Distributed power energy scheduling method and system based on adaptive penalty coefficient

Abstract

The invention discloses a distributed power energy scheduling method based on an adaptive penalty coefficient. The method comprises the following steps: S1, collecting power operation data and environment data of distributed energy nodes; s2, constructing an electric quantity prediction model based on a long short-term memory neural network, and training the prediction model by using the preprocessed power operation data and environment data to obtain an electric quantity prediction model; s3, combining the preprocessing model with the electric quantity prediction model, and deploying the combined model to edge equipment or a cloud to obtain a real-time electric quantity prediction model; s4, on the basis of real-time electric quantity prediction results of all distributed nodes, constructing a target function by taking the minimum scheduling cost as a target: solving the target function by adopting a Lagrangian relaxation method of an adaptive penalty coefficient to obtain a scheduling scheme; and S5, according to the scheduling scheme, generating a scheduling instruction, sending the scheduling instruction to the distributed nodes, and executing scheduling. According to the design, comprehensive scheduling of power generation, energy storage and power transmission is achieved, the calculation complexity is simplified, and the calculation amount is effectively reduced.

Description

Technical Field

[0001] The present invention relates to a distributed electric energy dispatching method and system based on an adaptive penalty coefficient, which are specifically suitable for reducing the dispatching calculation amount and improving the utilization rate of distributed energy. Background Art

[0002] With the continuous growth of global energy demand and the rapid development of new energy technologies, the electric energy system is undergoing a profound transformation from traditional centralized dispatching to distributed energy management. This transformation is not only an inevitable trend in the transformation of the energy industry, but also an important measure to build a green and low-carbon energy system. The traditional power system is centered on large centralized power plants, which transmit electricity from the power generation side to the load side through the transmission and distribution network. However, this centralized model is facing a series of severe challenges in the current complex energy demand and new energy access environment: (1) In order to cope with the environmental pollution and climate change problems caused by the use of fossil fuels, the world is accelerating the transformation of energy structure and vigorously developing clean energy represented by wind and solar energy. Although these distributed new energy power generation technologies have significant environmental benefits, they also have significant intermittency and volatility. For example, solar power generation is directly affected by the alternation of day and night and weather conditions, while wind power generation depends on changes in wind speed. This uncertainty makes it difficult to accurately predict the power generation output of new energy, which can easily cause an imbalance in the supply and demand of the power grid and even affect the stability of the power grid. More importantly, the distributed characteristics of new energy sources do not match the traditional centralized power grid structure. A single centralized dispatching system is difficult to cope with the complexity of multi-source distributed power systems, which limits the efficient absorption and flexible utilization of new energy sources. (2) With the accelerated advancement of urbanization and the continuous improvement of social electrification, the demand for power loads has become diversified and dynamic. For example, during peak hours, residential and industrial power demand may surge at the same time, while during off-peak hours, the power grid's power supply capacity may be idle. This significant demand volatility and uncertainty puts higher demands on the dispatching capacity of the power system, which not only increases the operating cost of the system, but also poses a threat to the stability of the power grid. In addition, modern society is increasingly dependent on power-driven smart devices and infrastructure, which further exacerbates the complexity of load changes and requires the system to have stronger dynamic response capabilities. (3) In a distributed energy system, each node may have the dual identities of "producer" and "consumer". For example, a distributed photovoltaic power station can output excess power to the power grid, but also needs to obtain power from the power grid to maintain its own operation. This two-way interactive mode makes the supply and demand balance problem of the system more complicated. Furthermore, distributed energy nodes are usually scattered over a wide geographical area, and their power generation efficiency, load demand, and ability to connect to the grid may vary significantly due to differences in geographical conditions and equipment performance. These factors not only increase the difficulty of system optimization, but also make it difficult for traditional centralized optimization algorithms to efficiently solve global problems. In addition, the demand for collaboration between distributed nodes has increased. How to achieve efficient energy scheduling without significantly increasing the communication and computing burden has become an important research direction in the design of distributed power systems.

[0003] In order to meet the above challenges, distributed energy power dispatching has emerged as an emerging technical solution. The distributed energy power dispatching method aims to achieve efficient management and coordination of multiple distributed energy nodes through intelligent means and optimization algorithms. Unlike traditional centralized power dispatching, the distributed dispatching method not only considers the power exchange and supply and demand balance between nodes, but also can flexibly adjust the working state of each node according to the actual needs and power generation capacity of different nodes, thereby achieving global optimization at the system level. The core idea of ​​distributed energy power dispatching is to transform power dispatching from a centralized "single-center control" mode to a distributed mode of "multi-center collaboration". Each node (such as distributed photovoltaic, wind power, energy storage equipment, load equipment, etc.) can make autonomous decisions based on local information and share information with neighboring nodes through local communication, thereby ensuring the global optimality of the system while avoiding the risks caused by single point failures or information transmission bottlenecks in centralized dispatching. Based on this, the present invention proposes a distributed power energy dispatching method and system based on the Lagrangian relaxation method with an adaptive penalty coefficient. First, real-time operating data of distributed nodes are collected, including power generation, load demand, energy storage status, weather information, etc. Then, based on the deep learning algorithm, the fluctuation trend of renewable energy power generation is predicted according to historical data and environmental parameters (such as weather and time period). After that, the Lagrangian relaxation method based on the adaptive penalty coefficient is used to realize the coordinated scheduling of distributed energy nodes, including distributed generation equipment, energy storage equipment and load equipment. Summary of the invention

[0004] The purpose of the present invention is to overcome the problem of large amount of distributed power dispatching calculation in the prior art, and to provide a distributed power energy dispatching method and system based on adaptive penalty coefficient to reduce the dispatching calculation amount and improve the utilization rate of distributed energy.

[0005] To achieve the above objectives, the technical solution of the present invention is:

[0006] In a first aspect, the present invention provides a distributed electric energy scheduling method based on an adaptive penalty coefficient, comprising the following steps:

[0007] S1. Collect power operation data and environmental data of distributed energy nodes, build a preprocessing model to preprocess the collected implementation data, and obtain preprocessed power operation data and environmental data;

[0008] S2. Build a power forecasting model based on long short-term memory neural network, and use the pre-processed power operation data and environmental data to train the forecasting model to obtain the power forecasting model;

[0009] S3. Combining the preprocessing model with the power prediction model and deploying them to the edge device or the cloud to obtain a real-time power prediction model, wherein the real-time power prediction model is used to collect real-time power operation data and environmental data for calculation and then output the real-time power prediction results of all distributed nodes;

[0010] S4. Based on the real-time power forecast results of all distributed nodes, an objective function is constructed with the minimum dispatch cost as the goal: the Lagrangian relaxation method with adaptive penalty coefficient is used to solve the objective function and obtain the dispatch plan;

[0011] S5. Generate scheduling instructions according to the scheduling plan and send them to distributed nodes to execute scheduling.

[0012] In S1, based on the installed instruments or IoT sensors, the power operation data and environmental data of the distributed energy nodes are collected in real time; the power operation data include: the generated power, load demand, energy storage status data of the distributed energy nodes and the monitoring voltage, frequency, and power transmission status data of the distributed energy nodes; the environmental data include: the weather information, temperature, and sunshine intensity information of the distributed energy nodes; the collected power operation data and environmental data are pre-processed by denoising and data completion to obtain the pre-processed power operation data and environmental data.

[0013] In S2, a power prediction model is constructed based on a long short-term memory neural network, and the prediction model is trained using pre-processed power operation data and environmental data to obtain a power prediction model; for node n:

[0014] Input feature construction: Using preprocessed historical power operation data: And the preprocessed historical environmental data {E(t-τ),...,E(t-1)} construct the time series input features:

[0015]

[0016] in, are the load requirements of node n at t-τ and t-1 respectively, are the power generation of node n at t-τ and t-1, respectively, and E(t-τ) and E(t-1) are the environmental data of node n at t-τ and t-1, respectively;

[0017] Load demand prediction: Use long short-term memory neural network to predict the load of node n at time t:

[0018]

[0019] Among them, f LSTM is the long short-term memory network prediction function, θ is the load prediction model parameter;

[0020] Power generation prediction: Use long short-term memory neural network to predict the renewable energy power generation of node n at time t:

[0021]

[0022] Among them, φ is the parameter of the power generation prediction model.

[0023] In S3, the preprocessing model is combined with the power prediction model and deployed to the edge device or the cloud to obtain a real-time power prediction model, which is used to collect real-time data and output real-time power prediction results of all N distributed nodes.

[0024] In S4, S41, based on the real-time power prediction results of all distributed nodes, an objective function is constructed with the minimum scheduling cost as the goal:

[0025]

[0026] Among them, C1 represents the power balance constraint; C2 and C3 represent the energy storage constraints; C4 is the transmission power constraint; represents the electricity production cost of node n, a n , b n 、c n All are coefficients of the electricity production cost formula; is the power generation of node n at time t; is the power demand of node n at time t, represents the energy storage cost, d n 、e n are coefficients of the energy storage cost formula; S n (t) is the remaining storage capacity of the energy storage device of node n at time t; They represent the upper and lower limits of the energy storage capacity of the energy storage device at node n respectively; represents the power transmission loss, where f n,j Represents the line transmission loss coefficient; P n,j (t) is the power transmission amount from node n to node j at time t. When its value is greater than 0, it means P n,j (t) is the amount of electricity transferred from node n to node j. When its value is less than 0, it means -P n,j (t) of electricity is transferred from node j to node n; η charge With η discharge are the charging and discharging efficiencies, respectively; They represent the charge and discharge amount of the energy storage device at node n at time t respectively; is the upper limit of power transmission from node n to j.

[0027] S42, solving the objective function P1 using a Lagrangian relaxation method with an adaptive penalty coefficient;

[0028] S421, reconstruct it based on the Lagrangian relaxation method of the adaptive penalty coefficient, and introduce the Lagrangian multiplier λ n Relax the power balance constraint:

[0029]

[0030] S422, then decompose the global problem into local problems of each node, and each node n optimizes its own variables:

[0031]

[0032] S423. Based on the optimization result of each node, the Lagrange multiplier λ of the adaptive penalty coefficient is obtained at the aggregation end. n The update rule is:

[0033]

[0034] Among them, A k Represents the constraint activation degree, which controls the increase of the update step size when the constraint deviates seriously to ensure rapid convergence; α (k) It is adaptively adjusted according to the number of iterations and the degree of constraint violation; definition is the difference between the actual value and the ideal value of the constraint function. k = 0 means the constraint has been satisfied and no adjustment is needed n , otherwise it means the constraint is violated and the system needs to n Update, α (k) is the update step size, which is used to adjust the optimization convergence speed, and k is the number of iterations. In order to further improve the convergence speed and avoid excessive adjustment steps, a dynamic adjustment mechanism is introduced to change the step size α (k) :

[0035]

[0036] where α 0 is the initial step length, β is the attenuation coefficient, which controls the speed of step length reduction to ensure gradual and stable convergence; constrain activation degree A k It can be defined as:

[0037]

[0038] When it is larger, it means the degree of default is higher, and the step length α (k) will increase, and the amplitude of the Lagrange multiplier update will also increase, thereby accelerating the adjustment of the imbalance. k Smaller means that the constraints are better satisfied and the step size gradually decreases;

[0039] Combining the above mechanisms, we get the Lagrange multiplier update rule:

[0040] Repeat S422 and S423 until the convergence condition |λ is met n When |≤ε, the iteration stops and the optimal solution is output as the scheduling solution.

[0041] In a second aspect, the present invention provides a distributed electric energy dispatching system based on an adaptive penalty coefficient, wherein the system is used to execute the aforementioned distributed electric energy dispatching method based on an adaptive penalty coefficient, and specifically comprises: a data acquisition and preprocessing module, an electric quantity prediction module, a real-time electric quantity prediction module, a dispatching calculation module, and a dispatching execution module;

[0042] Data collection and preprocessing module: used to collect power operation data and environmental data of distributed energy nodes, build a preprocessing model to preprocess the collected implementation data, and obtain preprocessed power operation data and environmental data;

[0043] Electricity prediction module: used to build an electricity prediction model based on long short-term memory neural network, and train the prediction model using pre-processed power operation data and environmental data to obtain an electricity prediction model;

[0044] Real-time power prediction module: used to combine the preprocessing model with the power prediction model and deploy them to edge devices or the cloud to obtain the real-time power prediction model, collect real-time power operation data and environmental data for calculation, and then output the real-time power prediction results of all distributed nodes;

[0045] Scheduling calculation module: used to construct the objective function based on the real-time power prediction results of all distributed nodes with the minimum scheduling cost as the goal: the Lagrangian relaxation method with adaptive penalty coefficient is used to solve the objective function and obtain the scheduling plan;

[0046] Scheduling execution module: used to generate scheduling instructions according to the scheduling plan and send them to distributed nodes to execute scheduling.

[0047] In the data acquisition and preprocessing module, the power operation data and environmental data of the distributed energy nodes are collected in real time based on the installed instruments or IoT sensors; the power operation data include: the power generation power, load demand, energy storage status data of the distributed energy nodes and the monitoring voltage, frequency, and power transmission status data of the distributed energy nodes; the environmental data include: weather information, temperature, and sunshine intensity information of the distributed energy nodes; the collected power operation data and environmental data are subjected to denoising and data completion preprocessing to obtain the preprocessed power operation data and environmental data.

[0048] In the power prediction module, a power prediction model is constructed based on the long short-term memory neural network, and the prediction model is trained using the preprocessed power operation data and environmental data to obtain the power prediction model; for node n:

[0049] Input feature construction: Using preprocessed historical power operation data: And the preprocessed historical environmental data {E(t-τ),...,E(t-1)} construct the time series input features:

[0050]

[0051] in, are the load requirements of node n at t-τ and t-1 respectively, are the power generation of node n at t-τ and t-1, respectively, and E(t-τ) and E(t-1) are the environmental data of node n at t-τ and t-1, respectively;

[0052] Load demand prediction: Use long short-term memory neural network to predict the load of node n at time t:

[0053]

[0054] Among them, f LSTM is the long short-term memory network prediction function, θ is the load prediction model parameter;

[0055] Power generation prediction: Use long short-term memory neural network to predict the renewable energy power generation of node n at time t:

[0056]

[0057] Among them, φ is the parameter of the power generation prediction model.

[0058] In the scheduling calculation module, the preprocessing model is combined with the power prediction model and deployed to the edge device or the cloud to obtain a real-time power prediction model, which is used to collect real-time data and output the real-time power prediction results of all N distributed nodes.

[0059] In the scheduling execution module, S41, based on the real-time power prediction results of all distributed nodes, an objective function is constructed with the minimum scheduling cost as the goal:

[0060]

[0061] Among them, C1 represents the power balance constraint; C2 and C3 represent the energy storage constraints; C4 is the transmission power constraint; represents the electricity production cost of node n, a n 、bn 、c n All are coefficients of the electricity production cost formula; is the power generation of node n at time t; is the power demand of node n at time t, represents the energy storage cost, d n 、e n are coefficients of the energy storage cost formula; S n (t) is the remaining storage capacity of the energy storage device of node n at time t; They represent the upper and lower limits of the energy storage capacity of the energy storage device at node n respectively; represents the power transmission loss, where f n,j Represents the line transmission loss coefficient; P n,j (t) is the power transmission amount from node n to node j at time t. When its value is greater than 0, it means P n,j (t) is the amount of electricity transferred from node n to node j. When its value is less than 0, it means -P n,j (t) of electricity is transferred from node j to node n; η charge With η discharge are the charging and discharging efficiencies, respectively; They represent the charge and discharge amount of the energy storage device at node n at time t respectively; is the upper limit of power transmission from node n to j.

[0062] S42, solving the objective function P1 using a Lagrangian relaxation method with an adaptive penalty coefficient;

[0063] S421, reconstruct it based on the Lagrangian relaxation method of the adaptive penalty coefficient, and introduce the Lagrangian multiplier λ n Relax the power balance constraint:

[0064]

[0065] S422, then decompose the global problem into local problems of each node, and each node n optimizes its own variables:

[0066]

[0067] S423. Based on the optimization result of each node, the Lagrange multiplier λ of the adaptive penalty coefficient is obtained at the aggregation end. n The update rule is:

[0068]

[0069] Among them, A k Represents the constraint activation degree, which controls the increase of the update step size when the constraint deviates seriously to ensure rapid convergence; α (k)It is adaptively adjusted according to the number of iterations and the degree of constraint violation; definition is the difference between the actual value and the ideal value of the constraint function. k = 0 means the constraint has been satisfied and no adjustment is needed n , otherwise it means the constraint is violated and the system needs to n Update, α (k) is the update step size, which is used to adjust the optimization convergence speed, and k is the number of iterations. In order to further improve the convergence speed and avoid excessive adjustment steps, a dynamic adjustment mechanism is introduced to change the step size α (k) :

[0070]

[0071] where α 0 is the initial step length, β is the attenuation coefficient, which controls the speed of step length reduction to ensure gradual and stable convergence; constrain activation degree A k It can be defined as:

[0072]

[0073] When it is larger, it means the degree of default is higher, and the step length α (k) will increase, and the amplitude of the Lagrange multiplier update will also increase, thereby accelerating the adjustment of the imbalance. k Smaller means that the constraints are better satisfied and the step size gradually decreases;

[0074] Combining the above mechanisms, we get the Lagrange multiplier update rule:

[0075] Repeat S422 and S423 until the convergence condition |λ is met n When |≤ε, the iteration stops and the optimal solution is output as the scheduling solution.

[0076] In a third aspect, the present invention provides a distributed electric energy dispatching device based on an adaptive penalty coefficient, comprising a memory and a processor, wherein the memory is used to store a computer program code and transmit the computer program code to the processor;

[0077] The processor is used to execute the aforementioned distributed electric energy scheduling method based on adaptive penalty coefficient according to the instructions in the computer program code.

[0078] In a fourth aspect, the present invention provides a computer program product, including a computer program, which is executed by a processor to implement the aforementioned distributed power energy scheduling method based on an adaptive penalty coefficient.

[0079] In a fifth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the aforementioned distributed electric power energy scheduling method based on an adaptive penalty coefficient is implemented.

[0080] Compared with the prior art, the present invention has the following beneficial effects:

[0081] 1. In a distributed power energy dispatching method based on an adaptive penalty coefficient, the present invention uses a distributed optimization algorithm to dispatch power energy, and optimizes its own variables for each node; then the Lagrange multiplier is updated at the aggregation end to achieve collaborative optimization combined with the local information of each node, significantly improving energy utilization efficiency and dispatching accuracy. Through the Lagrange relaxation method based on the adaptive penalty coefficient, the comprehensive dispatching of power generation, energy storage and power transmission is optimized, so that the system can adapt to complex load demands and energy supply changes.

[0082] 2. The distributed power energy dispatching method based on adaptive penalty coefficient of the present invention combines load forecasting with power generation planning, adopts advanced machine learning algorithm to predict load demand, and optimizes power generation and energy storage dispatching in real time, thereby achieving optimal dispatching of the system under variable load demand, reducing the response time and computational complexity of traditional dispatching methods.

[0083] 3. The distributed power energy dispatching method based on adaptive penalty coefficient of the present invention ensures that the system can operate safely and reliably in the face of uncertainty and emergencies by introducing multi-dimensional constraints (such as power generation limit, energy storage capacity, power transmission upper limit, etc.) and real-time adjustment strategies. In particular, under the volatility and instability of the new energy system, the stability and anti-interference ability of the system are improved.

[0084] 4. A distributed electric energy dispatching system based on an adaptive penalty coefficient of the present invention includes a data acquisition and preprocessing module, an electric quantity prediction module, a real-time electric quantity prediction module, a dispatching calculation module, and a dispatching execution module; the system is used to implement the steps of the distributed electric energy dispatching method based on an adaptive penalty coefficient as provided in any of the above technical solutions. Therefore, the system also includes all the beneficial effects of the distributed electric energy dispatching method based on an adaptive penalty coefficient as provided in any of the above technical solutions, which will not be repeated here.

[0085] 5. A distributed electric energy dispatching device based on an adaptive penalty coefficient of the present invention includes a processor and a memory, the memory is used to store computer program code, and transmit the computer program code to the processor, and the processor is used to execute the distributed electric energy dispatching method based on an adaptive penalty coefficient provided in any of the above technical solutions according to the instructions in the computer program code. Therefore, the device also includes all the beneficial effects of the distributed electric energy dispatching method based on an adaptive penalty coefficient provided in any of the above technical solutions, which will not be repeated here.

[0086] 6. A computer program product of the present invention, when executed by a processor, implements the steps of the distributed electric energy dispatching method based on adaptive penalty coefficient as provided in any of the above technical solutions. Therefore, the computer program product also includes all the beneficial effects of the distributed electric energy dispatching method based on adaptive penalty coefficient as provided in any of the above technical solutions, which will not be repeated here.

[0087] 7. The present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the distributed electric energy dispatching method based on the adaptive penalty coefficient provided in any of the above technical solutions are implemented. Therefore, the computer program product also includes all the beneficial effects of the distributed electric energy dispatching method based on the adaptive penalty coefficient provided in any of the above technical solutions, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 It is a flow chart of the method of the present invention.

[0089] Figure 2 It is a system module diagram of the present invention.

[0090] Figure 3 It is a schematic diagram of the device of the present invention. DETAILED DESCRIPTION

[0091] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0092] Embodiment 1:

[0093] See also Figure 1 , a distributed power energy scheduling method based on adaptive penalty coefficient, comprising the following steps:

[0094] S1. Based on the installed instruments or IoT sensors, collect the power operation data and environmental data of the distributed energy nodes, build a preprocessing model to preprocess the collected implementation data, and obtain the preprocessed power operation data and environmental data;

[0095] Assume that the distributed system contains N distributed nodes, and the power generation of node n at time t is The power demand of node n at time t is The remaining energy of the energy storage device of node n at time t is S n (t). Environmental variables E(t) represent light intensity, wind speed, and temperature, etc., and are used to predict the amount of electricity generated by renewable energy. The power transmission state from node n to node j is P n,j The invention minimizes the total dispatch cost by optimizing the power generation, energy storage device state variables, and power transmission power variables, which includes the power generation cost Energy storage costs and electricity transmission costs

[0096] The overall process of the use principle is as follows: Based on the installed meters or IoT sensors, the data such as the power generation, load demand and energy storage status of the distributed energy nodes are collected in real time. The real-time operation status is visualized through the monitoring platform, including the dynamic changes of voltage, frequency and power flow, which can be called by the dispatching system at any time. The data can be transmitted based on the existing 5G network.

[0097] The power operation data includes: the power generation power, load demand, energy storage status data of the distributed energy nodes and the monitoring voltage, frequency, and power transmission status data of the distributed energy nodes; the environmental data includes: the weather information, temperature, and sunshine intensity information of the distributed energy nodes; the collected power operation data and environmental data are pre-processed by denoising and data completion to obtain the pre-processed power operation data and environmental data.

[0098] S2. Build a power forecasting model based on long short-term memory neural network, and use the pre-processed power operation data and environmental data to train the forecasting model to obtain the power forecasting model;

[0099] For node n:

[0100] Input feature construction: Using preprocessed historical power operation data: And the preprocessed historical environmental data {E(t-τ),...,E(t-1)} construct the time series input features:

[0101]

[0102] in, are the load requirements of node n at t-τ and t-1 respectively, are the power generation of node n at t-τ and t-1, respectively, and E(t-τ) and E(t-1) are the environmental data of node n at t-τ and t-1, respectively;

[0103] Load demand prediction: Use long short-term memory neural network to predict the load of node n at time t:

[0104]

[0105] Among them, f LSTM is the long short-term memory network prediction function, θ is the load prediction model parameter;

[0106] Power generation prediction: Use long short-term memory neural network to predict the renewable energy power generation of node n at time t:

[0107]

[0108] Among them, φ is the parameter of the power generation prediction model.

[0109] S3. Combine the preprocessing model with the power prediction model and deploy them to the edge device or the cloud to obtain a real-time power prediction model, which is used to collect real-time power operation data and environmental data for calculation and then output the real-time power prediction results of all N distributed nodes.

[0110] S4. Based on the real-time power forecast results of all distributed nodes, an objective function is constructed with the minimum dispatch cost as the goal: the Lagrangian relaxation method with adaptive penalty coefficient is used to solve the objective function and obtain the dispatch plan;

[0111] S41. Based on the real-time power prediction results of all distributed nodes, an objective function is constructed with the minimum scheduling cost as the goal:

[0112]

[0113] Among them, C1 represents the power balance constraint; C2 and C3 represent the energy storage constraints; C4 is the transmission power constraint; represents the electricity production cost of node n, a n , b n 、c n All are coefficients of the electricity production cost formula; is the power generation of node n at time t; is the power demand of node n at time t, represents the energy storage cost, d n 、e n are coefficients of the energy storage cost formula; S n (t) is the remaining storage capacity of the energy storage device of node n at time t; They represent the upper and lower limits of the energy storage capacity of the energy storage device at node n respectively; represents the power transmission loss, where f n,jRepresents the line transmission loss coefficient; P n,j (t) is the power transmission amount from node n to node j at time t. When its value is greater than 0, it means P n,j (t) is the amount of electricity transferred from node n to node j. When its value is less than 0, it means -P n,j (t) of electricity is transferred from node j to node n; η charge With η discharge are the charging and discharging efficiencies, respectively; They represent the charge and discharge amount of the energy storage device at node n at time t respectively; is the upper limit of power transfer from node n to node j.

[0114] S42, using the Lagrangian relaxation method with adaptive penalty coefficient to solve the objective function P1,

[0115] S421, reconstruct it based on the Lagrangian relaxation method of the adaptive penalty coefficient, and introduce the Lagrangian multiplier λ n Relax the power balance constraint:

[0116]

[0117] S422, then decompose the global problem into local problems of each node, and each node n optimizes its own variables:

[0118]

[0119] S423, based on the optimization results of each node, update the Lagrange multiplier λ at the aggregation end n :

[0120]

[0121] where the definition is the difference between the actual value and the ideal value of the constraint function. k = 0 means the constraint has been satisfied and no adjustment is needed n , otherwise it means the constraint is violated and the system needs to n Update, α (k) is the update step size, which is used to adjust the optimization convergence speed, and k is the number of iterations. In order to further improve the convergence speed and avoid excessive adjustment steps, a dynamic adjustment mechanism is introduced to change the step size α (k) :

[0122]

[0123] where α 0is the initial step size, β is the attenuation coefficient, which controls the speed of step size reduction to ensure gradual and stable convergence; in order to ensure a faster response to severe constraint violations, the constraint activation is used to determine the adjustment speed of the penalty coefficient. The constraint activation A k is an indicator to measure the deviation from the constraint and can be defined as:

[0124]

[0125] When A k When it is larger, it means the degree of default is higher, and the step length is α (k) will increase, and the amplitude of the Lagrange multiplier update will also increase, thereby accelerating the adjustment of the imbalance. k Smaller means that the constraints are better satisfied and the step size gradually decreases;

[0126] Combining the above mechanisms, we get the Lagrange multiplier update rule for the adaptive penalty coefficient:

[0127]

[0128] Among them, A k Represents the constraint activation degree, which controls the increase of the update step size when the constraint deviates seriously to ensure rapid convergence; α (k) It is then adaptively adjusted according to the number of iterations and the degree of constraint violation;

[0129] Repeat S422 and S423 until the convergence condition |λ is met n When |≤ε, the iteration stops and the optimal solution is output as the scheduling solution.

[0130] S5. Generate scheduling instructions according to the scheduling plan and send them to the distributed nodes to execute scheduling, such as adjusting power generation, charging and discharging operations, power transmission paths, etc. through the intelligent controller.

[0131] By jointly optimizing the power generation, energy storage and power transmission of distributed energy nodes, the Lagrangian relaxation method with adaptive penalty coefficient is used to achieve energy balance between multiple nodes. By rationally scheduling power resources, the system not only reduces the loss of power transmission and the cost of power generation, but also improves the stability and energy efficiency of the power system. While optimizing power production, storage and transmission, it also effectively solves the coordination problem between nodes in the distributed system, improves the adaptability and operation efficiency of the power system, and can cope with different load demands and environmental changes, and has broad application prospects.

[0132] Embodiment 2:

[0133] See also Figure 2, a distributed electric energy dispatching system based on adaptive penalty coefficient, the system is used to execute the aforementioned distributed electric energy dispatching method based on adaptive penalty coefficient, specifically comprising: a data acquisition and preprocessing module, an electric quantity prediction module, a real-time electric quantity prediction module, a dispatching calculation module, and a dispatching execution module;

[0134] Data collection and preprocessing module: used to collect power operation data and environmental data of distributed energy nodes based on installed meters or IoT sensors, build a preprocessing model to preprocess the collected implementation data, and obtain preprocessed power operation data and environmental data;

[0135] Electricity prediction module: used to build an electricity prediction model based on long short-term memory neural network, and train the prediction model using pre-processed power operation data and environmental data to obtain an electricity prediction model;

[0136] Real-time power prediction module: used to combine the preprocessing model with the power prediction model and deploy them to edge devices or the cloud to obtain the real-time power prediction model, collect real-time power operation data and environmental data for calculation, and then output the real-time power prediction results of all distributed nodes;

[0137] Scheduling calculation module: used to construct the objective function based on the real-time power prediction results of all distributed nodes with the minimum scheduling cost as the goal: the Lagrangian relaxation method with adaptive penalty coefficient is used to solve the objective function and obtain the scheduling plan;

[0138] Scheduling execution module: used to generate scheduling instructions according to the scheduling plan and send them to distributed nodes to execute scheduling.

[0139] Embodiment 3:

[0140] See also Figure 3 A distributed electric energy dispatching device based on an adaptive penalty coefficient comprises a memory and a processor, wherein the memory is used to store computer program code and transmit the computer program code to the processor; the processor is used to execute the aforementioned distributed electric energy dispatching method based on an adaptive penalty coefficient according to the instructions in the computer program code.

[0141] Embodiment 4:

[0142] A computer program product includes a computer program, wherein the computer program is executed by a processor to implement the aforementioned distributed electric power energy scheduling method based on an adaptive penalty coefficient.

[0143] Embodiment 5:

[0144] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the aforementioned distributed electric power energy scheduling method based on an adaptive penalty coefficient.

Claims

1. A distributed power energy scheduling method based on adaptive penalty coefficient, characterized in that: The steps include: S1. Collect power operation data and environmental data of distributed energy nodes, build a preprocessing model to preprocess the collected implementation data, and obtain preprocessed power operation data and environmental data; S2. Build a power forecasting model based on long short-term memory neural network, and use the pre-processed power operation data and environmental data to train the forecasting model to obtain the power forecasting model; S3. Combining the preprocessing model with the power prediction model and deploying them to the edge device or the cloud to obtain a real-time power prediction model, wherein the real-time power prediction model is used to collect real-time power operation data and environmental data for calculation and then output the real-time power prediction results of all distributed nodes; S4. Based on the real-time power forecast results of all distributed nodes, an objective function is constructed with the minimum dispatch cost as the goal: the Lagrangian relaxation method with adaptive penalty coefficient is used to solve the objective function and obtain the dispatch plan; S5. Generate scheduling instructions according to the scheduling plan and send them to distributed nodes to execute scheduling.

2. A distributed power energy scheduling method based on adaptive penalty coefficient according to claim 1, characterized in that: In S1, based on the installed meters or IoT sensors, the power operation data and environmental data of the distributed energy nodes are collected in real time; The power operation data includes: the power generation power, load demand, energy storage status data of the distributed energy nodes and the monitoring voltage, frequency, and power transmission status data of the distributed energy nodes; the environmental data includes: the weather information, temperature, and sunshine intensity information of the distributed energy nodes; the collected power operation data and environmental data are pre-processed by denoising and data completion to obtain the pre-processed power operation data and environmental data.

3. The distributed power energy scheduling method based on adaptive penalty coefficient according to claim 1 is characterized in that: In S2, a power prediction model is constructed based on a long short-term memory neural network, and the prediction model is trained using pre-processed power operation data and environmental data to obtain a power prediction model; for node n: Input feature construction: Using preprocessed historical power operation data: And the preprocessed historical environmental data {E(t-τ),...,E(t-1)} construct the time series input features: in, are the load requirements of node n at t-τ and t-1 respectively, are the power generation of node n at t-τ and t-1, respectively, and E(t-τ) and E(t-1) are the environmental data of node n at t-τ and t-1, respectively; Load demand prediction: Use long short-term memory neural network to predict the load of node n at time t: Among them, f LSTM is the long short-term memory network prediction function, θ is the load prediction model parameter; Power generation prediction: Use long short-term memory neural network to predict the renewable energy power generation of node n at time t: Among them, φ is the parameter of the power generation prediction model.

4. The distributed power energy scheduling method based on adaptive penalty coefficient according to claim 1 is characterized by: In S3, the preprocessing model is combined with the power prediction model and deployed to the edge device or the cloud to obtain a real-time power prediction model, which is used to collect real-time data and output real-time power prediction results of all N distributed nodes.

5. The distributed power energy scheduling method based on adaptive penalty coefficient according to claim 1 is characterized in that: In S4, S41, based on the real-time power prediction results of all distributed nodes, an objective function is constructed with the minimum scheduling cost as the goal: Among them, C1 represents the power balance constraint; C2 and C3 represent the energy storage constraints; C4 is the transmission power constraint; represents the electricity production cost of node n, a n 、b n 、c n All are coefficients of the electricity production cost formula; is the power generation of node n at time t; is the power demand of node n at time t, represents the energy storage cost, d n 、e n are coefficients of the energy storage cost formula; S n (t) is the remaining storage capacity of the energy storage device of node n at time t; They represent the upper and lower limits of the energy storage capacity of the energy storage device at node n respectively; represents the power transmission loss, where f n,j Represents the line transmission loss coefficient; P n,j (t) is the power transmission amount from node n to node j at time t. When its value is greater than 0, it means P n,j (t) is the amount of electricity transferred from node n to node j. When its value is less than 0, it means -P n,j (t) of electricity is transferred from node j to node n; η charge With η discharge are the charging and discharging efficiencies, respectively; They represent the charge and discharge amount of the energy storage device at node n at time t respectively; is the upper limit of power transfer from node n to node j.

6. The distributed power energy dispatching method based on adaptive penalty coefficient according to claim 5 is characterized by: In S4, S42, solving the objective function P1 using a Lagrangian relaxation method with an adaptive penalty coefficient; S421, reconstruct it based on the Lagrangian relaxation method of the adaptive penalty coefficient, and introduce the Lagrangian multiplier λ n Relax the power balance constraint: S422, then decompose the global problem into local problems of each node, and each node n optimizes its own variables: S423. Based on the optimization result of each node, the Lagrange multiplier λ of the adaptive penalty coefficient is obtained at the aggregation end. n The update rule is: Among them, A k Represents the constraint activation degree, which controls the increase of the update step size when the constraint deviates seriously to ensure rapid convergence; α (k) It is adaptively adjusted according to the number of iterations and the degree of constraint violation; definition is the difference between the actual value and the ideal value of the constraint function. k = 0 means the constraint has been satisfied and no adjustment is needed n , otherwise it means the constraint is violated and the system needs to n Update, α (k) is the update step size, which is used to adjust the optimization convergence speed, and k is the number of iterations. In order to further improve the convergence speed and avoid excessive adjustment steps, a dynamic adjustment mechanism is introduced to change the step size α (k) : Where α0 is the initial step size, β is the attenuation coefficient, which controls the speed of step size reduction to ensure gradual and stable convergence; constrained activation degree A k It can be defined as: When it is larger, it means the degree of default is higher, and the step length α (k) will increase, and the amplitude of the Lagrange multiplier update will also increase, thereby accelerating the adjustment of the imbalance. k Smaller means that the constraints are better satisfied and the step size gradually decreases; Combining the above mechanisms, we get the Lagrange multiplier update rule: Repeat S422 and S423 until the convergence condition |λ is met n When |≤ε, the iteration stops and the optimal solution is output as the scheduling solution.

7. A distributed power energy dispatching system based on adaptive penalty coefficient, characterized in that: The system is used to execute the distributed electric energy dispatching method based on the adaptive penalty coefficient as described in any one of claims 1 to 6, specifically comprising: a data acquisition and preprocessing module, an electric quantity prediction module, a real-time electric quantity prediction module, a dispatching calculation module, and a dispatching execution module; Data collection and preprocessing module: used to collect power operation data and environmental data of distributed energy nodes, build a preprocessing model to preprocess the collected implementation data, and obtain preprocessed power operation data and environmental data; Electricity prediction module: used to build an electricity prediction model based on long short-term memory neural network, and train the prediction model using pre-processed power operation data and environmental data to obtain an electricity prediction model; Real-time power prediction module: used to combine the preprocessing model with the power prediction model and deploy them to edge devices or the cloud to obtain the real-time power prediction model, collect real-time power operation data and environmental data for calculation, and then output the real-time power prediction results of all distributed nodes; Scheduling calculation module: used to construct the objective function based on the real-time power prediction results of all distributed nodes with the minimum scheduling cost as the goal: the Lagrangian relaxation method with adaptive penalty coefficient is used to solve the objective function and obtain the scheduling plan; Scheduling execution module: used to generate scheduling instructions according to the scheduling plan and send them to distributed nodes to execute scheduling.

8. A distributed power energy dispatching device based on adaptive penalty coefficient, characterized in that: The method comprises a memory and a processor, wherein the memory is used to store computer program code and transmit the computer program code to the processor; The processor is used to execute the distributed electric power energy scheduling method based on adaptive penalty coefficient as described in any one of claims 1 to 6 according to the instructions in the computer program code.

9. A computer program product, comprising a computer program, characterized in that The computer program is executed by a processor to implement a distributed electric power energy scheduling method based on an adaptive penalty coefficient as described in any one of claims 1 to 6.

10. A computer storable medium, wherein a computer program is stored in the computer storable medium, characterized in that: The computer program is executed by a processor to implement a distributed electric power energy scheduling method based on an adaptive penalty coefficient as described in any one of claims 1 to 6.

Images