Optimization method and system for control strategy of optical storage based on health degree in energy system

Abstract

The application discloses a health degree-based photovoltaic storage control strategy optimization method and system in an energy system, relates to the source network load storage technical field, and is applied to the scheduling side and comprises the following steps: receiving a prediction boundary condition of each photovoltaic storage equipment in a photovoltaic storage scene after power and load of the photovoltaic storage equipment are predicted by a prediction model in multiple time scales on the edge side, wherein the prediction model is obtained by optimizing a local model based on corresponding global model parameters on the edge side; based on the prediction boundary condition, an initial control strategy of a photovoltaic storage power station in each time scale is generated under photovoltaic fluctuation conditions by using a reinforcement learning algorithm; the health degree of each target equipment is evaluated based on a preset hidden Markov model, a deep state space model and the remaining life of the target equipment involved in the initial control strategy; and the initial control strategy is optimized based on the reinforcement learning algorithm and the health degree, so that a target control strategy of the photovoltaic storage power station is obtained. The method improves the collaborative level of source network load storage in equipment life and operation safety.

Description

Technical Field

[0001] This application relates to the field of power generation, grid, load and energy storage technology, and in particular to a method and system for optimizing photovoltaic and energy storage control strategies based on health status in energy systems. Background Technology

[0002] Against the backdrop of promoting the integration of power generation, grid, load and storage, photovoltaic and energy storage, as a key hub integrating photovoltaic power generation, energy storage regulation, load response and grid interaction, are increasingly becoming the core components of energy management platforms and energy-carbon synergy systems in low-carbon / zero-carbon parks. Their operational efficiency is directly related to the safety and economy of the new power system. Currently, the scheduling and control of photovoltaic and energy storage power stations mainly focus on short-term power balance targets on the scheduling side, ignoring the cumulative damage to the health status of core equipment such as energy storage batteries and converters caused by photovoltaic power output fluctuations and frequent charging and discharging. Furthermore, the health assessment of the equipment is also completed on the edge side, which means that the assessed health status cannot be integrated into the scheduling decision on the scheduling side in real time. This results in a serious disconnect between the equipment health assessment and control strategy of photovoltaic and energy storage power stations, leading to insufficient coordination between power generation, grid, load and energy storage.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a method and system for optimizing the control strategy of photovoltaic-storage based on health in energy systems, which aims to solve the technical problem of insufficient coordination among power generation, grid, load and storage.

[0005] To achieve the above objectives, this application proposes a health-based photovoltaic-storage control strategy optimization method for energy systems, applied to the dispatching side. The dispatching side belongs to a photovoltaic-storage scenario modeled using relational graph convolutional networks and graph attention networks based on the multidimensional heterogeneous parameters of real-world photovoltaic-storage power plants. The photovoltaic-storage scenario also includes edge sides representing each photovoltaic-storage power plant, each of which is communicatively connected to the dispatching side. The method includes: The edge side receives the predicted boundary conditions after performing multi-time-scale predictions of the power and load of each photovoltaic storage device in the photovoltaic storage scenario through a prediction model. The prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters. Based on the predicted boundary conditions, a reinforcement learning algorithm is used to generate the initial control strategy of the photovoltaic-storage power station at each time scale under photovoltaic fluctuation conditions. Based on the preset hidden Markov model, deep state space model, and the remaining lifetime of the target devices involved in the initial control strategy, the health of each target device is evaluated. Based on the reinforcement learning algorithm and the health status, the initial control strategy is optimized to obtain the target control strategy of the photovoltaic-storage power station.

[0006] In one embodiment, the step of optimizing the initial control strategy based on the reinforcement learning algorithm and the health status to obtain the target control strategy for the photovoltaic-storage power station includes: Based on the health status and multi-timescale regulation attributes, the hierarchical reinforcement learning of the reinforcement learning algorithm is used to incorporate the health degradation trend of the equipment, maintenance arrangements and spare parts configuration requirements into the strategy optimization consideration dimensions. Under photovoltaic fluctuation conditions, the reinforcement learning algorithm is used to constrain and modify the initial control strategy at each time scale based on the equipment health and the optimization consideration dimensions. Based on the reinforcement learning algorithm, the constraint correction process is iteratively optimized to obtain the target control strategy of the photovoltaic-storage power station that adapts to the health status of the equipment, photovoltaic fluctuations, and multi-timescale operation requirements.

[0007] In one embodiment, the hierarchical reinforcement learning includes a two-level decision architecture consisting of a policy layer and an execution layer; The strategy layer outputs control strategy commands based on the health status of the target device; The execution layer, in conjunction with the multi-timescale control attributes, works with the control strategy instructions to complete the constraint correction and iterative optimization of the initial control strategy.

[0008] In one embodiment, the multiple time scales include three levels of control time scales: day-ahead, intraday, and real-time. Based on the health status and the control attributes of the multiple time scales, the step of incorporating the equipment's health degradation trend, maintenance schedule, and spare parts configuration requirements into the strategy optimization consideration dimensions using the hierarchical reinforcement learning algorithm of the reinforcement learning algorithm includes: Based on the rolling window optimization mechanism, within each rolling window, the control attributes of the three-level control time scale are combined, and the hierarchical reinforcement learning of the reinforcement learning algorithm is used to synchronously associate the health degradation trend, maintenance schedule and spare parts configuration requirements of the target equipment, so as to obtain the preliminary dimensions of strategy optimization considerations under the current rolling window. Using the initial dimension as the basis for optimization, the strategy optimization solution is performed within the corresponding scrolling window to obtain the first step scheduling instruction for the current scale. The first step scheduling instruction is then sent to the edge side, where the edge side executes the first step scheduling instruction to obtain the latest predicted boundary conditions. The latest predicted boundary conditions from the edge-side backhaul are incorporated into the subsequent rolling window, and the multi-timescale constraints and the preliminary dimensions are dynamically iterated and updated to obtain the strategy optimization consideration dimensions that adapt to the temporal changes of photovoltaic power load fluctuations and equipment health degradation.

[0009] In one embodiment, the step of assessing the health of each target device based on a preset hidden Markov model, a deep state-space model, and the remaining lifetime of the target devices involved in the initial control strategy includes: Based on a preset hidden Markov model, the inherent operating attributes of each device in the photovoltaic-storage power station are benchmarked, and the degradation state range corresponding to the target device is determined from the inherent operating attributes. Based on a preset deep state space model, the real-time operational observation data of the target device are fitted, and the current degree of degradation of the target device is extracted from the degradation state range; Based on the remaining lifespan of the target device associated with the initial control strategy, the current degradation level is adjusted by parameters to obtain the target degradation level; The health status of the target device is generated by quantifying the degree of degradation of the target.

[0010] Furthermore, to achieve the above objectives, this application also proposes a health-based photovoltaic-storage control strategy optimization method for energy systems, applied to the edge side. The edge side belongs to a photovoltaic-storage scenario modeled using relational graph convolutional networks and graph attention networks based on the multidimensional heterogeneous parameters of real-world photovoltaic-storage power plants. The photovoltaic-storage scenario also includes a scheduling side, and each edge side is communicatively connected to the scheduling side. The method includes: Based on the prediction model, the power and load of each photovoltaic and energy storage device in the photovoltaic and energy storage scenario are predicted at multiple time scales to obtain the prediction boundary conditions; the prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters. The predicted boundary conditions are sent to the scheduling side, which then uses a reinforcement learning algorithm to generate an initial control strategy for the photovoltaic-storage power station at various time scales under photovoltaic fluctuation conditions. Based on the preset hidden Markov model, deep state-space model, and the remaining lifetime of the target devices involved in the initial control strategy, the health of each target device is evaluated. Based on the reinforcement learning algorithm and the health status, the initial control strategy is optimized to obtain the target control strategy for the photovoltaic-storage power station.

[0011] In one embodiment, the training step of the prediction model includes: Receive the global initial parameters of the source-network-load-storage prediction model, which is a fusion of DCRNN (Diffusion Convolutional Recurrent Neural Network) and ST-GCN (Spatial-Temporal Graph Convolutional Network) network structures, issued by the scheduling layer. Using local operating data as training samples, the local deployed source-grid-load-storage prediction model incorporating DCRNN and ST-GCN is independently trained locally to obtain local model parameters. The local model parameters are sent to the scheduling layer, which then performs weighted aggregation of all local model parameters uploaded from the edge side based on the FedAvg (Federated Averaging) algorithm to obtain global model parameters. Receive the corresponding target global model parameters issued by the scheduling layer, and update the source-grid-load-storage prediction model based on the target global model parameters; The steps of obtaining local model parameters, sending the local model parameters to the scheduling layer, and updating the source-grid-load-storage prediction model are executed repeatedly until the accuracy of the source-grid-load-storage prediction model meets the preset convergence condition, thus obtaining the prediction model of source-grid-load-storage.

[0012] In addition, to achieve the above objectives, this application also proposes a health-based photovoltaic-storage control strategy optimization system in an energy system, which includes a scheduling side and an edge side. The scheduling side is used to receive the predicted boundary conditions sent by the edge side; based on the predicted boundary conditions, it uses a reinforcement learning algorithm to generate the initial control strategy of the photovoltaic-storage power station at each time scale under photovoltaic fluctuation conditions; based on the preset hidden Markov model, deep state space model and the remaining lifetime of the target equipment involved in the initial control strategy, it evaluates the health of each target equipment; based on the reinforcement learning algorithm and the health, it optimizes the initial control strategy to obtain the target control strategy of the photovoltaic-storage power station. The edge side is used to perform multi-time-scale prediction of the power and load of each photovoltaic storage device in the photovoltaic storage scenario based on the prediction model, obtain the prediction boundary conditions, and send the prediction boundary conditions to the scheduling side; the prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters.

[0013] Furthermore, to achieve the above objectives, this application also proposes a health-based photovoltaic-storage control strategy optimization device for an energy system. The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the health-based photovoltaic-storage control strategy optimization method for an energy system as described above.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the energy system health-based photovoltaic-storage control strategy optimization method as described above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: By receiving locally optimized predicted boundary conditions uploaded from each edge device, the system achieves refined perception of the operational status of photovoltaic and energy storage equipment at the edge devices and real-time information exchange with the scheduling decision-making layer. At the scheduling side, reinforcement learning algorithms are used to generate initial control strategies considering multiple time scales under photovoltaic fluctuation conditions. These initial control strategies are further jointly optimized by combining health assessments based on Hidden Markov Models, deep state-space models, and the remaining lifetime of the target equipment. This ensures that the final target control strategy meets operational requirements while proactively avoiding scheduling behaviors that could negatively impact the health of the target equipment, thereby reducing cumulative damage caused by regulation or power fluctuations. Therefore, by deeply integrating the equipment perception capabilities of the edge devices with the global optimization capabilities of the scheduling side, health becomes an endogenous variable in scheduling decisions, preventing the separation of equipment health assessment and operational control, and thus improving the coordination level of all aspects of the power generation, grid, load, and energy storage system in terms of equipment lifetime and operational safety. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating an embodiment of the photovoltaic-storage control strategy optimization method based on health in the energy system of this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the energy system health-based photovoltaic-storage control strategy optimization method provided in this application; Figure 3 This is a flowchart illustrating Embodiment 3 of the energy system health-based photovoltaic-storage control strategy optimization method provided in this application; Figure 4 This is a flowchart illustrating Embodiment 4 of the energy system health-based photovoltaic-storage control strategy optimization method provided in this application; Figure 5 This is a schematic diagram of the module structure of the photovoltaic-storage control strategy optimization system based on health status in the energy system of this application embodiment; Figure 6This is a schematic diagram of the equipment structure of the hardware operating environment involved in the health-based photovoltaic-storage control strategy optimization method in the energy system of this application embodiment.

[0018] Figure label: Processing device 1001, ROM 1002, storage device 1003, RAM 1004, bus 1005, I / O interface 1006, input device 1007, output device 1008, communication device 1009.

[0019] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0021] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0022] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or power grid control device capable of performing the above functions. The following description uses a power grid control device as an example to illustrate this embodiment and the subsequent embodiments.

[0023] Based on this, embodiments of this application provide a method for optimizing photovoltaic-storage control strategies based on health in energy systems, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the photovoltaic-storage control strategy optimization method based on health in the energy system of this application.

[0024] In this embodiment, the photovoltaic-storage power station is a type of new energy-storage coupled system. As a key component of low-carbon / zero-carbon parks, integrated energy management platforms, and energy-carbon collaborative management systems, the photovoltaic-storage power station undertakes the core functions of renewable energy consumption, load regulation, and carbon emission optimization. It should be understood that this application uses a photovoltaic-storage power station as an example to illustrate the core principles of constructing a multi-timescale, hierarchical collaborative control strategy optimization framework based on equipment health assessment, rather than limiting the scope of protection of this application. Those skilled in the art can seamlessly transfer the method to other types of new energy-storage coupled systems or broader regional energy management scenarios based on the teachings of this application. That is, this core idea is also applicable to new power systems such as wind-storage power stations, wind-solar-storage integrated power stations, and microgrids, and can serve as a bottom-level intelligent dispatch engine, integrated into energy management platforms and energy-carbon integrated systems for low-carbon parks, zero-carbon communities, or industrial parks, to achieve deep collaboration between equipment health management and operation control in terms of time scale and health status.

[0025] The aforementioned energy system's health-based photovoltaic-storage control strategy optimization method is applied to the dispatching side. The dispatching side belongs to a photovoltaic-storage scenario modeled using relational graph convolutional networks and graph attention networks based on the multidimensional heterogeneous parameters of real-world photovoltaic-storage power plants. The photovoltaic-storage scenario also includes edge sides representing each photovoltaic-storage power plant, and each edge side is communicatively connected to the dispatching side. The method includes steps S10-S40: Step S10: Receive the predicted boundary conditions after the edge side performs multi-time-scale prediction of the power and load of each photovoltaic storage device in the photovoltaic storage scenario using the prediction model. The prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters. It should be noted that photovoltaic (PV) and energy storage (ESS) power stations are integrated PV and ESS power stations, including but not limited to centralized PV and ESS power stations and park-level PV and ESS systems. The scheduling side is the centralized control center of the PV and ESS power station cluster, responsible for receiving prediction information uploaded from each edge side, generating and issuing multi-timescale control strategies based on global optimization objectives, and possessing reinforcement learning decision-making, health assessment, and strategy optimization functions. The edge side refers to intelligent terminals or edge computing nodes deployed locally at individual PV and ESS power stations, capable of local data acquisition, prediction model training, and execution of scheduling commands, and interacting with the scheduling side through a communication network. Multidimensional heterogeneous parameters are multi-source, multi-type data describing the operating status of PV and ESS power stations. Relationship graph convolutional networks are used to process the graph structure of PV and ESS power stations with multiple edge types. By assigning independent weight matrices to each type of relationship, it models the complex coupling relationships between different power stations. Graph attention networks are used to dynamically calculate the influence weights of neighboring nodes on the central node in the PV and ESS scenario graph, thereby adaptively focusing on related power stations that are more important to the current prediction or control task. Predictive boundary conditions are upper and lower bound parameters of the feasible region of the optimization problem on the dispatch side, which are output by the local prediction model on the edge side. They can include the maximum / minimum output of photovoltaic power, the load demand range, and the range of available energy storage capacity in each future period.

[0026] Understandably, by employing a federated learning framework, each edge device can optimize the prediction model based on global parameter initialization using its own data locally. This allows for the effective capture of unique local features of each photovoltaic and energy storage power station without uploading raw, sensitive operational data, thereby generating high-precision multi-timescale power and load prediction boundary conditions. These local features can include geographical location, equipment aging level, and micro-meteorological environment.

[0027] Understandably, predicting boundary conditions can provide accurate and privacy-secure input constraints for global optimization on the dispatching side, so as to avoid prediction bias caused by using a uniform model or historical average data. This allows the generated control strategy to more closely fit the actual physical operating limits of each power station, thus laying the foundation for achieving refined and personalized collaborative dispatching.

[0028] In practical implementation, the dispatch center can first issue a global initial parameter set for a source-grid-load-storage prediction model that integrates DCRNN and ST-GCN structures to the edge computing nodes of each photovoltaic-storage power station. Each edge node independently fine-tunes its locally deployed prediction model using local historical operating data (such as local weather station data, photovoltaic panel tilt angle, and energy storage battery health status). After training, the edge node uploads its local model parameters to the dispatch center. The dispatch center uses the FedAvg federated learning algorithm to perform a weighted average of the local model parameters of all edge nodes, generating updated global model parameters, which are then issued back to the corresponding edge nodes. This process is iterated until the prediction model accuracy of the edge nodes converges. Finally, each edge node uses the converged local model to predict photovoltaic output and load demand for the next 24 hours (day-ahead), 4 hours (intraday), and 15 minutes (real-time), and sends the upper and lower limits of the prediction results as prediction boundary conditions to the dispatch center.

[0029] Step S20: Based on the predicted boundary conditions, use a reinforcement learning algorithm to generate the initial control strategy of the photovoltaic-storage power station at each time scale under photovoltaic fluctuation conditions. It should be noted that the multiple time scales include three levels of control time scales: day-ahead, intraday, and real-time.

[0030] Understandably, reinforcement learning algorithms can learn optimal or suboptimal control strategies in complex and uncertain environments by continuously interacting and trying out photovoltaic fluctuation scenarios in a simulated environment, thereby improving control strategies in the face of highly random environmental changes.

[0031] Understandably, generating the initial control strategy directly under the predicted boundary conditions at multiple time scales can ensure the feasibility and security of the strategy. At the same time, it can embed the regulation requirements at different time scales into the design of the reinforcement learning reward function, so that the generated initial strategy has the characteristics of multi-time scale coordination, thereby improving the adaptability of source, grid, load and storage to photovoltaic fluctuations.

[0032] Because photovoltaic (PV) power output is affected by factors such as cloud cover and sudden weather changes, it exhibits strong randomness, intermittency, and volatility, making it a major source of disturbance to the stable operation of the power grid. In this embodiment, PV fluctuation conditions are used as a core component of the training and decision-making environment for the reinforcement learning algorithm. This ensures that the generated initial control strategy incorporates robustness to PV fluctuations from the design stage. Furthermore, the reinforcement learning agent learns through trial and error in numerous simulated PV fluctuation scenarios, autonomously exploring optimal or suboptimal control action sequences that can both mitigate power fluctuations and ensure equipment safety. Therefore, the final generated control strategy demonstrates strong adaptability and anti-interference capabilities when facing the complex and ever-changing PV fluctuations of the real world.

[0033] Step S30: Based on the preset hidden Markov model, deep state space model and the remaining lifetime of the target devices involved in the initial control strategy, assess the health of each target device. It should be noted that remaining lifespan refers to the estimated time or number of cycles that the target device (such as an energy storage battery or inverter) can operate safely under the current control strategy.

[0034] Understandably, by integrating the prior knowledge of equipment degradation state from the Hidden Markov Model, the dynamic fitting capability of the Deep State Space Model to high-dimensional real-time observation data, and the remaining lifetime information deduced from the initial control strategy, a multi-source information fusion health assessment system can be constructed to avoid one-sided health assessments of the target equipment.

[0035] Understandably, the integrated health assessment system not only considers the inherent historical attributes and current real-time status of the equipment, but also takes into account the impact of future control behaviors on its lifespan. This allows the health assessment results to reflect the current true state of the equipment and predict future health trends, thereby providing accurate and reliable quantitative basis for preventive maintenance and strategy optimization.

[0036] Step S40: Based on the reinforcement learning algorithm and the health status, optimize the initial control strategy to obtain the target control strategy of the photovoltaic-storage power station.

[0037] It should be noted that the target control strategy is the optimal set of instructions that is finally determined after health assessment and optimization, and is used to guide the operation of each target device in the photovoltaic-storage power station at each time scale.

[0038] Understandably, incorporating equipment health as a key constraint or negative reward factor into the optimization objective of reinforcement learning can enable the reinforcement learning algorithm to actively avoid high-risk control actions that would accelerate equipment aging or cause failures while seeking to maximize efficiency, thereby achieving a balance between efficiency and equipment health management.

[0039] In this embodiment, the control strategy is a power scheduling command generated by a reinforcement learning algorithm under photovoltaic fluctuation conditions, covering three time scales: day-ahead, intraday, and real-time. The operation and maintenance strategy is a quantitative health assessment result composed of the remaining equipment life inferred from the hidden Markov model, the deep state-space model, and the initial control strategy.

[0040] Specifically, the power grid control equipment incorporates health status as a key constraint or penalty into the optimization objective of reinforcement learning. This ensures that the control strategies at each level—day-ahead, intraday, and real-time—actively avoid actions that accelerate equipment aging or cause equipment failures during the generation process. In other words, the control of photovoltaic-storage power stations must comply with the requirements of long-term equipment health. Furthermore, based on the health status of each device predicted by the operation and maintenance strategy, the strategy search space of reinforcement learning can be dynamically adjusted to guide the generation of sustainable control strategies. Thus, through a two-way mechanism of control influencing health and health feedback correcting control, deep coupling and collaborative optimization of control strategies and operation and maintenance strategies are achieved across both the time scale and health status dimensions.

[0041] This embodiment provides a health-based photovoltaic (PV) and energy storage control strategy optimization method for energy systems, applied to the dispatching side. It receives locally optimized predicted boundary conditions uploaded from each edge device, enabling refined perception of the PV and energy storage equipment's operating status at the edge devices and real-time information exchange between the edge devices and the dispatching side's decision-making layer. At the dispatching side, a reinforcement learning algorithm is used to generate an initial control strategy considering multiple time scales under PV fluctuation conditions. This initial control strategy is further jointly optimized by combining the health assessment based on Hidden Markov Models, deep state-space models, and the remaining lifetime of the target equipment. This ensures that the final target control strategy meets operational requirements while proactively avoiding dispatching behaviors that negatively impact the health of the target equipment, thereby reducing cumulative damage caused by regulation or power fluctuations. Therefore, by deeply integrating the equipment perception capabilities of the edge devices with the global optimization capabilities of the dispatching side, health becomes an endogenous variable in dispatching decisions, avoiding the separation between equipment health assessment and operational control, and thus improving the coordination level of each link in the power generation, grid, load, and energy storage system across equipment lifetime and operational safety dimensions.

[0042] Based on Embodiment 1 of this application, in Embodiment 2 of this application, the content that is the same as or similar to that in Embodiment 1 can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 Step S40 also includes steps S01 to S03: Step S01: Based on the health status and the multi-timescale regulation attributes, the hierarchical reinforcement learning of the reinforcement learning algorithm is used to incorporate the health degradation trend of the equipment, maintenance arrangements and spare parts configuration requirements into the strategy optimization consideration dimensions. Step S02: Under photovoltaic fluctuation conditions, the initial control strategy at each time scale is constrained and modified by the reinforcement learning algorithm based on the equipment health and the optimization consideration dimensions. Step S03: Based on the reinforcement learning algorithm, iteratively optimize the constraint correction process to obtain the target control strategy of the photovoltaic-storage power station that adapts to the health status of the equipment, photovoltaic fluctuations, and multi-timescale operation requirements.

[0043] It should be noted that health status is a quantitative indicator of the current health state of target equipment within a photovoltaic-storage power station. Regulation attributes refer to the characteristics and requirements of the photovoltaic-storage power station's operation and control across different time dimensions. Hierarchical reinforcement learning is a reinforcement learning framework that decomposes complex decision-making tasks into multiple levels. Health degradation trend is a dynamic trajectory of equipment health changes over a future period, deduced through predictive models (such as deep state-space models) based on historical operating data and health assessment results.

[0044] Understandably, by using health as a core constraint, excessive operational stress can be effectively avoided when the equipment is in a vulnerable state, thereby suppressing the occurrence of unexpected failures from the source and extending the service life of the target equipment.

[0045] Understandably, the hierarchical reinforcement learning framework can organically combine long-term health management goals with short-term operational control goals. That is, when faced with random fluctuations in photovoltaic output, grid control equipment can not only maintain power balance, but also simultaneously execute protection equipment strategies, thereby improving the overall operational robustness of photovoltaic power generation under complex operating conditions.

[0046] Understandably, by predicting health degradation trends and incorporating them into optimization dimensions, maintenance and spare parts management can be proactively planned based on data, thereby reducing losses from emergency downtime and optimizing the allocation efficiency of maintenance resources.

[0047] In practical implementation, when the power grid control equipment detects a sharp decline in the health of a target device (e.g., a weekly decline in health exceeding 10%), it can automatically trigger an adaptive coupling mechanism. This mechanism can temporarily shorten the time interval between day-ahead and intraday optimizations and delegate some decision-making authority from the day-ahead layer to the intraday layer. This allows the control strategy to respond more agilely to sudden health risks, thereby avoiding equipment damage caused by long-cycle decision lags.

[0048] In practical implementation, when a power station's regulation capacity is insufficient due to equipment health issues, the higher layers of the hierarchical reinforcement learning framework can automatically identify neighboring power stations with good health and generate a cross-station support request. To incentivize this cooperative behavior, the power grid control equipment can also introduce a blockchain-based contribution record, i.e., health points that provide additional regulation services to the supporting party. These health points can be used for future priority maintenance or resource allocation, thereby building a healthy collaborative ecosystem at the regional power grid level.

[0049] In the specific implementation, for each time period Construct node power balance constraints, where N is the set of nodes. The power balance constraint formula is: in, Let be the photovoltaic power generation at node i at time t; Let be the energy storage charging and discharging power at node i at time t (charging is positive, discharging is negative); Let be the power exchanged between node i and the power grid at time t; Let be the load demand power at node i at time t.

[0050] The formula for satisfying the capacity and voltage constraints of the power flow in a line is: in, Let be the power flowing through node i at time t; The maximum allowed transmission power for node i; Let be the voltage at node i at time t; Let be the minimum voltage allowed at node i; This represents the maximum voltage allowed at node i.

[0051] Specifically, the hierarchical reinforcement learning includes a two-level decision architecture of policy layer and execution layer; The strategy layer outputs control strategy commands based on the health status of the target device; The execution layer, in conjunction with the multi-timescale control attributes, works with the control strategy instructions to complete the constraint correction and iterative optimization of the initial control strategy.

[0052] It should be noted that the policy layer is a high-level agent in the hierarchical reinforcement learning framework, used to output discrete, macroscopic control policy instructions based on a comprehensive perception of the current and predicted health status of target equipment within the photovoltaic-storage power station. The execution layer is a low-level agent in the hierarchical reinforcement learning framework, used to receive control policy instructions from the policy layer, and, in conjunction with the regulation attributes of the photovoltaic-storage power station at multiple time scales such as day-ahead, intraday, and real-time, dynamically constrain and modify the initial control policy aimed at operational efficiency, and ultimately generate an executable target control policy through iterative optimization.

[0053] Understandably, in the strategy and execution layer architecture, the strategy layer focuses on long-term, macro-level health maintenance, while the execution layer focuses on short-term, fine-grained operational optimization. The two work together through a clear instruction interface, which not only ensures the flexibility of control but also guarantees the effectiveness of health management.

[0054] Understandably, using health as the core decision-making basis at the strategy level and transforming health into a hard constraint on control strategies through the execution level fundamentally avoids applying excessive operational stress to equipment in a vulnerable state, effectively preventing unexpected failures and thus ensuring the long-term reliable operation of the power plant.

[0055] In practical implementation, if the execution layer detects in real time that the actual operating parameters of a target device deviate significantly from the health model prediction, it indicates that the target device has a sudden health risk that has not been identified by the strategy layer. At this time, the execution layer can proactively send a health risk alarm signal to the strategy layer, triggering the strategy layer to immediately reassess and update the control strategy instructions, thereby forming a bottom-up rapid feedback closed loop to enhance the proactive defense capability of the source-grid-load-storage system.

[0056] Furthermore, step S01 also includes: Based on the rolling window optimization mechanism, within each rolling window, the control attributes of the three-level control time scale are combined, and the hierarchical reinforcement learning of the reinforcement learning algorithm is used to synchronously associate the health degradation trend, maintenance schedule and spare parts configuration requirements of the target equipment, so as to obtain the preliminary dimensions of strategy optimization considerations under the current rolling window. Using the initial dimension as the basis for optimization, the strategy optimization solution is performed within the corresponding scrolling window to obtain the first step scheduling instruction for the current scale. The first step scheduling instruction is then sent to the edge side, where the edge side executes the first step scheduling instruction to obtain the latest predicted boundary conditions. The latest predicted boundary conditions from the edge-side backhaul are incorporated into the subsequent rolling window, and the multi-timescale constraints and the preliminary dimensions are dynamically iterated and updated to obtain the strategy optimization consideration dimensions that adapt to the temporal changes of photovoltaic power load fluctuations and equipment health degradation.

[0057] It should be noted that the rolling window optimization mechanism divides the entire optimization cycle into a series of continuous and partially overlapping time windows. The three-level control time scale refers to the three time dimensions upon which the operation and control of the photovoltaic-storage power station are based: the day-ahead scale, the intraday scale, and the real-time scale. The initial dimension considered in strategy optimization is a set of constraint or guidance information on equipment health management output by the high-level agent (policy layer) in the hierarchical reinforcement learning framework within the current rolling window; that is, the key parameters affecting the feasible domain of the lower-level control strategy. Multi-time scale constraints are a set of operational constraints originating from different time scales that must be simultaneously satisfied during the rolling optimization process.

[0058] Understandably, the rolling window mechanism can closely integrate the latest predicted boundary conditions fed back after execution on the edge side with the temporal changes in the health status of the equipment, forming a dynamic closed loop of optimization-execution-feedback-re-optimization. This ensures that the control strategy is always based on the most realistic and timely system state, thereby improving the effectiveness and security of the strategy.

[0059] Understandably, using the latest predicted boundary conditions as the core input for rolling optimization can avoid strategy failure caused by the accumulation of long-term prediction errors, so that the source-grid-load-storage system can continuously track and adapt to the random fluctuations in photovoltaic and energy storage power and load, thereby ensuring stable and reliable operation in scenarios with a high proportion of renewable energy access.

[0060] Understandably, by synchronously linking health degradation trends, maintenance schedules, and spare parts requirements within each rolling window at the strategy layer, and generating refined preliminary dimensions, health management can be integrated into proactive protection in every operational decision, thereby fundamentally extending the service life of critical equipment and reducing the risk of unplanned downtime.

[0061] In practical implementation, at the day-ahead scale Intraday scale and real-time scale Above, construct a hierarchical objective function: in, Let τ be the optimization cost / objective at the day-ahead scale. Let τ be the optimization cost / objective at the intraday scale; Let τ be the optimization cost / objective at the real-time scale. Within each rolling window, after solving the problem, only the first step of the scheduling instruction at the current time is executed.

[0062] Based on Embodiments 1 and 2 of this application, the same or similar content in Embodiment 3 of this application can be referred to the above description, and will not be repeated hereafter. Please refer to [the above description]. Figure 3 Step S30 also includes steps S31 to S34: Step S31: Based on the preset hidden Markov model, the inherent operating attributes of each device in the photovoltaic-storage power station are compared, and the degradation state range corresponding to the target device is determined from the inherent operating attributes. Step S32: Fit the real-time operation observation data of the target device based on the preset deep state space model, and extract the current degradation degree of the target device from the degradation state range; Step S33: Based on the remaining lifespan of the target device associated with the initial control strategy, the current degradation level is adjusted by parameters to obtain the target degradation level; Step S34: Quantify the degree of degradation of the target to generate the health status of the target device.

[0063] It should be noted that inherent operating attributes are the set of stable operating characteristics exhibited by each device in a photovoltaic-storage power station under a specific degradation state, distinguishing it from other states. The degradation state range is, for a specific target device, identified using a Hidden Markov Model based on its inherent operating attributes, as one or more consecutive healthy state intervals to which the target device is most likely currently located. The target degradation degree is a quantitative indicator, after parameter correction, that accurately reflects the current health status of the target device.

[0064] Understandably, through the two-level fusion of HMM and DSSM, HMM provides macroscopic, historical data-based prior knowledge of the degradation state, while DSSM utilizes real-time data for fine-grained, continuous tracking of the degree of degradation. Therefore, leveraging the complementary advantages of both can effectively avoid limitations when dealing with complex or non-stationary data, thereby improving the accuracy and robustness of health assessment.

[0065] Understandably, incorporating the future operational stresses implied by the initial control strategy into the health correction process allows for forward-looking predictions of future health evolution, thereby ensuring that scheduling decisions are based on real and instructive health information.

[0066] In practical implementation, this paper proposes a photovoltaic power station operation and maintenance decision-making technology based on hidden Markov degradation and hierarchical reinforcement learning to address the long-term degradation and operation and maintenance decision-making problems of photovoltaic modules, inverters and energy storage batteries.

[0067] Hidden Markov Modeling of Equipment Degradation: Constructing a Hidden Markov Model; Calculate the posterior probability of the state given the observation sequence: in, Given a complete sequence of observations at time t. Hidden state The posterior probability; The hidden state (potential degradation state of the device) at time t. The index for the hidden state; This represents the entire observation sequence from time 0 to T.

[0068] The operational strategy structure for hierarchical reinforcement learning; Power plant operation and maintenance decisions are divided into a strategy layer and an execution layer: in, The macro-control command output by the strategy layer (higher level) at time t; The specific executable action output by the execution layer (lower layer) at time t; For the policy function of the policy layer; For the strategy function of the execution layer; The device health index at time t; Let t be the macroscopic environmental conditions at time t (such as electricity price, weather level, etc.). This represents the detailed running status at time t.

[0069] Health-based policy adaptation and constraint-based operation and maintenance; Forming a constrained hierarchical reinforcement learning problem allows the strategy to achieve balance.

[0070] Based on the above embodiments of this application, in Embodiment 4 of this application, the content that is the same as or similar to the above embodiments can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 The energy system's health-based photovoltaic-storage control strategy optimization method is also applied to the edge side. The edge side belongs to a photovoltaic-storage scenario modeled using relational graph convolutional networks and graph attention networks based on the multidimensional heterogeneous parameters of real-world photovoltaic-storage power plants. The photovoltaic-storage scenario also includes a scheduling side, and each edge side is communicatively connected to the scheduling side. The method further includes steps S1-S2: Step S1: Based on the prediction model, perform multi-time-scale prediction of the power and load of each photovoltaic storage device in the photovoltaic storage scenario to obtain the prediction boundary conditions; the prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters. Step S2: The predicted boundary conditions are sent to the scheduling side, enabling the scheduling side to generate an initial control strategy for the photovoltaic-storage power station at various time scales based on the predicted boundary conditions using a reinforcement learning algorithm under photovoltaic fluctuation conditions. Based on a preset hidden Markov model, a deep state-space model, and the remaining lifetime of the target devices involved in the initial control strategy, the health of each target device is evaluated. Based on the reinforcement learning algorithm and the health status, the initial control strategy is optimized to obtain the target control strategy for the photovoltaic-storage power station.

[0071] This embodiment provides a health-based optimization method for photovoltaic (PV) and energy storage (ESS) control strategies in an energy system, applied to the edge side. By using an optimized local model at the edge side to predict the power and load of PV and ESS devices across multiple time scales, high-precision predictive boundary conditions reflecting the dynamic characteristics of local devices are generated and uploaded to the dispatch side in real time. This enables refined perception of the operating status of PV and ESS devices at the edge side and real-time information exchange between the dispatch side and the decision-making layer. At the dispatch side, reinforcement learning algorithms are used to generate an initial control strategy considering multiple time scales based on the predicted boundary conditions under PV fluctuation conditions. This initial control strategy is further jointly optimized by combining the health assessment based on Hidden Markov Models, Deep State Space Models, and the remaining lifetime of the target devices. This ensures that the final target control strategy meets operational requirements while proactively avoiding dispatch behaviors that could negatively impact the health of the target devices, thereby reducing cumulative damage caused by regulation or power fluctuations. Therefore, by deeply integrating the device perception capabilities at the edge side with the global optimization capabilities at the dispatch side, health becomes an endogenous variable in dispatch decisions, avoiding the separation between device health assessment and operational control, and improving the coordination level of all aspects of the power generation, grid, load, and energy storage systems in terms of device lifetime and operational safety.

[0072] Furthermore, the training process of the prediction model is as follows: Receive the global initial parameters of the source-grid-load-storage prediction model that integrates the DCRNN and ST-GCN network structures, issued by the scheduling layer; Using local operating data as training samples, the local deployed source-grid-load-storage prediction model incorporating DCRNN and ST-GCN is independently trained locally to obtain local model parameters. The local model parameters are sent to the scheduling layer, which then performs weighted aggregation of all local model parameters uploaded from the edge side based on the FedAvg algorithm to obtain global model parameters. Receive the corresponding target global model parameters issued by the scheduling layer, and update the source-grid-load-storage prediction model based on the target global model parameters; The steps of obtaining local model parameters, sending the local model parameters to the scheduling layer, and updating the source-grid-load-storage prediction model are executed repeatedly until the accuracy of the source-grid-load-storage prediction model meets the preset convergence condition, thus obtaining the prediction model of source-grid-load-storage.

[0073] It should be noted that local operational data refers to high-dimensional, multi-source, time-series raw data collected by sensors and monitoring systems deployed at the edge of the photovoltaic-storage power station. This data includes information on local distributed power sources (such as photovoltaic output), local loads, local energy storage status, and grid parameters (such as voltage and frequency) at the connection point. The preset convergence criteria are quantitative standards used to determine whether the training process of the source-grid-load-storage prediction model is complete.

[0074] Understandably, the composite model structure that integrates DCRNN and ST-GCN can accurately characterize the complex spatiotemporal coupling dynamics of the power system. Furthermore, the aggregation mechanism of federated learning enables the final model to learn the common patterns and unique characteristics under different regions, climates, and equipment configurations, thereby enhancing the model's generalization ability when facing unknown scenarios.

[0075] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the health-based photovoltaic-storage control strategy optimization method in the energy system of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0076] This application also provides a health-based photovoltaic-storage control strategy optimization system for energy systems. Please refer to [link / reference]. Figure 5 The energy system based on health status for photovoltaic and energy storage control strategy optimization includes: a dispatch side (central node) and an edge side, wherein the edge side includes grid components, energy storage, grid and users, etc. The scheduling side is used to receive the predicted boundary conditions sent by the edge side; based on the predicted boundary conditions, it uses a reinforcement learning algorithm to generate the initial control strategy of the photovoltaic-storage power station at each time scale under photovoltaic fluctuation conditions; based on the preset hidden Markov model, deep state space model and the remaining lifetime of the target equipment involved in the initial control strategy, it evaluates the health of each target equipment; based on the reinforcement learning algorithm and the health, it optimizes the initial control strategy to obtain the target control strategy of the photovoltaic-storage power station. The edge side is used to perform multi-time-scale prediction of the power and load of each photovoltaic storage device in the photovoltaic storage scenario based on the prediction model, obtain the prediction boundary conditions, and send the prediction boundary conditions to the scheduling side; the prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters.

[0077] The energy system health-based photovoltaic-storage control strategy optimization system provided in this application adopts the energy system health-based photovoltaic-storage control strategy optimization method in the above embodiments, which can solve the technical problem of insufficient coordination between power generation, grid, load, and storage. Compared with the prior art, the beneficial effects of the energy system health-based photovoltaic-storage control strategy optimization system provided in this application are the same as the beneficial effects of the energy system health-based photovoltaic-storage control strategy optimization method in the above embodiments, and other technical features of the energy system health-based photovoltaic-storage control strategy optimization system are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0078] This application provides a health-based photovoltaic-storage control strategy optimization device for an energy system. The health-based photovoltaic-storage control strategy optimization device for an energy system includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the health-based photovoltaic-storage control strategy optimization method for an energy system as described in Embodiment 1 above.

[0079] The following is for reference. Figure 6 The diagram illustrates a structural schematic of a health-based photovoltaic-storage control strategy optimization device suitable for implementing embodiments of this application in an energy system. The health-based photovoltaic-storage control strategy optimization device in the energy system of this application can include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, personal digital assistants, tablets, portable multimedia players, and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 6 The energy system-based photovoltaic-storage control strategy optimization device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments in this application.

[0080] like Figure 6As shown, the energy system's health-based photovoltaic-storage control strategy optimization device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the energy system's health-based photovoltaic-storage control strategy optimization device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. An I / O (input / output) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the health-based photovoltaic-storage control strategy optimization device in the energy system to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a health-based photovoltaic-storage control strategy optimization device in an energy system with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0081] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0082] The energy system health-based photovoltaic-storage control strategy optimization device provided in this application adopts the energy system health-based photovoltaic-storage control strategy optimization method in the above embodiments, which can solve the technical problem of insufficient coordination between power generation, grid, load, and storage. Compared with the prior art, the beneficial effects of the energy system health-based photovoltaic-storage control strategy optimization device provided in this application are the same as the beneficial effects of the energy system health-based photovoltaic-storage control strategy optimization method provided in the above embodiments, and other technical features in the energy system health-based photovoltaic-storage control strategy optimization device are the same as the features disclosed in the previous embodiment method, and will not be repeated here.

[0083] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0084] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0085] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the energy system based on health-related photovoltaic-storage control strategy optimization method in the above embodiments.

[0086] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0087] The aforementioned computer-readable storage medium may be included in the health-based photovoltaic-storage control strategy optimization device in the energy system; or it may exist independently and not be installed in the health-based photovoltaic-storage control strategy optimization device in the energy system.

[0088] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a health-based photovoltaic-storage control strategy optimization device in an energy system, the health-based photovoltaic-storage control strategy optimization device in the energy system performs the following: receives prediction boundary conditions from the edge side after multi-time-scale prediction of the power and load of each photovoltaic-storage device in the photovoltaic-storage scenario using a prediction model, wherein the prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters; based on the prediction boundary conditions, generates an initial control strategy for the photovoltaic-storage power station at each time scale using a reinforcement learning algorithm under photovoltaic fluctuation conditions; evaluates the health of each target device based on a preset hidden Markov model, a deep state-space model, and the remaining lifetime of the target devices involved in the initial control strategy; and optimizes the initial control strategy based on the reinforcement learning algorithm and the health, thereby obtaining the target control strategy for the photovoltaic-storage power station.

[0089] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0090] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0091] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0092] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described energy system health-based photovoltaic-storage control strategy optimization method, thereby solving the technical problem of insufficient coordination between power generation, grid, load, and storage. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the energy system health-based photovoltaic-storage control strategy optimization method provided in the above embodiments, and will not be repeated here.

[0093] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A method for optimizing a photovoltaic-storage control strategy based on health in an energy system, characterized in that, Applied to the scheduling side, the scheduling side belongs to a photovoltaic-storage scenario modeled using relational graph convolutional networks and graph attention networks based on the multidimensional heterogeneous parameters of real photovoltaic-storage power stations. The photovoltaic-storage scenario also includes edge sides representing each photovoltaic-storage power station, and each edge side is communicatively connected to the scheduling side. The method includes: The edge side receives the predicted boundary conditions after performing multi-time-scale predictions of the power and load of each photovoltaic storage device in the photovoltaic storage scenario through a prediction model. The prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters. Based on the predicted boundary conditions, a reinforcement learning algorithm is used to generate the initial control strategy of the photovoltaic-storage power station at each time scale under photovoltaic fluctuation conditions. Based on the preset hidden Markov model, deep state space model, and the remaining lifetime of the target devices involved in the initial control strategy, the health of each target device is evaluated. Based on the reinforcement learning algorithm and the health status, the initial control strategy is optimized to obtain the target control strategy of the photovoltaic-storage power station.

2. The method for optimizing photovoltaic-storage control strategy based on health in an energy system as described in claim 1, characterized in that, The steps of optimizing the initial control strategy based on the reinforcement learning algorithm and the health status to obtain the target control strategy for the photovoltaic-storage power station include: Based on the health status and multi-timescale regulation attributes, the hierarchical reinforcement learning of the reinforcement learning algorithm is used to incorporate the health degradation trend of the equipment, maintenance arrangements and spare parts configuration requirements into the strategy optimization consideration dimensions. Under photovoltaic fluctuation conditions, the reinforcement learning algorithm is used to constrain and modify the initial control strategy at each time scale based on the equipment health and the optimization consideration dimensions. Based on the reinforcement learning algorithm, the constraint correction process is iteratively optimized to obtain the target control strategy of the photovoltaic-storage power station that adapts to the health status of the equipment, photovoltaic fluctuations, and multi-timescale operation requirements.

3. The method for optimizing photovoltaic-storage control strategy based on health in an energy system as described in claim 2, characterized in that, The hierarchical reinforcement learning includes a two-level decision-making architecture of policy layer and execution layer; The strategy layer outputs control strategy commands based on the health status of the target device; The execution layer, in conjunction with the multi-timescale control attributes, works with the control strategy instructions to complete the constraint correction and iterative optimization of the initial control strategy.

4. The method for optimizing photovoltaic-storage control strategy based on health in an energy system as described in claim 2, characterized in that, The multiple time scales include three levels of control time scales: day-ahead, intraday, and real-time. Based on the aforementioned health status and multi-timescale regulation attributes, the steps of incorporating the equipment's health degradation trend, maintenance schedule, and spare parts configuration requirements into the strategy optimization considerations using the hierarchical reinforcement learning algorithm include: Based on the rolling window optimization mechanism, within each rolling window, the control attributes of the three-level control time scale are combined, and the hierarchical reinforcement learning of the reinforcement learning algorithm is used to synchronously associate the health degradation trend, maintenance schedule and spare parts configuration requirements of the target equipment, so as to obtain the preliminary dimensions of strategy optimization considerations under the current rolling window. Using the initial dimension as the basis for optimization, the strategy optimization solution is performed within the corresponding scrolling window to obtain the first step scheduling instruction for the current scale. The first step scheduling instruction is then sent to the edge side, where the edge side executes the first step scheduling instruction to obtain the latest predicted boundary conditions. The latest predicted boundary conditions from the edge-side backhaul are incorporated into the subsequent rolling window, and the multi-timescale constraints and the preliminary dimensions are dynamically iterated and updated to obtain the strategy optimization consideration dimensions that adapt to the temporal changes of photovoltaic power load fluctuations and equipment health degradation.

5. The method for optimizing photovoltaic-storage control strategy based on health in an energy system as described in claim 1, characterized in that, The steps for assessing the health of each target device based on a preset hidden Markov model, a deep state-space model, and the remaining lifetime of the target devices involved in the initial control strategy include: Based on a preset hidden Markov model, the inherent operating attributes of each device in the photovoltaic-storage power station are benchmarked, and the degradation state range corresponding to the target device is determined from the inherent operating attributes. Based on a preset deep state space model, the real-time operational observation data of the target device are fitted, and the current degree of degradation of the target device is extracted from the degradation state range; Based on the remaining lifespan of the target device associated with the initial control strategy, the current degradation level is adjusted by parameters to obtain the target degradation level; The health status of the target device is generated by quantifying the degree of degradation of the target.

6. A method for optimizing a photovoltaic-storage control strategy based on health in an energy system, characterized in that, Applied to the edge side, the edge side belongs to a photovoltaic-storage scenario modeled using relational graph convolutional networks and graph attention networks based on the multidimensional heterogeneous parameters of real photovoltaic-storage power plants. The photovoltaic-storage scenario also includes a scheduling side, and each edge side is communicatively connected to the scheduling side. The method includes: Based on the prediction model, the power and load of each photovoltaic and energy storage device in the photovoltaic and energy storage scenario are predicted at multiple time scales to obtain the prediction boundary conditions; the prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters. The predicted boundary conditions are sent to the scheduling side, which then uses a reinforcement learning algorithm to generate an initial control strategy for the photovoltaic-storage power station at various time scales under photovoltaic fluctuation conditions. Based on the preset hidden Markov model, deep state-space model, and the remaining lifetime of the target devices involved in the initial control strategy, the health of each target device is evaluated. Based on the reinforcement learning algorithm and the health status, the initial control strategy is optimized to obtain the target control strategy for the photovoltaic-storage power station.

7. The method for optimizing photovoltaic-storage control strategy based on health in an energy system as described in claim 6, characterized in that, The training steps of the prediction model include: Receive the global initial parameters of the source-grid-load-storage prediction model that integrates the DCRNN and ST-GCN network structures, issued by the scheduling layer; Using local operating data as training samples, the local deployed source-grid-load-storage prediction model incorporating DCRNN and ST-GCN is independently trained locally to obtain local model parameters. The local model parameters are sent to the scheduling layer, which then performs weighted aggregation of all local model parameters uploaded from the edge side based on the FedAvg algorithm to obtain global model parameters. Receive the corresponding target global model parameters issued by the scheduling layer, and update the source-grid-load-storage prediction model based on the target global model parameters; The steps of obtaining local model parameters, sending the local model parameters to the scheduling layer, and updating the source-grid-load-storage prediction model are executed repeatedly until the accuracy of the source-grid-load-storage prediction model meets the preset convergence condition, thus obtaining the prediction model of source-grid-load-storage.

8. A photovoltaic-storage control strategy optimization system based on health status in an energy system, characterized in that, The system includes: a scheduling side and an edge side; The scheduling side is used to receive the predicted boundary conditions sent by the edge side; based on the predicted boundary conditions, it uses a reinforcement learning algorithm to generate the initial control strategy of the photovoltaic-storage power station at each time scale under photovoltaic fluctuation conditions; based on the preset hidden Markov model, deep state space model and the remaining lifetime of the target equipment involved in the initial control strategy, it evaluates the health of each target equipment; based on the reinforcement learning algorithm and the health, it optimizes the initial control strategy to obtain the target control strategy of the photovoltaic-storage power station. The edge side is used to perform multi-time-scale prediction of the power and load of each photovoltaic storage device in the photovoltaic storage scenario based on the prediction model, obtain the prediction boundary conditions, and send the prediction boundary conditions to the scheduling side; the prediction model is obtained by the edge side after optimizing the local model based on the corresponding global model parameters.

9. A photovoltaic-storage control strategy optimization device based on health status in an energy system, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the method for optimizing a health-based photovoltaic-storage control strategy in an energy system as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the energy system based on health-related photovoltaic-storage control strategy optimization method as described in any one of claims 1 to 7.

Images