Photovoltaic reverse power consumption system based on mobile energy storage vehicle

Abstract

The invention provides a photovoltaic reverse power consumption system based on a mobile energy storage vehicle. A photovoltaic power generation unit is configured to output photovoltaic electric energy, supply power to a load and realize grid-connected operation through a grid-connected point of a power grid; the reverse power detection and protection unit dynamically generates a self-adaptive reverse power protection threshold value through a time sequence prediction model, compares a real-time reverse power value with the reverse power protection threshold value, generates a reverse power trigger signal when the real-time reverse power value exceeds the threshold value, and sends the reverse power trigger signal to the central control unit; the central control unit distributes a consumption task for the mobile energy storage vehicle cluster through a multi-agent collaborative decision model, and plans a driving path fusing the own energy state and the driving energy efficiency for each mobile energy storage vehicle participating in the consumption; and each mobile energy storage vehicle in the mobile energy storage vehicle cluster receives the scheduling instruction, travels to the corresponding charging facility to access the charging loop, and consumes the reverse power electric energy generated by the photovoltaic power generation unit.

Description

Technical Field

[0001] This invention belongs to the field of distributed photovoltaic power generation grid-connected operation control and energy storage system optimization scheduling technology, specifically involving a photovoltaic reverse power consumption system based on a mobile energy storage vehicle. Background Technology

[0002] As the proportion of photovoltaic (PV) power generation in the energy structure continues to increase, PV systems are prone to triggering reverse power protection under low load or grid faults, leading to a growing problem of curtailment. Traditional solutions mainly rely on fixed energy storage systems or off-load resistors, which have significant limitations: Fixed energy storage devices typically use energy storage media such as lead-acid batteries and lithium-ion batteries, deployed in containerized or cabinet-like forms in specific locations. Their capacity and power configuration are relatively fixed, making it difficult to match the spatiotemporal fluctuations of distributed PV power generation. In scenarios with large differences in PV output peaks and valleys and dispersed reverse power locations, an imbalance often occurs where local capacity is insufficient while other areas have idle capacity. While off-load resistor solutions are low-cost and have a rapid response, they convert valuable electrical energy into heat dissipation, not only wasting energy and reducing the overall economic efficiency of the system, but also potentially causing safety hazards such as heat dissipation and fire prevention, which does not conform to the basic principle of efficient use of green energy. Especially in scenarios such as large-scale PV power plants, industrial parks, and commercial building complexes, the wide distribution of PV arrays and the large differences in load characteristics, along with the instantaneous, intermittent, and spatially dispersed nature of reverse power, place higher demands on absorption technologies.

[0003] In existing technologies, reverse power detection often employs a judgment mechanism based on a fixed threshold, typically setting a static power percentage or absolute value as the trigger threshold. This simple threshold method cannot adapt to rapid power fluctuations caused by sudden changes in sunlight intensity or cloud cover, nor can it respond to dynamic changes in load. In actual operation, it is prone to two typical problems: First, if the threshold setting is too conservative, the reverse power may actually exceed the grid's capacity without triggering protection in time, affecting the stable operation of the grid. Second, if the threshold setting is too sensitive, it may frequently malfunction due to short-term fluctuations in sunlight or instantaneous changes in load, causing unnecessary system shutdowns or frequent start-ups and shutdowns of energy storage devices, accelerating equipment aging and reducing the effective power generation time of photovoltaics. In addition, traditional detection methods are mostly based on simple comparisons of real-time power sampling values, lacking in-depth analysis of historical operating data and the ability to predict future trends, making it difficult to provide early warning and preprocessing in the early stages of reverse power formation.

[0004] In energy storage scheduling, existing solutions generally employ static strategies based on rules or simple priorities, such as assigning tasks according to the remaining capacity of energy storage devices, distance, or a fixed rotation order. These methods lack a holistic optimization perspective for the coordinated operation of multiple energy storage units and cannot make dynamic decisions based on multi-dimensional information such as real-time reverse power distribution, the status of each energy storage device (e.g., state of charge, health, current location), and the availability of charging facilities. When multiple photovoltaic nodes simultaneously experience reverse power and there is competition for grid connection demand, static scheduling strategies often lead to overuse of some energy storage devices while others remain idle, or uneconomical behavior such as energy storage devices moving long distances to absorb small amounts of reverse power, making it difficult to maximize the overall system grid connection efficiency and minimize operating costs. Although some research has attempted to introduce optimization algorithms for task allocation, these have mostly focused on single-vehicle or fixed-path scenarios, and effective solutions are still lacking for the coordinated scheduling problem of multiple mobile energy storage units in dynamic environments.

[0005] Path planning is a crucial component of mobile energy storage and utilization systems. Existing technologies mostly employ traditional graph search algorithms (such as A* and Dijkstra's algorithms) or geometric path planning methods, primarily optimizing for the shortest travel distance or time. These algorithms typically treat energy storage vehicles as ordinary mobile robots or transportation tools, failing to fully consider their unique attributes as energy carriers. On one hand, the energy state of the energy storage vehicle (such as battery state of charge) directly affects its travel distance and task execution capability; a vehicle with low battery power may be unable to complete long-distance travel tasks. On the other hand, energy consumption during vehicle operation is closely related to factors such as speed, road conditions, and load. Simply pursuing the shortest path may lead the vehicle to choose routes with steep inclines and poor road conditions, thereby increasing overall energy consumption. More importantly, traditional path planning algorithms often assume a static and unchanging environment, lacking the ability to adapt to real-time environmental information such as dynamic obstacles, temporary traffic control, and changes in the status of charging facilities. When a planned path becomes unavailable due to unforeseen circumstances, it often needs to be completely replanned, resulting in high computational costs and significant response delays.

[0006] The aforementioned technical limitations are interconnected and mutually restrictive, collectively leading to the core problems of low efficiency and insufficient adaptability in current photovoltaic reverse power consumption systems. The rigidity of fixed threshold detection limits the system's perception accuracy in complex operating environments, the one-sidedness of static scheduling strategies hinders the realization of the collaborative potential of multiple devices, and path planning that ignores energy characteristics affects the overall energy efficiency of mobile energy storage units. These shortcomings are particularly prominent against the backdrop of the continuous expansion of distributed photovoltaic scale and the increasing requirements for grid interaction, urgently requiring a systematic solution that integrates intelligent detection, collaborative scheduling, and energy efficiency optimization. Summary of the Invention

[0007] To address the shortcomings and deficiencies of existing technologies, this invention provides a photovoltaic reverse power consumption system based on mobile energy storage vehicles, aiming to solve the problem of curtailment caused by reverse power protection in photovoltaic systems. The core of this invention lies in constructing a closed-loop consumption architecture that integrates intelligent detection, collaborative scheduling, and energy self-consistency. The system collects photovoltaic output, load, and environmental data in real time through a reverse power detection and protection unit, and uses a Long Short-Term Memory Recurrent Neural Network (LSTM) model to dynamically predict adaptive reverse power protection thresholds, replacing the traditional fixed threshold judgment to improve detection accuracy and reduce false alarms. When reverse power exceeds the dynamic threshold, the central control unit receives a trigger signal and, based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, performs collaborative task allocation for the mobile energy storage vehicle cluster. This algorithm constructs a global state space including photovoltaic nodes, charging facilities, and vehicle dynamic parameters, and optimizes the scheduling strategy by integrating a multi-objective reward function that integrates photovoltaic consumption rate, battery state-of-charge balance, vehicle travel distance, and charging efficiency. The weight ratio of each objective is determined through Pareto front analysis. Meanwhile, the central control unit uses the D*Lite algorithm to plan the driving path for each energy storage vehicle. This algorithm incorporates the vehicle's real-time battery state of charge and driving speed into a dynamic cost function to achieve energy-efficient path planning and has the ability to incrementally replan in response to environmental changes. The mobile energy storage vehicle cluster moves to the corresponding charging facility according to the scheduling instructions to connect to the charging circuit and consume reverse power energy. The charging facility provides real-time feedback on the vehicle's charging and battery status, and the energy consumption of the energy storage vehicle is entirely supplied by its own battery pack, thus forming a self-sufficient closed-loop energy management system. This invention significantly improves the flexibility, efficiency, and economy of photovoltaic reverse power consumption through the organic combination of the above-mentioned dynamic threshold prediction, multi-vehicle collaborative scheduling, and energy self-consistent closed loop.

[0008] The specific technical solution adopted by this invention to solve its technical problem is as follows:

[0009] A photovoltaic reverse power consumption system based on a mobile energy storage vehicle, wherein the reverse power detection and protection unit is electrically connected to the photovoltaic power generation unit, the central control unit is communicatively connected to the reverse power detection and protection unit, the charging facility, and the mobile energy storage vehicle cluster, and the charging facility is electrically connected to the power output side of the photovoltaic power generation unit.

[0010] The photovoltaic power generation unit is configured to output photovoltaic power to supply power to the load and achieve grid-connected operation through the grid connection point;

[0011] The reverse power detection and protection unit is configured to collect historical operating data, real-time output status, load status and real-time environmental parameters of the photovoltaic power generation unit, dynamically generate an adaptive reverse power protection threshold through a time-series prediction model, compare the real-time reverse power value with the reverse power protection threshold, and generate a reverse power trigger signal and send it to the central control unit when the real-time reverse power value exceeds the threshold.

[0012] The central control unit is configured to receive the reverse power trigger signal, and based on the real-time reverse power status, the availability status of charging facilities, and the real-time status of each mobile energy storage vehicle, allocate consumption tasks to the mobile energy storage vehicle cluster through a multi-agent collaborative decision-making model, and plan a driving path that integrates its own energy status and driving energy efficiency for each mobile energy storage vehicle participating in the consumption, and generate corresponding scheduling instructions to be sent to the mobile energy storage vehicle cluster.

[0013] Each mobile energy storage vehicle in the mobile energy storage vehicle cluster is equipped with an energy storage battery pack, a communication module, a positioning module, and a driving unit. It is configured to receive the scheduling command, drive to the corresponding charging facility, connect to the charging circuit, and consume the reverse power energy generated by the photovoltaic power generation unit.

[0014] Furthermore, the inverse power detection and protection unit includes a data processing submodule and a dynamic threshold prediction submodule. The data processing submodule performs moving average filtering and irradiance coupling correction preprocessing on the collected raw data in sequence. The moving average filtering is used to suppress high-frequency noise in the raw data and output a smooth photovoltaic power signal. The irradiance coupling correction is used to correct the photovoltaic power according to the irradiance deviation and eliminate environmental interference. The corrected photovoltaic output power satisfies the following expression:

[0015]

[0016] In the formula, Let be the photovoltaic output power value after moving average processing at time t. These are the correction coefficients obtained through regression fitting of historical irradiance data. Let be the real-time irradiance at time t. This is the average radiation intensity over a previously preset 24-hour period. This is the standard deviation of the radiation intensity over a previously preset 24-hour period. The value of the photovoltaic output power after irradiance coupling correction at time t is t.

[0017] Furthermore, the time-series prediction model used by the reverse power detection and protection unit is an LSTM model; the LSTM model constructs a time-series input vector based on the preprocessed corrected photovoltaic power, power change rate, and real-time irradiance, and outputs an adaptive reverse power protection threshold, the threshold output satisfying the following expression:

[0018]

[0019] In the formula, Let t be the adaptive inverse power protection threshold. Let be the hidden state vector of the LSTM model at time t. The weight parameters of the output layer of the LSTM model. These are the bias parameters for the output layer of the LSTM model. To modify the activation function of the linear unit, This is the adjustment coefficient for the rate of change of power. This represents the real-time rate of change of photovoltaic output power.

[0020] Furthermore, the reverse power detection and protection unit determines whether to generate a reverse power trigger signal through the following logic:

[0021]

[0022] In the formula, This indicates the generation of an inverse power trigger signal. This indicates that no reverse power trigger signal is generated. The value of the photovoltaic output power at time t is the preprocessed and corrected value. Let be the real-time load power of the system at time t. This is the maximum reverse power value allowed by the power grid. Let t be the adaptive inverse power protection threshold output by the LSTM model at time t.

[0023] Furthermore, the central control unit employs the MADDPG algorithm as its multi-agent collaborative decision-making model. The MADDPG algorithm integrates photovoltaic point parameters, charging facility state parameters, and the dynamic parameters of the mobile energy storage vehicle into a global state space, and the global state vector satisfies the following expression:

[0024]

[0025] In the formula, Let be the global state vector at time t. For the battery state of charge of all mobile energy storage vehicles, This represents the real-time reverse power value of each photovoltaic node. The availability status of charging facilities. The coordinates of obstacles within the scheduling area, This represents the real-time output power of the photovoltaic power generation unit. The charging power for mobile energy storage vehicles, This refers to the real-time driving speed of the mobile energy storage vehicle. The charging efficiency of charging facilities.

[0026] Furthermore, the MADDPG algorithm constructs a multi-objective reward function to optimize the scheduling strategy, and the multi-objective reward function satisfies the following expression:

[0027]

[0028] In the formula, Let be the reward value at time t. This represents the total photovoltaic reverse power actually consumed by all mobile energy storage vehicles at the current moment. This represents the total power generation of the photovoltaic power generation unit at the current moment. Let represent the state of charge (SOC) of the battery of the i-th mobile energy storage vehicle. Let be the distance traveled by the i-th mobile energy storage vehicle at the current moment. This is a weighting factor for charging efficiency. The actual charging efficiency of the charging facilities. This refers to the real-time charging power of the mobile energy storage vehicle.

[0029] Furthermore, in the multi-objective reward function, the weight ratios of the photovoltaic absorption term, the state of charge equilibrium penalty term, and the travel distance penalty term are determined through Pareto front analysis.

[0030] Furthermore, the central control unit plans the driving path of the mobile energy storage vehicle by using the D*Lite algorithm, which integrates the vehicle's own energy state and driving efficiency. The algorithm sets a dynamic cost function for the driving path, incorporating the vehicle's battery state of charge and real-time driving speed into the path cost calculation. The dynamic cost function satisfies the following expression:

[0031]

[0032] In the formula, Let be the final dynamic cost from vertex v to vertex v'. Let v be the basic movement cost from vertex v to vertex v'. These are the weighting coefficients for the state of charge. This provides the real-time battery state of charge for mobile energy storage vehicles. For speed weighting coefficients, This refers to the real-time driving speed of the mobile energy storage vehicle.

[0033] Furthermore, the D*Lite algorithm detects environmental changes within the scheduling area in real time. When a path cost change caused by an obstacle is detected, the path cost and priority of the affected vertex are updated, and the local driving path is corrected through incremental replanning. When no feasible path exists, an exception handling mechanism is triggered.

[0034] Furthermore, the charging facility collects the charging status of the mobile energy storage vehicle and the status of the energy storage battery pack in real time and feeds it back to the central control unit; the driving unit of the mobile energy storage vehicle is powered by its own energy storage battery pack.

[0035] Compared to existing technologies, this invention and its preferred solution, from the perspective of overall system architecture, adopt a cluster of mobile energy storage vehicles to replace traditional fixed energy storage systems. This overcomes the inherent limitations of fixed energy storage equipment, such as rigid layout and poor capacity adaptability. It can flexibly match the temporal and spatial fluctuation characteristics of distributed photovoltaic reverse power, effectively solving the problems of low capacity utilization and difficulty in scaling up to multi-node distributed photovoltaic scenarios. At the same time, it avoids the waste of renewable energy and safety hazards caused by traditional offloading resistor schemes, achieving efficient recovery and utilization of excess photovoltaic power while ensuring grid connection safety. In the reverse power detection and protection stage, it abandons the traditional fixed threshold judgment logic and first processes the collected raw data. Preprocessing is performed to eliminate environmental interference and data noise. Then, an adaptive reverse power protection threshold is dynamically generated using a time-series prediction model. This allows for real-time adaptation to complex operating conditions such as sudden changes in sunlight and dynamic load changes, effectively reducing the probability of false triggering of reverse power protection and avoiding unnecessary photovoltaic curtailment losses. This achieves a balanced optimization between grid-connected operation safety and photovoltaic absorption capacity. In the energy storage cluster scheduling stage, a multi-agent collaborative decision-making model is used to achieve collaborative optimization scheduling of multiple mobile energy storage vehicles. A multi-objective optimization system covering photovoltaic absorption, battery health, driving energy consumption, and charging efficiency is constructed. Pareto front analysis is used to determine the weight ratio of each optimization objective, solving the problems of poor adaptability and multi-objective limitations of traditional static scheduling strategies. The weighting system addresses the issues of subjective setting and mismatch of energy storage resources. It maximizes the efficiency of photovoltaic reverse power absorption while balancing the lifespan of energy storage batteries and energy consumption control during dispatch, achieving optimal allocation of energy storage resources. In the path planning stage, it overcomes the limitations of traditional path planning that only focuses on geometric distance. By integrating the energy state, speed, and energy consumption characteristics of the mobile energy storage vehicle into the path cost calculation, it can respond in real-time to environmental changes in the dispatch area, performing incremental local path corrections. This results in the planning of optimal paths that combine accessibility and driving energy efficiency, effectively reducing ineffective energy consumption during dispatch while ensuring the real-time performance and reliability of path planning, and avoiding range constraints during energy storage unit dispatch. Risks; In addition, this solution achieves closed-loop optimization of the scheduling strategy through real-time status feedback of charging facilities. At the same time, it relies on the energy storage battery of the mobile energy storage vehicle to supply the driving energy consumption, forming a self-sufficient closed-loop energy management system. No additional energy supply facilities are required, which further improves the self-consistency and economy of system operation. Overall, this solution has both spatial deployment flexibility and operational energy self-consistency, which can effectively reduce the downtime frequency of photovoltaic systems caused by reverse power protection, improve the comprehensive utilization rate of renewable energy, and adapt to the reverse power consumption needs of high-proportion photovoltaic grid-connected scenarios such as distributed photovoltaic power stations and industrial parks, providing stable and reliable technical support for high-proportion renewable energy to access the distribution network. Attached Figure Description

[0036] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0037] Figure 1 This is an overall architecture diagram of a photovoltaic reverse power consumption system based on a mobile energy storage vehicle, according to an embodiment of the present invention.

[0038] Figure 2 This is a flowchart of the mobile energy storage vehicle path planning process based on the D*Lite algorithm in an embodiment of the present invention.

[0039] Figure 3 The figures show a performance comparison of the photovoltaic reverse power absorption system based on a mobile energy storage vehicle according to an embodiment of the present invention. In the figures, (a) is a comparison of reverse power detection effect, (b) is a comparison of false trigger rate, (c) is a demonstration of multi-energy storage vehicle collaborative scheduling, and (d) is a comparison of absorption efficiency. Detailed Implementation

[0040] To make the features and advantages of the present invention more apparent and understandable, specific embodiments are described below in detail:

[0041] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0042] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0043] To address the problems of existing technologies, this invention proposes a dynamic energy consumption system based on mobile energy storage vehicles. This system aims to solve the problem of curtailment caused by reverse power protection in photovoltaic power generation systems under low load or grid fault conditions, and to overcome the limited coverage of fixed energy storage. It reduces false triggering rates, improves photovoltaic energy consumption rates, and provides key technical support for high-proportion renewable energy integration. First, an LSTM neural network is used to predict the dynamic threshold of reverse power, replacing the traditional fixed threshold detection method and significantly reducing the false triggering rate. Second, the MADDPG algorithm is introduced to achieve collaborative scheduling of the energy storage vehicle cluster, simultaneously optimizing energy consumption rates, battery life, and mobile energy consumption through a multi-objective reward function. Finally, the D* Lite algorithm is combined to achieve dynamic path planning, incorporating parameters such as energy storage vehicle SOC and speed into the cost function, making path selection both energy-efficient and real-time.

[0044] The system mainly consists of photovoltaic power generation units, mobile energy storage vehicles, charging piles, and a central controller. When the reverse power protection module detects that the reverse power exceeds a set threshold, the central controller dispatches the energy storage vehicle to a charging pile connected to the photovoltaic power generation unit. The vehicle's battery is then charged via a charging interface to absorb the reverse power. Simultaneously, the energy consumption of the energy storage vehicle is supplied by its own energy storage battery, forming a closed-loop energy management system. The system employs a multi-energy storage vehicle collaborative scheduling algorithm to optimize charging task allocation based on the magnitude of the reverse power, thereby improving absorption efficiency.

[0045] Compared to traditional fixed energy storage solutions, this system combines spatial flexibility and energy self-sufficiency, which can reduce the downtime frequency of photovoltaic systems and improve the utilization rate of renewable energy. It is suitable for distributed photovoltaic power stations, industrial parks and other scenarios.

[0046] The specific implementation of the present invention will be further illustrated and described below with reference to the accompanying drawings:

[0047] Figure 1 As shown, the system provided in this embodiment includes a photovoltaic power generation layer, a central control layer, and a mobile energy storage layer.

[0048] The photovoltaic (PV) power generation layer includes a PV array and a reverse power protection module. The reverse power protection module includes a data processing module and a dynamic threshold long short-term memory (LSTM) recurrent neural network prediction model. When the reverse power protection module detects that the PV array's power generation does not exceed the rated power of the reverse power detection, the current generated by the PV array flows into the grid at the grid connection point. When the power generation exceeds the rated power, reverse power protection is triggered, and a signal is transmitted to the central controller.

[0049] The central control layer consists of a central controller. After receiving the reverse power protection trigger signal, the central controller uses a dynamic scheduling algorithm to allocate charging tasks for the energy storage vehicles in the charging area and performs path planning for the energy storage vehicles.

[0050] The mobile energy storage layer includes charging areas and a cluster of mobile energy storage vehicles. After receiving a dispatch signal from the central controller, the energy storage vehicle cluster travels to the charging area to receive electrical energy generated by reverse power protection. While charging the energy storage vehicles, the charging piles in the charging area simultaneously report the charging status of the energy storage vehicles and the status of the energy storage batteries in the vehicles to the central controller.

[0051] The following provides a detailed explanation of the implementation methods of each core module of this system:

[0052] As a preferred implementation, the reverse power detection module collects the output power of the photovoltaic array in real time through sensors. Load power and radiation intensity The sampling interval is Seconds. The obtained raw data is input into the data processing module and preprocessed as follows:

[0053] (1) Moving average filtering is used to suppress high noise in the original data and output a smooth power signal. Its expression is as follows:

[0054]

[0055] Where M is the width of the filter window, which represents the number of data points considered when calculating the moving average. It is used to balance noise immunity and response speed and can be dynamically adjusted according to the requirements of noise immunity and response speed on site. The original power signal represents the time... The power value at any given time; The time interval represents the time difference between adjacent data points.

[0056] (2) Irradiance coupling correction: The photovoltaic power is corrected according to the irradiance deviation to eliminate environmental interference. Its expression is as follows:

[0057]

[0058] in, These are correction coefficients obtained through regression fitting of historical irradiance data, used to adjust the correction magnitude to ensure that the corrected power signal more accurately reflects the actual photovoltaic output. This represents the original power value after moving average processing at time t. The mean and standard deviation of irradiance over the past 24 hours. This represents the actual irradiance value at time t.

[0059] As a preferred implementation, the reverse power protection module inputs the preprocessed data into the LSTM model to predict the dynamic threshold of reverse power at the current moment. The prediction process is as follows:

[0060] (1) Input construction: The time series input includes corrected power, power change, and irradiance, as shown in the following expression:

[0061]

[0062] in, The input vector at time t is used for prediction by the LSTM model. In time The corrected power value, where k is an integer from 0 to N−1, representing the past N time points; The time interval represents the time difference between adjacent data points; This is the corrected power value at the current time t; G(t) is the rate of change of power over time, used to capture dynamic changes in power; G(t) is the irradiance value at time t, representing the intensity of sunlight.

[0063] (2) Based on the constructed LSTM model prediction, the stability and prediction accuracy of the photovoltaic system are improved. The expressions for hidden state update and threshold input are as follows:

[0064] Hidden state update:

[0065]

[0066] in, It is the hidden state vector at time t, which contains the internal information of the LSTM network at the current time step and is used to capture long-term dependencies in time series data. It is a function of the LSTM unit used to receive the input of the current time step. The hidden state of the previous step and the weight parameters of LSTM Output the hidden state at the current time step. . Let be the input vector at time t, which includes the corrected power, the rate of change of power, and the irradiance. Let be the hidden state vector at time t−1, representing the LSTM hidden state at the previous time step. These are the weight parameters of the LSTM model, used to learn patterns in sequence data.

[0067] Threshold input:

[0068]

[0069] in, The inverse power dynamic threshold at time t is the output predicted by the LSTM model and is used to set the threshold for inverse power protection to prevent power reversal during the operation of the photovoltaic system. To correct the linear unit function, the output of the LSTM is... Through weight and bias After performing a linear transformation, the ReLU activation function is applied. The function is usually defined as max(0,x), which sets all negative values ​​to 0 and retains positive values. These are the weight parameters for the output layer, used to map the hidden states of the LSTM to the predicted output. These are the bias parameters for the output layer; This is the adjustment coefficient, used to adjust the rate of power change. Impact on predicted output; This is the rate of change of power over time, used to reflect the dynamic changes in power.

[0070] As a further preferred implementation method, the LSTM model is trained using historical photovoltaic operation data from the past 30 days with a sampling interval of 1 second. The model input sequence length N=60 corresponds to historical time series data with a duration of 1 minute, ensuring accurate capture of the fluctuation trend of photovoltaic power output.

[0071] (3) The dynamic threshold of the LSTM model prediction output Maximum reverse power allowed by the power grid Comparison, triggering protection actions:

[0072]

[0073] in, The dynamic threshold predicted by the LSTM model represents the allowable range of variation of the inverse power at time t. The dynamic threshold can be adaptively adjusted based on historical data and the current power change rate and irradiance to more accurately reflect the inverse power demand of the system. This represents the output power of the photovoltaic inverter after irradiance coupling correction. The corrected power value takes into account the impact of environmental factors on the photovoltaic system output, making the power value more accurate and reliable. Let be the load power demand at time t, used to represent the power consumed by all loads in the system at that time. The maximum reverse power value allowed by the grid is a fixed threshold set by the grid operator to limit the maximum power that a photovoltaic system can deliver to the grid.

[0074] Trigger(t) is a logic function used to determine whether the reverse power protection mechanism needs to be triggered. Its logic is as follows: If... If the reverse power exceeds the grid's allowable range, then Trigger(t) = 1, indicating that a protection mechanism needs to be triggered to prevent reverse power from exceeding the grid's allowable range. Otherwise, Trigger(t) = 0, indicating that a protection mechanism does not need to be triggered.

[0075] When the reverse power protection module triggers the protection mechanism, it immediately sends a trigger signal to the central control module. The central control module then starts the scheduling algorithm and generates instructions to control the mobile energy storage vehicle to go to the charging area to collect electrical energy.

[0076] As a preferred implementation, the central control module uses the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm to achieve the coordinated scheduling of multiple mobile energy storage vehicles. The construction process is as follows:

[0077] (1) Define a global state space, and integrate the dynamic parameters of photovoltaic points, charging facilities and energy storage vehicles into a state vector, the expression of which is shown below:

[0078]

[0079] in, State of charge for all energy storage vehicle batteries; The reverse power value of each photovoltaic node; The charging station is in an available state; The coordinates of the obstacle; This represents the power output of the current photovoltaic power station. The charging power for energy storage vehicles; This is the current speed of the energy storage vehicle; For charging efficiency.

[0080] (2) A multi-objective reward function is designed. By combining four weighted factors, the four core objectives of photovoltaic power consumption, battery health, driving energy consumption, and charging efficiency are quantified in a unified manner. This enables the MADDPG algorithm to output a collaborative scheduling strategy with the shortest path, the least loss, the most power consumption, and the longest lifespan under real-time changing photovoltaic power output, charging pile status, and obstacle environment. Its expression is as follows:

[0081]

[0082] in, This represents the photovoltaic (PV) grid integration rate, which encourages greater PV grid integration and prevents curtailment. This represents the actual photovoltaic power consumed by all energy storage vehicles at the current moment. This represents the total power generation of all photovoltaic power sources at the current moment. This represents the SOC balancing penalty, which penalizes vehicles whose SOC deviates too much by 0.5, preventing overcharging or over-discharging of a particular vehicle and improving battery life and system stability. Let be the remaining battery charge of the i-th vehicle, and 0.5 be the ideal SOC center value; The total distance traveled is penalized to prevent ineffective or redundant movements, thereby reducing driving energy consumption. Let be the distance traveled by the i-th vehicle at the current moment; This is a charging efficiency reward program. By incentivizing the selection of high-efficiency charging stations, it guides vehicles to prioritize high-efficiency charging areas, reducing energy loss and improving both economic efficiency and grid friendliness. This is a weighting coefficient for charging efficiency. This refers to the actual charging efficiency of the charging station. The current charging efficiency is used. The weighting ratio of 0.6:0.2:0.2 was determined through Pareto front analysis, which solves the subjective problem of manually setting weights for multiple objectives.

[0083] As a further preferred implementation, during the training process, the MADDPG algorithm determines that the model has converged when the global average reward change rate is less than 0.1% over 1000 consecutive iterations, thus completing the training and solidifying the scheduling strategy network parameters.

[0084] (3) Based on the state vector s t The system generates scheduling instructions using a reward function, and the path planning module plans the route for the mobile energy storage vehicle after receiving the instructions. Once the mobile energy storage vehicle reaches the designated charging area, it sends a charging instruction to the charging station.

[0085] like Figure 2 As shown, in a preferred implementation, the path planning module in this embodiment uses the D*Lite algorithm to plan the path for the mobile energy storage vehicle. The construction steps are as follows:

[0086] (1) Initialization phase: Construct the initial path planning environment and pre-calculate the heuristic cost. The construction steps are as follows:

[0087] Define the graph structure: Model the scheduling region as a grid graph G=(V,E), where vertices v∈V represent coordinate points, V represents the set of vertices, and each vertex v∈V represents a coordinate point in the grid graph. These coordinate points can be points on a two-dimensional plane, typically used to represent the position of a robot or moving body in the environment. Edges e∈E represent the movement relationships between adjacent grids. The existence of edges means that it is reachable from one vertex to another. Each vertex v stores a dynamic cost c(v,v′), initially set to the unobstructed Manhattan distance, as shown in the formula below:

[0088]

[0089] Here, c(v,v′) is used to calculate the cost of moving from one vertex v to another vertex v′; and This represents the coordinates of vertex v in the raster graph, used to determine the vertex's position in two-dimensional space; and This represents the coordinates of vertex v′ in the raster graph.

[0090] Pre-computed heuristic values: For all v∈V, compute the heuristic value up to the target point v. goal The heuristic estimate of h(v), using the Manhattan distance, is shown in the following formula:

[0091]

[0092] Among them, v goalThe target vertex in the graph represents the endpoint of the path planning. The target vertex is predefined in the path planning problem. The goal of all path planning algorithms is to find the optimal path from the starting point to the target point. and These represent the target vertex v. goal The x and y coordinates in the raster image are used to determine the position of the target point in two-dimensional space.

[0093] Initialize the priority queue: Create a priority queue U, sorted by priority k(v), initially k(v) start )=h(v start ).

[0094] (2) After constructing the initial environment, perform the first path planning and generate the optimal path based on the initial environment. The steps are as follows:

[0095] Priority calculation: For vertex v, the priority k(v) is defined as the minimum of the following two:

[0096]

[0097] Where g(v) is the current path cost, the initial cost is ∞, and it is the actual cost from the starting point to vertex v; The lookahead cost is used to estimate the minimum cost from vertex v to the target point, and h(v) is the heuristically estimated cost from vertex v to the target point. If v is the target point, then... ,otherwise:

[0098]

[0099] In the formula, Succ(v) is the successor node of v, and c(v,v′) is the cost from vertex v to vertex v′.

[0100] Update vertex state after calculating priority: if Mark v as “inconsistent” and insert it into queue U.

[0101] After updating the vertex state, the main loop begins, and its steps are as follows:

[0102] Loop condition: Queue U is not empty and the minimum k(v) is less than the current starting point.

[0103] Processing vertices: Extracting The vertex v with the lowest priority k(v) is selected from queue U. This vertex is considered the most promising vertex to be guided to the target point.

[0104] like When the estimate is overestimated, update. and for all predecessors Recalculate And update the queue; if When underestimation occurs, temporarily set g(v) = ∞ and recalculate. And update the predecessor node.

[0105] Generation path: From Start by selecting the smallest value at each step. successor node until arrival .

[0106] (3) After the path is generated, dynamic obstacles are processed to respond to real-time environmental changes. The steps are as follows:

[0107] Cost update: Edge detected Cost changes, updates .like It is the predecessor node; recalculate rhs(v):

[0108]

[0109] Here, v is the current vertex, which may be the predecessor node. It is one of the successor nodes of v. From the starting point to the vertex The actual cost. For the updated vertex v to vertex The cost reflects environmental changes. Succ(v) is the set of successor nodes of vertex v, that is, all vertices that can be directly reached from v.

[0110] Queue reordering: For the affected vertex v, update its priority k(v) and re-insert it into queue U, reducing computation by updating only the affected region.

[0111] Incremental replanning: Repeat the main loop but only process vertices where k(v) is lower than the current path cost, quickly correcting local paths.

[0112] (4) After processing the dynamic obstacles, the energy storage vehicle constraints are integrated, and the SOC and speed are dynamically incorporated into the path cost. The steps are as follows:

[0113] Dynamic cost function: Modifying the edge cost function Add SOC weights and speed weight As shown below:

[0114]

[0115] in, The basic cost of movement can be Manhattan distance or other cost metrics suitable for specific application scenarios; SOC weights. It is a positive coefficient used to increase the impact of battery state of charge on route selection; a larger one... The value indicates that the battery state has a greater impact on route selection; SOC is the current state of charge of the battery, which directly affects route selection to avoid depleting the battery during the trip; Speed ​​weight is a positive coefficient used to increase the impact of speed on path selection; a larger value indicates higher speed. The value indicates that speed has a greater impact on route selection; speed is the vehicle's speed, which affects energy consumption and travel time.

[0116] Heuristic adjustment: Maintain consistency of h(v), i.e. This is to ensure the optimality of the algorithm.

[0117] (5) Termination conditions

[0118] Success condition: When the starting point of Output a feasible path when the algorithm converges and there are no lower-priority vertices in the queue.

[0119] Failure handling: If queue U is empty and If no feasible path is found, the exception handling mechanism is triggered.

[0120] As a further preferred implementation method, the execution flow of the anomaly handling mechanism is as follows: when it is determined that there is no feasible path, the path planning module immediately feeds back the path failure signal to the central control module. Based on the current real-time reverse power state and the real-time state of other mobile energy storage vehicles, the central control module dynamically adjusts the consumption task allocation strategy through the multi-agent collaborative decision-making model, reassigns other mobile energy storage vehicles with reachability conditions to perform the consumption task, and at the same time replans the optimal driving path for the newly assigned mobile energy storage vehicle.

[0121] Compared with the prior art, the beneficial effects of the solution provided by the present invention include:

[0122] 1. At the detection level, an LSTM dynamic threshold prediction model is used to replace the traditional fixed threshold detection method. By analyzing historical power sequences and real-time irradiance data, adaptive judgment of inverse power trends is achieved, effectively overcoming the problem of high false trigger rate of fixed threshold under sudden change in illumination conditions.

[0123] 2. At the scheduling level, the MADDPG multi-agent reinforcement learning algorithm is introduced to construct a state space containing multi-dimensional parameters such as photovoltaic locations, charging pile status, and vehicle SOC. Through Pareto optimal weight allocation, multi-objective collaborative optimization of absorption efficiency, battery life, and mobile energy consumption is achieved.

[0124] 3. At the path planning level, the improved D*Lite algorithm uses the battery SOC and real-time speed of the energy storage vehicle as key parameters of the dynamic cost function, so that the path planning not only considers the geometric distance, but also takes into account the optimal solution of energy consumption, which solves the shortcomings of traditional A-type algorithms in energy efficiency optimization.

[0125] The test results of this embodiment are as follows: Figure 3 As shown, Figure 3 The comparison of inverse power detection results in (a) shows that the traditional fixed threshold method is prone to misjudgment in scenarios with fluctuating illumination. However, the LSTM dynamic threshold algorithm used in this invention can adaptively track the trend of inverse power changes, significantly improving detection accuracy. Its core advantage lies in dynamically adjusting the threshold through a time-series prediction model, avoiding unnecessary protective actions caused by sudden weather changes. Figure 3 The false trigger rate statistics in (b) further verify the reliability of the algorithm. The dynamic threshold significantly reduces the false trigger rate to less than 60% of that of the fixed threshold method. This is due to the LSTM model's ability to extract features from historical power sequences, effectively distinguishing between normal fluctuations and real inverse power events. Figure 3 (c) demonstrates the spatial flexibility of the system through multi-energy storage vehicle collaborative scheduling. The MADDPG algorithm enables intelligent path planning for the energy storage vehicle group, allowing vehicles to quickly respond to the reverse power signals of different photovoltaic points and optimize the charging pile allocation strategy, reducing response delay by about 67% compared to traditional fixed energy storage systems. Figure 3 The comparison of grid integration efficiency in (d) comprehensively reflects the overall performance improvement. The grid integration efficiency of the present invention is 92%, which is 27 percentage points higher than that of traditional methods. At the same time, the response time is reduced to 1.3 seconds. This leapfrog improvement stems from the organic integration of three major modules: dynamic detection, collaborative scheduling, and path planning. These experimental results, from the three dimensions of algorithm response speed, resource utilization, and system stability, confirm the breakthrough progress of the present invention in solving the problem of photovoltaic curtailment, and provide key technical support for high-proportion renewable energy grid connection.

[0126] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0127] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

[0128] This invention is not limited to the preferred embodiment described above. Anyone inspired by this invention can derive various other forms of photovoltaic reverse power consumption systems based on mobile energy storage vehicles. All equivalent variations and modifications made within the scope of the claims of this invention should be included within the scope of this invention.

Claims

1. A photovoltaic reverse power consumption system based on a mobile energy storage vehicle, characterized in that: The reverse power detection and protection unit is electrically connected to the photovoltaic power generation unit. The central control unit is communicatively connected to the reverse power detection and protection unit, the charging facility, and the mobile energy storage vehicle cluster. The charging facility is electrically connected to the power output side of the photovoltaic power generation unit. The photovoltaic power generation unit is configured to output photovoltaic power to supply power to the load and achieve grid-connected operation through the grid connection point; The reverse power detection and protection unit is configured to collect historical operating data, real-time output status, load status and real-time environmental parameters of the photovoltaic power generation unit, dynamically generate an adaptive reverse power protection threshold through a time-series prediction model, compare the real-time reverse power value with the reverse power protection threshold, and generate a reverse power trigger signal and send it to the central control unit when the real-time reverse power value exceeds the threshold. The central control unit is configured to receive the reverse power trigger signal, and based on the real-time reverse power status, the availability status of charging facilities, and the real-time status of each mobile energy storage vehicle, allocate consumption tasks to the mobile energy storage vehicle cluster through a multi-agent collaborative decision-making model, and plan a driving path that integrates its own energy status and driving energy efficiency for each mobile energy storage vehicle participating in the consumption, and generate corresponding scheduling instructions to be sent to the mobile energy storage vehicle cluster. Each mobile energy storage vehicle in the mobile energy storage vehicle cluster is equipped with an energy storage battery pack, a communication module, a positioning module, and a driving unit. It is configured to receive the scheduling command, drive to the corresponding charging facility, connect to the charging circuit, and consume the reverse power energy generated by the photovoltaic power generation unit.

2. The photovoltaic reverse power consumption system based on a mobile energy storage vehicle according to claim 1, characterized in that: The reverse power detection and protection unit includes a data processing submodule and a dynamic threshold prediction submodule. The data processing submodule performs moving average filtering and irradiance coupling correction preprocessing on the collected raw data in sequence. The moving average filtering is used to suppress high-frequency noise in the raw data and output a smooth photovoltaic power signal. The irradiance coupling correction is used to correct the photovoltaic power according to the irradiance deviation and eliminate environmental interference. The corrected photovoltaic output power satisfies the following expression: In the formula, Let be the photovoltaic output power value after moving average processing at time t. These are the correction coefficients obtained through regression fitting of historical irradiance data. Let be the real-time irradiance at time t. This is the average radiation intensity over a previously preset 24-hour period. This is the standard deviation of the radiation intensity over a previously preset 24-hour period. The value of the photovoltaic output power after irradiance coupling correction at time t is t.

3. A photovoltaic reverse power consumption system based on a mobile energy storage vehicle according to claim 2, characterized in that: The reverse power detection and protection unit uses an LSTM model for time-series prediction. The LSTM model constructs a time-series input vector based on preprocessed corrected photovoltaic power, power change rate, and real-time irradiance, and outputs an adaptive reverse power protection threshold. The threshold output satisfies the following expression: In the formula, Let t be the adaptive inverse power protection threshold. Let be the hidden state vector of the LSTM model at time t. The weight parameters of the output layer of the LSTM model. These are the bias parameters for the output layer of the LSTM model. To modify the activation function of the linear unit, This is the adjustment coefficient for the rate of change of power. This represents the real-time rate of change of photovoltaic output power.

4. A photovoltaic reverse power consumption system based on a mobile energy storage vehicle according to claim 3, characterized in that: The reverse power detection and protection unit determines whether to generate a reverse power trigger signal using the following logic: In the formula, This indicates the generation of an inverse power trigger signal. This indicates that no reverse power trigger signal is generated. The value of the photovoltaic output power at time t is the preprocessed and corrected value. Let be the real-time load power of the system at time t. This is the maximum reverse power value allowed by the power grid. Let t be the adaptive inverse power protection threshold output by the LSTM model at time t.

5. A photovoltaic reverse power consumption system based on a mobile energy storage vehicle according to claim 1, characterized in that: The central control unit employs the MADDPG algorithm as its multi-agent cooperative decision-making model. The MADDPG algorithm integrates photovoltaic point parameters, charging facility status parameters, and the dynamic parameters of the mobile energy storage vehicle into a global state space. The global state vector satisfies the following expression: In the formula, Let be the global state vector at time t. For the battery state of charge of all mobile energy storage vehicles, This represents the real-time reverse power value of each photovoltaic node. The availability status of charging facilities. The coordinates of obstacles within the scheduling area, This represents the real-time output power of the photovoltaic power generation unit. The charging power for mobile energy storage vehicles, This refers to the real-time driving speed of the mobile energy storage vehicle. The charging efficiency of charging facilities.

6. A photovoltaic reverse power consumption system based on a mobile energy storage vehicle according to claim 5, characterized in that: The MADDPG algorithm constructs a multi-objective reward function to optimize the scheduling strategy, and the multi-objective reward function satisfies the following expression: In the formula, Let be the reward value at time t. This represents the total photovoltaic reverse power actually consumed by all mobile energy storage vehicles at the current moment. This represents the total power generation of the photovoltaic power generation unit at the current moment. Let represent the state of charge (SOC) of the battery of the i-th mobile energy storage vehicle. Let be the distance traveled by the i-th mobile energy storage vehicle at the current moment. This is a weighting factor for charging efficiency. The actual charging efficiency of the charging facilities. This refers to the real-time charging power of the mobile energy storage vehicle.

7. A photovoltaic reverse power consumption system based on a mobile energy storage vehicle according to claim 6, characterized in that: In the multi-objective reward function, the weight ratios of the photovoltaic consumption term, the state of charge equilibrium penalty term, and the travel distance penalty term are determined through Pareto front analysis.

8. A photovoltaic reverse power consumption system based on a mobile energy storage vehicle according to claim 1, characterized in that: The central control unit plans the driving path of the mobile energy storage vehicle by using the D*Lite algorithm, which integrates the vehicle's own energy state and driving efficiency. The algorithm sets a dynamic cost function for the driving path, incorporating the vehicle's battery state of charge and real-time driving speed into the path cost calculation. The dynamic cost function satisfies the following expression: In the formula, Let be the final dynamic cost from vertex v to vertex v'. Let v be the basic movement cost from vertex v to vertex v'. These are the weighting coefficients for the state of charge. This provides the real-time battery state of charge for mobile energy storage vehicles. For speed weighting coefficients, This refers to the real-time driving speed of the mobile energy storage vehicle.

9. A photovoltaic reverse power consumption system based on a mobile energy storage vehicle according to claim 8, characterized in that: The D*Lite algorithm detects environmental changes within the scheduling area in real time. When it detects changes in path cost caused by obstacles, it updates the path cost and priority of the affected vertices and corrects the local driving path through incremental replanning. When no feasible path exists, it triggers an exception handling mechanism.

10. A photovoltaic reverse power consumption system based on a mobile energy storage vehicle according to claim 1, characterized in that: The charging facility collects the charging status and battery status of the mobile energy storage vehicle in real time and feeds them back to the central control unit; the driving unit of the mobile energy storage vehicle is powered by its own battery pack.

Images