Chemical industrial park adjustable load dynamic identification and prediction method based on machine learning

Abstract

The invention relates to the field of electric power, in particular to a chemical industry park adjustable load dynamic identification and prediction method based on machine learning, and the method comprises the steps: classifying electrical loads based on an application scene; acquiring real-time load and historical load data according to the classified electrical loads, and constructing a park load curve prediction model; based on the adjustable load prediction dynamic identification and prediction system and the load data, constructing an optimal scheduling model; constructing a power flow topology net rack according to the net rack information; and inputting the predicted load value output by the park load curve prediction model and the optimal regulation and control value output by the optimal scheduling model into the power flow topology grid, and outputting a regulation and control scheme. According to the scheme, the trend of the load in the park can be predicted, and the dispatching capability of the power grid can be optimized. And the difference between the actual condition of load regulation and the scheduling demand is subjected to fitting correction, so that the problem of low investment utilization rate is solved, the operation of the park is better served, and the load capacity of the power grid is ensured.

Description

Technical Field

[0001] The present invention relates to the field of electric power, and in particular to a method for dynamically identifying and predicting adjustable loads in a chemical park based on machine learning. Background Art

[0002] Load curve forecasting is an extremely important part of the electricity market. The forecast results of the load curve provide a basis for decision-making on installed power capacity and resource scheduling. The accuracy of the forecast directly determines the compliance and rationality of grid layout, investment and scheduling.

[0003] With the continuous development of the power market, the traditional load curve forecasting level can no longer meet the growing needs of the power market. In addition, with the development of social economy, more and more factors affect the planning, operation and dispatch of power grids, and it is difficult to use a single forecasting method to ensure the accuracy of load curve forecasting.

[0004] In the chemical park scenario, since the power system must always maintain a balance between supply and demand, the traditional approach is to increase the output of generators during peak load demand. Most peak load units are thermal power units with high costs and low environmental protection. However, peak load periods are often short-lived, and the increased power generation and transmission and distribution investment to meet this part of the demand has a low utilization rate. In view of the above problems, the optimization and dispatching of the power grid is also difficult to achieve, which brings challenges to the development of the power grid.

[0005] Therefore, those skilled in the art provide a method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning to solve the problems raised in the above background technology. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning, comprising the following steps:

[0007] S1. Based on the application scenarios of the power grid in the chemical park, the power loads in the park are classified according to their characteristics;

[0008] S2. Collect the real-time load data and historical load data of the park according to the classified power load, and build a load curve prediction model for the park;

[0009] S3. Based on the adjustable load forecasting dynamic identification and prediction system and the above load data, an optimization scheduling model in the chemical park is constructed;

[0010] S4. Construct a tidal topology grid based on the grid information of the chemical park;

[0011] S5. Input the predicted load value output by the park load curve prediction model and the optimal control value output by the optimization scheduling model into the power flow topology grid, and output a control plan.

[0012] Preferably, the loads in the chemical park include: controllable loads for industrial equipment, including generator sets, gas turbines, gas boilers, refrigerators, standby generators, etc.; adjustable loads for user electricity demand, including but not limited to air conditioning systems and electric vehicle charging needs.

[0013] Preferably, the construction of the optimization scheduling model includes the following steps:

[0014] Combining the real-time operation data of the power grid and the offline operation model to form a decision model;

[0015] Using power grid input and output parameters as input and output parameters of the optimization dispatch model;

[0016] Using controllable load constraint parameters, adjustable load constraint parameters, storage constraint parameters and system constraint parameters as constraint conditions of the optimization scheduling model;

[0017] Taking economic dispatch parameters, environmental impact parameters and safe operation risk parameters as the objective function of the optimization dispatch model;

[0018] Collect and analyze the real-time operation data to obtain current operation status information, last moment status information and corresponding strategy return value to form an experience sample;

[0019] The decision model is trained by the experience samples to update corresponding parameters, thereby training an optimized scheduling model.

[0020] Preferably: the park load curve prediction model adopts LSTM for time series data prediction; after extracting characteristic values ​​from real-time load data and historical load data, normalization is performed and then data division is performed.

[0021] Preferably: the step of generating the control scheme includes: inputting the regional grid information into the regional load curve prediction model to obtain the predicted load value of the controllable load and the adjustable load reactive compensation in the region;

[0022] The predicted load values ​​are respectively input into the optimization scheduling model, and the optimization scheduling model uses the scheduling value as a decision variable to perform optimization and solve, and outputs the optimal scheduling value;

[0023] The optimal scheduling value is input into the power flow calculation topology grid to perform power flow calculation, so as to obtain the charging and discharging power and time of each compensation point in the next 24 hours, and the power flow calculation result of the first hour is output as the control scheme.

[0024] Preferably: the adjustable load forecasting dynamic identification and prediction system comprises a dispatching subunit, a dispatching demand unit, an adjustable load forecasting unit, a dispatching algorithm output unit and an interactive feedback unit;

[0025] The dispatching subunit is used to obtain the grid information and real-time load information of the regional power grid, and obtain the dispatching information and the load demand information in the chemical park;

[0026] The dispatch information includes load response starting time, expected duration and planned control amount information;

[0027] The dispatching demand unit generates a real-time response demand based on the dispatching information, the demand information of the load in the chemical park, and the power demand information feedback issued by the user;

[0028] The adjustable load prediction unit is used to perform modeling and parameter configuration according to the resource load situation in the park to obtain an adjustable load;

[0029] Adjustable load refers to the load whose usage time can be changed but the usage power cannot be changed during the operation of industrial equipment in the chemical park. The adjustable load can be expressed as follows:

[0030]

[0031] In the above formula, i∈N, N represents the total number of adjustable load devices, P i cu (t) represents the actual load of the adjustable load at time t, Indicates the regulation rate of the adjustable load. When it is 0, it means that at the current moment, the adjustable load does not change. It means that industrial equipment i must ensure production requirements or maintain normal working conditions.

[0032] The scheduling algorithm output unit is interactively connected with the adjustable load prediction unit, predicts and calculates the task allocation under different load requirements based on the adjustable load, controllable load, and real-time response requirements, and sends the task allocation to the scheduling center for load adjustment;

[0033] The interactive feedback unit is located at the top of the system and interacts with the scheduling algorithm output unit to realize feedback display of scheduling information and load regulation.

[0034] Preferably: the scheduling algorithm output unit also performs fitting correction based on the actual load regulation situation and the difference between scheduling requirements to obtain load regulation information feedback, and sends the load regulation information feedback to the scheduling requirement unit.

[0035] Preferably, the difference is adjusted by closed-loop control through PID negative feedback.

[0036] Technical effects and advantages of the present invention:

[0037] Based on the input and output parameters related to the power grid, an optimization scheduling model is established. It is possible to combine the specific conditions of the power grid, modify the model according to actual results, and improve the ability of optimization scheduling.

[0038] Using the load curve to predict trends and then reversely correcting the load curve based on traditional load forecasting results can not only predict the load trend within the park, but also optimize the dispatching capacity of the power grid.

[0039] Aiming at adjustable loads, controllable loads and real-time response needs, the task allocation under different load demands is predicted and calculated, and the difference between the actual load regulation and the scheduling demand is fitted and corrected, which solves the problem of low investment utilization, better serves the operation of the park, and ensures the load capacity of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the system composition of a method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning provided in an embodiment of the present application;

[0041] Figure 2 It is a schematic diagram of the steps of a method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

[0043] Example 1

[0044] See also Figures 1-2 In this embodiment, a method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning is provided, comprising the following steps:

[0045] S1. Based on the application scenarios of the power grid in the chemical park, the power loads in the park are classified according to their characteristics;

[0046] S2. Collect the real-time load data and historical load data of the park according to the classified power load, and build a load curve prediction model for the park;

[0047] S3. Based on the adjustable load forecasting dynamic identification and prediction system and the above load data, an optimization scheduling model in the chemical park is constructed;

[0048] S4. Construct a tidal topology grid based on the grid information of the chemical park;

[0049] S5. Input the predicted load value output by the park load curve prediction model and the optimal control value output by the optimization scheduling model into the power flow topology grid, and output a control plan.

[0050] Among them, the loads in the chemical park include: controllable loads for industrial equipment, including generator sets, gas turbines, gas boilers, refrigerators, standby generators, etc.; adjustable loads for user electricity needs, including but not limited to air-conditioning systems and electric vehicle charging needs.

[0051] In step S2, the park load curve prediction model adopts LSTM for time series data prediction; after extracting characteristic values ​​from real-time load data and historical load data, normalization is performed and data division is performed.

[0052] In step S3, the construction of the optimization scheduling model includes the following steps:

[0053] Combining the real-time operation data of the power grid and the offline operation model to form a decision model;

[0054] Using power grid input and output parameters as input and output parameters of the optimization dispatch model;

[0055] Using controllable load constraint parameters, adjustable load constraint parameters, storage constraint parameters and system constraint parameters as constraint conditions of the optimization scheduling model;

[0056] Taking economic dispatch parameters, environmental impact parameters and safe operation risk parameters as the objective function of the optimization dispatch model;

[0057] Collect and analyze the real-time operation data to obtain current operation status information, last moment status information and corresponding strategy return value to form an experience sample;

[0058] The decision model is trained by the experience samples to update corresponding parameters, thereby training an optimized scheduling model.

[0059] In step S5, the step of generating the control scheme includes: inputting the regional grid information into the regional load curve prediction model to obtain the predicted load value of the controllable load and the adjustable load reactive compensation in the region;

[0060] The predicted load values ​​are respectively input into the optimization scheduling model, and the optimization scheduling model uses the scheduling value as a decision variable to perform optimization and solve, and outputs the optimal scheduling value;

[0061] The optimal scheduling value is input into the power flow calculation topology grid to perform power flow calculation, so as to obtain the charging and discharging power and time of each compensation point in the next 24 hours, and the power flow calculation result of the first hour is output as the control scheme.

[0062] As another embodiment, the present solution also provides an adjustable load forecasting dynamic identification and prediction system including a scheduling subunit, a scheduling demand unit, an adjustable load forecasting unit, a scheduling algorithm output unit, and an interactive feedback unit, wherein the above units are arranged in order from bottom to top;

[0063] The dispatching subunit is used to obtain the grid information and real-time load information of the regional power grid, and obtain the dispatching information and the load demand information in the chemical park;

[0064] The dispatch information includes load response starting time, expected duration and planned control amount information;

[0065] The dispatching demand unit generates a real-time response demand based on the dispatching information, the demand information of the load in the chemical park, and the power demand information feedback issued by the user;

[0066] The adjustable load prediction unit is used to perform modeling and parameter configuration according to the resource load situation in the park to obtain an adjustable load;

[0067] Adjustable load refers to the load whose usage time can be changed but the usage power cannot be changed during the operation of industrial equipment in the chemical park. The adjustable load can be expressed as follows:

[0068]

[0069] In the above formula, i∈N, N represents the total number of adjustable load devices, P i cu (t) represents the actual load of the adjustable load at time t, Indicates the regulation rate of the adjustable load. When it is 0, it means that at the current moment, the adjustable load does not change. It means that industrial equipment i must ensure production requirements or maintain normal working conditions.

[0070] The scheduling algorithm output unit is interactively connected with the adjustable load prediction unit, predicts and calculates the task allocation under different load requirements based on the adjustable load, controllable load, and real-time response requirements, and sends the task allocation to the scheduling center for load adjustment;

[0071] In this unit, the scheduling algorithm output unit also performs fitting correction based on the actual load regulation and the difference between the scheduling requirements, obtains load regulation information feedback, and sends the adjustment information feedback to the scheduling requirement unit. The difference is adjusted by closed-loop control through PID negative feedback.

[0072] The interactive feedback unit is located at the top of the system and interacts with the scheduling algorithm output unit to realize feedback display of scheduling information and load regulation.

[0073] Aiming at adjustable loads, controllable loads and real-time response needs, the task allocation under different load demands is predicted and calculated, and the difference between the actual load regulation and the scheduling demand is fitted and corrected, which solves the problem of low investment utilization, better serves the operation of the park, and ensures the load capacity of the power grid.

[0074] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. A method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning, characterized in that: The following steps are involved: S1. Based on the application scenarios of the power grid in the chemical park, the power loads in the park are classified according to their characteristics; S2. Collect the real-time load data and historical load data of the park according to the classified power load, and build a load curve prediction model for the park; S3. Based on the adjustable load forecasting dynamic identification and prediction system and the above load data, an optimization scheduling model in the chemical park is constructed; S4. Construct a tidal topology grid based on the grid information of the chemical park; S5. Input the predicted load value output by the park load curve prediction model and the optimal control value output by the optimization scheduling model into the power flow topology grid, and output a control plan.

2. The method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning according to claim 1 is characterized in that: The loads in the chemical park include: uncontrollable loads for industrial equipment, including generator sets, gas turbines, gas boilers, refrigerators, and standby generators; and controllable loads for user electricity needs, including air-conditioning systems and electric vehicle charging needs.

3. The method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning according to claim 1 is characterized in that: The construction of the optimization scheduling model includes the following steps: Combining the real-time operation data of the power grid and the offline operation model to form a decision model; Using power grid input and output parameters as input and output parameters of the optimization dispatch model; Using uncontrollable load constraint parameters, controllable load constraint parameters, storage constraint parameters and system constraint parameters as constraint conditions of the optimization scheduling model; Taking economic dispatch parameters, environmental impact parameters and safe operation risk parameters as the objective function of the optimization dispatch model; Collect and analyze the real-time operation data to obtain current operation status information, last moment status information and corresponding strategy return value to form an experience sample; The decision model is trained by the experience samples to update corresponding parameters, thereby training an optimized scheduling model.

4. The method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning according to claim 1 is characterized in that: The park load curve prediction model adopts LSTM for time series data prediction; after extracting characteristic values ​​from real-time load data and historical load data, normalization is performed and then data division is performed.

5. The method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning according to claim 1 is characterized in that: The steps of generating the control scheme include: Inputting the regional grid information into the regional load curve prediction model to obtain the predicted load value of the uncontrollable load and the reactive power compensation of the controllable load in the region; The predicted load values ​​are respectively input into the optimization scheduling model, and the optimization scheduling model uses the scheduling value as a decision variable to perform optimization and solve, and outputs the optimal scheduling value; The optimal scheduling value is input into the power flow calculation topology grid to perform power flow calculation, so as to obtain the charging and discharging power and time of each compensation point in the next 24 hours, and the power flow calculation result of the first hour is output as the control plan.

6. The method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning according to claim 1 is characterized in that: The adjustable load forecasting dynamic identification and prediction system includes a dispatching subunit, a dispatching demand unit, an adjustable load forecasting unit, a dispatching algorithm output unit, and an interactive feedback unit; The dispatching subunit is used to obtain the grid information and real-time load information of the regional power grid, and obtain the dispatching information and the load demand information in the chemical park; The dispatching demand unit generates a real-time response demand based on the dispatching information, the demand information of the load in the chemical park, and the power demand information feedback issued by the user; The adjustable load prediction unit is used to perform modeling and parameter configuration according to the resource load situation in the park to obtain a controllable load; The scheduling algorithm output unit is interactively connected with the adjustable load prediction unit, predicts and calculates the task allocation under different load demands based on the controllable load, uncontrollable load, and real-time response demand, and sends the task allocation to the scheduling center for load adjustment; The interactive feedback unit is located at the top of the system and interacts with the scheduling algorithm output unit to realize feedback display of scheduling information and load regulation.

7. The method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning according to claim 6 is characterized in that: The dispatch information includes load response starting time, expected duration and planned control amount information.

8. The method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning according to claim 6 is characterized in that: Controllable load refers to the load whose usage time can be changed but the usage power cannot be changed during the operation of industrial equipment in the chemical park. The controllable load can be expressed as follows: In the above formula, , N represents the total number of adjustable load devices, It represents the actual load of the adjustable load at time t. Indicates the regulation rate of the adjustable load. When it is 0, it means that at the current moment, the adjustable load does not change. It means that industrial equipment i must ensure production requirements or maintain normal working conditions.

9. The method for dynamic identification and prediction of adjustable loads in a chemical park based on machine learning according to claim 6, characterized in that: The dispatch algorithm output unit also performs fitting correction based on the actual load regulation situation and the difference between the dispatch requirements to obtain load regulation information feedback, and sends the load regulation information feedback to the dispatch requirement unit; The difference is adjusted through PID negative feedback for closed-loop control.

Images