A method and system for managing batteries in an intelligent shared battery swapping cabinet

Abstract

This application discloses a battery management method and system for intelligent shared battery swapping cabinets, relating to the field of intelligent Internet of Things (IoT) technology. The method includes: real-time acquisition of battery status, user demand, and grid signals to form global status perception data; construction of a dynamic graph network prediction model to output predicted battery demand and grid price signals for each network node within a preset future time period; calculation of the power constraint envelope for each region corresponding to each intelligent battery swapping cabinet using a hierarchical multi-agent trade-off algorithm, and the target trade-off factors for dynamically balancing user waiting time, battery life loss, and grid cost; generation of a real-time charging power allocation list for each battery within the swapping cabinet based on the real-time status and prediction results; and distribution of this list to the corresponding intelligent battery swapping cabinet. This application aims to address the problem that traditional or single-dimensional optimization methods are almost incapable of simultaneously optimizing user waiting time, battery life loss, and grid cost.

Description

Technical Field

[0001] This invention relates to the field of smart Internet of Things (IoT) technology, and in particular to a smart shared battery swapping cabinet battery management method and system. Background Technology

[0002] Existing shared battery swapping station scheduling systems mostly employ rule-based or single-objective optimization algorithms (such as minimizing user waiting time only). These methods have significant drawbacks: 1) Fast charging to meet peak demand accelerates battery degradation and increases maintenance costs; 2) Ignoring real-time electricity prices and loads leads to high electricity costs and impacts the grid; 3) Independent decision-making by each swapping station lacks global coordination, failing to achieve optimal overall network efficiency. This is the "impossible triangle" problem of "user waiting time - battery life - grid cost." Traditional single-agent reinforcement learning or centralized optimization struggles to achieve real-time, multi-dimensional optimal trade-offs in dynamic, large-scale networks. Summary of the Invention

[0003] This invention provides a method for managing batteries in an intelligent shared battery swapping cabinet, comprising:

[0004] Real-time acquisition of battery status, user needs, and power grid signals forms global status awareness data;

[0005] A dynamic graph network prediction model is constructed based on global state perception data, and the predicted values ​​of battery demand and grid price signals for each network node are output within a preset time period in the future.

[0006] Using the predicted value as input, the power constraint envelope of the area corresponding to each smart battery swapping cabinet is calculated through a hierarchical multi-agent trade-off algorithm, as well as the target trade-off factor for dynamically balancing user waiting time, battery life loss and grid cost.

[0007] Based on the power constraint envelope and target trade-off factor, a real-time charging power allocation list for each battery in the battery swapping cabinet is generated, combined with the real-time status of the corresponding battery swapping cabinet and the prediction results.

[0008] The real-time charging power allocation list is sent to the corresponding smart battery swapping cabinet, and new global state perception data is collected and fed back to the dynamic graph network prediction model.

[0009] The aforementioned intelligent shared battery swapping cabinet battery management method constructs a dynamic graph network prediction model based on global state perception data, and outputs predicted battery demand values ​​and grid price signal values ​​for each network node within a preset future time period, including:

[0010] Using the battery swapping cabinet as a node, historical allocation records as edges, and the state vector of the battery swapping cabinet as a node feature, the recent historical feature sequence is fused to obtain enhanced node features.

[0011] By calculating the attention coefficients between nodes to aggregate neighbor features and enhance node features, and then inputting the spatially aggregated node features into a time-series decoder, a node battery demand prediction matrix and a grid price signal prediction matrix are generated for the future within a preset period.

[0012] The aforementioned intelligent shared battery swapping cabinet battery management method uses the predicted value as input, calculates the power constraint envelope of the corresponding area of ​​each intelligent battery swapping cabinet through a hierarchical multi-agent trade-off algorithm, and determines the target trade-off factors for dynamically balancing user waiting time, battery life loss, and grid cost, including:

[0013] The global floating power pool is calculated using the predicted value through a hierarchical multi-agent trade-off algorithm, and the upper and lower limits of its own power are calculated accordingly. Finally, the dynamic power constraint envelope of each smart cabinet for the corresponding time period is output.

[0014] The urgency indicators corresponding to user waiting time, battery life loss and grid cost are calculated by a hierarchical multi-agent trade-off algorithm, and the target trade-off factor is obtained by combining the dynamic power constraint envelope.

[0015] The aforementioned intelligent shared battery swapping cabinet battery management method calculates the urgency index of each objective through a hierarchical multi-agent trade-off algorithm and obtains the objective trade-off factor by combining the dynamic power constraint envelope, including:

[0016] Based on the predicted values ​​and the real-time battery status in the global state perception data, the urgency indicators corresponding to user waiting time, battery life loss and grid cost are calculated.

[0017] A dynamic trade-off factor is calculated by combining the urgency indicators corresponding to power constraint envelope tension, user waiting time, battery life loss, and grid cost.

[0018] The aforementioned intelligent shared battery swapping cabinet battery management method generates a real-time charging power allocation list for each battery within the swapping cabinet based on the power constraint envelope, target trade-off factor, and the real-time status and prediction results of the corresponding swapping cabinet. The method includes:

[0019] Based on the current objective trade-off factors within the power constraint envelope, determine the total charging power quota for the region in the next control cycle, and generate a multi-objective allocation strategy set under the total charging power quota constraint.

[0020] Based on a multi-objective allocation strategy set, a real-time battery charging power allocation list is generated for each smart battery swapping cabinet.

[0021] A smart shared battery swapping cabinet battery management system includes:

[0022] The global status awareness module is used to acquire battery status, user needs and power grid signals in real time to form global status awareness data;

[0023] The dynamic prediction modeling module is used to construct a dynamic graph network prediction model based on global state perception data, and output the predicted values ​​of battery demand and grid price signals for each network node in the future within a preset time period.

[0024] The multi-agent trade-off module is used to take the predicted value as input and calculate the power constraint envelope of the area corresponding to each smart battery swapping cabinet through a hierarchical multi-agent trade-off algorithm, as well as the target trade-off factor for dynamically balancing user waiting time, battery life loss and grid cost.

[0025] The charging power allocation module is used to generate a real-time charging power allocation list for each battery in the battery swapping cabinet based on the power constraint envelope, the target trade-off factor, the real-time status of the corresponding battery swapping cabinet and the prediction results.

[0026] The instruction issuance and feedback module is used to issue the real-time charging power allocation list to the corresponding smart battery swapping cabinet and collect new global state perception data to feed back to the dynamic graph network prediction model.

[0027] The beneficial effects achieved by this invention are as follows:

[0028] By employing a hierarchical multi-agent collaborative framework and a dynamic trade-off mechanism, the "impossible trinity" problem is effectively solved. Dynamic trade-off factors are used to intelligently adjust the priorities of multiple objectives, and flexible power envelopes are combined for real-time resource allocation. This significantly reduces user waiting time and slows down battery life loss while ensuring grid security and economy, achieving dynamic synergistic optimization of the three conflicting objectives. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0030] Figure 1 This is a flowchart of a smart shared battery swapping cabinet battery management method provided in Embodiment 1 of this application.

[0031] Figure 2 This is a schematic diagram of an intelligent shared battery swapping cabinet battery management system provided in Embodiment 2 of this application. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example 1

[0034] like Figure 1 As shown, Embodiment 1 of this application provides a battery management method for an intelligent shared battery swapping cabinet, including:

[0035] S1: Real-time acquisition of battery status, user needs and power grid signals to form global status awareness data;

[0036] Data is acquired through sensors in the battery management system (BMS) built into each battery and the battery swapping station, and uploaded in real time via an IoT communication module. Specific data collected includes core parameters such as voltage, current, remaining charge (SOC), state of health (SOH), temperature, and operating conditions for each battery. Real-time capture of user-initiated "scan-to-swap" requests includes the request time, requested battery swapping station location, and user ID. Nodal marginal price (LMP) forecast curves or time-of-use price signals for the next 24 hours or longer are obtained from the local electricity spot market trading platform or virtual power plant operation platform.

[0037] The aforementioned multi-source, heterogeneous real-time data streams are aggregated into the system's data platform. After processing, structured global state-aware data is formed: Specifically, the raw data streams are parsed, outliers and erroneous data are removed, and timestamps are aligned. Using the "battery swapping cabinet" as the basic node unit, all battery states, current user requests, and local grid price signals belonging to that node are aggregated. A standardized state vector is generated for each node. The state vectors of all nodes are arranged in topological order to form an M x N global state-aware matrix, where M is the total number of nodes and N is the state feature dimension.

[0038] S2: Construct a dynamic graph network prediction model based on global state perception data, and output the predicted battery demand and grid price signal of each network node within a preset time period in the future;

[0039] The process involves constructing a dynamic graph network prediction model based on global state awareness data, and outputting predicted battery demand and grid price signals for each network node within a preset future time period. This includes the following sub-steps:

[0040] S21: Using the battery swapping cabinet as a node, historical allocation records as edges, and the state vector of the battery swapping cabinet as a node feature, the recent historical feature sequence is fused to obtain enhanced node features.

[0041] Define the entire battery swapping network as a dynamic graph. Specifically, each smart battery swapping station is treated as a graph node. If there are two battery swapping cabinets and If a battery transfer record exists in the historical data, or if the geographical distance between the two is within a preset threshold (e.g., 3 kilometers), then an undirected edge is established between them. The weights of the edges are dynamically initialized based on the historical allocation frequency or the reciprocal of the distance.

[0042] For each node The state vector generated in step S1 As a characteristic of this node at time t The features of all nodes are stacked to form the full graph node feature matrix. ,in The total number of nodes. This represents the feature dimension of the state vector. The initial features of the node are used to capture spatiotemporal dependencies. This will be integrated with recent historical features. Store the past of each node. Features of a time slice (e.g., the past 12 15-minute intervals) A temporal context feature vector is generated by encoding the historical feature sequence of a node itself through a lightweight recurrent neural network (such as GRU) or a one-dimensional temporal convolution. .

[0043] Node features at the current time With temporal context features By splicing the data, enhanced node features are obtained. This serves as the input to the graph neural network layers. This process ensures that each node's features not only contain its instantaneous state but also its short-term trends.

[0044] S22: By calculating the attention coefficients between nodes, neighboring nodes are aggregated to enhance node features, and the spatially aggregated node features are input into the time-series decoder to generate a node battery demand prediction matrix and a grid price signal prediction matrix for the future preset period.

[0045] Joint prediction is performed based on the encoded graph structure using a graph attention network and a sequence prediction model. Specifically,

[0046] Use Graph Attention Network (GAT) layers to process enhanced node features Specifically, for nodes and any of its neighbors The GAT layer calculates an attention coefficient. :

[0047] ,in, The attention coefficient represents the value at time t. Neighbor nodes Status information for the target node The importance of performing state forecasting (such as demand forecasting); This represents the leakage linear rectifier function, introducing nonlinear computational capabilities to the model; The attention vector maps the concatenated high-dimensional features to a real scalar (i.e., the attention score). The transpose operator represents converting a vector into its original form. Convert a row vector to a column vector; A learnable shared weight matrix is ​​used to apply the input features to all nodes. Perform a shared linear transformation; These are node feature vectors, representing the target nodes respectively. and neighboring nodes At any moment Enhanced feature vectors; Indexing neighbor nodes It is traversing the set The index of each element in the array; Representative in the diagram In and target node The set of all directly connected neighboring nodes.

[0048] Next, the nodes All neighbor features are aggregated using a formula, as follows:

[0049]

[0050] in, This represents the features of the aggregated nodes. itself (through) (Information implicit in the calculation) and all its neighboring nodes; These are non-linear activation functions, such as ELU and ReLU. The attention coefficient represents the features of neighboring nodes with high importance. Their contribution was even greater; For shared weight matrix; It is the feature vector of the neighboring nodes, that is, the neighboring nodes. At any moment The enhanced feature vector.

[0051] By stacking multiple layers of GAT, multi-hop spatial dependencies are captured, allowing the prediction of a node to be indirectly influenced by the state of distant nodes. Then, the features of each node are spatially aggregated. Input a temporal decoder. This decoder consists of an attention-based sequence model (such as a Transformer decoder or an LSTM with attention). The decoder uses... Using the initial context and referencing historical time series patterns, it autoregressively generates a predicted sequence for the next preset period (e.g., the next 24 15-minute intervals).

[0052] The decoder outputs two parallel prediction channels:

[0053] Battery demand forecasting channel: Output node In the future, at various times The expected number of battery swap requests (continuous value).

[0054] Grid Price Forecasting Channel: Output Node The projected electricity price for the corresponding electricity price region during the same future period.

[0055] The final output consists of two prediction matrices: a battery demand prediction matrix. With the power grid price forecast matrix .

[0056] S3: Using the predicted value as input, calculate the power constraint envelope of the area corresponding to each smart battery swapping cabinet through a hierarchical multi-agent trade-off algorithm, as well as the target trade-off factor for dynamically balancing user waiting time, battery life loss and grid cost.

[0057] Through the upper and lower layer collaboration between the distribution network coordination agent and multiple regional management agents, two core decision variables are calculated in parallel: first, a dynamically adjustable power constraint envelope set for each smart battery swapping cabinet area; second, a real-time trade-off factor used to dynamically balance multiple objectives (user waiting time, battery life loss, and grid interaction cost) in subsequent real-time optimization.

[0058] The process involves using the predicted value as input, calculating the power constraint envelope for each area corresponding to the smart battery swapping station through a hierarchical multi-agent trade-off algorithm, and determining the target trade-off factors for dynamically balancing user waiting time, battery life loss, and grid costs. This includes the following sub-steps:

[0059] S31: Calculate the global floating power pool using the predicted value through a hierarchical multi-agent trade-off algorithm, and calculate its own power upper and lower limits accordingly, and finally output the dynamic power constraint envelope of each smart cabinet for the corresponding time period.

[0060] Based on a two-layer intelligent agent architecture, the power constraint envelope is calculated for the area covered by each intelligent battery swapping station. The upper layer is the distribution network coordination intelligent agent, responsible for the integration of global grid constraints and overall resource allocation; the lower layer is the regional management intelligent agent, responsible for local state assessment and demand calculation. The inputs are the battery demand forecast of each node, the grid price signal forecast, the real-time grid load status signal, and the battery status of each battery swapping station.

[0061] Next, each regional management agent i calculates a basic power demand based on local battery demand forecasts and the current state of energy storage. This is then reported as an initial power request to the distribution network coordinating agent. The distribution network coordinating agent aggregates requests from all regions and, based on the total grid load constraints,... Based on safety margins and grid regulation requirements, the globally allocatable floating power pool is calculated. .

[0062] The floating power pool is connected to the global load rate through the distribution network coordination agent. The signal is broadcast; each regional management agent calculates its own power constraint envelope in parallel based on the same adaptive boundary function, and the specific calculation formula is as follows:

[0063]

[0064] in, Let be the upper limit of the power in region i at time t; The basic predicted power demand for region i at time t; This represents the floating power pool, i.e., the floating power capacity of the power grid; L(t) is the real-time load rate of the upstream power grid node. Its critical load rate; The average state of charge of the batteries in the battery swapping cabinets within region i; Indicates the weighting coefficient of the power grid load sub-item; The kurtosis coefficient represents the slope of the power grid load response curve; This represents the weighting coefficient of the local resource status sub-item.

[0065] The formula enables each region to coordinately respond to the global power grid state and adaptively adjust its available power space by utilizing local resources, while meeting its local basic predicted power demand. Next, the power lower limit is calculated using the formula. The specific formula is as follows:

[0066]

[0067] in, Let be the lower limit of the power in region i at time t; Indicates the absolute safe power lower limit; The basic predicted power demand for region i at time t; This represents the basic demand retention ratio when the power grid is fully relaxed. The coefficient representing the attenuation sensitivity of the grid load pressure; L(t) is the real-time load rate of the upstream grid node. Its critical load rate;

[0068] The final output is the dynamic power constraint envelope of each region managed by agent i in each time slice within a preset future time period. .

[0069] S32: The urgency indicators corresponding to user waiting time, battery life loss and grid cost are calculated by using a hierarchical multi-agent trade-off algorithm, and the target trade-off factor is obtained by combining the dynamic power constraint envelope.

[0070] The process involves calculating the urgency index of each objective using a hierarchical multi-agent tradeoff algorithm and obtaining the objective tradeoff factor by combining it with the dynamic power constraint envelope. This includes the following sub-steps:

[0071] S321: Calculate the urgency indicators corresponding to user waiting time, battery life loss and grid cost based on the predicted value and the real-time battery status in the global state perception data.

[0072] After obtaining the power constraint envelope, the system employs a virtual multi-agent game framework within the region management agent to determine how to balance conflicting optimization objectives under current and future prediction scenarios. This framework creates a virtual agent for each optimization objective. Specifically,

[0073] The inputs are the power constraint envelope of each region, the predicted value of battery demand, the predicted value of grid price signal, and the real-time status of batteries in each battery swapping cabinet.

[0074] In regional management intelligent agents Internally, three virtual agent agents (representing user waiting time, battery life depletion, and grid cost, respectively) calculate in parallel the normalized urgency indicators for their respective objectives based on the current input. For example, the delay agent assesses the peak-to-valley demand difference, the loss agent assesses battery charging pressure, and the cost agent assesses the risk of electricity price fluctuations during certain periods. The specific calculation formulas are as follows:

[0075] , Indicators representing the urgency of user waiting time; The number of batteries in region i at time t represents the predicted total demand. The available service capacity of region i at time t is defined as follows: ,in Indicates the number of available batteries; This indicates the average service rate per battery per hour; This represents the average waiting time in region i within the most recent ΔT time period; This indicates a reference waiting time (such as the value specified in the service level agreement). Represents the delay sensitivity coefficient; This represents the historical waiting weight coefficient.

[0076] , Indicators of urgency; This represents the total number of batteries in region i; k represents the k-th battery. This represents the weight of the k-th battery. ;in This indicates the rated capacity of battery k; This represents the aging coefficient of the k-th battery; The total rated capacity of the j-th battery within region i; This represents the aging coefficient of the j-th battery; The loss rate of battery k at time t is represented by the Arrhenius-type temperature acceleration model, which is used for calculation. ,in, Indicators representing the urgency of power grid costs; This indicates the current real-time electricity price has been normalized. Indicates the benchmark electricity price; Indicates electricity price volatility; Indicates the decision-making timeframe; Indicates the power grid frequency deviation; Indicates the strength of the demand response signal; Indicates the sensitivity coefficient; This indicates the contribution of the logarithmic ratio of current power to baseline power to the urgency indicator, reflecting the sensitivity of power ratio changes to grid costs. This represents the actual power of region i interacting with the power grid at time t; Represents the reference power of region i; The standard deviation of the power grid signal volatility reflects the amplitude of the power grid excitation signal fluctuation over time. The greater the fluctuation, the more significant its impact on urgency indicators. This indicates the length of the predicted time period.

[0077] S322: Calculate dynamic trade-off factors by combining the urgency indicators corresponding to power constraint envelope tension, user waiting time, battery life loss, and grid cost.

[0078] Next, the regional management agent i receives urgency indicators from all virtual agents; and calculates the concentration of adaptive tradeoff factors based on the current power envelope tension and grid stability requirements. This parameter is negatively correlated with the power envelope width, reflecting the pressure of external constraints. For example, the power envelope width represents the size of the operating space in region i at time t where the power can be freely adjusted. A large width means that the system can adjust the power over a larger range, with sufficient resources to alleviate the urgency of multiple objectives simultaneously (e.g., it can use slightly higher power to shorten the wait time, or choose to charge when the electricity price is low).

[0079] The tradeoff factor is then calculated using the formula, which is as follows:

[0080] ,in, Indicates the trade-off factors for the objectives; ， Indicates the target index. , ;in, Indicators representing the urgency of user waiting time; Indicators indicating the urgency of battery life depletion; Indicators representing the urgency of power grid costs; This represents the sensitivity coefficient of target k; the larger the value, the higher the likelihood that the target will be given priority in the competition. This represents a nonlinear mapping function for target k, used to adjust the dynamic range of the urgency index and enhance its discriminative power. This indicates the urgency of target k in region i at time t; This represents the time accumulation coefficient of target k, relative to time. Multiplication is used to prevent a target from being ignored for a long time and to simulate the cumulative effect of an emergency. Indicates the current moment; Indicate target The sensitivity coefficient; the larger the value, the higher the likelihood that the target will be given priority in the competition; Indicates targeting The nonlinear mapping function is used to adjust the dynamic range of urgency indicators and enhance their discriminative power. Indicate target The urgency index in region i at time t; Indicate target Time accumulation factor; Adaptive trade-off concentration, controlling the degree of concentration of trade-off factors: When the value is very large, all exponential terms approach 1, and the weights... Approaching uniform distribution; when the power envelope is tight ( When the time is short, it tends to support the most pressing goal; it ensures that goals engage in autonomous and dynamic game-playing and weighing based on the real-time internal and external environment.

[0081] The final output is the dynamic target trade-off factor for each region managed by agent i in each time slice within a preset future time period. .

[0082] S4: Based on the power constraint envelope and target trade-off factor, and combined with the real-time status and prediction results of the corresponding battery swapping cabinet, generate a real-time charging power allocation list for each battery in the battery swapping cabinet.

[0083] The process of generating a real-time charging power allocation list for each battery in the battery swapping cabinet, based on the power constraint envelope, the target trade-off factor, and the real-time status and prediction results of the corresponding battery swapping cabinet, includes the following sub-steps:

[0084] S41: Based on the current target trade-off factor within the power constraint envelope, determine the total charging power quota for this region in the next control cycle, and generate a multi-objective allocation strategy set under the total charging power quota constraint;

[0085] The system is based on basic power requirements Based on this benchmark, and according to the direction emphasized by the current target trade-off factors, the total charging power quota for this region in the next control cycle is determined by dynamically adjusting within the power constraint envelope. For example, if user waiting time increases significantly, the system will tend to move towards the maximum power to increase charging speed; if grid costs dominate, it may move towards the minimum power to reduce grid costs or respond to dispatch. Meanwhile, any real-time override commands from the grid (such as forced power curtailment) will have the highest priority, directly constraining the total charging power quota. The final value.

[0086] In total power quota Under the hard constraints, the system generates specific allocation principles based on trade-off factors to guide power distribution among batteries within the cabinet. For example, high battery life loss will generate a strategy that prioritizes protecting high-temperature and severely aged batteries (e.g., by reducing their charging current or adopting a gentle charging mode). The final output is the approved total charging power for each region in the next control cycle, along with a dynamic multi-objective allocation strategy set to guide power distribution among batteries within that region. This includes: dominant objective, emergency charging threshold, battery power allocation, non-emergency battery handling, aged battery protection, and power smoothing requirements.

[0087] S42: Based on the multi-objective allocation strategy set, generate a real-time battery charging power allocation list for each smart battery swapping cabinet.

[0088] Based on a multi-objective allocation strategy set, all batteries awaiting charging within the cabinet are dynamically prioritized and ranked. This ranking is the result of comprehensive decision-making: for example, considering the objective of "reducing waiting time," batteries nearing depletion and those with associated reservation requests are prioritized; considering the objective of "extending battery life," charging power requests from high-temperature or aged batteries are suppressed or marked as requiring special charging curves. Based on this priority order, the system allocates a certain amount of power to the area. Within the budget, initially assign a power value or charging current level to each battery.

[0089] Next, the initial allocation scheme was checked against multiple constraints, including: ensuring that the total electrical power of a single battery swapping cabinet does not exceed its equipment safety limits; ensuring that the allocated power of a single battery complies with the safe charging range allowed by its BMS (Battery Management System); and checking whether the allocation scheme as a whole meets the requirements. The requirements are as follows. Based on the verification results, fine-tuning and optimization are performed to ultimately generate a real-time battery charging power allocation list for each smart battery swapping cabinet. This list clearly includes the unique identifier of each battery in the cabinet, its corresponding charging power (or current) command to be executed, the expected charging mode, and the next state check time.

[0090] S5: Send the real-time charging power allocation list to the corresponding smart battery swapping cabinet, and collect new global state perception data to feed back to the dynamic graph network prediction model.

[0091] The generated refined charging power allocation list is securely and reliably sent to each battery swapping station for execution, and its reception and execution status is confirmed to ensure that the control intentions are accurately implemented. Subsequently, the system immediately collects the latest multi-source data generated after execution, including data on batteries, battery swapping stations, the power grid, and user needs. After aggregation and cleaning, an updated global state snapshot is generated and fed back to the system's predictive model as a new round of input.

[0092] Example 2

[0093] like Figure 2 As shown, Embodiment 2 of this application provides an intelligent shared battery swapping cabinet battery management system, including:

[0094] Global Status Awareness Module 21: Real-time acquisition of battery status, user needs and power grid signals to form global status awareness data;

[0095] Dynamic Prediction Modeling Module 22: Constructs a dynamic graph network prediction model based on global state perception data, and outputs the predicted battery demand and grid price signal of each network node within a preset future time period;

[0096] Multi-agent trade-off module 23: Using the predicted value as input, it calculates the power constraint envelope of the area corresponding to each smart battery swapping cabinet through a hierarchical multi-agent trade-off algorithm, as well as the target trade-off factor for dynamically balancing user waiting time, battery life loss and grid cost.

[0097] Charging power allocation module 24: Based on the power constraint envelope and target trade-off factor, and combined with the real-time status and prediction results of the corresponding battery swapping cabinet, it generates a real-time charging power allocation list for each battery in the battery swapping cabinet.

[0098] Command issuance and feedback module 25: issues the real-time charging power allocation list to the corresponding smart battery swapping cabinet and collects new global state perception data to feed back to the dynamic graph network prediction model.

[0099] Corresponding to the above embodiments, the present invention provides a computer storage medium, including: at least one memory and at least one processor;

[0100] The memory is used to store one or more program instructions;

[0101] A processor is used to run one or more program instructions to execute a smart shared battery swapping cabinet battery management method.

[0102] Corresponding to the above embodiments, this embodiment of the invention provides a computer-readable storage medium containing one or more program instructions, which are executed by a processor to provide a smart shared battery swapping cabinet battery management method.

[0103] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions. When the computer program instructions are executed on a computer, the computer performs the above-described intelligent shared battery swapping cabinet battery management method.

[0104] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0105] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0106] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0107] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0108] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0109] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0110] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0111] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for managing batteries in an intelligent shared battery swapping cabinet, characterized in that, include: Real-time acquisition of battery status, user needs, and power grid signals forms global status awareness data; A dynamic graph network prediction model is constructed based on global state perception data, and the predicted values ​​of battery demand and grid price signals for each network node are output within a preset time period in the future. Specifically, this includes defining the entire battery swapping network as a dynamic graph. Each smart battery swapping cabinet is treated as a graph node. If there are two power swapping cabinets and If a battery transfer record exists in the historical data, or if the geographical distance between the two is within a preset threshold, then an undirected edge is established between them. The weights of the edges are dynamically initialized based on the historical allocation frequency or the reciprocal of the distance. For each node Its state vector As the node at time Features The features of all nodes are stacked to form the full graph node feature matrix. ,in The total number of nodes. The feature dimension of the state vector; the initial features of the node to capture spatiotemporal dependencies. It will be integrated with recent historical features; storing the past of each node. Features of a time slice A lightweight recurrent neural network or a one-dimensional temporal convolution is used to encode the historical feature sequence of a node, generating a temporal context feature vector. ; Node features at the current time With temporal context features By splicing the data, enhanced node features are obtained. , as input to the graph neural network layer; Joint prediction is performed based on the encoded graph structure using graph attention networks and sequence prediction models; specifically, Use graph attention network layers to enhance node features. Specifically, for nodes and any of its neighbors The GAT layer calculates an attention coefficient. : ,in, The attention coefficient represents the value at time t. Neighbor nodes Status information for the target node The importance of performing state prediction; This represents the leakage linear rectifier function, introducing nonlinear computational capabilities into the model; The attention vector maps the concatenated high-dimensional features to a real scalar, i.e., the attention score. The transpose operator represents converting a vector to its original state. Convert a row vector to a column vector; A learnable shared weight matrix is ​​used to apply the input features to all nodes. Perform a shared linear transformation; These are node feature vectors, representing the target nodes respectively. and neighboring nodes At any moment Enhanced feature vectors; Indicates the first The neighboring nodes at time... Enhanced feature vectors; Indexing neighbor nodes It is traversing the set The index of each element in the array; Representative in the diagram In and target node The set of all directly connected neighboring nodes; Next, the nodes All neighbor features are aggregated using a formula, as follows: ,in, This represents the features of the aggregated nodes. Information about itself and all its neighboring nodes; It is a non-linear activation function; The attention coefficient represents the features of neighboring nodes with high importance. Their contribution was even greater; It is a learnable shared weight matrix; It is the feature vector of the neighboring nodes, that is, the neighboring nodes. At any moment Enhanced feature vectors; By stacking multiple layers of GAT, multi-hop spatial dependencies are captured, allowing the prediction of a node to be indirectly influenced by the state of distant nodes; then, the features of each node are spatially aggregated. Input a temporal decoder; this decoder consists of a sequence model based on an attention mechanism; the decoder uses... Using the initial context and referring to historical time series patterns, the system regressively generates a predicted sequence for a future preset period. The decoder outputs two parallel prediction channels: a battery demand prediction channel and an output node. Expected number of battery swapping requests in different time periods in the future; Grid Price Forecasting Channel: Output Node The projected electricity price for the corresponding electricity price region during the same period in the future; The final outputs are a battery demand forecast matrix and a grid price forecast matrix; Using the predicted value as input, the power constraint envelope of the area corresponding to each smart battery swapping cabinet is calculated through a hierarchical multi-agent trade-off algorithm, as well as the target trade-off factor for dynamically balancing user waiting time, battery life loss and grid cost. Based on the power constraint envelope and target trade-off factor, a real-time charging power allocation list for each battery in the battery swapping cabinet is generated, combined with the real-time status of the corresponding battery swapping cabinet and the prediction results. The real-time charging power allocation list is sent to the corresponding smart battery swapping cabinet, and new global state perception data is collected and fed back to the dynamic graph network prediction model.

2. The intelligent shared battery swapping cabinet battery management method according to claim 1, characterized in that, Using the predicted value as input, a hierarchical multi-agent tradeoff algorithm is used to calculate the power constraint envelope of the area corresponding to each smart battery swapping station, as well as the target tradeoff factors for dynamically balancing user waiting time, battery life loss, and grid cost, including: The global floating power pool is calculated using the predicted value through a hierarchical multi-agent trade-off algorithm, and the upper and lower limits of its own power are calculated accordingly. Finally, the dynamic power constraint envelope of each smart cabinet for the corresponding time period is output. The urgency indicators corresponding to user waiting time, battery life loss and grid cost are calculated by a hierarchical multi-agent trade-off algorithm, and the target trade-off factor is obtained by combining the dynamic power constraint envelope.

3. The intelligent shared battery swapping cabinet battery management method according to claim 1, characterized in that, The urgency index of each objective is calculated using a hierarchical multi-agent tradeoff algorithm, and the objective tradeoff factors are obtained by combining the dynamic power constraint envelope, including: Based on the predicted values ​​and the real-time battery status in the global state perception data, the urgency indicators corresponding to user waiting time, battery life loss and grid cost are calculated. A dynamic trade-off factor is calculated by combining the urgency indicators corresponding to power constraint envelope tension, user waiting time, battery life loss, and grid cost.

4. The intelligent shared battery swapping cabinet battery management method according to claim 1, characterized in that, Based on the power constraint envelope and target trade-off factor, and combined with the real-time status and prediction results of the corresponding battery swapping cabinet, a real-time charging power allocation list for each battery in the battery swapping cabinet is generated, including: Based on the current objective trade-off factors within the power constraint envelope, determine the total charging power quota for the region in the next control cycle, and generate a multi-objective allocation strategy set under the total charging power quota constraint. Based on a multi-objective allocation strategy set, a real-time battery charging power allocation list is generated for each smart battery swapping cabinet.

5. A smart shared battery swapping cabinet battery management system, characterized in that, include: The global status awareness module is used to acquire battery status, user needs and power grid signals in real time to form global status awareness data; The dynamic prediction modeling module is used to construct a dynamic graph network prediction model based on global state perception data, and output the predicted values ​​of battery demand and grid price signals for each network node in the future within a preset time period. Specifically, this includes defining the entire battery swapping network as a dynamic graph. Each smart battery swapping cabinet is treated as a graph node. If there are two power swapping cabinets and If a battery transfer record exists in the historical data, or if the geographical distance between the two is within a preset threshold, then an undirected edge is established between them. The weights of the edges are dynamically initialized based on the historical allocation frequency or the reciprocal of the distance. For each node Its state vector As the node at time Features The features of all nodes are stacked to form the full graph node feature matrix. ,in The total number of nodes. The feature dimension of the state vector; the initial features of the node to capture spatiotemporal dependencies. It will be integrated with recent historical features; storing the past of each node. Features of a time slice A lightweight recurrent neural network or a one-dimensional temporal convolution is used to encode the historical feature sequence of a node, generating a temporal context feature vector. ; Node features at the current time With temporal context features By splicing the data, enhanced node features are obtained. , as input to the graph neural network layer; Joint prediction is performed based on the encoded graph structure using graph attention networks and sequence prediction models; specifically, Use graph attention network layers to enhance node features. Specifically, for nodes and any of its neighbors The GAT layer calculates an attention coefficient. : ,in, The attention coefficient represents the value at time t. Neighbor nodes Status information for the target node The importance of performing state prediction; This represents the leakage linear rectifier function, introducing nonlinear computational capabilities into the model; The attention vector maps the concatenated high-dimensional features to a real scalar, i.e., the attention score. The transpose operator represents converting a vector to its original state. Convert a row vector to a column vector; A learnable shared weight matrix is ​​used to apply the input features to all nodes. Perform a shared linear transformation; These are node feature vectors, representing the target nodes respectively. and neighboring nodes At any moment Enhanced feature vectors; Indicates the first The neighboring nodes at time... Enhanced feature vectors; Indexing neighbor nodes It is traversing the set The index of each element in the array; Representative in the diagram In and target node The set of all directly connected neighboring nodes; Next, the nodes All neighbor features are aggregated using a formula, as follows: ,in, This represents the features of the aggregated nodes. Information about itself and all its neighboring nodes; It is a non-linear activation function; The attention coefficient represents the features of neighboring nodes with high importance. Their contribution was even greater; It is a learnable shared weight matrix; It is the feature vector of the neighboring nodes, that is, the neighboring nodes. At any moment Enhanced feature vectors; By stacking multiple layers of GAT, multi-hop spatial dependencies are captured, allowing the prediction of a node to be indirectly influenced by the state of distant nodes; then, the features of each node are spatially aggregated. Input a temporal decoder; this decoder consists of a sequence model based on an attention mechanism; the decoder uses... Using the initial context and referring to historical time series patterns, the system regressively generates a predicted sequence for a future preset period. The decoder outputs two parallel prediction channels: Battery demand forecasting channel: Output node Expected number of battery swapping requests in different time periods in the future; Grid Price Forecasting Channel: Output Node The projected electricity price for the corresponding electricity price region during the same period in the future; The final outputs are a battery demand forecast matrix and a grid price forecast matrix; The multi-agent trade-off module is used to take the predicted value as input and calculate the power constraint envelope of the area corresponding to each smart battery swapping cabinet through a hierarchical multi-agent trade-off algorithm, as well as the target trade-off factor for dynamically balancing user waiting time, battery life loss and grid cost. The charging power allocation module is used to generate a real-time charging power allocation list for each battery in the battery swapping cabinet based on the power constraint envelope, the target trade-off factor, the real-time status of the corresponding battery swapping cabinet and the prediction results. The instruction issuance and feedback module is used to issue the real-time charging power allocation list to the corresponding smart battery swapping cabinet and collect new global state perception data to feed back to the dynamic graph network prediction model.

6. A computer-readable storage medium, characterized in that, It includes one or more program instructions, which are executed by a processor as described in any one of claims 1-4, for a smart shared battery swapping cabinet battery management method.

Images