represents the position of the current optimal individual in the y direction, δ y represents the random perturbation parameter in the y direction, rand y Represents a random number in the y direction.

[0038] A second aspect of an embodiment of the present invention provides a power plant optimization scheduling system based on a machine learning model, comprising:

[0039] The first unit is used to obtain historical power grid load demand, generate a data set based on the historical power grid load demand, divide the data set into a training set and a validation set, establish a memory network model based on deep learning, determine the structure of the memory network model and define a hyperparameter grid, adjust the hyperparameter value range of the memory network model based on the hyperparameter grid, cross-validate the memory network model based on the data set and determine the hyperparameter, and deploy the memory network model;

[0040] The second unit is used to determine the input parameters in the memory network model based on the memory network model, set the probability distribution, generate random samples through Monte Carlo simulation, randomly extract samples according to the set probability distribution for each input parameter and solve the model to obtain uncertainty, evaluate the power grid load loss risk based on the uncertainty, and construct an objective function with minimization of the power grid load loss risk and minimization of operating costs as goals;

[0041] The third unit is used to generate a fruit fly population and set the number of individuals based on the objective function, and to simulate the initial temperature and temperature attenuation coefficient of annealing, wherein each individual in the fruit fly population represents a solution of the objective function, and to obtain the optimal individual based on repeated iterations of the trained fruit fly algorithm, and to determine whether the current optimal individual is accepted by combining the initial temperature and the temperature attenuation coefficient through the simulated annealing algorithm, and if so, determine the scheduling optimization plan according to the objective function.

[0042] According to a third aspect of the embodiments of the present invention,

[0043] A virtual control system is provided, comprising:

[0044] The upper monitoring system uses MACS software and includes engineer station, operator station and history station functions;

[0045] The virtual controller deployed in the server consists of hardware resources, virtualization platform software, operating system and application software. It uses virtualization technology to deploy multiple controllers on one server, supports controller redundancy, and communicates with the host computer software through dual Ethernet.

[0046] System network, used for communication between the upper monitoring system and the virtual controller;

[0047] Control network, used for data communication between virtual controller and I / O modules;

[0048] Gateway, used to complete the protocol conversion between Ethernet and DP between the virtual controller and the I / O module, to realize the communication between the virtual controller and the I / O module. A single gateway supports 4 independent DP links, communicates with the virtual controller through redundant Ethernet, and supports the use of 5G network between the virtual controller and the gateway;

[0049] I / O modules for field data collection and execution.

[0050] A fourth aspect of the embodiments of the present invention is:

[0051] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0052] In the present invention, historical data and deep learning models are combined to predict power grid load demand and evaluate uncertainty, thereby improving the accuracy and reliability of the prediction. Random samples are generated through Monte Carlo simulation, and each input parameter is randomly sampled and the model is solved. The diversity and uncertainty of power grid load loss risks are fully considered, providing a more comprehensive evaluation and prediction for power grid operation. Based on the objective function, the fruit fly algorithm can effectively generate high-quality solutions for the optimized scheduling scheme, so that the power grid can minimize the risk of power grid load loss and operating costs while meeting the load demand, thereby improving the operating efficiency and stability of the power grid. In summary, the present invention can effectively improve the quality and efficiency of the optimized scheduling scheme of the power plant and reduce the operating risks of the power plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of a flow chart of a power plant optimization scheduling method based on a machine learning model according to an embodiment of the present invention;

[0054] Figure 2 It is a structural schematic diagram of a power plant optimization scheduling system based on a machine learning model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0057] Figure 1 Schematic diagram of a flow chart of a power plant optimization scheduling method based on a machine learning model according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0058] S1. Obtain historical power grid load demand, generate a data set based on the historical power grid load demand, divide the data set into a training set and a validation set, establish a memory network model based on deep learning, determine the structure of the memory network model and define a hyperparameter grid, adjust the hyperparameter value range of the memory network model based on the hyperparameter grid, cross-validate the memory network model based on the data set and determine the hyperparameters, and deploy the memory network model;

[0059] The hyperparameter grid refers to an arrangement of hyperparameter combinations in a machine learning model. Hyperparameter grid search is a common method for systematically trying various hyperparameter combinations to find the best hyperparameter combination to improve model performance. The cross-validation is a technique for evaluating the performance of a machine learning model by dividing the dataset into multiple subsets, training the model multiple times, and using different subsets as validation sets in each training for evaluation. The hyperparameters are parameters set before the machine learning model is trained, and they control the learning process and complexity of the model.

[0060] In an optional embodiment,

[0061] The method of obtaining historical power grid load demand, generating a data set based on the historical power grid load demand, dividing the data set into a training set and a validation set, establishing a memory network model based on deep learning, determining the structure of the memory network model and defining a hyperparameter grid, adjusting the hyperparameter value range of the memory network model based on the hyperparameter grid, cross-validating the memory network model based on the data set and determining the hyperparameter, and deploying the memory network model includes:

[0062] Obtain historical power grid load demand, save it in the form of timestamp and specific value and generate a data set, and divide the data set into a training set and a validation set, wherein the training set accounts for 80% of the data set and the validation set accounts for 20%;

[0063] Constructing a memory network model, determining the number of gated recurrent unit layers in the memory network model, determining the corresponding number of neurons for each gated recurrent unit layer according to computing requirements, and setting the values ​​of initial hyperparameters;

[0064] Based on the initial hyperparameters, a parameter grid is defined, and each hyperparameter combination in the parameter grid is traversed. Based on the structure and task requirements of the memory network model, a value range of each hyperparameter is determined. Based on the value range, the performance of the memory network model is evaluated through cross-validation. The validation is repeated until all hyperparameter combinations are validated. The hyperparameter combination that enables the memory network model to have the highest performance is selected and recorded as the optimal hyperparameter combination. The memory network model is deployed based on the optimal hyperparameter combination.

[0065] The timestamp refers to a mark or index of a specific time point, which is usually used to record the time when an event occurs. The gated recurrent unit is a variant of a recurrent neural network, which is used to process sequence data. The neuron is the basic unit of an artificial neural network. The neuron simulates the function of a biological neuron, receives an input signal, performs weighted summation on it, applies an activation function, and then passes the result to the next layer of neurons or the output layer. The hyperparameter combination refers to a set of hyperparameters selected before model training.

[0066] Obtain historical grid load demand data from reliable data sources, record the timestamp and corresponding grid load demand value at each time point, save the data in a suitable format, and divide the data set into a training set and a validation set in chronological order, where the first 80% of the data in chronological order is taken as the training set, and the last 20% of the data is taken as the validation set;

[0067] Build a memory network model, determine the number of layers of gated recurrent units, determine the number of neurons in each layer of GRU according to the computational requirements and the complexity of the model, set the initial hyperparameters of the model, such as learning rate, regularization parameters, etc., train the model using the training set, minimize the loss function by iteratively optimizing the model parameters, monitor the training process using the validation set, evaluate the performance of the model on unseen data, so as to adjust the hyperparameters or model structure, and stop training when the performance on the validation set no longer improves to prevent overfitting.

[0068] According to the initial hyperparameters, a hyperparameter grid is defined, including the hyperparameters that need to be adjusted and their value ranges, each hyperparameter combination in the parameter grid is traversed, a model is trained for each combination and a cross-validation evaluation is performed, for each hyperparameter combination, the model is evaluated using a cross-validation technique, and performance indicators of the model on the validation set are calculated, such as the mean absolute error and the root mean square error, and according to the cross-validation results, the hyperparameter combination that makes the model perform best is selected as the optimal hyperparameter combination, and the memory network model is deployed based on the optimal hyperparameter combination;

[0069] Exemplarily, assume that the initial hyperparameters are set to: learning rate: [0.001, 0.01, 0.1], batch size range: [16, 32, 64], number of gated recurrent unit layers: [1, 2, 3], number of neurons in each layer of gated recurrent units: [32, 64, 128], initialize the parameter grid, assume that the learning rate of the parameter grid is 0.001, the batch size is 16, the gated recurrent unit is 1 layer, and the number of neurons is 32, verify the performance of this parameter grid, repeatedly verify the performance of parameter grids of different specifications, select the hyperparameters corresponding to the parameter grid with the best performance, set the memory network model and deploy it.

[0070] In this embodiment, by obtaining historical power grid load demand data and saving it in the form of timestamps and specific values, a data set is constructed to ensure that there is enough data for effective learning and verification during model training and evaluation, while avoiding overfitting of the model to the same data. The introduction of gated recurrent units can help the model better capture long-term dependencies in time series data, improve the model's ability to predict power grid load demand, define a parameter grid, and traverse each hyperparameter combination. Evaluating the performance of the model under each hyperparameter combination through cross-validation can help find the best model hyperparameters, thereby improving the model's prediction accuracy and generalization ability. In summary, this embodiment can improve the accuracy and reliability of power grid load demand prediction, thereby bringing good technical effects and contributing to the operation and planning of the power system.

[0071] S2. Based on the memory network model, determine the input parameters in the memory network model, set the probability distribution, generate random samples through Monte Carlo simulation, randomly extract samples for each input parameter according to the set probability distribution and solve the model to obtain uncertainty, evaluate the grid load loss risk based on the uncertainty, and construct an objective function with the minimization of the grid load loss risk and the minimization of the operating cost as the goal;

[0072] The probability distribution is a distribution method that describes the possible values ​​of a random variable. In the prediction of power grid load demand, probability distribution can be used to describe the probability distribution of power grid load at different time points. The random sample is sample data randomly extracted from the probability distribution. The uncertainty refers to the uncertainty caused by incomplete information or difficulty in accurate prediction. The power grid load loss risk refers to the potential risk caused by power grid losses in the power system (such as transmission line impedance, transformer losses, etc.).

[0073] In an optional embodiment,

[0074] Based on the memory network model, the input parameters in the memory network model are determined, the probability distribution is set, random samples are generated through Monte Carlo simulation, for each input parameter, samples are randomly drawn according to the set probability distribution and the model is solved to obtain uncertainty, the grid load loss risk is evaluated based on the uncertainty, and the objective function is constructed with the minimization of the grid load loss risk and the minimization of the operating cost as the goal, including:

[0075] For the deployed memory network model, obtaining input parameters of the memory network model, namely, weather conditions and economic indicators that affect power grid load demand;

[0076] For the input parameters, determine the probability distribution corresponding to each parameter through discrete probability distribution or continuous probability distribution according to the data type, and set the corresponding normal distribution for the model error;

[0077] For each input parameter, random sampling is performed based on a probability density function according to a set probability distribution to generate a random sample, the random sample is added to the memory network model for solving, a first prediction result is obtained, a first standard deviation corresponding to the first prediction result is calculated, and the first standard deviation is used as the uncertainty;

[0078] Based on the uncertainty, combined with the predicted value of the grid load demand and the rated capacity, the overload time is calculated, wherein the longer the overload time, the greater the risk of grid unload. The objective function is constructed with the goal of minimizing the risk of grid unload and minimizing the operating cost.

[0079] The weather conditions refer to the meteorological conditions in a certain area within a specific time range, including but not limited to temperature, humidity, precipitation, wind speed, air pressure, etc. In the prediction of power grid load demand, weather conditions are an important influencing factor, because weather changes will directly affect people's electricity consumption behavior. The economic indicators are data used to measure economic activities and economic conditions. In the prediction of power grid load demand, economic indicators can be used to predict the level of economic activity in a certain area, thereby indirectly affecting electricity demand. For example, when the economy is booming, industrial production activities increase, which may lead to an increase in electricity demand. The normal distribution, also known as the Gaussian distribution, is a common probability distribution model with the characteristics of a bell curve. The random sampling refers to the process of randomly selecting samples from a population to represent the characteristics of the population. The overload time refers to the duration of the power grid being in an overloaded state.

[0080] Determine the weather conditions and economic indicators that affect grid load demand as the model’s input parameters, including temperature, humidity, wind speed, economic growth rate, etc.;

[0081] For each input parameter, determine the corresponding probability distribution according to its data type. If the parameter is discrete, you can use discrete probability distribution, such as Bernoulli distribution, multinomial distribution, etc. If the parameter is continuous, you can use continuous probability distribution, such as normal distribution, uniform distribution, etc. For parameters that may have cyclical or seasonal changes such as economic indicators, you can consider using cyclical or seasonal probability distribution models, set the corresponding probability distribution for the error part of the model, and set the model error to obey the normal distribution by default;

[0082] For each input parameter, random sampling is performed based on the probability density function according to the set probability distribution to generate random samples, and the generated random samples are added to the memory network model for solution to obtain the first prediction result, and the standard deviation corresponding to the first prediction result is calculated as a measure of uncertainty. Based on the uncertainty, the predicted value of the grid load demand and the rated capacity, the overload time is calculated, wherein the longer the overload time, the greater the risk of grid load loss. According to the operation rules and safety standards of the grid, the threshold of the overload time is determined, and the grid load loss risk and the operation cost are quantified as mathematical expressions to construct an objective function that comprehensively considers the grid load loss risk and the operation cost;

[0083] Exemplarily, temperature and economic growth rate are selected as input parameters affecting grid load demand. For temperature, it is assumed that its data is continuous and described by normal distribution. Assume that the mean is 25 degrees Celsius and the standard deviation is 5 degrees Celsius. Assume that the economic growth rate is continuous and described by normal distribution. Assume that the mean is 0.03 and the standard deviation is 0.01. For each input parameter, based on the set probability distribution, a probability density function is used for random sampling to generate a certain number of random samples. Assume that 100 temperature samples and 100 economic growth rate samples are generated. The generated random samples are added to the memory network model for solution to obtain the corresponding grid load demand forecast value. For each forecast result, the corresponding standard deviation is calculated. Assuming the standard deviation is 100 and the rated capacity is 10,000 units, the overload time is calculated based on uncertainty and forecast value. If the forecast value exceeds the rated capacity, it is considered that overload has occurred. Assuming that the overload time is 10 hours, the grid load loss risk and operating cost are combined to construct an objective function.

[0084] In this embodiment, the use of a probability distribution model can better capture the impact of different parameter values ​​on the grid load demand, improve the accuracy and reliability of the model, calculate the standard deviation corresponding to each prediction result, effectively quantify the uncertainty of the model, and more clearly understand the possible fluctuation range of the model's prediction results, thereby better assessing the risk of grid load demand. Based on uncertainty and predicted values, the overload time is calculated, which can help power system operators better assess the stability and security of the grid. In summary, this embodiment helps power system operators to better manage and maintain the grid and ensure the stable operation and power supply reliability of the grid.

[0085] In an optional embodiment,

[0086] The objective function is constructed with the goal of minimizing the risk of grid loss and the operating cost as shown in the following formula:

[0087] ;

[0088] Where F represents the objective function value, E[] + represents the expected value operation, D(t) represents the predicted load demand in time period t, and P gen,i (t) represents the power generation of the ith power generation unit in time period t, T represents the total number of time periods, C i represents the unit power generation cost of the i-th power generation unit, ∑ t It means to sum the expected values ​​over all time periods t, ∑ i It indicates summing the power generated by the power generation units.

[0089] In this embodiment, by calculating the difference between the actual load demand and the power generation and comparing it with zero to ensure that the actual load demand does not exceed the power generation, it helps to ensure the stability and reliability of the power grid and reduce the risk of power outages. By multiplying the power generation of each power generation unit by its unit power generation cost and adding the operating costs of all power generation units to obtain the total operating cost in a given time period, it can help optimize the scheduling and operation of the power generation units to reduce the overall operating cost, and can more effectively utilize energy resources and improve energy utilization, thereby reducing the total cost of electricity production. In summary, this embodiment combines the risk of power grid unloading and operating costs, and can minimize operating costs while ensuring the stability of the power grid, achieving more balanced and sustainable power grid operations, and at the same time improving the overall efficiency and reliability of the power grid.

[0090] S3. Based on the objective function, a fruit fly population is generated and the number of individuals is set, and the initial temperature and temperature attenuation coefficient of simulated annealing are set, wherein each individual in the fruit fly population represents a solution of the objective function, and the optimal individual is obtained by repeated iteration based on the trained fruit fly algorithm. In combination with the initial temperature and the temperature attenuation coefficient, a simulated annealing algorithm is used to determine whether the current optimal individual is accepted. If so, a scheduling optimization plan is determined according to the objective function.

[0091] The temperature attenuation coefficient is a key parameter that determines how the temperature decreases over time, thereby affecting the algorithm's search process and efficiency. The simulated annealing algorithm is a heuristic optimization algorithm used to solve global optimization problems and to find the global optimal solution in a complex search space. The fruit fly algorithm is a heuristic algorithm that imitates the behavior of fruit flies when looking for food sources and is used to solve optimization problems.

[0092] In an optional embodiment,

[0093] Based on the objective function, a fruit fly population is generated and the number of individuals is set, and the initial temperature and temperature attenuation coefficient of simulated annealing are set, wherein each individual in the fruit fly population represents a solution of the objective function, and the optimal individuals are obtained by repeated iteration based on the trained fruit fly algorithm, including:

[0094] Generate a fruit fly population and set the number of individuals, the initial temperature and temperature attenuation coefficient corresponding to the simulated annealing algorithm, and represent each individual in the fruit fly population as a binary vector through binary coding;

[0095] For each individual in the fruit fly population, a corresponding first fitness value is calculated, an individual with the highest fitness value is selected as the optimal individual, and offspring individuals are randomly generated in the neighborhood by adding random perturbations in each dimension of the binary vector;

[0096] For the offspring individuals, according to a preset dynamic step size, the offspring individuals are moved toward the optimal individual, and the positions of the offspring individuals are repeatedly updated until the position of each offspring individual is updated, and the second fitness value corresponding to each offspring individual after the update is calculated, and the minimum second fitness value corresponding to the individual in the offspring individuals is compared with the first fitness value corresponding to the optimal individual. If the second fitness value is greater than the first fitness value, the current offspring individual is used to replace the optimal individual.

[0097] The binary encoding is a method of using binary numbers to represent information, and is used to represent candidate solutions. The binary vector is a vector composed of binary numbers, and is used to represent solutions or other information in the algorithm. The random perturbation is a technology that explores the solution space by adding randomness to the solution or parameters. It is often used in optimization algorithms to increase the diversity of solutions and avoid the algorithm from falling into the local optimal solution prematurely. The dynamic step size refers to the practice of dynamically adjusting the search step size according to a certain strategy or performance feedback during the iteration process of the optimization algorithm.

[0098] A certain number of fruit fly individuals are randomly generated, where each fruit fly individual represents a solution, which is encoded using a binary vector to define the initial temperature and the temperature attenuation coefficient;

[0099] For each individual in the fruit fly population, the corresponding fitness value is calculated according to the solution represented by its binary vector, and the individual with the highest fitness value is selected from the population as the optimal individual. Random perturbations are added to the binary vector of the optimal individual in each dimension to generate new offspring individuals.

[0100] For each child individual, according to a preset dynamic step size mechanism, the position of the child individual is adjusted to move toward the optimal individual. Exemplarily, this can be achieved by adjusting the amplitude of the disturbance or changing the mutation probability of certain bits. The above process is repeated until the position of each child individual is updated, and the second fitness value of each child individual after the update is calculated. The minimum second fitness value among the child individuals is compared with the first fitness value of the current optimal individual. If the second fitness value of a child individual is greater than the first fitness value of the optimal individual, the child individual is replaced with the new optimal individual.

[0101] In this embodiment, through the application of binary coding and dynamic step size, the algorithm can perform adaptive exploration in the solution space, so that the algorithm can adaptively adjust the dynamic step size of exploration according to the current search situation, and can perform both extensive exploration and fine search. By adding random perturbations to each dimension of the binary vector to generate offspring individuals, the algorithm increases the diversity of solutions, helps to explore neglected areas of the solution space, and reduces the risk of premature convergence of the algorithm. In summary, this embodiment achieves an effective balance between global and local searches, and enhances the search efficiency of the algorithm and the quality of the solution.

[0102] In an optional embodiment,

[0103] Combining the initial temperature and the temperature attenuation coefficient, determining whether the current optimal individual is accepted by a simulated annealing algorithm, and if so, determining a scheduling optimization solution according to the objective function includes:

[0104] For the optimal individual generated in each iteration in the fruit fly population, calculating the corresponding objective function value according to the objective function, and applying the simulated annealing algorithm;

[0105] For each objective function value, starting from the initial temperature, an alternative result is generated through neighborhood search and an alternative objective function value corresponding to the alternative result is calculated. If the alternative objective function value is better than the objective function value corresponding to the optimal individual, the alternative result is used to replace the current optimal individual, and the update is repeated by lowering the temperature until the preset minimum temperature is reached to obtain the scheduling optimization plan;

[0106] If the alternative objective function value is not better than the objective function value corresponding to the optimal individual, the alternative result is deleted. If the alternative results obtained by lowering the temperature search three times in a row are not selected, the temperature attenuation coefficient is doubled and repeatedly updated until the preset minimum temperature is reached to obtain the scheduling optimization plan.

[0107] Setting parameters such as initial temperature, minimum temperature, temperature attenuation coefficient, and the number of consecutive non-selected alternative results, obtaining the best individual generated in each iteration in the fruit fly population, determining the objective function value based on the best individual, and applying the simulated annealing algorithm based on the objective function value;

[0108] At the current temperature, a neighborhood search is performed to generate alternative results, the objective function value of the alternative results is calculated, and the objective function value of the alternative results is compared with that of the current optimal individual. If the alternative result is better than the current optimal individual, it is replaced with the current optimal individual. If the alternative result is not better than the current optimal individual, it is decided whether to accept the result based on a certain probability, and the temperature is lowered according to the temperature attenuation coefficient. Check whether the stopping condition is met (for example, the minimum temperature is reached). If it is met, the algorithm is terminated. If no alternative result is selected for three consecutive neighborhood searches, the temperature attenuation coefficient is doubled to accelerate the cooling process. When the preset minimum temperature is reached, the algorithm is stopped and the current optimal scheduling plan is returned.

[0109] In this embodiment, the simulated annealing algorithm can perform a global search in the solution space, which helps to find a better solution. The algorithm can accept a solution that is worse than the current solution during the search process, with a certain probability, which helps to avoid falling into the local optimal solution. By adjusting parameters such as the initial temperature and the temperature attenuation coefficient, the search range and speed of the algorithm can be controlled, which helps to better balance exploration and utilization, and improves the efficiency of scheduling optimization. By adopting the strategy of increasing the temperature attenuation coefficient when the alternative results are not selected for three consecutive times, it can effectively deal with difficult situations near the local optimal solution and improve the robustness and adaptability of the algorithm. In summary, this embodiment can effectively solve complex scheduling problems.

[0110] In an optional embodiment,

[0111] According to the preset dynamic step length, the offspring individuals are moved toward the optimal individuals as shown in the following formula:

[0112] ;

[0113] in, represents the position of individual n in the x direction at time step t+1, x n represents the position of individual n in the x direction at time step t, x best Indicates the position of the current optimal individual in the x direction, S x represents the dynamic step size in the x direction, δ x represents the random perturbation parameter in the x direction, rand x represents a random number in the x direction, represents the position of individual n in the y direction at the t+1 time step, y n represents the position of individual n in the x direction at time step t, S y Indicates the dynamic step size in the y direction, y best represents the position of the current optimal individual in the y direction, δ y represents the random perturbation parameter in the y direction, rand y Represents a random number in the y direction.

[0114] In this embodiment, the dynamic step size and moving in the direction of the optimal individual can promote the individual to move in the solution space towards the global optimal solution, thereby improving the global search ability and convergence speed of the algorithm. The random perturbation parameters and the introduction of a certain degree of randomness can help individuals jump out of the local optimal solution, increase the diversity of the algorithm, and thus make it more likely to find the global optimal solution. The dynamic step size allows individuals to adopt different step sizes at different positions, thereby improving the adaptability of the algorithm to different problems. In summary, this embodiment combines the strategies of dynamic step size and random perturbation, has good global search ability, avoids local optimal solutions, adaptability and robustness, and can effectively solve complex optimization problems.

[0115] Figure 2 Schematic diagram of the structure of the power plant optimization scheduling system based on the machine learning model according to the embodiment of the present invention. Figure 2 As shown, the system comprises:

[0116] The first unit is used to obtain historical power grid load demand, generate a data set based on the historical power grid load demand, divide the data set into a training set and a validation set, establish a memory network model based on deep learning, determine the structure of the memory network model and define a hyperparameter grid, adjust the hyperparameter value range of the memory network model based on the hyperparameter grid, cross-validate the memory network model based on the data set and determine the hyperparameter, and deploy the memory network model;

[0117] The second unit is used to determine the input parameters in the memory network model based on the memory network model, set the probability distribution, generate random samples through Monte Carlo simulation, randomly extract samples according to the set probability distribution for each input parameter and solve the model to obtain uncertainty, evaluate the power grid load loss risk based on the uncertainty, and construct an objective function with minimization of the power grid load loss risk and minimization of operating costs as goals;

[0118] The third unit is used to generate a fruit fly population and set the number of individuals based on the objective function, and to simulate the initial temperature and temperature attenuation coefficient of annealing, wherein each individual in the fruit fly population represents a solution of the objective function, and to obtain the optimal individual based on repeated iterations of the trained fruit fly algorithm, and to determine whether the current optimal individual is accepted by combining the initial temperature and the temperature attenuation coefficient through the simulated annealing algorithm, and if so, determine the scheduling optimization plan according to the objective function.

[0119] According to a third aspect of the embodiments of the present invention,

[0120] A virtual control system is provided, comprising:

[0121] The upper monitoring system uses MACS software and includes engineer station, operator station and history station functions;

[0122] The virtual controller deployed in the server consists of hardware resources, virtualization platform software, operating system and application software. It uses virtualization technology to deploy multiple controllers on one server, supports controller redundancy, and communicates with the host computer software through dual Ethernet.

[0123] System network, used for communication between the upper monitoring system and the virtual controller;

[0124] Control network, used for data communication between virtual controller and I / O modules;

[0125] Gateway, used to complete the protocol conversion between Ethernet and DP between the virtual controller and the I / O module, to realize the communication between the virtual controller and the I / O module. A single gateway supports 4 independent DP links, communicates with the virtual controller through redundant Ethernet, and supports the use of 5G network between the virtual controller and the gateway;

[0126] I / O modules for field data collection and execution.

[0127] For example, the virtualized control system adopts the virtualization technology based on DCS controller, which can be deployed on the virtual machine of the power plant's private cloud computing platform, and utilizes the computing power resources of the cloud server. The real-time operating system is loaded through the redundant virtual machine environment provided by the cloud server, and the operation of the DCS DPU in the virtualized environment is realized. The virtualized deployment of DCS control logic and monitoring screen is realized, and finally the interactive and comprehensive utilization of data and the rational allocation of resources are achieved.

[0128] The system consists of a virtual controller, control network, remote gateway and IO module, and is equipped with an operator station, engineer and history station and related supporting software. The system operation conforms to the conventional DCS operation habits. The virtual controller is a fully virtualized controller. The virtual controller supports deployment in a single virtual machine and can be installed on the cloud platform. The cloud platform allocates virtual machines for installing cloud controllers, network communications and their dependent environments. The logical configuration and data storage of the cloud controller are implemented in the virtual server. The logical operation relies on the virtual controller. The scanning cycle of the virtual controller of the entire cloud platform, the transmission cycle of the signal loop, and the reliability should meet the production delay requirements of the auxiliary workshop of the power plant (less than or equal to 200ms).

[0129] The upper monitoring system includes functions such as engineer station, operator station, history station, etc., and its overall functions are basically consistent with those of traditional DCS software.

[0130] Virtual controllers are deployed in servers and consist of hardware resources, virtualization platform software, operating systems, and application software. They perform the same functions as traditional controllers. Through virtualization technology, multiple controllers can be deployed on one server. Virtual controllers support controller redundancy and communicate with host computer software through dual Ethernet.

[0131] The whole system contains three kinds of networks: system network, control network and redundant network. System network: the network for DCS engineer station, operator station, history station and other stations to communicate with the controller. Control network: the network for data communication between the controller and IO module.

[0132] The gateway of this application is used to complete the protocol conversion between Ethernet and DP between the virtual controller and the IO module, and realize the communication between the cloud controller and the I / O module. A single gateway supports 4 independent DP links. Communication with the controller is through Ethernet communication, supporting two link redundancy; supporting the use of 5G network between the virtual controller and the gateway.

[0133] According to a fourth aspect of the embodiments of the present invention,

[0134] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0135] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power plant optimization scheduling method based on a machine learning model, characterized in that: include: Obtain historical power grid load demand, generate a data set based on the historical power grid load demand, divide the data set into a training set and a validation set, establish a memory network model based on deep learning, determine the structure of the memory network model and define a hyperparameter grid, adjust the hyperparameter value range of the memory network model based on the hyperparameter grid, cross-validate the memory network model based on the data set and determine the hyperparameters, and deploy the memory network model; Based on the memory network model, input parameters in the memory network model are determined, probability distribution is set, random samples are generated through Monte Carlo simulation, for each input parameter, samples are randomly drawn according to the set probability distribution and the model is solved to obtain uncertainty, the grid load loss risk is evaluated based on the uncertainty, and an objective function is constructed with minimization of the grid load loss risk and minimization of operating costs as goals; Based on the objective function, a fruit fly population is generated and the number of individuals is set, and the initial temperature and temperature attenuation coefficient of simulated annealing are set, wherein each individual in the fruit fly population represents a solution of the objective function, and the optimal individual is obtained by repeated iteration based on the trained fruit fly algorithm, and the initial temperature and the temperature attenuation coefficient are combined to determine whether the current optimal individual is accepted through the simulated annealing algorithm, and if so, determine the scheduling optimization plan according to the objective function; Based on the memory network model, the input parameters in the memory network model are determined, the probability distribution is set, random samples are generated through Monte Carlo simulation, for each input parameter, samples are randomly drawn according to the set probability distribution and the model is solved to obtain uncertainty, the grid load loss risk is evaluated based on the uncertainty, and the objective function is constructed with the minimization of the grid load loss risk and the minimization of the operating cost as the goal, including: For the deployed memory network model, obtaining input parameters of the memory network model, namely, weather conditions and economic indicators that affect power grid load demand; For the input parameters, determine the probability distribution corresponding to each parameter through discrete probability distribution or continuous probability distribution according to the data type, and set the corresponding normal distribution for the model error; For each input parameter, random sampling is performed based on a probability density function according to a set probability distribution to generate a random sample, the random sample is added to the memory network model for solving, a first prediction result is obtained, a first standard deviation corresponding to the first prediction result is calculated, and the first standard deviation is used as the uncertainty; Based on the uncertainty, combined with the predicted value of the load demand of the power grid and the rated capacity, the overload time is calculated, wherein the longer the overload time is, the greater the risk of power grid load loss is, and the objective function is constructed with the goal of minimizing the risk of power grid load loss and minimizing the operating cost; The objective function is constructed with the goal of minimizing the risk of grid loss and the operating cost as shown in the following formula: ; in, F represents the objective function value, E [] + represents an expected value operation, D(t) Indicates the time period t The predicted load demand, P gen,i (t) Indicates the time period t Neidi i The power generation of each power generation unit, T Indicates the total number of time periods. C i Indicates i The unit power generation cost of each power generation unit, ∑ t For all time periods t Sum the expected values ​​on i It indicates summing the power generated by the power generation units.

2. The method according to claim 1, characterized in that The method of obtaining historical power grid load demand, generating a data set based on the historical power grid load demand, dividing the data set into a training set and a validation set, establishing a memory network model based on deep learning, determining the structure of the memory network model and defining a hyperparameter grid, adjusting the hyperparameter value range of the memory network model based on the hyperparameter grid, cross-validating the memory network model based on the data set and determining the hyperparameter, and deploying the memory network model includes: Obtain historical power grid load demand, save it in the form of timestamp and specific value and generate a data set, and divide the data set into a training set and a validation set, wherein the training set accounts for 80% of the data set and the validation set accounts for 20%; Constructing a memory network model, determining the number of gated recurrent unit layers in the memory network model, determining the corresponding number of neurons for each gated recurrent unit layer according to computing requirements, and setting the values ​​of initial hyperparameters; Based on the initial hyperparameters, a parameter grid is defined, and each hyperparameter combination in the parameter grid is traversed. Based on the structure and task requirements of the memory network model, a value range of each hyperparameter is determined. Based on the value range, the performance of the memory network model is evaluated through cross-validation. The validation is repeated until all hyperparameter combinations are validated. The hyperparameter combination that enables the memory network model to have the highest performance is selected and recorded as the optimal hyperparameter combination. The memory network model is deployed based on the optimal hyperparameter combination.

3. The method according to claim 1, characterized in that Based on the objective function, a fruit fly population is generated and the number of individuals is set, and the initial temperature and temperature attenuation coefficient of simulated annealing are set, wherein each individual in the fruit fly population represents a solution of the objective function, and the optimal individuals are obtained by repeated iteration based on the trained fruit fly algorithm, including: Generate a fruit fly population and set the number of individuals, the initial temperature and temperature attenuation coefficient corresponding to the simulated annealing algorithm, and represent each individual in the fruit fly population as a binary vector through binary coding; For each individual in the fruit fly population, a corresponding first fitness value is calculated, an individual with the highest fitness value is selected as the optimal individual, and offspring individuals are randomly generated in the neighborhood by adding random perturbations in each dimension of the binary vector; For the offspring individuals, according to a preset dynamic step size, the offspring individuals are moved toward the optimal individual, and the positions of the offspring individuals are repeatedly updated until the position of each offspring individual is updated, and the second fitness value corresponding to each offspring individual after the update is calculated, and the minimum second fitness value corresponding to the individual in the offspring individuals is compared with the first fitness value corresponding to the optimal individual. If the second fitness value is greater than the first fitness value, the current offspring individual is used to replace the optimal individual.

4. The method according to claim 1, characterized in that Combining the initial temperature and the temperature attenuation coefficient, determining whether the current optimal individual is accepted by a simulated annealing algorithm, and if so, determining a scheduling optimization solution according to the objective function includes: For the optimal individual generated in each iteration in the fruit fly population, calculating the corresponding objective function value according to the objective function, and applying the simulated annealing algorithm; For each objective function value, starting from the initial temperature, an alternative result is generated through neighborhood search and an alternative objective function value corresponding to the alternative result is calculated. If the alternative objective function value is better than the objective function value corresponding to the optimal individual, the alternative result is used to replace the current optimal individual, and the update is repeated by lowering the temperature until the preset minimum temperature is reached to obtain the scheduling optimization plan; If the alternative objective function value is not better than the objective function value corresponding to the optimal individual, the alternative result is deleted. If the alternative results obtained by lowering the temperature search three times in a row are not selected, the temperature attenuation coefficient is doubled and repeatedly updated until the preset minimum temperature is reached to obtain the scheduling optimization plan.

5. The method according to claim 3, characterized in that: According to the preset dynamic step length, the offspring individuals are moved toward the optimal individuals as shown in the following formula: ; in, Represents an individual n exist t+1 The position in the x direction at the time step, x n Represents an individual n exist t Time step x Position in direction, x best Indicates that the current optimal individual is x Position in direction, S x Indicated in x Dynamic step size in direction, δ x express x The random perturbation parameter in the direction, rand x express x Random numbers in the direction, Represents an individual n exist t+1 Time step y Position in direction, y n Represents an individual n exist t Time step x Position in direction, S y Indicated in y Dynamic step size in direction, y best Indicates that the current optimal individual is y Position in direction, δ y express y The random perturbation parameter in the direction, rand y express y A random number in the direction.

6. A power plant optimization dispatching system based on a machine learning model, used to implement the method according to any one of claims 1 to 5, characterized in that: include: The first unit is used to obtain historical power grid load demand, generate a data set based on the historical power grid load demand, divide the data set into a training set and a validation set, establish a memory network model based on deep learning, determine the structure of the memory network model and define a hyperparameter grid, adjust the hyperparameter value range of the memory network model based on the hyperparameter grid, cross-validate the memory network model based on the data set and determine the hyperparameter, and deploy the memory network model; The second unit is used to determine the input parameters in the memory network model based on the memory network model, set the probability distribution, generate random samples through Monte Carlo simulation, randomly extract samples according to the set probability distribution for each input parameter and solve the model to obtain uncertainty, evaluate the power grid load loss risk based on the uncertainty, and construct an objective function with minimization of the power grid load loss risk and minimization of operating costs as goals; The third unit is used to generate a fruit fly population and set the number of individuals based on the objective function, and to simulate the initial temperature and temperature attenuation coefficient of annealing, wherein each individual in the fruit fly population represents a solution of the objective function, and to obtain the optimal individual based on repeated iterations of the trained fruit fly algorithm, and to determine whether the current optimal individual is accepted by combining the initial temperature and the temperature attenuation coefficient through the simulated annealing algorithm, and if so, determine the scheduling optimization plan according to the objective function.

7. A virtual control system, used to implement the method according to any one of claims 1 to 5, characterized in that: include: The upper monitoring system uses MACS software and includes engineer station, operator station and history station functions; The virtual controller deployed in the server consists of hardware resources, virtualization platform software, operating system and application software. It uses virtualization technology to deploy multiple controllers on one server, supports controller redundancy, and communicates with the host computer software through dual Ethernet. System network, used for communication between the upper monitoring system and the virtual controller; Control network, used for data communication between virtual controller and I / O modules; Gateway, used to complete the protocol conversion between Ethernet and DP between the virtual controller and the I / O module, to realize the communication between the virtual controller and the I / O module. A single gateway supports 4 independent DP links, communicates with the virtual controller through redundant Ethernet, and supports the use of 5G network between the virtual controller and the gateway; I / O modules for field data collection and execution.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.