Multi-resource aggregation regulation and control method and system oriented to source network load storage

Abstract

The invention provides a source network load storage oriented multi-resource aggregation regulation and control method and system, and relates to the technical field of electric power intelligent scheduling, and the method comprises the steps: obtaining the real-time electric power operation data of each electric power decision point to be subjected to optimal scheduling in a power grid system; constructing power operation time sequence data for each power decision point, and inputting the power operation time sequence data into a preset power operation prediction model to determine corresponding estimated power operation data; constructing a target graph network based on the estimated power operation data corresponding to each power decision point; and inputting the target graph network into a preset reinforcement learning model, so that a corresponding target management scheduling scheme is determined for each node in the target graph network based on the reinforcement learning model. Therefore, by performing fusion analysis on the data of various power decision points of the source network load storage in the power grid system, multi-resource aggregation regulation and control oriented to the source network load storage are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power intelligent dispatching, and in particular to a multi-resource aggregation control method and system for source, grid, load and storage. Background Art

[0002] With the development of smart grids and the large-scale access of renewable energy, the power system is facing unprecedented regulation challenges. In traditional power systems, power sources, power grids, loads and energy storage devices usually operate independently, lacking effective coordination and optimization mechanisms. With the widespread application of renewable energy such as wind power and solar energy, the volatility and uncertainty of the power system have increased significantly.

[0003] In addition, the rapid changes in power market demand and the improvement of safe operation standards of power grids have also put forward higher requirements for dispatching systems. Existing control systems often fail to achieve effective coordination between sources (power generation), networks (power transmission and distribution), loads (power consumption), and storage (energy storage), resulting in low resource utilization and slow response speed, which cannot meet the needs of rapid regulation and efficient operation.

[0004] In response to the above problems, the industry has not yet proposed a better technical solution. Summary of the invention

[0005] The present invention provides a multi-resource aggregation control method, system, electronic device and storage medium for source, grid, load and storage, which are used to at least solve the problem that the source, grid, load and storage cannot be considered in a coordinated manner in the prior art, resulting in low resource utilization or slow response speed in the current scheduling scheme.

[0006] The present invention provides a multi-resource aggregation control method for source, grid, load and storage, comprising: obtaining real-time power operation data of each power decision point to be optimized in the power grid system; each power decision point includes a power generation source, a transmission network, a load aggregation point and an energy storage device; the power generation source includes a thermal power generation source and a renewable power generation source; the real-time power operation data includes the power generation of the power generation source, the load rate of the transmission network, the load demand curve and the energy storage use curve for each power decision point, and the real-time power operation data also includes weather forecast information, power supply and distribution cost function and power grid peak and valley electricity price floating curve of the corresponding power decision point; for each power decision point, constructing power operation time series data according to the real-time power operation data corresponding to the power decision point and the historical power operation data corresponding to the first preset time period in the past, and inputting the power operation time series data into a preset power operation prediction model to determine the corresponding estimated power operation data; based on the estimated power operation data corresponding to each power decision point The target graph network is constructed based on the power operation data; each node in the target graph network is used to represent the corresponding power decision point, each edge connection is used to connect the power decision points with an associated relationship, and each edge connection is also configured with a corresponding node association degree; the target graph network is input into a preset reinforcement learning model, so that based on the reinforcement learning model, a corresponding target management scheduling scheme is determined for each node in the target graph network; the state space of the reinforcement learning model is defined by the real-time power operation data of each node, the estimated power operation data and the overall scheduling requirements of the power grid, and the action space of the reinforcement learning model is defined by the power generation adjustment operation, the transmission path opening and closing operation of the transmission network, the load aggregation point control plan and the energy storage equipment charging and discharging operation; the overall scheduling requirements of the power grid include maximizing the utilization rate of renewable energy and minimizing the power supply operation cost of the power grid system when the total power supply of the power grid system is greater than the total load demand of each load aggregation point.

[0007] The present invention also provides a multi-resource aggregation control system for source, grid, load and storage, including: a real-time data acquisition unit, configured to acquire real-time power operation data of each power decision point to be optimized and scheduled in the power grid system; each power decision point includes a power generation source, a transmission network, a load aggregation point and an energy storage device; the power generation source includes a thermal power generation source and a renewable power generation source; the real-time power operation data includes the power generation of the power generation source, the load rate of the transmission network, the load demand curve and the energy storage usage curve for each power decision point, and the real-time power operation data also includes weather forecast information, power supply and distribution cost function and power grid peak and valley electricity price floating curve of the corresponding power decision point; an operation data prediction unit, configured to construct power operation time series data for each power decision point according to the real-time power operation data corresponding to the power decision point and the historical power operation data corresponding to the first preset time period in the past, and input the power operation time series data into a preset power operation prediction model to determine the corresponding estimated power operation data; a graph network construction unit, configured to construct power operation time series data based on the real-time power operation data corresponding to the power decision point and the historical power operation data corresponding to the first preset time period in the past, and input the power operation time series data into the ... real The estimated power operation data corresponding to each power decision point constructs a target graph network; each node in the target graph network is used to represent the corresponding power decision point, each edge connection is used to connect the power decision points with an associated relationship, and each edge connection is also configured with a corresponding node association degree; a reinforcement learning scheduling unit is configured to input the target graph network into a preset reinforcement learning model, so that based on the reinforcement learning model, a corresponding target management scheduling scheme is determined for each node in the target graph network; the state space of the reinforcement learning model is defined by the real-time power operation data, the estimated power operation data and the overall scheduling requirements of the power grid of each node, and the action space of the reinforcement learning model is defined by the power generation adjustment operation, the transmission path opening and closing operation of the transmission network, the load aggregation point control plan and the energy storage equipment charging and discharging operation; the overall power grid scheduling requirements include maximizing the renewable energy utilization rate and minimizing the power supply operation cost of the power grid system when the total power supply of the power grid system is greater than the total load demand of each load aggregation point.

[0008] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any of the above-described multi-resource aggregation control methods for source, grid, load and storage.

[0009] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the multi-resource aggregation control method for source, grid, load and storage as described in any one of the above is implemented.

[0010] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned multi-resource aggregation control methods for source, grid, load and storage.

[0011] The multi-resource aggregation control method and system for source, grid, load and storage provided by the present invention can produce at least the following technical effects:

[0012] This technical solution mainly solves the control challenges of traditional power systems in the large-scale access of renewable energy and rapidly changing market demands through a multi-resource aggregation control method for sources, grids, loads and storage. The specific technical effects include:

[0013] (1) Through the acquisition and analysis of real-time power operation data, combined with historical data to build accurate power operation time series data, the power operation prediction model uses the power operation time series data to estimate the future power operation of each power decision point, so that power generation, transmission, load and energy storage equipment can be dispatched with higher accuracy. Therefore, the optimization of dispatch strategies based on real-time and estimated data can effectively reduce the waste caused by volatility and uncertainty, improve the utilization efficiency of renewable energy, and improve the resource utilization of the power grid system.

[0014] (2) A target graph network is constructed based on the estimated power operation data corresponding to each power decision point. Through the fine configuration of nodes and edges, the target graph network can provide real-time feedback on the operation status and demand of each power decision point. Then, the target graph network is input into the reinforcement learning model to stimulate the reinforcement learning model to quickly generate the optimal scheduling strategy based on this information, so that each power decision point can achieve a high degree of collaborative operation and realize resource scheduling flexibility and rapid response under the overall decision of the power grid system.

[0015] (3) In the grid dispatch calculation process of this technical solution, the fluctuation of electricity cost and peak and valley electricity price is comprehensively considered, and the operation of each decision point is optimized through the reinforcement learning model to minimize the cost. Therefore, when the total power supply is sufficient to cover the total load demand, the optimized dispatch strategy can maximize the use of renewable energy while reducing the use of expensive traditional energy, effectively reducing the power supply operation cost.

[0016] (4) Through the integrated multi-resource control and intelligent prediction system, this technical solution can better cope with the volatility and uncertainty brought by renewable energy and reduce power supply interruptions caused by extreme weather or emergencies. In addition, the interconnection relationship between nodes in the target graph network provides a mechanism so that when a problem occurs at a certain node, the system can quickly readjust the operation strategy of the surrounding nodes to maintain the stable operation of the entire power grid and improve the safety and stability of the power grid system operation*.

[0017] Through this technical solution, based on the optimization of intelligent dispatching strategies, it is possible to reduce energy consumption and operating costs, improve the economic benefits and competitiveness of the power system, promote the sustainable development of the power grid system, and significantly improve the economy, safety, stability and environmental friendliness of the power grid system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of an example of a multi-resource aggregation control method for source, grid, load and storage according to an embodiment of the present invention is shown;

[0020] Figure 2 A schematic diagram of state transition of an example of a reinforcement learning model according to an embodiment of the present invention is shown;

[0021] Figure 3 A schematic diagram showing a structural connection of an example of a power operation prediction model according to an embodiment of the present invention is shown;

[0022] Figure 4 A structural block diagram of an example of a multi-resource aggregation control system for source, grid, load and storage according to an embodiment of the present invention is shown;

[0023] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] In the technical solution of the present invention, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.

[0026] Figure 1 A flowchart of an example of a multi-resource aggregation control method for source, grid, load and storage according to an embodiment of the present invention is shown.

[0027] Regarding the execution subject of the method of the embodiment of the present invention, it can be any electronic device with processing and computing capabilities, such as a computer terminal, a mobile phone terminal, etc. More specifically, in combination with the business service scenario of power grid system dispatch, its execution subject can be a power grid resource dispatch server, which integrates and analyzes the data of different power decision points of source, grid, load and storage in the power grid system to achieve multi-resource aggregation and control for source, grid, load and storage, effectively reduce the waste caused by volatility and uncertainty, improve the utilization efficiency of renewable energy, and at the same time improve the resource utilization rate and power supply stability of the power grid system.

[0028] like Figure 1 As shown, in step S110, real-time power operation data of each power decision point to be optimized and scheduled in the power grid system is obtained.

[0029] Here, each power decision point includes a power generation source, a transmission network, a load aggregation point, and an energy storage device. The power generation source includes a thermal power generation source and a renewable power generation source, so it is necessary to effectively optimize the system resource scheduling of the hybrid power grid. The real-time power operation data includes the power generation of the power generation source, the load rate of the transmission network, the load demand curve, and the energy storage usage curve for each power decision point, which can effectively feedback the real-time operation status of each power decision point of the source network load storage. In addition, in order to facilitate the system to better implement resource scheduling analysis, the real-time power operation data also includes the weather forecast information, the power supply and distribution cost function, and the peak and valley electricity price floating curve of the corresponding power decision point. It should be noted that weather forecast information has a great impact on the energy output of renewable power sources (for example, wind power stations, photovoltaic power stations, etc.). Therefore, weather forecast information can help the system more accurately predict the upper limit of the power output of power decision points involving renewable energy in the corresponding time period in the future, so as to facilitate the reasonable allocation and use of renewable power resources by the power grid system. Through the peak and valley electricity price floating curve of the power grid, it reflects the price fluctuation of each power decision point, which is conducive to the power grid system to calculate the revenue based on the power supply data of each power decision point. In addition, the cost function of power supply and distribution can be preset, such as fuel cost, line maintenance cost, tax cost and CO2 emission cost, etc., which is conducive to the cost settlement of the power grid system based on the power supply data of each power decision point. More specifically, corresponding cost models are constructed for each different type of power decision point. For more details, please refer to the introduction in various relevant papers, which will not be repeated here.

[0030] In step S120, for each power decision point, power operation sequence data is constructed based on the real-time power operation data corresponding to the power decision point and the historical power operation data corresponding to the first preset time period in the past, and the power operation sequence data is input into the preset power operation prediction model to determine the corresponding estimated power operation data.

[0031] Here, the power operation prediction model can adopt various deep learning models, which should not be limited here. More specifically, some experts and scholars use LSTM architecture when dealing with similar time series analysis problems. Furthermore, the power operation prediction model can adopt a more novel architecture, and more details of this innovative power operation prediction model will be expanded in conjunction with other examples below.

[0032] In step S130, a target graph network is constructed based on the estimated power operation data corresponding to each power decision point.

[0033] Here, each node in the target graph network is used to represent a corresponding power decision point, each edge connection is used to connect power decision points with an associated relationship, and each edge connection is also configured with a corresponding node association degree.

[0034] It should be noted that in the target graph network, the "association" of edge connections refers to the measure of mutual dependence or influence between two power decision points. This association can be defined based on a variety of factors, including but not limited to physical connection (such as whether two decision points are directly connected by transmission lines), economic impact (for example, an oversupply of electricity in one area may affect electricity prices in adjacent areas), or operational coordination (specifically, whether two decision points need to operate synchronously to optimize overall grid performance, such as coordinating the output of power generation and energy storage facilities to cope with high load demand).

[0035] Specifically, the "association" of edge connections can be quantitatively evaluated based on the transmission capacity, economic cost, and line loss between nodes. More preferably, in some embodiments, in order to enhance the priority of renewable energy nodes and promote full utilization of the power of renewable energy nodes, the association calculation algorithm can also be optimized, and more details will be expanded in conjunction with other examples below.

[0036] In step S140, the target graph network is input into a preset reinforcement learning model, so that based on the reinforcement learning model, a corresponding target management scheduling scheme is determined for each node in the target graph network.

[0037] Here, the state space of the reinforcement learning model is defined by the real-time power operation data of each node, the estimated power operation data and the overall dispatching requirements of the power grid, and the action space of the reinforcement learning model is defined by the power generation adjustment operation, the transmission path opening and closing operation of the transmission network, the load aggregation point control plan and the energy storage equipment charging and discharging operation.

[0038] It should be noted that, in an embodiment of the present invention, each state in the state space includes the overall dispatching requirements of the power grid. Specifically, the overall dispatching requirements of the power grid include maximizing the utilization rate of renewable energy and minimizing the power supply operating costs of the power grid system when the total power supply of the power grid system is greater than the total load demand of each load aggregation point. Therefore, using the overall dispatching requirements of the power grid to define the state space of the reinforcement learning model can help the model better understand and predict the dynamic changes of the entire power network, thereby providing a global perspective for enhanced decision-making, allocating power resources more efficiently, and reducing waste. In this way, even in the event of an emergency or a sudden change in demand, the model can adjust its strategy more quickly and achieve rapid response to ensure the stable operation of the power grid system.

[0039] In an embodiment of the present invention, the target management scheduling scheme determined by the reinforcement learning model can efficiently allocate resources and optimize equipment operation modes, increase the proportion of renewable energy such as wind energy and solar energy, effectively reduce the operation and maintenance costs of the entire system, and ensure that the power grid can operate stably under various load and supply conditions.

[0040] In terms of business scenarios, for urban power grid systems, it includes multiple power sources (thermal, solar, wind), a wide transmission network, multiple load aggregation points (residential, commercial and industrial areas) and energy storage equipment. For thermal power sources, increase output during peak demand periods and reduce operation during low demand periods. For solar and wind power sources, adjust output expectations based on weather forecast information and give priority to these power generation methods to reduce carbon emissions and operating costs. For the management of the transmission network, the transmission plan can be dynamically adjusted based on real-time load rate data, such as reducing transmission in areas with low demand and increasing transmission capacity in high-load areas to reduce transmission losses and improve efficiency. For the regulation of load aggregation points, increase power supply to industrial areas during low electricity price periods, and encourage high-power-consuming industries to operate when electricity is sufficient to balance the system load. For the operation of energy storage equipment, store excess electricity when renewable energy production is high, and release the stored electricity when production is low or demand is high to ensure the continuity and stability of power supply.

[0041] Figure 2 A schematic diagram of state transition of an example of a reinforcement learning model according to an embodiment of the present invention is shown.

[0042] like Figure 2 As shown, the state transition diagram involves multiple states f1~f nThe basic state set may have state transitions between different states, for example, a1 represents the state transition action from f1 to f2, a2 ​​represents the state transition action from f2 to f1, a3 represents the state transition action from f1 to f3, etc. Here, the corresponding state transitions may occur based on the state transition strategy, and each state transition strategy is used to indicate different state transitions.

[0043] It should be noted that the reward function (or reward function) is a core component in reinforcement learning, which is used to quantify the immediate benefit or cost of a specific action in a given state. It is obtained directly from the environment and describes the feedback given by the environment to a certain operation of the agent. In the scenario of power system regulation, the reward function can be designed based on goals such as reducing energy consumption, reducing operating costs, and improving the efficiency of renewable energy use. For example, if more renewable energy is used to reduce carbon emissions and costs, the reward function will return a higher positive value.

[0044] In addition, each state transition has a corresponding state transition reward, and each state transition reward can be determined based on a preset state transition reward function. Generally, the larger the state transition reward, the more valuable the state transition action is, for example, the more stable the supply and demand balance of the corresponding power grid system, the higher the comprehensive benefit of the power grid system, the lower the power supply operation cost of the power grid system, or the higher the utilization rate of renewable energy in the power grid system, etc.

[0045] In some examples of the embodiments of the present invention, the reward function R(s) of the reinforcement learning model t ,a t ) is defined as follows:

[0046] R(s t ,a t )=α1R cost +α2R renewable +α3R balance , Formula (1)

[0047] R cost =-(GC(s t ,a t )+TLC(s t ,a t )), Formula (2)

[0048]

[0049] R balance =SS(s t ,a t )-PFD(s t ,a t ), Formula (4)

[0050] In the formula, s t represents the state at time step t, a t represents the action at time step t, GC(s t ,a t ) represents the cost of electricity generation, which is calculated based on the amount of electricity generated and the unit cost of different energy types. For example, it can be set that the unit cost of renewable energy may be lower than that of fossil fuels. TLC(s t ,a t ) represents the transmission loss cost, which is based on the transmission distance and grid efficiency, where the transmission loss is usually proportional to the amount of electricity transmitted and the distance.

[0051] In formula (3), the renewable energy utilization function uses a square function to further encourage the use of a high proportion of renewable energy, where REP(s t ,a t ) indicates that in action a t The amount of renewable energy generated under t ,a t ) represents the total electricity generation, including renewable and non-renewable energy sources.

[0052] In formula (4), the system supply and demand balance function is used to evaluate whether the power supply of the power grid meets the demand, and at the same time, the situation of oversupply is encouraged to increase the reliability and flexibility of the system. Specifically, SS(s t ,a t ) indicates the amount by which the total power supply exceeds the total demand, and surplus power supply is encouraged. PFD(s t ,a t ) represents the amount of penalty imposed if the supply is insufficient to meet demand, in order to further prevent power shortages.

[0053] Furthermore, in some examples of the embodiments of the present invention, the reinforcement learning model adopts a PPO (Proximal Policy Optimization) model.

[0054] It should be noted that in the process of practicing the present invention, considering that the power system scheduling problem involves the problem of continuous action space, potential model solutions include TRPO (Trust Region Policy Optimization) model and PPO model. TRPO uses a mathematically rigorous method to update the strategy, ensuring that a certain proximity is maintained between the new strategy and the old strategy, and ensuring the stability of training by controlling the trust region, but its complexity and computational overhead are not suitable for situations where computing resources are limited or rapid iteration is required. In contrast, PPO simplifies the implementation of TRPO and provides a similar stable and reliable policy update mechanism, but in a more easily implemented and computationally efficient manner. PPO avoids the complex constrained optimization problem in TRPO by using a clipped probability ratio target, and is more conducive to quickly updating the prediction results based on real-time data in the system, which can support the realization of more balanced performance.

[0055] More specifically, the PPO model belongs to an Actor-Critic reinforcement learning architecture, which uses two main components: a policy network (Actor) and a value network (Critic). The policy network is responsible for generating actions in a given state. Its input is the current state of the environment, and its output is the probability distribution of actions (for discrete action space) or action values ​​(for continuous action space). The value network estimates the value of the state under the current policy, that is, the expected return that can be obtained by acting according to the current policy starting from the current state, which helps to evaluate the quality of the actions selected by the policy network.

[0056] The goal of PPO is to t (θ) maximizes an objective function and avoids updates that are too large by clipping policy updates. The objective function of the PPO algorithm is the core component used in the training phase of the model. In the training phase, the objective function is used to guide parameter updates to optimize the performance of the model. In addition, in the model prediction phase (or inference phase), the PPO model uses the trained policy network to make decisions, and the objective function is not used at this time.

[0057] More specifically, the network parameters θ of PPO are obtained by the following trimmed version of the objective function L CLIP (θ) and update:

[0058]

[0059]

[0060]

[0061] η t =R t +τV(s t+1)-V(s t ), Formula (8)

[0062] In the formula, r t (θ) represents the new strategy π θ (a t |s t ) and the old strategy The probability ratio of choosing the same action. Represents the expected value of the reward calculated at all time steps for the data sample set of the reinforcement learning model. clip(r t (θ),1-ξ,1+ξ) represents the clipping operation, which is used to convert the probability ratio r t (θ) is restricted to a specified interval [1-ξ,1+ξ], where ξ represents the clipping threshold, which indicates the magnitude of the policy update. represents the advantage function estimate, which is used to quantify action a t The expected additional reward relative to the average policy action. η t represents the temporal difference (TD) error at time step t. s t represents the state at time step t, R t Indicates that at time step t, due to executing action a t The instant reward obtained. V(s t ) is the state value function estimate at time step t, representing the transition from state s t The discounted sum of future rewards expected from starting to follow the current strategy. γ represents the reward discount factor. τ represents the GAE (Generalized Advantage Estimation) smoothing parameter. T represents the total number of time steps corresponding to the time series data. Here, the advantage function estimate is expressed by the GAE formula, and τ is used to effectively balance the bias and variance of the estimate.

[0063] The TD error is used to express the difference between the observed reward plus the discounted value of the next state minus the value of the current state to evaluate the goodness of the result after the current strategy executes a specific action. In addition, by accumulating the discounted TD error, GAE takes into account the impact of each future step, but achieves variance reduction and estimation smoothing through the decreasing weighting of τ.

[0064] In conjunction with the example of this technical scenario, assume that in the power system dispatch, the state includes the current load level, power generation and weather conditions. If an action is to increase the output of a wind power station, the immediate reward R t May include calculations based on cost savings and efficiency gains. The system is able to evaluate the advantages of this action relative to other possible actions, thereby better guiding future generation scheduling decisions to optimize overall system performance and respond to environmental changes.

[0065] In an embodiment of the present invention, PPO natively supports continuous action space and is suitable for continuous decision-making problems in power systems such as power generation regulation and transmission path configuration. In addition, in actual power systems, obtaining a large amount of interaction data may be expensive or difficult to implement. By adopting the PPO model, using multi-step advantage estimation and multiple learning with the same samples, it supports rapid adaptation to policy changes in fewer sample updates. On the other hand, the reward function of the reinforcement learning model may involve multiple objectives (such as cost, efficiency, and environmental impact), and PPO can effectively handle these complex reward designs and optimize overall performance.

[0066] Figure 3 A schematic diagram of structural connections of an example of a power operation prediction model according to an embodiment of the present invention is shown.

[0067] like Figure 3 As shown, the power operation prediction model 300 adopts a time series Transformer architecture, which includes an input layer 310, a position encoding layer 320, an enhanced Transformer network 330 and an output layer 340, wherein the enhanced Transformer network 330 includes an improved multi-head self-attention subnetwork 331 and an improved time-aware feedforward subnetwork 333.

[0068] The input layer 310 is used to perform data standardization preprocessing on the real-time power operation data. By performing standardization preprocessing on the information of each dimension in the real-time power operation data, the stability and convergence speed of the model can be effectively ensured.

[0069] The position encoding layer 320 is used to add corresponding position information to each time step involved in the real-time power operation data to maintain the sequential characteristics of the time series:

[0070]

[0071]

[0072] Where, PE (pos,2i) and PE (pos,2i+1) Represents two consecutive elements in the position encoding vector, PE (pos,2i) Represents the 2ith element of the position encoding calculated using the sine function, PE (pos,2i+1) The 2i+1th element of the position code representing the cosine function calculation. model is the input feature dimension.

[0073] It should be noted that since Transformer does not have the ability to process the position information of sequence data, position encoding is added to the input features to maintain the continuity of time information. Therefore, on the basis of traditional Transformer, the added time-aware position encoding enables the model to better understand the temporal relationships and patterns in time series data.

[0074] The improved multi-head self-attention subnetwork 331 allows the model to learn different features of the input data in multiple subspaces simultaneously:

[0075] MultiHead(Q,K,V)=Concat(head1,...,head h )W O , Formula (11)

[0076] head i =Attention(QW i Q ,KW i K ,VW i V ), Formula (12)

[0077]

[0078] Where Q, K, and V represent the query matrix, key matrix, and value matrix, respectively. k Indicates the dimension of the key; QK T Represents the dot product between the query and the key. head i W represents the i-th attention head. i Q ,W i K ,W i V W denotes the weight matrices of query, key, and value for the i-th attention head, respectively. O Represents the weight matrix of the feature concatenation operation.

[0079] It should be noted that traditional sequence prediction models (such as LSTM) may encounter information loss problems when processing long sequences. Here, in the time series transformer, the time series transformer can capture more fine-grained time dependencies and nonlinear relationships from power operation data through a multi-head self-attention mechanism, enabling the model to distinguish and utilize data features at different time scales, such as the impact of short-term weather changes on renewable energy output, or long-term load trends, improving the prediction accuracy of complex power demand and supply patterns.

[0080] The improved time-aware feedforward subnetwork 333 includes multiple time-aware feedforward layers (not shown), each of which is connected to a corresponding attention head:

[0081] FFN(x)=max(0,xW1+b1)W2+b2, Equation (14)

[0082] x new =x+δ(t,W time ), Formula (15)

[0083] δ(t,W time )=sin(W time ·t+b time ), Formula (16)

[0084] Where x represents the input of the current time-aware feedforward layer, W1, W2 and b1, b2 represent the weight matrix and bias term of the two cascaded linear transformation layers in the feedforward neural network, and FFN(x) represents the corresponding output feature. new Represents the output features of the current time-aware feedforward layer. δ(t,W time ) represents the time adjustment function based on time step t, W time represents the weight of the time adjustment function, b time Represents the bias of the time adjustment function.

[0085] It should be noted that time-related parameters are added to the traditional feedforward layer, and a time-aware feedforward network is added after each attention mechanism to enhance the responsiveness to time changes. This optimized feedforward network is able to adjust its processing based on time changes, especially in capturing and responding to peak and valley changes in power demand.

[0086] The output layer 340 is used to output the corresponding estimated power operation data:

[0087] y=W y h last +b y , Formula (17)

[0088] In the formula, h last W represents the feature vector output by the last Transformer layer, and y represents the estimated power operation data. y and b y They represent the weight and bias of the output layer respectively, both of which are rectangular matrices with a dimension of [d output ,d model ],d output Represents the output dimension determined by the prediction target. For example, if the hourly power operation data for the next 24 hours is predicted, then d outputShould be set to 24.

[0089] Through the structure of the power operation prediction model provided by the embodiment of the present invention, due to the parallel processing capability of Transformer, the model can quickly adapt to real-time updated data, which is crucial for power system regulation, especially in the face of emergencies (such as extreme weather or equipment failures), the system can quickly adjust strategies to ensure the stability of the power grid. In addition, the model can process and integrate data from multiple sources (such as load data, meteorological information, market prices, etc.). By integrating this information, the model can not only predict power demand, but also predict electricity price fluctuations and power supply costs, so that power grid managers (or provide a basis for the precise scheduling plan of the reinforcement learning model) can pre-identify potential supply and demand imbalances, perform scheduling optimization in advance, and increase the robustness and fault prevention capabilities of the system.

[0090] In some examples of the embodiments of the present invention, the loss function L of the power operation prediction model is total for:

[0091] L total =L accuracy +L cost +L renewable , Formula (18)

[0092]

[0093]

[0094]

[0095] Where, L accuracy , L cost and L renewable They represent the classification accuracy loss term, cost efficiency loss term, and renewable energy utilization optimization loss term respectively. N represents the total number of samples in the data sample set of the power operation prediction model, y t represents the sample label value of the data sample, Indicates the predicted value corresponding to the training sample. t represents the unit electricity cost at time t, and λ represents the cost adjustment coefficient. t Indicates the proportion of actual renewable energy in the total power generation in the grid system, represents the predicted proportion of renewable energy, and μ represents the optimized regulation coefficient of renewable energy.

[0096] It should be noted that the loss function for the power operation prediction model provided in the embodiment of the present invention involves a classification accuracy loss term, a cost efficiency loss term, and a renewable energy utilization optimization loss term. The classification accuracy loss term promotes the high accuracy of the prediction results of the operation data of the power decision point, the cost efficiency loss term emphasizes that the model optimizes cost efficiency while satisfying the prediction accuracy, and the renewable energy utilization optimization loss term rewards the model for tending to predict and optimize the use of renewable energy.

[0097] It should be noted that the loss function of the model no longer uses prediction accuracy as the only metric, and incorporates comprehensive considerations of cost efficiency and renewable energy utilization. In particular, by adding items that encourage the use of renewable energy to the loss function, the forecasting model can support these policy goals and help the power grid better plan and utilize renewable energy, thereby increasing its proportion in the energy mix. By rewarding the forecasting model for using more renewable energy, it can promote the development of power grid management technologies, such as demand response and energy storage technology, further improve the reliability and flexibility of the power grid, and help support the achievement of environmental protection goals, while also helping to reduce power grid operating costs.

[0098] Regarding the implementation details of the above step S130, in some implementations, the node association degree is determined by:

[0099]

[0100] Where P(i,j), C(i,j) and L(i,j) are the transmission capacity, line maintenance cost and power transmission loss rate from node i to node j respectively. G(i,j) indicates that when node i or node j corresponds to a renewable power source, G(i,j) is a preset positive value, and when node i or node j does not belong to a renewable power source, it is zero. ∈ represents a stability coefficient used to prevent division by zero.

[0101] In the embodiment of the present invention, in order to enhance the priority of renewable energy nodes, we can add a priority coefficient G(i,j) in the correlation calculation, which provides a weighted advantage for renewable energy nodes. If node i or node j is a renewable energy power generation source, a positive value is assigned, such as 0.2 or a value adjusted according to the policy; if it is a non-renewable energy source, it is 0. Therefore, by reflecting the preference for renewable energy through the priority coefficient, the utilization rate of renewable energy in power dispatch can be increased to support environmentally friendly and sustainable power system strategies.

[0102] Through the embodiments of the present invention, the priority of renewable energy power generation sources in power grid scheduling is improved, which not only promotes the utilization of renewable energy, but also helps to optimize power supply costs and improve power grid operation efficiency, thereby ensuring the safety and reliability of power supply while supporting the implementation of environmental protection policies.

[0103] The multi-resource aggregation control system for source, grid, load and storage provided by the present invention is described below. The multi-resource aggregation control system for source, grid, load and storage described below and the multi-resource aggregation control method for source, grid, load and storage described above can be referenced to each other.

[0104] Figure 4 A structural block diagram of an example of a multi-resource aggregation control system for source, grid, load and storage according to an embodiment of the present invention is shown.

[0105] like Figure 4 As shown, the multi-resource aggregation control system 400 for source, grid, load and storage includes a real-time data acquisition unit 410, an operation data prediction unit 420, a graph network construction unit 430 and a reinforcement learning scheduling unit 440.

[0106] The real-time data acquisition unit 410 is configured to acquire real-time power operation data of each power decision point to be optimized and scheduled in the power grid system; each power decision point includes a power generation source, a transmission network, a load aggregation point and an energy storage device; the power generation source includes a thermal power generation source and a renewable power generation source; the real-time power operation data includes the power generation of the power generation source, the load rate of the transmission network, the load demand curve and the energy storage utilization curve for each power decision point respectively, and the real-time power operation data also includes weather forecast information of the corresponding power decision point, the power supply and distribution cost function and the power grid peak and valley electricity price floating curve.

[0107] The operation data prediction unit 420 is configured to construct power operation sequence data for each of the power decision points according to the real-time power operation data corresponding to the power decision point and the historical power operation data corresponding to the first preset time period in the past, and input the power operation sequence data into a preset power operation prediction model to determine the corresponding estimated power operation data.

[0108] The graph network construction unit 430 is configured to construct a target graph network based on the estimated power operation data corresponding to each power decision point; each node in the target graph network is used to represent the corresponding power decision point, each edge connection is used to connect the power decision points with an associated relationship, and each edge connection is also configured with a corresponding node association degree.

[0109] The reinforcement learning scheduling unit 440 is configured to input the target graph network into a preset reinforcement learning model, so that based on the reinforcement learning model, a corresponding target management scheduling scheme is determined for each node in the target graph network; the state space of the reinforcement learning model is defined by the real-time power operation data, estimated power operation data and the overall scheduling requirements of the power grid of each node, and the action space of the reinforcement learning model is defined by the power generation adjustment operation, the transmission path opening and closing operation of the transmission network, the load aggregation point control plan and the energy storage equipment charging and discharging operation; the overall power grid scheduling requirements include maximizing the renewable energy utilization rate and minimizing the power supply operation cost of the power grid system when the total power supply of the power grid system is greater than the total load demand of each load aggregation point.

[0110] Figure 5 is a schematic diagram of the hardware structure of an electronic device for executing a multi-resource aggregation control method for source, network, load and storage provided by another embodiment of the present invention, such as Figure 5 As shown, the device includes:

[0111] One or more processors 510 and memory 520, Figure 5 A processor 510 is taken as an example.

[0112] The device for executing the source-grid-load-storage oriented multi-resource aggregation control method may also include: an input device 530 and an output device 540 .

[0113] The processor 510, the memory 520, the input device 530 and the output device 540 may be connected via a bus or other means. Figure 5 The example of connecting through bus is taken in the following.

[0114] The memory 520, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the source-grid-load-storage multi-resource aggregation control method in the embodiment of the present invention. The processor 510 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 520, that is, realizing the source-grid-load-storage multi-resource aggregation control method in the above method embodiment.

[0115] The memory 520 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 520 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 520 may optionally include a memory remotely arranged relative to the processor 510, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0116] The input device 530 may receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 540 may include a display device such as a display screen.

[0117] The one or more modules are stored in the memory 520, and when executed by the one or more processors 510, the multi-resource aggregation control method for source, grid, load and storage in any of the above method embodiments is executed.

[0118] The above product can execute the multi-resource aggregation control method for source, grid, load and storage provided by the embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0119] The electronic devices of the embodiments of the present invention exist in various forms, including but not limited to:

[0120] (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0121] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.

[0122] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0123] (4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.

[0124] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0125] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-resource aggregation control method for source, grid, load and storage, comprising: Obtain real-time power operation data of each power decision point to be optimized in the power grid system; The various power decision points include power generation sources, transmission networks, load aggregation points and energy storage equipment; The power generation sources include thermal power generation sources and renewable power generation sources; the real-time power operation data include the power generation of the power generation sources, the load rate of the transmission network, the load demand curve and the energy storage utilization curve for each power decision point, and the real-time power operation data also includes the weather forecast information of the corresponding power decision point, the power supply and distribution cost function and the power grid peak and valley electricity price floating curve; For each of the power decision points, construct power operation time series data according to the real-time power operation data corresponding to the power decision point and the historical power operation data corresponding to the first preset time period in the past, and input the power operation time series data into a preset power operation prediction model to determine the corresponding estimated power operation data; Construct a target graph network based on the estimated power operation data corresponding to each power decision point; Each node in the target graph network is used to represent a corresponding power decision point, each edge connection is used to connect power decision points with an associated relationship, and each edge connection is also configured with a corresponding node association degree; Inputting the target graph network into a preset reinforcement learning model, so that based on the reinforcement learning model, a corresponding target management scheduling scheme is determined for each node in the target graph network; The state space of the reinforcement learning model is defined by the real-time power operation data of each node, the estimated power operation data and the overall dispatching requirements of the power grid, and the action space of the reinforcement learning model is defined by the power generation adjustment operation, the transmission path opening and closing operation of the transmission network, the load aggregation point control plan and the energy storage equipment charging and discharging operation; the overall dispatching requirements of the power grid include maximizing the renewable energy utilization rate of the power grid system and minimizing the power supply operation cost when the total power supply of the power grid system is greater than the total load demand of each load aggregation point.

2. The method according to claim 1, wherein: The power operation prediction model adopts a time series Transformer architecture, which includes an input layer, a position encoding layer, an enhanced Transformer network and an output layer, wherein the enhanced Transformer network includes an improved multi-head self-attention sub-network and an improved time-aware feedforward sub-network; The input layer is used to perform data standardization preprocessing on the real-time power operation data; The position encoding layer is used to add corresponding position information to each time step involved in the real-time power operation data to maintain the sequential characteristics of the time series: Where, PE (pos,2i) and PE (pos,2i+1) Represents two consecutive elements in the position encoding vector, PE (pos,2i) Represents the 2ith element of the position encoding calculated using the sine function, PE (pos,2i+1) The 2i+1th element of the position code representing the cosine function calculation; d model is the input feature dimension; The improved multi-head self-attention sub-network allows the model to learn different features of the input data in multiple subspaces simultaneously: MultiHead(Q,K,V)=Concat(head1,...,head h )W O head i =Attention(QW i Q ,KW i K ,VW i V ) Where Q, K, and V represent the query matrix, key matrix, and value matrix, respectively; k Indicates the dimension of the key; QK T Represents the dot product between the query and the key; head i represents the i-th attention head; W i Q ,W i K ,W i V denotes the weight matrices of query, key, and value for the i-th attention head respectively; W O Represents the weight matrix of the feature concatenation operation; The improved time-aware feedforward subnetwork comprises a plurality of time-aware feedforward layers, each of which is connected to a corresponding attention head: FFN(x)=max(0,xW1+b1)W2+b2 x new =x+δ(t,W time ) δ(t,W time )=sin(W time ·t+b time ) Where x represents the input of the current time-aware feedforward layer, W1, W2 and b1, b2 represent the weight matrix and bias term of the two cascaded linear transformation layers in the feedforward neural network, and FFN(x) represents the corresponding output feature; x new represents the output feature of the current time-aware feedforward layer; δ(t,W time ) represents the time adjustment function based on time step t, W time represents the weight of the time adjustment function, b time represents the bias of the time adjustment function; The output layer is used to output the corresponding estimated power operation data: y=W y h last +b y In the formula, h last represents the feature vector output by the last Transformer layer, y represents the estimated power operation data; W y and b y They represent the weight and bias of the output layer respectively, both of which are rectangular matrices with a dimension of [d output ,d model ],d output Represents the output dimension determined according to the prediction target.

3. The method according to claim 2, wherein: The loss function L of the power operation prediction model is total for: L total =L accuracy +L cost +L renewable Where, L accuracy , L cost and L renewable They represent the classification accuracy loss term, cost efficiency loss term and renewable energy utilization optimization loss term respectively; N represents the total number of samples in the data sample set of the power operation prediction model, y t represents the sample label value of the data sample, Indicates the predicted value corresponding to the training sample; c t represents the unit electricity cost at time t, λ represents the cost adjustment coefficient; e t Indicates the proportion of actual renewable energy in the total power generation in the grid system, represents the predicted proportion of renewable energy, and μ represents the optimized regulation coefficient of renewable energy.

4. The method according to claim 1, wherein: The node association degree is determined by: Where P(i,j), C(i,j) and L(i,j) are the transmission capacity, line maintenance cost and power transmission loss rate from node i to node j respectively; G(i,j) is a preset positive value when node i or node j corresponds to a renewable power source, and is zero when node i or node j does not belong to a renewable power source; ∈ represents a stability coefficient used to prevent division by zero.

5. The method according to claim 1, wherein: The reinforcement learning model adopts the PPO model, and its network parameter θ is obtained by the following trimmed version of the objective function L CLIP (θ) and update: or t =R t +τV(s t+1 )-V(s t ) In the formula, r t (θ) represents the new strategy π θ (a t |s t ) and the old strategy The probability ratio of choosing the same action under represents the expected value of the reward calculated at all time steps for the data sample set of the reinforcement learning model; clip(r t (θ),1-ξ,1+ξ) represents the clipping operation, which is used to convert the probability ratio r t (θ) is limited to a specified interval [1-ξ,1+ξ]; ξ represents the clipping threshold, which indicates the magnitude of the policy update; represents the advantage function estimate, which is used to quantify action a t The expected additional return relative to the average strategy action; η t represents the time difference error at time step t; s t represents the state at time step t, R t Indicates that at time step t, due to executing action a t The immediate reward obtained; V(s t ) is the state value function estimate at time step t, representing the transition from state s t The discounted sum of future rewards expected from starting to follow the current strategy; γ represents the reward discount factor; τ represents the GAE smoothing parameter; and T represents the total number of time steps corresponding to the time series data.

6. A multi-resource aggregation control system for source, grid, load and storage, comprising: A real-time data acquisition unit is configured to acquire real-time power operation data of each power decision point to be optimized and dispatched in the power grid system; The various power decision points include power generation sources, transmission networks, load aggregation points and energy storage equipment; The power generation sources include thermal power generation sources and renewable power generation sources; the real-time power operation data include the power generation of the power generation sources, the load rate of the transmission network, the load demand curve and the energy storage utilization curve for each power decision point, and the real-time power operation data also includes the weather forecast information of the corresponding power decision point, the power supply and distribution cost function and the power grid peak and valley electricity price floating curve; an operation data prediction unit configured to construct, for each of the power decision points, power operation time series data according to the real-time power operation data corresponding to the power decision point and the historical power operation data corresponding to the first preset time period in the past, and input the power operation time series data into a preset power operation prediction model to determine corresponding estimated power operation data; A graph network construction unit is configured to construct a target graph network based on estimated power operation data corresponding to each power decision point; Each node in the target graph network is used to represent a corresponding power decision point, each edge connection is used to connect power decision points with an associated relationship, and each edge connection is also configured with a corresponding node association degree; A reinforcement learning scheduling unit is configured to input the target graph network into a preset reinforcement learning model so that a corresponding target management scheduling scheme is determined for each node in the target graph network based on the reinforcement learning model; The state space of the reinforcement learning model is defined by the real-time power operation data of each node, the estimated power operation data and the overall dispatching requirements of the power grid, and the action space of the reinforcement learning model is defined by the power generation adjustment operation, the transmission path opening and closing operation of the transmission network, the load aggregation point control plan and the energy storage equipment charging and discharging operation; the overall dispatching requirements of the power grid include maximizing the renewable energy utilization rate of the power grid system and minimizing the power supply operation cost when the total power supply of the power grid system is greater than the total load demand of each load aggregation point.

7. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the multi-resource aggregation control method for source, grid, load and storage as described in any one of claims 1 to 5 is implemented.

Images