Wheel type AI battery replacement robot power station energy management system
By using a cloud-based energy management platform and LSTM neural network for electricity price prediction and battery health assessment, the problems of grid electricity price utilization, battery health management and equipment collaborative control in traditional battery swapping stations have been solved. This has enabled efficient, intelligent and economical energy management for the new energy vehicle and energy storage battery industries, reducing operating costs and extending battery life.
Patent Information
- Application Number
- CN202510921150.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-04
- Publication Date
- 2025-11-14
AI Technical Summary
Traditional battery swapping stations and charging facilities have many problems in terms of grid electricity pricing, battery health management, power supply priority adjustment, and equipment collaborative control, resulting in high electricity costs, short battery life, and low system operating efficiency, which cannot meet the energy management needs of the new energy vehicle and energy storage battery industries for high efficiency, intelligence, and economy.
A cloud-based energy management platform combined with LSTM neural networks is used for electricity price prediction and battery health assessment. The charging strategy is dynamically adjusted through a multi-level energy management system to optimize power resource allocation and achieve collaborative control of equipment. This includes the integration of the cloud-based energy management platform, edge controller, and terminal device layers. LSTM neural networks are used to predict future electricity prices, generate charging task queues during low-price periods, and optimize battery usage strategies through a battery SOH assessment model.
It improved electricity costs by 40%, extended battery life by 30%, and achieved 99.9% system availability, realizing efficient utilization of power resources and coordinated optimization of equipment.
Smart Images
Figure CN120952890A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery swapping technology, and in particular to an energy management system for a wheeled AI battery swapping robot power station. Background Technology
[0002] In the current global trend of energy transition, my country has set ambitious goals for carbon peaking and carbon neutrality, aiming to promote a comprehensive green transformation of its economy and society and achieve sustainable development. Against this backdrop, industries such as new energy vehicles, energy storage batteries, and industrial batteries have ushered in a period of rapid development, becoming key forces driving energy transformation and industrial upgrading. However, with the rapid development of these industries, traditional battery swapping stations and charging facilities have gradually exposed a series of problems that urgently need to be addressed, severely restricting the efficient development of the industry and the improvement of energy utilization efficiency.
[0003] First, traditional battery swapping stations mostly adopt a "first-come, first-served" charging strategy. This simple charging sequence arrangement does not fully consider the significant difference between peak and off-peak electricity prices. During peak electricity consumption periods, grid electricity prices are high, but battery swapping stations still charge according to the conventional mode, leading to a substantial increase in electricity costs. This not only increases operating costs but also weakens the economic advantages of industries such as new energy vehicles. Conversely, during off-peak hours, electricity prices are relatively low, but traditional battery swapping stations fail to effectively utilize this period for charging, missing a valuable opportunity to reduce electricity costs.
[0004] Secondly, the simplistic nature of battery charging and discharging strategies is another prominent issue. Currently, most battery swapping stations and charging facilities operate based solely on fixed charging and discharging parameters, lacking dynamic assessment of battery health. Over long-term use, the internal chemical properties of batteries change, and their health gradually declines. However, traditional strategies fail to flexibly adjust charging and discharging parameters according to the actual health status of the battery. This not only fails to effectively extend battery life but may also accelerate battery aging, increase battery replacement frequency, and further raise operating costs.
[0005] Furthermore, traditional battery swapping stations and charging facilities have significant shortcomings in adjusting power supply priorities. In actual operation, they do not dynamically adjust based on real-time robot task demands. During peak periods, robots have heavy workloads and high power demands, but traditional battery swapping stations cannot rationally allocate power resources according to the urgency and importance of tasks, easily leading to power shortages and affecting the normal execution of robot tasks. Conversely, during off-peak periods, power resources may be idle and wasted, failing to fully utilize the equipment's efficiency.
[0006] Finally, the control of charging piles, robots, and energy storage devices lacks global coordinated scheduling. These devices operate independently, lacking effective information exchange and coordinated control mechanisms. In complex operating scenarios, optimal resource allocation and efficient utilization cannot be achieved, resulting in low overall system efficiency and failing to meet the energy management needs of industries such as new energy vehicles and energy storage batteries for high efficiency, intelligence, and economy.
[0007] In summary, traditional battery swapping stations and charging facilities suffer from numerous problems in areas such as grid electricity pricing, battery health management, power supply priority adjustment, and equipment collaborative control, making them unable to meet the demands of the rapidly developing new energy vehicle and energy storage battery industries. Therefore, there is an urgent need for a new energy management technology that can comprehensively consider grid electricity price fluctuations, battery health status, real-time task requirements, and equipment collaborative scheduling to improve energy utilization efficiency, reduce operating costs, extend battery life, enhance overall system performance, and promote the sustainable development of the new energy vehicle industry and related industries. Summary of the Invention
[0008] The purpose of this invention is to provide an energy management system for a wheeled AI battery swapping robot power station to solve the problems mentioned in the background art.
[0009] To achieve the above-mentioned objectives, this invention provides an energy management system for a wheeled AI battery swapping robot power station, comprising a cloud-based energy management platform, an edge controller, and a terminal device layer, wherein:
[0010] The cloud-based energy management platform is used to provide dynamic electricity price response strategies, predict the time-of-use electricity price for the next 24 hours based on LSTM neural networks, generate charging task queues during low-price periods, and collaboratively manage battery health.
[0011] The edge controller is deployed locally at the battery swapping station to collect real-time data on robot battery level, charging pile status, and grid load.
[0012] The terminal device layer is used to execute commands from the cloud-based energy management platform. The terminal devices include the battery compartment of the battery swapping robot, the smart charging pile, and the battery equipment.
[0013] Furthermore, the cloud-based energy management platform employs an LSTM neural network for electricity price prediction. The LSTM neural network achieves high-precision prediction in electricity price forecasting through time-series feature capture and gating mechanisms.
[0014] Furthermore, the use of LSTM neural networks for electricity price forecasting includes the following steps:
[0015] Step S1: Load historical data using pandas + numpy technology;
[0016] Step S2: Data standardization is performed using pandas+MinMaxScaler. The data standardization process includes data denoising, data cleaning, special data processing, and standardized output.
[0017] Step S3: Model training is performed using LSTM+Dense+Dropout technology;
[0018] Step S4: Use an LSTM neural network model to predict electricity prices for the next 24 hours;
[0019] Step S5: Based on Euclidean distance matching of historical low price patterns, periodic patterns are learned through LSTM; the output layer prediction value is combined with quantile statistics to automatically divide the low price range.
[0020] Step S6: Create a charging queue task using the schedule;
[0021] Step S7: Send charge / discharge control commands via MQTT.
[0022] Furthermore, step S4 includes the following steps:
[0023] Step S401: Preprocess the data by constructing supervised learning data using a sliding window and combining RobustScaler to enhance normalization to handle outliers.
[0024] Step S402: The input layer integrates multi-dimensional features, filters key time-series information through a forget gate or input gate, and performs feature fusion. The multi-dimensional features include historical electricity prices, load, and weather.
[0025] Step S403: Optimize the hybrid model by combining XGBoost to improve nonlinear fitting ability and reduce RMSE to 0.74.
[0026] Furthermore, the cloud-based energy management platform employs a battery SOH assessment model for collaborative management of battery health, calculating health status in real time through battery parameters, including charge-discharge cycle count, internal resistance change, and temperature rise rate.
[0027] Furthermore, the method for constructing the battery SOH evaluation model includes the following steps:
[0028] Step S601: Collect time-series data such as voltage, current, and temperature during the battery charging and discharging process, and perform data cleaning and time alignment. The data cleaning includes missing value filling and outlier handling.
[0029] Step S602: Extract indirect health factors and obtain internal resistance change characteristics through EIS spectrum analysis. The indirect health factors include the duration of the constant current / constant voltage phase and the peak value of the capacity increment curve.
[0030] Step S603: Construct the model architecture, adopting a hybrid model, lightweight deployment, and multi-task learning mode;
[0031] Step S604: Training is performed using a small sample scenario, with pre-training on laboratory data and fine-tuning on real vehicle data;
[0032] Step S605: After model quantization and compression, the model is deployed to the BMS chip.
[0033] Furthermore, the cloud-based energy management platform allocates power according to the health status of batteries, prioritizing high-power tasks for batteries with high SOH and limiting the charging and discharging rates of batteries with low SOH.
[0034] Furthermore, the cloud-based energy management platform distributes energy in a multi-tiered manner, divided into three layers:
[0035] The first layer is used for global optimization. The cloud generates a charge and discharge plan based on electricity price and SOH to minimize the total cost.
[0036] The second layer is used for real-time adjustment, with the edge controller dynamically allocating charging power based on grid load fluctuations.
[0037] The third layer is used for equipment execution. The charging pile adjusts the output current according to the instructions, and the robot's battery compartment enables flexible charging.
[0038] Compared with existing technologies, this system and method have the following advantages:
[0039] 1. Improved economic efficiency: Charging through off-peak electricity pricing increases the proportion to 75%, reducing overall electricity costs by 40%;
[0040] 2. Extended battery life: The rate of battery capacity decay is reduced by 30% (the number of cycles with SOH>80% increases to over 2000).
[0041] 3. Fault tolerance capability: Automatically switches to backup power in the event of a single node failure, with system availability ≥99.9%. Attached Figure Description
[0042] Figure 1 This is a diagram illustrating the architecture of an energy management system for a wheeled AI battery swapping robot power station.
[0043] Figure 2 This is a flowchart for electricity price prediction using an LSTM neural network.
[0044] Figure 3 This diagram illustrates the energy charging and discharging management strategy of the energy management system for a wheeled AI battery swapping machine power station. Detailed Implementation
[0045] 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 embodiments of the present invention, and not all embodiments. 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.
[0046] like Figure 1 The diagram shown illustrates the system architecture of this invention. This embodiment provides a multi-level energy management system based on dynamic electricity price prediction and battery health assessment, enabling adaptive scheduling of robot charging and discharging tasks within a battery swapping station, grid load balancing, and extended battery life. The system includes a cloud-based energy management platform, an edge controller, and a terminal device layer. The cloud-based energy management platform integrates an electricity price prediction model, a battery SOH assessment algorithm, and a multi-objective optimization engine. The edge controller is deployed locally at the battery swapping station and is responsible for real-time collection of robot SOC, charging pile status, and grid load data. The terminal device layer includes the battery compartment of the battery swapping robot, intelligent charging piles, and battery equipment (such as cascaded battery packs), used to execute commands issued by the cloud-based energy management platform.
[0047] The cloud-based energy management platform employs a dynamic electricity price response strategy, using an LSTM neural network to predict time-of-use electricity prices for the next 24 hours and generate a charging task queue for low-price periods. It dynamically adjusts charging priorities based on the urgency of robot tasks (e.g., allowing urgent tasks to jump the queue during high-price periods). The LSTM neural network achieves high-precision prediction in electricity price forecasting through temporal feature capture and gating mechanisms. Figure 2 The flowchart shown is for electricity price prediction using an LSTM neural network, including the following steps:
[0048] Step S1: Use pandas + numpy technology to load historical data.
[0049] Step S2 uses pandas+MinMaxScaler technology to perform data standardization processing, including data denoising, data cleaning, special data processing, and standardized output.
[0050] Step S3: Use LSTM+Dense+Dropout technology to train the model.
[0051] Step S4 involves forecasting the electricity price for the next 24 hours, specifically including steps S401 to S403:
[0052] Step S401: Data preprocessing. Supervised learning data is constructed using a sliding window (e.g., predicting time t from time t-30), and outliers are processed using RobustScaler for enhanced normalization.
[0053] Step S402: Perform feature fusion. The input layer integrates multi-dimensional features such as historical electricity prices, load, and weather, and filters key time-series information through forget gate / input gate.
[0054] Step S403: Hybrid model optimization, combining XGBoost to improve nonlinear fitting ability, reducing RMSE to 0.74.
[0055] Step S5: Based on Euclidean distance matching of historical low price patterns, the periodic patterns are learned through LSTM; the output layer prediction value is combined with quantile statistics to automatically divide the low price range.
[0056] Step S6: Create a queue task using schedule.
[0057] Step S7: Send charge / discharge control commands via MQTT.
[0058] Battery health is managed collaboratively by constructing a battery State of Health (SOH) assessment model, which calculates battery health in real time using parameters such as charge / discharge cycle count, internal resistance change, and temperature rise rate. Health-based tiered scheduling is implemented, prioritizing high-power tasks for batteries with high SOH and limiting the charge / discharge rate for batteries with low SOH.
[0059] The construction of a battery SOH evaluation model includes data preparation and feature engineering, model architecture, training strategies, and deployment. Specifically, it includes the following steps:
[0060] Step S601, data preparation stage: collect time-series data such as voltage, current, and temperature during battery charging and discharging, and perform data cleaning (filling in missing values and handling outliers) and time alignment.
[0061] Step S602, Feature Extraction Stage: Extract indirect health factors such as the duration of the constant current / constant voltage stage and the peak value of the capacity increment curve (ICA), and obtain the internal resistance change characteristics through EIS spectrum analysis.
[0062] Step S603: The model architecture can adopt various modes such as hybrid model, lightweight deployment, and multi-task learning;
[0063] Step S604 involves small-sample scenario training using transfer learning. Pre-training data is first collected in the lab, followed by fine-tuning on a real vehicle. Regularization employs a Dropout layer and an L2 penalty term to handle noise in the real vehicle data. The loss function is a hybrid MAE+MSE loss that balances convergence speed and robustness.
[0064] Step S605: After model quantization and compression, the model is deployed to the BMS chip. Model parameters are incrementally updated based on OTA technology.
[0065] The cloud-based energy management platform employs a multi-tiered energy allocation system. The first tier is responsible for global optimization, generating charge / discharge plans based on electricity prices and State of Health (SOH) to minimize total costs (electricity cost + battery depreciation cost). The second tier handles real-time adjustments, with edge controllers dynamically allocating charging power based on grid load fluctuations (e.g., switching to energy storage power when the load exceeds limits). The third tier is responsible for device execution, with charging piles adjusting output current according to instructions, and the robot's battery compartment employing flexible charging (pulse charging to protect the battery).
[0066] This system integrates hardware and software using a web-based architecture, employing Spring Boot, Saniljob, EMQX, and Vue for intelligent power station energy management. The host computer architecture includes C#, CAN bus, Socket, and OPC communication to achieve real-time data interaction with hardware devices. The hardware design includes an intelligent charging pile supporting bidirectional power adjustment (10-150kW) and an integrated electrochemical impedance spectroscopy (EIS) detection module. The robot's battery compartment incorporates multiple sensors (temperature, voltage, current) and supports active balancing management.
[0067] like Figure 3 The diagram shows the energy charging and discharging management strategy of the energy management system for a wheeled AI battery swapping machine power station. The edge controller controls the edge computing gateway to push real-time data. Devices such as the PLC, battery, and charger establish connections with the host computer via Socket sockets and synchronize data in real time. After receiving the real-time synchronized data, the host computer synchronizes it to the cloud-based energy management platform via MQTT.
[0068] The cloud-based energy management platform updates device status data in real time via scheduled tasks, stores it in historical data, and provides multi-dimensional reports. An LSTM neural network model loads historical real-time device data, performs data denoising and cleaning, and stores it in a standardized format. The LSTM computing module is trained on the data to predict prices and State of Health (SOH) in real time and provides the optimal charging and discharging task queue.
[0069] The cloud-based energy management platform executes charging and discharging queue tasks in real time and sends charging and discharging commands to the edge controller via MQTT. The edge controller then controls the edge computing gateway to synchronize data to the execution module based on boundary conditions such as cloud charging and discharging strategies and power limits.
[0070] The device execution module at the terminal device layer charges and discharges the battery, charger, and AGV power module according to instructions.
[0071] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. An energy management system for a wheeled AI battery swapping robot power station, characterized in that, This includes a cloud-based energy management platform, edge controllers, terminal device layer, and data acquisition layer, among which: The cloud-based energy management platform is used to provide dynamic electricity price response strategies, predict the time-of-use electricity price for the next 24 hours based on LSTM neural networks, generate charging task queues during low-price periods, and collaboratively manage battery health. The edge controller is deployed locally at the battery swapping station to collect real-time data on robot battery level, charging pile status, and grid load. The terminal equipment is used to execute commands from the cloud-based energy management platform. The terminal equipment includes the battery compartment of the battery swapping robot, the smart charging pile, and the battery device.
2. The energy management system for a wheeled AI battery swapping robot power station according to claim 1, characterized in that, The cloud-based energy management platform uses an LSTM neural network for electricity price prediction. The LSTM neural network achieves high-precision prediction in electricity price prediction through time-series feature capture and gating mechanisms.
3. The energy management system for a wheeled AI battery swapping robot power station according to claim 1 or 2, characterized in that, Electricity price forecasting using LSTM neural networks includes the following steps: Step S1: Load historical data using pandas + numpy technology; Step S2: Data standardization is performed using pandas+MinMaxScaler. The data standardization process includes data denoising, data cleaning, special data processing, and standardized output. Step S3: Model training is performed using LSTM+Dense+Dropout technology; Step S4: Use an LSTM neural network model to predict electricity prices for the next 24 hours; Step S5: Based on Euclidean distance matching of historical low price patterns, periodic patterns are learned through LSTM; the output layer prediction value is combined with quantile statistics to automatically divide the low price range. Step S6: Create a charging queue task using the schedule; Step S7: Send charge / discharge control commands via MQTT.
4. The energy management system for a wheeled AI battery swapping robot power station according to claim 3, characterized in that, Step S4 includes the following steps: Step S401: Preprocess the data by constructing supervised learning data using a sliding window and combining RobustScaler to enhance normalization to handle outliers. Step S402: The input layer integrates multi-dimensional features, filters key time-series information through a forget gate or input gate, and performs feature fusion. The multi-dimensional features include historical electricity prices, load, and weather. Step S403: Optimize the hybrid model by combining XGBoost to improve nonlinear fitting ability and reduce RMSE to 0.
74.
5. The energy management system for a wheeled AI battery swapping robot power station according to claim 1, characterized in that, The cloud-based energy management platform uses a battery SOH assessment model to collaboratively manage battery health. It calculates battery health in real time through battery parameters, including the number of charge-discharge cycles, internal resistance change, and temperature rise rate.
6. The energy management system for a wheeled AI battery swapping robot power station according to claim 5, characterized in that, The method for constructing the battery SOH evaluation model includes the following steps: Step S601: Collect time-series data such as voltage, current, and temperature during the battery charging and discharging process, and perform data cleaning and time alignment. The data cleaning includes missing value filling and outlier handling. Step S602: Extract indirect health factors and obtain internal resistance change characteristics through EIS spectrum analysis. The indirect health factors include the duration of the constant current / constant voltage phase and the peak value of the capacity increment curve. Step S603: Construct the model architecture, adopting a hybrid model, lightweight deployment, and multi-task learning mode; Step S604: Training is performed using a small sample scenario, with pre-training on laboratory data and fine-tuning on real vehicle data; Step S605: After model quantization and compression, the model is deployed to the BMS chip.
7. The energy management system for a wheeled AI battery swapping robot power station according to claim 1, characterized in that, The cloud-based energy management platform allocates power according to the health status of batteries, prioritizing high-power tasks for batteries with high SOH and limiting the charging and discharging rates for batteries with low SOH.
8. The energy management system for a wheeled AI battery swapping robot power station according to claim 1, characterized in that, The cloud-based energy management platform distributes energy in a multi-tiered manner, consisting of three layers: The first layer is used for global optimization. The cloud generates a charge and discharge plan based on electricity price and SOH to minimize the total cost. The second layer is used for real-time adjustment, with the edge controller dynamically allocating charging power based on grid load fluctuations. The third layer is used for equipment execution. The charging pile adjusts the output current according to the instructions, and the robot's battery compartment enables flexible charging.