Multi-type load collaborative management and control method, system and device based on time domain analysis and medium

Abstract

The invention relates to the field of load management, in particular to a multi-type load collaborative management and control method, system and device based on time domain analysis and a medium. The method comprises two parts of load prediction and load coordination control, the load prediction integrates a load prediction algorithm of time domain analysis into load regulation and control, and the load value of a subsequent time step is predicted on the time level based on historical data. In the aspect of load coordination control, each load control object is regarded as an intelligent agent, then a load regulation and control mechanism based on reinforcement learning is constructed, rewards, state spaces and action spaces of reinforcement learning are designed, all action spaces meeting load constraint targets are found in a traversal mode, and therefore load coordination control is achieved. According to the method, independent load constraint and control are carried out on each load control object, the overall load control requirement is met, regulation and control distribution are carried out according to the characteristics of the loads, the overall requirement of load regulation and control is met, and the load distribution effect is maximized.

Description

Technical Field

[0001] The present invention relates to the field of load management, and specifically to a method, system, device and medium for collaborative control of multiple types of loads based on time domain analysis. Background Art

[0002] With the transformation of energy structures and the intensification of the contradiction between electricity supply and demand, load management, as a key component of the new power system, is gradually upgrading from traditional orderly electricity consumption to refined and intelligent methods. At the same time, new business models and models such as distributed power sources, electric vehicles, and virtual power plants are booming. Users are gradually shifting from traditional rigid, purely consumer-oriented demand to flexible, production-consumption-based demand, increasing the uncertainty of load characteristics and significantly improving adjustability.

[0003] Traditional load forecasting typically targets a single energy source (such as electricity). However, integrated energy systems encompass the coordinated demands of multiple energy sources, including electricity, heat, cooling, and gas. Existing models struggle to effectively capture the coupling relationships between these multiple energy loads. Furthermore, the randomness and time-varying nature of flexible loads, such as electric vehicles and distributed energy storage, increase the uncertainty of load characteristics. Existing methods are often trained based on fixed historical data and lack online learning mechanisms, making models susceptible to degradation over time. Furthermore, the high proportion of renewable energy connected to the grid has led to a decrease in system inertia, making it difficult for traditional control architectures to achieve real-time coordination between the power source, grid, load, and storage, exacerbating pressure on power balance and frequency regulation. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention proposes a method, system, device and medium for collaborative control of multiple types of loads based on time domain analysis. Based on the time domain analysis results and load weights, each load control object is individually constrained and controlled. While meeting the overall load control needs, the loads are separately regulated and distributed according to their characteristics to meet the overall needs of load control and maximize the effect of load distribution.

[0005] This invention addresses the application scenario where peak power loads require regulation of the following load devices to ensure safe and stable grid operation. These loads include adjustable, flexible loads such as air conditioners, energy storage systems, and charging stations. By reducing the power of these loads, coordinated control of the loads under load regulation is achieved to meet the overall load regulation requirements.

[0006] The collaborative control method for multiple types of loads described in the present invention includes two parts: one is load forecasting, and the other is load control allocation for multiple objects based on load information, so as to maximize the effectiveness of load control and reduce the impact on users.

[0007] Specifically, the multi-type load collaborative control method based on time domain analysis described in the present invention includes: Load forecasting: collect historical load data, obtain load curves of different load control objects, pre-process the data, build a load forecast database, analyze the load data, handle missing values ​​and outliers, and build a time domain analysis model to predict load information; For load coordination control, we first model the load control objects and regard each load control object as an intelligent agent. Then, based on the load forecast information, we design a load coordination control mechanism based on reinforcement learning. In the load coordination control mechanism, the reinforcement learning reward r is set as: , in represents the user satisfaction of the i-th load control object, and N represents the number of load control objects; Set the state space s of reinforcement learning to: , in represents the current load of load control object i, represents the load forecast value of load control object i at time t output by the time domain analysis model, represents the weight of load control object i; Set the action space a of reinforcement learning to a percentile, that is, a certain degree of reduction in the original load; By traversing, we can find all action spaces that meet the load constraint target. The load constraint target is ,in is the load control target of load control object i, is the total load after optimization, requiring .

[0008] Furthermore, the ARMA algorithm is used to construct a time domain analysis model. The prediction results output by the time domain analysis model are: , in 、 、 is the load value at time t, t-1, and tp, 、 、 are independent and identically distributed random variables at time t, t-1, and tp, respectively. is the autoregressive parameter, is the moving average parameter, p is the order of the autoregressive model, and q is the order of the moving average model. Based on historical load data, the model order is determined by the ACF autocorrelation coefficient and the PACF partial autocorrelation coefficient, and the order p of the autoregressive model and the order of the moving average model are determined. The least squares method or moment estimation method is used to determine the order of the model. 、 Parameters are set to complete the construction of the time domain analysis model.

[0009] Furthermore, when load coordinated control models the load control object, only the power and controllable power of the load control object are considered.

[0010] Furthermore, when traversing, the total load after optimization The total load is reduced by 5% relative to the original total load until all action spaces that meet the load constraint target are found.

[0011] Furthermore, the action value function of a deep convolutional neural network is defined as , where s represents the state space, a represents the dynamic space, Represents the neural network parameters at iteration number j; Action-value function neural network The loss function is defined as: , in and is the state or action of the next time step, are the parameters of the target network, and For two networks with exactly the same structure but different parameters, The parameters are several time steps ago Network parameters, It is an experience pool that stores the environment and action data at each time step t Deposit one In the process of learning and updating the network each time, a batch of data is randomly extracted from the experience pool and the parameters are updated; in the process of updating the reinforcement learning algorithm, the reinforcement learning algorithm is optimized by the gradient descent method, and the optimal load distribution strategy is obtained by minimizing the loss function. At each time step, the load is distributed through the action value function to maximize the load distribution effect.

[0012] Furthermore, the multiple types of loads include air conditioners, energy storage systems, and charging piles.

[0013] Furthermore, when collecting historical load data, data preprocessing includes eliminating unreasonable data. When analyzing data, missing values ​​and outliers are handled by using interpolation methods or directly deleting abnormal data in units of hours, days, or months.

[0014] The present invention also discloses a multi-type load forecasting system based on time domain analysis, comprising a load forecasting module and a load coordination control module; The load forecasting module is used to collect historical load data, obtain load curves of different load control objects, pre-process the data, build a load forecasting database, analyze the load data, process missing values ​​and outliers, and build a time domain analysis model to predict load information; The load coordination control module is used to control and distribute multiple load control objects based on load information to meet the requirements of overall load regulation; The working process of the load coordination control module is: First, the load control objects are modeled and each load control object is regarded as an intelligent agent. Then, based on the load forecast information, a load coordination control mechanism based on reinforcement learning is designed. In the load coordination control mechanism, the reinforcement learning reward r is set as: , in represents the user satisfaction of the i-th load control object, and N represents the number of load control objects; Set the state space s of reinforcement learning to: , in represents the current load of load control object i, represents the load forecast value of load control object i at time t output by the time domain analysis model, represents the weight of load control object i; Set the action space a of reinforcement learning to a percentile, that is, a certain degree of reduction in the original load; By traversing, we can find all action spaces that meet the load constraint target. The load constraint target is ,in is the load control target of load control object i, is the total load after optimization, requiring .

[0015] The present invention also discloses a multi-type load collaborative control device based on time domain analysis, including a processor and a memory storing program instructions, and the processor is configured to execute the multi-type load collaborative control method based on time domain analysis as described above when running the program instructions.

[0016] The present invention also discloses a storage medium storing program instructions, which, when running, execute the multi-type load collaborative control method based on time domain analysis as described above.

[0017] Beneficial effects of the present invention: The present invention combines the time domain analysis algorithm, and targets various loads such as energy storage, air conditioning, and charging piles. Based on the time domain analysis results and load weights, each load control object is individually constrained and controlled. While meeting the overall load management and control needs, the load is separately regulated and distributed according to its characteristics to meet the overall needs of load regulation, maximize the effect of load distribution, and achieve overall coordinated control of the load under load regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of the method described in Example 1. DETAILED DESCRIPTION

[0019] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0020] Example 1 During peak power consumption, due to line and overall load constraints, the load requirements of each consumer cannot be met. Therefore, it is necessary to reduce the overall load and impose load limits on each adjustable load to meet the overall load requirements. During load regulation, different load control targets exist, such as charging stations, energy storage systems, and other adjustable loads like air conditioners. These loads vary in importance and have different characteristics. For example, air conditioners and other heating equipment can maintain their effectiveness for a period of time after power is reduced, and the same applies to energy storage systems. Therefore, for different load targets, while ensuring that overall load regulation meets requirements, load regulation targets must be set for each control target to maximize load regulation effectiveness.

[0021] Based on this, this embodiment proposes a multi-type load collaborative control method based on time domain analysis, such as Figure 1 As shown, the load forecasting algorithm of time domain analysis is integrated into load regulation and distribution. In this way, in load regulation, the load demand of the control object in the future is taken into consideration, thereby increasing the flexibility and reliability of load control.

[0022] The load forecasting method based on time-domain analysis predicts load values ​​for subsequent time steps based on historical data. First, historical load data is collated and collected to obtain load curves for different load-controlled objects. This data is then preprocessed to remove inappropriate data and construct a complete load forecasting database. Second, the load data is analyzed in units of hours, days, or months, and missing and outliers are addressed using methods such as interpolation or direct deletion.

[0023] In the prediction model, an ARMA algorithm time domain analysis algorithm model is constructed to predict the load information. The prediction results output by the time domain analysis model are: , in 、 、 is the load value at time t, t-1, and tp, 、 、 are independent and identically distributed random variables at time t, t-1, and tp, respectively. is the autoregressive parameter, is the moving average parameter, p is the order of the autoregressive model, and q is the order of the moving average model. Based on historical load data, the model order is determined by the ACF autocorrelation coefficient and the PACF partial autocorrelation coefficient, and the order p of the autoregressive model and the order of the moving average model are determined. The least squares method or moment estimation method is used to determine the order of the model. 、 Parameters are set to complete the construction of the time domain analysis model.

[0024] In the load coordination control method, the load control object is first modeled. Here, only the key parameters (power and controllable power) of the load control object are considered, without considering the specific object differences. Only quantitative modeling is performed through load information, and each load control object is regarded as an intelligent entity. , i represents the load control objects, among which , N is the number of load control objects.

[0025] Deep reinforcement learning is a highly effective method for dealing with complex control systems. In particular, when combined with neural networks, the model can consider multi-dimensional parameters and implement more complex control logic. This embodiment designs a load coordination control mechanism based on reinforcement learning based on load forecast information. In the load coordination control mechanism, the reinforcement learning reward r is set to: , in represents the user satisfaction of the i-th load control object; In the state space s, all indicators that may affect user satisfaction are considered, so the state space s of reinforcement learning is set to: , in represents the current load of load control object i, represents the load forecast value of load control object i at time t output by the time domain analysis model, Represents the weight of load control object i, indicating the importance of this user.

[0026] In the action space a, because the purpose of load control is to reduce the overall load and readjust the load, the action space a of reinforcement learning is set to a percentile, that is, a certain degree of reduction in the original load.

[0027] In this embodiment, the original total load is set to P, and the optimized total load is ,set up . is the load condition of control object i, , is the load control target of the control object i. In the present invention, in order to reduce the learning curve, it is set to for The percentage reduction, in units of 5%, can be understood as , and so on. That is, the total load after optimization The total load is reduced by 5% relative to the original total load until all action spaces that meet the load constraint target are found.

[0028] After setting the reward, dynamic space and state space, traverse and find all action spaces that meet the load constraint target through traversal, so as to achieve load control. The load constraint target is ,in is the load control target of load control object i, is the total load after optimization, requiring .

[0029] In this embodiment, the action value function of a deep convolutional neural network is defined as , where s represents the state space, a represents the dynamic space, Represents the neural network parameters at iteration number j; Action-value function neural network The loss function is defined as: , in and is the state or action of the next time step, are the parameters of the target network, and For two networks with exactly the same structure but different parameters, The parameters are several time steps ago Network parameters are designed to maintain network stability.

[0030] In order to alleviate the problem of excessive data correlation caused by reinforcement learning exploration, an experience pool is established , at each time step t, the environment and action data Deposit one In the process of learning and updating the reinforcement learning algorithm network, a batch of data is randomly extracted from the experience pool and the parameters are updated. During the updating process of the reinforcement learning algorithm, the reinforcement learning algorithm is optimized by the gradient descent method, and the optimal load distribution strategy is obtained by minimizing the loss function. At each time step, the load is distributed through the action value function to maximize the load distribution effect.

[0031] In this embodiment, the action of action space a is to reduce the power of the load control object, that is, to reduce the load power by a set ratio of 5%, thereby finding all action spaces that meet the load constraint target and realizing load control.

[0032] Example 2 This embodiment discloses a multi-type load forecasting system based on time domain analysis, including a load forecasting module and a load coordination control module.

[0033] The load forecasting module is used to collect historical load data, obtain load curves of different load control objects, preprocess the data, build a load forecasting database, analyze the load data, process missing values ​​and outliers, and build a time domain analysis model to predict load information.

[0034] The load coordination control module is used to control and distribute multiple load control objects based on load information to meet the requirements of overall load regulation; The working process of the load coordination control module is: First, the load control objects are modeled and each load control object is regarded as an intelligent agent. Then, based on the load forecast information, a load coordination control mechanism based on reinforcement learning is designed. In the load coordination control mechanism, the reinforcement learning reward r is set as: , in represents the user satisfaction of the i-th load control object, and N represents the number of load control objects; Set the state space s of reinforcement learning to: , in represents the current load of load control object i, represents the load forecast value of load control object i at time t output by the time domain analysis model, represents the weight of load control object i; Set the action space a of reinforcement learning to a percentile, that is, a certain degree of reduction in the original load; By traversing, we can find all action spaces that meet the load constraint target. The load constraint target is ,in is the load control target of load control object i, is the total load after optimization, requiring .

[0035] Example 3 This embodiment discloses a multi-type load collaborative control device based on time domain analysis, including a processor and a memory storing program instructions. The processor is configured to execute the multi-type load collaborative control method based on time domain analysis as described in Example 1 when running the program instructions.

[0036] Example 4 This embodiment discloses a storage medium storing program instructions. When the program instructions are run, the method for collaborative control of multiple types of loads based on time domain analysis as described in Example 1 is executed.

[0037] The above description is only the basic principle and preferred embodiments of the present invention. Improvements and substitutions made by those skilled in the art based on the present invention fall within the protection scope of the present invention.

Claims

1. A method for collaborative control of multiple types of loads based on time domain analysis, characterized by: include: Load forecasting: collect historical load data, obtain load curves of different load control objects, pre-process the data, build a load forecast database, analyze the load data, handle missing values ​​and outliers, and build a time domain analysis model to predict load information; For load coordination control, we first model the load control objects and regard each load control object as an intelligent agent. Then, based on the load forecast information, we design a load coordination control mechanism based on reinforcement learning. In the load coordination control mechanism, the reinforcement learning reward r is set as: ， in represents the user satisfaction of the i-th load control object, and N represents the number of load control objects; Set the state space s of reinforcement learning to: ， in represents the current load of load control object i, represents the load forecast value of load control object i at time t output by the time domain analysis model, represents the weight of load control object i; Set the action space a of reinforcement learning to a percentile, that is, a certain degree of reduction in the original load; By traversing, we can find all action spaces that meet the load constraint target. The load constraint target is ,in is the load control target of load control object i, is the total load after optimization, requiring .

2. The multi-type load collaborative control method based on time domain analysis according to claim 1 is characterized by: The ARMA algorithm is used to build a time domain analysis model. The prediction results output by the time domain analysis model are: ， in 、 、 are the load values ​​at time t, t-1, and tp respectively, 、 、 are independent and identically distributed random variables at time t, t-1, and tp, respectively. is the autoregressive parameter, is the moving average parameter, p is the order of the autoregressive model, and q is the order of the moving average model. Based on historical load data, the model order is determined by the ACF autocorrelation coefficient and the PACF partial autocorrelation coefficient, and the order p of the autoregressive model and the order of the moving average model are determined. The least squares method or moment estimation method is used to determine the order of the model. 、 Parameters are set to complete the construction of the time domain analysis model.

3. The multi-type load collaborative control method based on time domain analysis according to claim 1 is characterized by: When load coordinated control models the load control object, only the power and controllable power of the load control object are considered.

4. The method for coordinated control of multiple loads based on time domain analysis according to claim 1 is characterized in that: When traversing, the total load after optimization The total load is reduced by 5% relative to the original total load until all action spaces that meet the load constraint target are found.

5. The multi-type load collaborative control method based on time domain analysis according to claim 1 is characterized in that: Define the action value function of a deep convolutional neural network as , where s represents the state space, a represents the dynamic space, Represents the neural network parameters at iteration number j; Action-value function neural network The loss function is defined as: ， in and is the state or action of the next time step, are the parameters of the target network, and For two networks with exactly the same structure but different parameters, The parameters are several time steps ago Network parameters, It is an experience pool that stores the environment and action data at each time step t Deposit one In the process of learning and updating the network each time, a batch of data is randomly extracted from the experience pool and the parameters are updated; in the process of updating the reinforcement learning algorithm, the reinforcement learning algorithm is optimized by the gradient descent method, and the optimal load distribution strategy is obtained by minimizing the loss function. At each time step, the load is distributed through the action value function to maximize the load distribution effect.

6. The multi-type load collaborative control method based on time domain analysis according to claim 1 is characterized in that: Multiple types of loads include air conditioners, energy storage systems, and charging piles.

7. The multi-type load collaborative control method based on time domain analysis according to claim 1 is characterized by: When collecting historical load data, data preprocessing includes eliminating unreasonable data. When analyzing data, missing values ​​and outliers are handled by using interpolation methods or directly deleting abnormal data in units of hours, days, or months.

8. A multi-type load forecasting system based on time domain analysis, characterized by: Including load forecasting module and load coordination control module; The load forecasting module is used to collect historical load data, obtain load curves of different load control objects, pre-process the data, build a load forecasting database, analyze the load data, process missing values ​​and outliers, and build a time domain analysis model to predict load information; The load coordination control module is used to control and distribute multiple load control objects based on load information to meet the requirements of overall load regulation; The working process of the load coordination control module is: First, the load control objects are modeled and each load control object is regarded as an intelligent agent. Then, based on the load forecast information, a load coordination control mechanism based on reinforcement learning is designed. In the load coordination control mechanism, the reinforcement learning reward r is set as: ， in represents the user satisfaction of the i-th load control object, and N represents the number of load control objects; Set the state space s of reinforcement learning to: ， in represents the current load of load control object i, represents the load forecast value of load control object i at time t output by the time domain analysis model, represents the weight of load control object i; Set the action space a of reinforcement learning to a percentile, that is, a certain degree of reduction in the original load; By traversing, we can find all action spaces that meet the load constraint target. The load constraint target is ,in is the load control target of load control object i, is the total load after optimization, requiring .

9. A multi-type load collaborative control device based on time domain analysis, comprising a processor and a memory storing program instructions, characterized in that: The processor is configured to execute the multi-type load collaborative control method based on time domain analysis as described in any one of claims 1 to 7 when running the program instructions.

10. A storage medium storing program instructions, characterized in that: When the program instructions are run, they execute the multi-type load collaborative control method based on time domain analysis as described in any one of claims 1 to 7.

Images