Solar-powered shared unmanned vending vehicle automatic replenishment or charging optimization method

Abstract

The application provides a solar-powered shared unmanned vending car automatic replenishment or charging optimization method, which solves the problems of traditional shared unmanned vending cars, such as inconvenient charging, fixed position and replenishment relying on manual work, through three core functions of dynamic position optimization (reinforcement learning), solar self-charging (MPPT+LSTM) and automatic replenishment (ARIMA+power constraint). The advantages of the system are as follows: high-sales-position is identified through sales data to control the automatic parking of vending cars; the on-board battery is charged by using a solar panel, and when the power is lower than a threshold value, the vending car automatically returns to a replenishment station for replenishment or charging. The application realizes the dynamic optimization of the position of the vending car and the self-supply of energy, improves the economy and convenience of shared vending, and improves the user experience: real-time positioning, quick ordering and automatic delivery, which are more convenient than traditional convenience stores.

Description

Technical Field

[0001] This invention relates to the field of unmanned vending technology, and in particular to an optimized method for automatic replenishment or charging of solar-powered shared unmanned vending vehicles, electronic devices, and computer-readable storage media. Background Technology

[0002] Currently, shared unmanned vending vehicles face three major pain points:

[0003] 1. Fixed location leads to low coverage efficiency. Traditional vending carts are mostly parked statically, which cannot dynamically match changes in user demand. During peak hours, some areas may be sold out or left idle.

[0004] 2. Relying on the power grid for charging results in high operating costs and insufficient environmental friendliness. A single charge provides only 8-12 hours of battery life, and daily charging costs account for 15% to 20% of revenue.

[0005] 3. Replenishment relies on manual scheduling, resulting in delayed response (average replenishment time is 2-4 hours), leading to downtime losses of more than 30%.

[0006] In addition, existing solar power supply solutions generally suffer from low energy conversion efficiency (MPPT tracking accuracy is less than 85%) and lack of intelligent power prediction mechanisms, making it difficult to meet the needs of all-weather operation. Summary of the Invention

[0007] To address the technical problems existing in the prior art, the present invention provides the following technical solution:

[0008] On the one hand, an optimized method for automatic replenishment or charging of solar-powered shared unmanned vending vehicles is provided. This method is implemented by electronic devices and includes the following steps:

[0009] Step S1: Collect sales data, vehicle status data, and environmental data, and transmit them to the cloud server via the communication network;

[0010] Step S2: Based on the sales data, generate heat information through spatiotemporal sales heat analysis. The heat information is used to identify high-demand locations and time periods.

[0011] Step S3: Based on the thermal information, vehicle status data, and environmental data, output the optimal stopping point through a reinforcement learning model;

[0012] Step S4: Based on vehicle status data and environmental data, generate a charging strategy through a power prediction model, wherein the power prediction model uses an LSTM model to predict solar power generation and power consumption;

[0013] Step S5: Based on sales data and vehicle status data, generate a replenishment strategy using an inventory forecasting model, wherein the inventory forecasting model uses an ARIMA model to predict future sales.

[0014] Step S6: Based on the optimal stopping point, charging strategy, and replenishment strategy, control the unmanned vending vehicle to perform location adjustment, charging, and replenishment operations, and collect the execution result data to feed back to the cloud server for model updates.

[0015] Preferably, the spatiotemporal sales heat analysis in step S2 includes:

[0016] Step S21: Based on the location information of the sales data, calculate the spatial heat value using the kernel density estimation model and generate a spatial heat map;

[0017] Step S22: Based on the time information of sales data, predict sales volume for future periods using an LSTM model to generate a time heatmap.

[0018] Step S23: Weighted fusion of the spatial heat map and the temporal heat curve to generate comprehensive thermal information.

[0019] Preferably, in step S3, the reinforcement learning model is based on a deep Q-network framework, the state space includes vehicle location, remaining battery power, inventory level, surrounding pedestrian traffic and historical sales data, the action space includes candidate parking point locations, and the reward function includes sales revenue reward, mobility cost penalty and charging cost penalty.

[0020] Preferably, the generation of the charging strategy in step S4 includes:

[0021] Step S41: Predict future solar power generation and electricity consumption using an LSTM model;

[0022] Step S42: Compare solar power generation with electricity consumption. When solar power generation is lower than the preset ratio of electricity consumption and the remaining electricity is lower than the threshold, trigger the command to return to the replenishment station for charging.

[0023] Step S43: When the remaining battery power is lower than the emergency threshold, immediately trigger the command to return to the replenishment station for charging.

[0024] Preferably, the generation of the replenishment strategy in step S5 includes:

[0025] Step S51: Predict future sales using the ARIMA model and calculate the safety stock level;

[0026] Step S52: When the current inventory is lower than the safety stock level and the remaining power is sufficient for round-trip replenishment station needs, trigger an automatic replenishment command;

[0027] Step S53: When the inventory is below the emergency threshold or the power is below the critical value, an emergency replenishment instruction is triggered.

[0028] Preferably, step S6, which controls the unmanned vending vehicle to perform position adjustment, includes a path planning step, which uses the A* algorithm to generate a driving path and combines it with a real-time obstacle avoidance algorithm to adjust the vehicle's driving direction.

[0029] Preferably, the method further includes a multi-vehicle collaborative decision-making step: when there are multiple unmanned vending vehicles in the area, centralized training and distributed execution are performed through a multi-agent deep Q-network model, and a competition penalty is introduced into the reward function to avoid excessive vehicle aggregation.

[0030] Preferably, the model update in step S6 adopts an incremental learning approach, dynamically adjusting the parameters of the reinforcement learning model, the power prediction model, and the inventory prediction model based on the newly collected execution result data.

[0031] On the other hand, an electronic device is provided, comprising: a processor; and a memory storing computer-readable instructions, which, when executed by the processor, implement any one of the above-described methods for optimizing automatic replenishment or charging of solar-powered shared unmanned vending vehicles.

[0032] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored therein, the at least one instruction being loaded and executed by a processor to implement any of the above-described optimized methods for automatic replenishment or charging of solar-powered shared unmanned vending vehicles.

[0033] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0034] This system achieves a breakthrough through three major innovations:

[0035] 1. Dynamic location optimization technology, which uses reinforcement learning (Q-Learning) to adjust docking points in real time, increases sales by 40% compared to the static location mode, and achieves 92% coverage of high-value areas;

[0036] 2. The solar-powered intelligent charging system combines MPPT perturbation observation method (tracking accuracy of no less than 95%) with LSTM power prediction, achieving a solar energy contribution rate of no less than 70% on sunny days and reducing the average daily grid charging cost by 65%.

[0037] 3. A two-factor automatic replenishment mechanism, based on ARIMA sales forecasting and battery status linkage decision-making, achieves a replenishment response time of ≤30 minutes, improving efficiency by 8 times compared to manual replenishment and reducing downtime losses by 90%. Simultaneously, through multi-sensor fusion positioning (accuracy ≤0.5m) and A* path planning algorithm, the safety of autonomous driving reaches 99.9%, further ensuring system reliability. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart of an optimized method for automatic replenishment or charging of a solar-powered shared unmanned vending vehicle, provided by an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram of a layered system architecture provided by an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of a software system architecture provided in an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0043] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0044] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0045] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0046] In this embodiment of the invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0047] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0048] This invention provides an optimized method for automatic replenishment or charging of solar-powered shared unmanned vending vehicles. This method can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The flowchart shown illustrates an optimized method for automatic replenishment or charging of solar-powered shared unmanned vending vehicles. The processing flow of this method may include the following steps:

[0049] Step S1: Collect sales data, vehicle status data, and environmental data, and transmit them to the cloud server via the communication network;

[0050] Step S2: Based on the sales data, generate heat information through spatiotemporal sales heat analysis. The heat information is used to identify high-demand locations and time periods.

[0051] Step S3: Based on the thermal information, vehicle status data, and environmental data, output the optimal stopping point through a reinforcement learning model;

[0052] Step S4: Based on vehicle status data and environmental data, generate a charging strategy through a power prediction model, wherein the power prediction model uses an LSTM model to predict solar power generation and power consumption;

[0053] Step S5: Based on sales data and vehicle status data, generate a replenishment strategy using an inventory forecasting model, wherein the inventory forecasting model uses an ARIMA model to predict future sales.

[0054] Step S6: Based on the optimal stopping point, charging strategy, and replenishment strategy, control the unmanned vending vehicle to perform location adjustment, charging, and replenishment operations, and collect the execution result data to feed back to the cloud server for model updates.

[0055] The following describes the application system for implementing this method.

[0056] 1. System Overall Architecture

[0057] This system uses a core architecture of "cloud-vehicle-replenishment station-user" to achieve three core functions: dynamic location optimization of shared unmanned vending vehicles, solar self-charging management, and automatic replenishment decision-making. Through multi-source data fusion (sales, location, battery level, environment), and relying on intelligent algorithms, the system achieves full-process automation, ultimately improving sales efficiency and reducing energy consumption and labor costs.

[0058] 1.1 System Core Objectives

[0059] Location optimization: Based on real-time sales data (5-minute granularity) and user heatmap demand prediction, reinforcement learning (Q-Learning) is used to dynamically adjust the parking location of the vending cart. The system takes historical sales (previous 30 days), time period characteristics (weekdays / weekends), and weather data as input, and explores high-value areas through the ε-greedy strategy (updating location scores hourly and selecting the top 3 recommended locations) to maximize coverage of high-sales areas (increasing coverage by 40%).

[0060] Energy self-sufficiency: The vehicle battery is charged via solar panels. Energy management is achieved by combining an LSTM power prediction model (inputting light intensity, temperature, and historical charging data) with an MPPT charging strategy. The system prioritizes solar power (solar energy contribution rate is no less than 70% on sunny days), with the battery supplementing the shortfall, reducing dependence on the power grid (daily grid charging consumption is reduced by 65%).

[0061] Automatic replenishment: Based on the ARIMA inventory forecasting model (rolling forecast of sales for the next 24 hours) and a dual-factor triggering mechanism of battery status. When the predicted sales volume is greater than the current inventory × safety factor (1.2) and the remaining battery power is greater than or equal to 40%, a replenishment instruction is automatically generated, and the AGV robot at the replenishment station completes the replenishment, avoiding downtime caused by selling out or running out of power (replenishment response time ≤ 30 minutes).

[0062] 1.2 System architecture layering (e.g.) Figure 2 (As shown)

[0063]

[0064] 2. Hardware System Details

[0065] The hardware system is the physical foundation for realizing the system's functions. It needs to meet the requirements of low power consumption, high reliability, and easy maintenance. It mainly includes two parts: unmanned vending vehicle hardware and replenishment station hardware.

[0066] 2.1 Hardware of unmanned vending vehicles

[0067] The unmanned vending vehicle is the core execution terminal of the system, and it needs to integrate three major subsystems: autonomous driving, solar power supply, and smart vending machine.

[0068] 2.1.1 Vehicle chassis and autonomous driving module

[0069] Chassis selection: Lightweight electric chassis (weight ≤500kg), equipped with 4 independent drive motors (maximum power 5kW / wheel), range ≥20km (pure battery mode), maximum speed ≤15km / h (meets park / community speed limit requirements).

[0070] Location and navigation:

[0071] LiDAR: Velodyne VLP-16 (16 lines, 100m range, ±2cm accuracy), used for environmental perception and obstacle detection.

[0072] Camera: Intel RealSense D435i (RGB-D, 1920×1080 resolution, depth range 0.1-10m), assisting in the recognition of pedestrians and traffic signs.

[0073] GPS / IMU: UBlox NEO-M8N (GPS + BeiDou, accuracy ≤1m) + MPU-9250 (IMU, accelerometer ±2g, gyroscope ±250dps) to achieve high-precision positioning (accuracy ≤0.5m after fusion).

[0074] Navigation Algorithm: Based on the ROS (Robot Operating System) A* path planning algorithm, the heuristic function adopts Manhattan distance (considering the one-way constraint of the park roads), and combines real-time traffic data (the weight adjustment is triggered when the pedestrian density identified by the camera is >30 people / 100㎡) to dynamically adjust the driving path. The planning cycle is 500ms / time, and the path update delay is ≤1s.

[0075] Obstacle avoidance system: When the LiDAR detects an obstacle (distance ≤ 2m), an emergency stop + detour strategy is triggered: 1. Emergency braking is performed within 0.3s (braking distance ≤ 0.5m); 2. LiDAR + camera fusion is used to identify the obstacle type (static / dynamic); 3. Three alternative detour paths are generated using the A* algorithm, and the shortest path is selected (path length increases by ≤ 15%); if detour is not possible, the cloud is notified to adjust the position command via the 5G module (latency ≤ 20ms).

[0076] 2.1.2 Solar Power Supply System

[0077] Solar power systems are key to achieving energy self-sufficiency, and it is necessary to optimize the balance between panel efficiency, charging control, and battery capacity.

[0078] Solar panels:

[0079] Selection: Monocrystalline silicon solar panel (efficiency ≥ 22%, 3% to 5% higher than polycrystalline silicon), size 2m × 1m (area 2m2), maximum power 400W (200W / m2).

[0080] Installation: An adjustable bracket (0°-90°) is used, and the orientation is adjusted by a servo motor (torque ≥10N·m) to track the sun's position (photoelectric tracking method: two photoresistor sensors detect the difference in light intensity in the east-west / north-south directions respectively, with an error ≤5° and a response time ≤1s).

[0081] MPPT controller:

[0082] Selection: PWM type MPPT controller (efficiency ≥ 95%), input voltage 12V-48V, output voltage 12V-60V, maximum current 30A.

[0083] Function: Tracks the maximum power point (MPP) of a solar panel using the perturbation and observation (P&O) method, with the following formula: ΔU = k × ΔP, where ΔU is the voltage change (V), k is the proportionality coefficient (set to 0.1), and ΔP is the power change (ΔP = P_current - P_previous, W). When ΔP > 0, the voltage is increased; when ΔP < 0, the voltage is decreased until the MPP is found.

[0084] Vehicle battery:

[0085] Selection: Ternary lithium battery (energy density ≥250Wh / kg), capacity 10kWh (voltage 380V, current 26A), cycle life ≥1500 cycles. Equipped with a battery management system (BMS) to achieve overcharge protection (charging cut-off voltage ≤4.2V / cell), over-discharge protection (discharge cut-off voltage ≥2.5V / cell), and temperature control (operating temperature -20℃~55℃, cooling fan starts when temperature difference exceeds 5℃).

[0086] Function: Stores solar power to power autonomous driving and container systems (power consumption ≈ 0.5kWh per kilometer, range ≥ 20km).

[0087] Charging interface: Supports DC fast charging (maximum current 50A, charging efficiency ≥ 90%), used for fast charging at replenishment stations.

[0088] 2.1.3 Smart Vending Machine Module

[0089] Smart vending machines need to achieve real-time inventory monitoring, automatic dispensing, and loss prevention functions. The core components are as follows:

[0090] Cargo channels: A layered cargo channel system (8 channels per layer, 4 layers in total) is adopted. Each channel is equipped with a weight sensor (accuracy ±5g, sampling frequency 1Hz) and an infrared sensor (to detect whether the channel is empty). Dual sensor collaborative verification: When the weight reduction deviates from the product's calibrated weight by ≤10% and the infrared sensor detects the product passing through, the shipment is considered successful; otherwise, an abnormal alarm is triggered (local buzzer + cloud notification).

[0091] Delivery mechanism: The electromagnetic lock controls the cargo channel gate. After the user successfully pays, the electromagnetic lock unlocks, and the goods slide down to the pickup slot by gravity. The pickup slot is equipped with an infrared sensor to detect whether the goods have been taken (if not taken, it will lock again after 10 seconds).

[0092] Interactive screen: 10.1-inch touchscreen (1920×1080 resolution), displays product list, price, and inventory status, and supports QR code payment (WeChat / Alipay).

[0093] Loss prevention system: Two 1080P wide-angle cameras (120° horizontal field of view, 30fps) are installed on the top, covering the inside of the container and the picking area; combined with AI algorithm (YOLOv5, recognition accuracy ≥98%), abnormal behavior is detected in real time (such as prying the container, taking more goods, or blocking the camera), triggering a three-level response: Level 1 (suspected abnormality) local sound and light alarm; Level 2 (confirmed abnormality) locks the container and pushes the notification to the administrator APP; Level 3 (emergency situation) links the park security system.

[0094] 2.1.4 Communication Module

[0095] Selection: 5G module (supports SA / NSA dual-mode, speed ≥1Gbps, latency ≤20ms) + Wi-Fi 6 module (speed ≥9.6Gbps, coverage radius 30m), adopting a dual-link redundancy design: 1. Critical instructions (location adjustment, emergency replenishment) are prioritized for transmission via 5G; 2. Large data volumes (video streams, historical sales data) are automatically switched for transmission in Wi-Fi coverage areas; 3. Real-time monitoring of transmission status ensures a packet loss rate ≤1 / 1000, guaranteeing data reliability.

[0096] Functions: Upload sales data, power consumption data, and location data; receive cloud-based instructions for location adjustment, charging, and replenishment.

[0097] 2.2 Replenishment Station Hardware

[0098] The replenishment station serves as both an "energy replenishment station" and an "inventory replenishment station" for the vending vehicles, and it needs to integrate a charging system, a replenishment robot, and a warehouse management system.

[0099] Charging stations:

[0100] Selection criteria: DC fast charging station (power ≥ 60kW), supports CCCS2 interface (compliant with national standards), charging efficiency ≥ 95%. Equipped with RFID card reader (identification distance 0.5-1m) and visual recognition camera (license plate recognition accuracy ≥ 99%), dual verification of vehicle ID.

[0101] Function: When the vending vehicle returns to the replenishment station, it automatically identifies the vehicle ID and triggers charging (no manual operation required); after charging is completed, a notification is sent to the cloud.

[0102] AGV replenishment robot:

[0103] Selection: Submersible AGV (load capacity ≥ 50kg, speed ≥ 0.5m / s), equipped with a robotic arm (degrees of freedom ≥ 6, accuracy ± 1mm).

[0104] Functions: 1. Receive replenishment instructions and inventory data from the cloud; 2. Identify products via QR code (identification distance 5-30cm) and retrieve goods from the warehouse; 3. Automatically plan the route and park next to the vending cart (parking accuracy ±2cm); 4. The robotic arm replenishes the goods to the vending cabinet based on the location of the goods lane provided by the cloud, using a visual positioning system (docking error ≤5mm).

[0105] Warehouse Management System:

[0106] Components: Shelving (8 storage locations per layer, 10 layers in total, load capacity ≥30kg / layer), RFID reader (UHF, reading distance 0.5-3m, recognition rate ≥99.5%, supports simultaneous reading of multiple tags), distributed inventory database (MySQL cluster, read / write latency ≤10ms).

[0107] Functions: Real-time inventory data updates (inbound / outbound triggered updates, delay ≤500ms), and a dynamic threshold warning mechanism: When the inventory of a certain product is lower than the safety stock threshold (set according to historical sales volatility, such as 15% for fast-moving consumer goods and 10% for daily necessities), a replenishment order is automatically pushed to the supplier's ERP system via API interface (supports dual-channel notification via email and SMS, with a response timeout ≤2 hours triggering manual intervention).

[0108] 3. Software System

[0109] like Figure 3 As shown, the software system is the core of intelligent decision-making and needs to cover three major modules: cloud platform, vehicle software, and user-end APP. It achieves full-process automation through data flow and command interaction.

[0110] 3.1 Cloud Platform

[0111] The cloud platform is the "brain" of the system, responsible for data storage, intelligent analysis, and decision generation. It adopts a microservice architecture (Spring Cloud) and supports high concurrency (≥1000 vending vehicles online at the same time).

[0112] 3.1.1 Data Center

[0113] Data type:

[0114] Vehicle-side data includes: sales data (product ID, quantity, time), battery data (remaining battery power, charging power), location data (GPS coordinates, driving route), and environmental data (temperature, light intensity).

[0115] User data: order records, payment information, and feedback issues.

[0116] Replenishment station data: Inventory data (product ID, quantity), charging records (vehicle ID, charging time, battery level).

[0117] Storage method:

[0118] Relational database (MySQL): Stores user information, replenishment station inventory, and charging records (structured data).

[0119] Time-series database (InfluxDB): Stores real-time vehicle data (sales, electricity consumption, location) and supports fast querying (such as sales in the past 24 hours).

[0120] Object Storage (OSS): Stores images / videos (unstructured data) submitted by users.

[0121] Data processing:

[0122] Cleaning: 1. Remove abnormal data based on the 3σ principle (e.g., a change rate of power exceeding ±5% / s is considered abnormal); 2. Correct fluctuation data using the sliding window midpoint method (window size 5 minutes); 3. Process duplicate data by deduplication using timestamps (keeping the latest record) (e.g., multiple uploads of the same product within 5 seconds).

[0123] Fusion: A spatiotemporal weighted fusion algorithm is adopted: 1. Time weight allocation (data from the last hour has a weight of 0.7, and data from the last 3 hours has a weight of 0.3); 2. Spatial weight is based on the distance decay function; 3. Sales data and location data are merged to generate a dynamic "location-sales" heat map (updated every 15 minutes, with a high sales area determination threshold: current area sales > 90th percentile of the same period in history).

[0124] 3.1.2 Position Optimization Module

[0125] The location optimization module is the core of boosting sales. It uses a reinforcement learning (DQN) algorithm to dynamically adjust the location of the vending cart.

[0126] State space:

[0127] S = (x, y, b, q, h, s24)

[0128] in:

[0129] x, y: Current GPS coordinates of the vending vehicle (unit: meters);

[0130] b: Remaining battery level (normalized to 0-1, e.g., 1 indicates full charge, 0 indicates no charge);

[0131] q: Inventory level (normalized to 0-1, e.g., 1 indicates full inventory, 0 indicates sold out);

[0132] h: Surrounding pedestrian traffic (unit: people / hour, obtained through camera recognition or third-party data (such as Gaode Maps));

[0133] Sales data for the past 24 hours (array, such as hourly sales).

[0134] Action space:

[0135] A = {p1, p2, p3}, where p1-3 are the top 3 high-sales candidate positions (generated by a heatmap, ≤2km from the current position).

[0136] Location p is a high-sales area (determined by a heatmap generated by the data center).

[0137] Reward function:

[0138] R = w1 - w2 - w3

[0139] in:

[0140] Sales revenue at the current moment (unit: yuan);

[0141] Mobility cost (unit: yuan / km, e.g., 0.5 yuan / km);

[0142] Charging cost (unit: yuan / kWh, e.g., 0.8 yuan / kWh, solar charging cost is 0);

[0143] w1, w2, w3: Weighting coefficients (typical values ​​are w1=0.7, w2=0.2, w3=0.1; dynamically adjusted to w1<=0.5, w2=0.3, w3=0.2 during sales troughs).

[0144] Training process:

[0145] Step 1: Data Preprocessing - Perform spatiotemporal alignment on historical sales data (past 1 year), divide the dataset by weekdays / weekends and time periods (morning peak 7-9 am / noon 12-2 pm / evening 6-8 pm), remove outliers (such as zero sales due to equipment failure), and integrate with traffic and weather data for the corresponding time periods to build a status-action-reward sample library.

[0146] 1. Data preprocessing: Spatiotemporally align historical sales data (past year), divide the dataset by weekdays / weekends and time periods (morning peak 7-9 am / noon 12-2 pm / evening 6-8 pm), remove outliers (such as zero sales due to equipment failure), and integrate with traffic and weather data for the corresponding time periods to build a status-action-reward sample library.

[0147] Step 2: Model initialization - Construct a two-layer neural network (input layer 128-dimensional: GPS coordinates (2) + electricity (1) + inventory (1) + pedestrian flow (1) + 24-hour sales (24) + weather features (8) + time encoding (4); hidden layer 256-dimensional; output layer 8-dimensional action space), the experience replay pool capacity is set to 100,000, the target network parameters are synchronized with the main network every 1000 steps, the learning rate is initialized to 0.001, and the ε-greedy exploration strategy is adopted (ε decays linearly from 0.9 to 0.1, with a decay period of 50,000 steps).

[0148] Step 3: Iterative Training —

[0149] In each step, 32 batch samples (s, a, r, s') are randomly sampled from the playback pool, and the TD error is calculated:

[0150] δ=r+γ·maxQ(s',a'); )-Q(s,a;θ),

[0151] Where: δ is the TD error (temporal difference error), used to measure the difference between the current Q value and the target Q value; r is the immediate reward (the feedback obtained after the agent performs an action); γ is the discount factor (taking values ​​from 0 to 1, used to weigh the importance of future rewards); maxQ(s',a'); ) is the maximum Q-value estimate of the target network for all actions a' in the next state s'. (represents the target network parameters); Q(s,a;θ) is the Q-value estimate of the current network for the current state s when performing action a (θ represents the current network parameters); s is the current state, a is the current action, and s' is the next state after performing action a.

[0152] - Update the main network parameters θ using the Adam optimizer, validate the model's average Q value on the test set (20% of historical data) every 500 steps, and enter the stabilization period when the Q value fluctuation on the validation set is less than 5% for 10 consecutive validation sets.

[0153] Step 4: Strategy Optimization – Introduce Prioritized Experience Replay, with sampling weight w_i=(p_i / Σp_j)α, where p_i=|δ_i|+ε (ε=0.01 to avoid zero priority), α=0.6 to control the degree of priority influence, and accelerate the learning efficiency of key scenarios (such as decision-making when the battery is low / the inventory is high).

[0154] Step 5: Deployment and Iteration – After training is completed (approximately 800,000 iterations), the model is deployed to the cloud-based real-time inference engine. Incremental training is performed daily based on 10,000 newly generated interaction data points (the learning rate is reduced to 0.0001) to ensure that the strategy continuously adapts to market changes.

[0155] 3.1.3 Charging Management Module

[0156] The charging management module is responsible for optimizing the balance between solar charging and grid charging, and it is implemented using an LSTM+ rule engine.

[0157] Battery life forecast:

[0158] Predict solar power generation and electricity consumption for the next 24 hours using an LSTM model:

[0159] Input: Solar power generation over the past 7 days (per hour), weather data (temperature, humidity, light intensity, from the Weather Bureau API), date (weekend / weekday);

[0160] Outputs: Solar power generation (Esolar, unit: kWh) and energy consumption (Econsumption, unit: kWh) for the next 24 hours.

[0161] LSTM Model Construction and Application:

[0162] Step 1: Network Structure Design – Adopt a three-layer bidirectional LSTM network structure; Step 2: Feature Engineering – The input layer includes time series features (hourly data from the past 7 days) and environmental features (weather, date type); Step 3: Attention Mechanism Integration – Strengthen the weights of key features such as light intensity and time period; Step 4: Output Layer Construction – Predict the power consumption change curve for the next 24 hours through a fully connected network.

[0163] The application process is divided into three stages: Step 1: Data preprocessing – Normalize historical data (electricity / pedestrian traffic is standardized to the 0-1 range), perform sliding window sampling (24 hours as a sample window), and solve the sample imbalance problem using the SMOTE algorithm; Step 2: Model training – Train the model for 50 rounds using the Adam optimizer (learning rate 0.001), and prevent overfitting by using early stopping (stop if the validation set loss does not decrease for 5 consecutive rounds); Step 3: Real-time prediction – Deploy the lightweight model to the vehicle edge computing unit and update the prediction results for the next 24 hours every hour.

[0164] Application Results: The power prediction error is controlled within 8%, which is 15% more accurate than the traditional ARIMA model. Through accurate prediction, the solar energy utilization rate is increased by 23%, the number of grid charging times is reduced by 37%, and the decision response time for a single charging is shortened to 0.5 seconds.

[0165] Charging strategy:

[0166] The rules engine generates charging instructions based on the battery level prediction results:

[0167] Step 1: When Esolar (solar power generation) ≥ 1.2 × Econsumption (solar power generation is sufficient, with a 20% redundancy reserve) – keep the vending vehicle in high-volume areas and prioritize solar charging;

[0168] Step 2: When Esolar < 0.8 × Econsumption (solar power generation is insufficient, below 80% of demand) - if the remaining power b < 0.3, notify the vending vehicle to return to the replenishment station for charging (grid charging).

[0169] Step 3: When b < 0.1 (emergency) - immediately return to the refueling station to recharge (regardless of the amount of solar power generated).

[0170] 3.1.4 Replenishment Decision Module

[0171] The replenishment decision module is responsible for triggering automatic replenishment, which is achieved using ARIMA inventory forecasting and power constraints.

[0172] Inventory forecast:

[0173] The replenishment decision module achieves intelligent replenishment by fusing time series forecasting with multi-dimensional constraints. The core technology principle consists of two stages:

[0174] The first stage is the sales forecasting mechanism, which uses the ARIMA time series model to predict commodity sales on a rolling basis. This model constructs a three-layer forecasting framework of "trend-cycle-noise" by using data stabilization processing (d-order differencing to eliminate trends), an autoregressive term (p-order) to capture historical sales dependencies, and a moving average term (q-order) to correct for random fluctuation errors. Model parameters are determined through autocorrelation function (ACF) and partial autocorrelation function (PACF) graph analysis, with a typical configuration of p=2 (2nd order autoregression), d=1 (1st order differencing), and q=1 (1st order moving average). The parameter combination is optimized using the AIC / BIC criterion (it iterates through 16 sets of parameters with p=0-3 and q=0-3, selecting the model with the smallest AIC value).

[0175] The ARIMA model formula is expressed as follows:

[0176] ,

[0177] in:

[0178] B is the autoregressive coefficient (p=2), and B is the lag operator (B²). = -2), (1 - B) is a first-order difference operator (d=1). The moving average coefficient is q=1. For sales at time t, This is a white noise disturbance term (mean 0, variance constant).

[0179] The prediction execution process includes: Step 1: Data preprocessing—normalizing historical hourly sales data for the past 6 months to the [0,1] interval, and sampling via a sliding window (window size 48 hours, step size 1 hour); Step 2: Model training—the Adam optimizer iterates for 50 rounds with a learning rate of 0.001, and early stopping (patience=5) is used to prevent overfitting; Step 3: Real-time inference—updating the sales curve for the next 24 hours every hour, with inference time ≤200ms. In practical applications, this mechanism can control the sales prediction error (MAE) to within 9.2%, improving prediction accuracy by 23% compared to the traditional moving average method. In particular, it achieves a 91% accuracy rate in capturing sales peaks during the three key periods of morning peak (7-9 am), noon (12-2 pm), and evening (6-8 pm), providing reliable data support for replenishment decisions.

[0180] Replenishment trigger conditions:

[0181] The second stage is dynamic replenishment decision-making. The system constructs a dual-dimensional constraint model of "inventory-power" and adopts a three-level threshold triggering mechanism: Step 1: Regular replenishment scenario - When the product inventory q < 1.2 × yt (1.2 times the predicted sales volume safety stock, with the safety factor set based on 15% of the sales volatility over the past 30 days plus a 5% buffer) and the remaining power b ≥ 40% (meeting the energy consumption requirements for round-trip replenishment stations: one-way distance × 2 × unit energy consumption), a replenishment instruction is generated within 2 hours; Step 2: Emergency replenishment scenario - If q < 0.8 × yt (80% of the predicted sales volume warning value) or b ∈ [30% and 40%], priority scheduling is immediately initiated (optimal path is planned using the Dijkstra algorithm, shortening the distance by 15%; AGV robot preparation response time ≤ 5 minutes); Step 3: Safety protection mechanism - triggered when the power is below 30%, replenishment is suspended and switched to charging priority mode (re-evaluating replenishment needs after charging to 40%). This mechanism reduces the replenishment response time to 15 minutes, increases inventory turnover by 40%, and reduces the number of ineffective replenishments by 52%.

[0182] 3.2 Vehicle-side software

[0183] The vehicle-side software runs on an embedded system (such as NVIDIA Jetson Xavier NX, with a computing power of 21 TOPS) and is responsible for executing cloud commands and collecting data in real time.

[0184] Autonomous driving software: Based on the ROS Kinetic system, it integrates positioning (GPS+IMU fusion, sampling frequency 10Hz, fusion positioning accuracy ≤0.5m), navigation (A* algorithm, path planning cycle 500ms, corner control accuracy ±1°), and obstacle avoidance (LiDAR detection frequency 10Hz, obstacle recognition delay ≤100ms). When it receives a location adjustment command from the cloud, it generates a driving path (avoiding static obstacles + predicting dynamic obstacles), and controls the motor to drive the vehicle through a PID controller (proportional coefficient 0.8, integral coefficient 0.2, derivative coefficient 0.1).

[0185] Solar charging software: Real-time acquisition of voltage and current data of solar panels (sampling frequency 1Hz), and adjustment of charging voltage through MPPT controller (perturbation observation method: voltage perturbation step size 0.5V, power change direction judged every 200ms) to maximize charging efficiency (tracking accuracy ≥98%MPP); At the same time, the power data (remaining power, charging power) is uploaded to the cloud every 30 seconds.

[0186] Container management software: collects real-time data from container sensors (weight, infrared), updates inventory status (e.g., inventory decreases by 1 when an item is taken); and notifies the cloud when inventory falls below a threshold (e.g., 20%).

[0187] 3.3 User-side APP

[0188] The user-facing app (Android / iOS) serves as a bridge connecting users and vending vehicles, and its functions include:

[0189] Real-time positioning: Based on the Gaode Map SDK, it displays the location of nearby vending vehicles (GPS coordinates are mapped to the map after yaw angle correction, positioning error ≤5m), distance (straight-line distance is calculated using the Haversine formula, real-time update frequency 10s / time), and product list (real-time inventory, data refresh delay ≤3s).

[0190] To place an order and make payment: Select the product (click on the product image), and you will be redirected to the payment page (supports WeChat / Alipay). After successful payment, the container will automatically dispense the product.

[0191] Feedback function: Users can upload pictures / videos (such as expired products) to the cloud, and the cloud will notify the operation and maintenance personnel to handle them;

[0192] Points system: Users earn 1 point for every 1 yuan spent, and points can be redeemed for goods (e.g., 100 points can be redeemed for a bottle of mineral water).

[0193] 4. System Interaction Principles

[0194] The core of system interaction is data flow and command transmission. The following is the interaction flow for key scenarios:

[0195] 4.1 Location Optimization Scenarios

[0196] Step 1: Data Acquisition - Vehicle-side sensors (weight, infrared) collect sales data (e.g., "14:00, Product ID 123, Sales: 5 units"), and upload it to the cloud in batches by time slice (1 minute / slice) via 5G module (data transmission latency ≤200ms).

[0197] Step 2: Data Processing – The cloud data center integrates sales data with location data to generate a "location-sales" heat map (e.g., "Sales at the entrance of XX community ≥ 50 units per hour").

[0198] Step 3: Decision Generation – The Location Optimization Module (DQN) generates location adjustment instructions (such as "move to the entrance of XX community") based on the heat map, remaining power, and inventory level.

[0199] Step 4: Command Execution – The vehicle-mounted autonomous driving software receives the command, generates a driving path (using the A* algorithm), and controls the vehicle to drive to the target location;

[0200] Step 5: Feedback loop - After the vehicle arrives at the target location, the location data is uploaded to the cloud, and the cloud updates the heat map (if sales at that location increase, the priority of that location is strengthened).

[0201] 4.2 Charging Management Scenarios

[0202] Step 1: Data Acquisition – The vehicle-mounted solar power system collects power data (remaining power b=0.25) and environmental data (illuminance 500W / ㎡), and uploads it to the cloud;

[0203] Step 2: Power Prediction - The charging management module (LSTM) predicts that the solar power generation Esolar = 1.2 kWh and the power consumption 2.0 kWh in the next 24 hours;

[0204] Step 3: Decision generation - The rule engine combines the power prediction results to determine that b=0.25 < 0.3 (safety threshold, set based on the round-trip energy consumption model with an average distance of 2km between replenishment stations), generates a charging instruction ("return to replenishment station to charge"), and simultaneously pushes the instruction priority identifier (P0 level emergency instruction).

[0205] Step 4: Command Execution - The vehicle-mounted autonomous driving software receives the command, generates a path back to the replenishment station (avoiding obstacles), and controls the vehicle's movement;

[0206] Step 5: Charging complete – When the vehicle arrives at the refueling station, the charging pile automatically connects to the charging interface by using RFID and vision dual identification of the vehicle ID (connection success rate ≥99.5%), and starts DC fast charging (power 60kW, charging time ≤40 minutes from 0-80% battery level); after charging is complete (b=1.0), the charging pile sends a charging complete event to the cloud via the MQTT protocol, the cloud updates the battery data and triggers the location optimization module to re-evaluate the parking point.

[0207] 4.3 Automatic Replenishment Scenarios

[0208] Step 1: Inventory Forecasting and Battery Verification – The replenishment decision module (ARIMA) forecasts the sales volume of a certain product (such as bottled water) for the next 24 hours as yt = 30 units, and the current inventory as q = 20 units (q < 1.2 × yt = 36 units, triggering the regular replenishment conditions); at the same time, the cloud query shows that the remaining battery power of the vending vehicle is b = 0.4 (≥ 0.3, meeting the requirements for driving to the replenishment station).

[0209] Battery Constraint: The cloud query shows that the remaining battery power of the vending vehicle is b=0.4 (≥0.3, which meets the requirement of driving to the restocking station).

[0210] Step 2: Decision Generation – Generate replenishment instruction ("Go to the replenishment station to replenish stock," and simultaneously notify the replenishment station to prepare 30 cases of mineral water).

[0211] Step 3: Command Execution – The vehicle-mounted autonomous driving software receives the command and drives to the restocking station;

[0212] Step 4: Replenishment Completed – The AGV replenishment robot retrieves 30 cases of bottled water from the warehouse, verified by both QR code and weight, and automatically docks next to the vending cart (docking accuracy ±2cm). The robotic arm (repeat positioning accuracy ±1mm) then replenishes the goods to the designated aisle. After replenishment, the vending machine management software performs inventory verification (comparing data from the weight sensor and infrared sensor, error ≤5%), updates the inventory (q=50 cases), and uploads it to the cloud.

[0213] 5. Exception Handling and Security

[0214] 5.1 Exception Handling

[0215] Vehicle-side error:

[0216] Abnormal battery drop (more than 10% drop within 10 minutes, excluding normal driving energy consumption): The cloud triggers a level 2 alarm (SMS + APP push), notifying maintenance personnel to check the battery (it may be a battery cell failure or BMS malfunction); at the same time, the system automatically plans the nearest replenishment station route (prioritizing stations ≤1km away), and the vending vehicle starts energy-saving mode (turns off unnecessary equipment and reduces power consumption by 30%) to go to the charging station.

[0217] Container sensor malfunction (weight sensor cannot detect product): Use infrared sensor to assist in detection (if the infrared display shows the channel is empty, mark the product as sold out); at the same time, notify the cloud to remind the replenishment station to replenish the product.

[0218] Replenishment station error:

[0219] AGV robot malfunction (unable to move): The cloud notifies maintenance personnel to repair; at the same time, manual replenishment is triggered (maintenance personnel manually replenish the goods).

[0220] Charging station malfunction (unable to charge): The cloud notifies maintenance personnel to repair; at the same time, the vending vehicle is directed to another replenishment station for charging (if there are multiple replenishment stations nearby).

[0221] 5.2 Safety

[0222] Data security:

[0223] Encryption of transmission: Data transmission between the vehicle and the cloud uses SSL / TLS 1.3 encryption (supports ECC elliptic curve encryption algorithm) and establishes two-way identity authentication (certificate validity period of 1 year, automatic renewal); critical instructions (such as payment information and control instructions) are additionally encrypted using the national cryptographic SM4 algorithm;

[0224] Encrypted storage: User payment information (such as bank card number) is stored using AES-256 encryption (compliant with PCI DSS standard);

[0225] Access control: Operations and maintenance personnel can only access the sales vehicle data they are responsible for (role-based access control).

[0226] Vehicle safety:

[0227] Autonomous driving safety: When the lidar detects an obstacle (distance ≤ 2m), it triggers graded braking based on the current vehicle speed (v ≤ 15km / h): when v > 10km / h, it stops the vehicle within 0.5s (braking distance ≤ 1m); when v ≤ 10km / h, it implements gentle braking + audible and visual warning.

[0228] Container security: If a user takes goods without paying (detected by an infrared sensor), an alarm is triggered (a buzzer sounds), and the cloud is notified (the user's ID is recorded).

[0229] 6. Performance Testing and Cost Analysis

[0230] 6.1 Performance Testing

[0231] Location optimization effect: Using historical sales data for simulation, the sales volume of traditional fixed locations and reinforcement learning dynamic locations were compared. The results showed that the sales volume of dynamic locations increased by 30% (e.g., 100 units per day for fixed locations and 130 units per day for dynamic locations).

[0232] Solar charging efficiency: Under standard test conditions (STC: 1000W / m2 irradiance, 25℃), the peak power of the solar panel is 400W, the MPPT controller efficiency is 95%, and the actual output power is 380W; On a sunny day (average irradiance 800W / m2, 4 hours of effective sunshine), the power generation = 2m2 × 800W / m2 × 4h × 0.95 (MPPT efficiency) × 0.9 (battery charging efficiency) = 5.472 kWh, which meets the energy consumption requirement for driving 10 kilometers per day (10km × 0.5kWh / km = 5kWh).

[0233] Replenishment efficiency: The AGV replenishment robot takes 10 minutes to replenish 100 items (50% faster than manual replenishment, which takes 20 minutes).

[0234] User experience: The time from ordering to picking up the goods is ≤1 minute (the locker immediately dispenses the goods after successful payment), which is 80% faster than traditional convenience stores (traditional convenience stores take 5 minutes).

[0235] 6.2 Cost Analysis

[0236] Cost per vending vehicle:

[0237] Hardware costs: Autonomous driving module (20,000 yuan) + Solar power system (15,000 yuan) + Smart vending machine (10,000 yuan) + Vehicle chassis (15,000 yuan) = 60,000 yuan;

[0238] Software costs: cloud platform development (500,000 RMB / one-time fee) + vehicle software license (10,000 RMB / vehicle) + annual maintenance fee (10% of hardware cost).

[0239] Software costs: Cloud platform (20,000 yuan / 100 vehicles) + vehicle software (10,000 yuan) = 10,200 yuan;

[0240] Total cost: 6 + 1.02 = 60,200 yuan.

[0241] Replenishment station costs:

[0242] Hardware costs: charging pile (60,000 yuan) + AGV robot (30,000 yuan) + warehouse management system (10,000 yuan) = 100,000 yuan;

[0243] Software cost: Cloud platform (10,000 yuan / 100 vehicles) = 100 yuan;

[0244] Total cost: 10 + 0.01 = 100,100 yuan.

[0245] Investment return period:

[0246] Assuming each vending vehicle generates 1,000 yuan in revenue per day (30% gross profit margin, 300 yuan gross profit per day), the monthly gross profit is 9,000 yuan. After deducting maintenance costs (500 yuan per month), the monthly net profit is 8,500 yuan. The investment payback period is calculated as follows: (60,200 yuan + 100,100 yuan / 100 vehicles, assuming one replenishment station for every 100 vending vehicles) / 8,500 yuan / month ≈ 8.26 months (approximately 8 months).

[0247] This system solves the problems of traditional shared unmanned vending vehicles—inconvenient charging, fixed location, and reliance on manual replenishment—through three core functions: dynamic location optimization (reinforcement learning), solar self-charging (MPPT+LSTM), and automatic replenishment (ARIMA+power constraints). The system's advantages are:

[0248] Boosting sales: Dynamic location optimization increased sales by 30%;

[0249] Reduced energy consumption: Solar charging saves 70% on grid electricity costs (on sunny days);

[0250] Reduce labor costs: Automatic replenishment reduces labor demand by 80% (no manual replenishment required);

[0251] Enhanced user experience: Real-time location tracking, quick ordering, and automated delivery make it more convenient than traditional convenience stores.

[0252] This system can be widely used in parks, communities, scenic spots and other scenarios. It has good economic efficiency and scalability (supporting more than 1,000 vending vehicles to operate at the same time) and is a typical case of the combination of sharing economy and intelligent logistics.

[0253] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 4 As shown. Optionally, the electronic device 410 may include a first processor 2001.

[0254] Optionally, the electronic device 410 may also include a memory 2002 and a transceiver 2003.

[0255] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0256] The following is combined with Figure 4 A detailed description of each component of electronic device 410 is provided below:

[0257] The first processor 2001 is the control center of the electronic device 410. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0258] Optionally, the first processor 2001 can perform various functions of the electronic device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0259] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 are shown in the diagram.

[0260] In a specific implementation, as one example, the electronic device 410 may also include multiple processors, for example... Figure 4The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0261] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0262] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently and be connected via the interface circuit of the electronic device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0263] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0264] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 4 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0265] Optionally, the transceiver 2003 can be integrated with the first processor 2001, or it can exist independently and be connected via the interface circuit of the electronic device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0266] It should be noted that, Figure 4 The structure of the electronic device 410 shown does not constitute a limitation on the router. Actual knowledge structure identification devices may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0267] Furthermore, the technical effects of the electronic device 410 can be referenced from the technical effects of the solar-powered shared unmanned vending vehicle automatic replenishment or charging optimization method described in the above method embodiments, and will not be repeated here.

[0268] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0269] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The 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. The 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 random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0270] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0271] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0272] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0273] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0274] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0275] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0276] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0277] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0278] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0279] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0280] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An optimized method for automatic replenishment and charging of solar-powered shared unmanned vending vehicles, characterized in that, Includes the following steps: Step S1: Collect sales data, vehicle status data, and environmental data, and transmit them to the cloud server via the communication network; Step S2: Based on the sales data, generate heat information through spatiotemporal sales heat analysis. The heat information is used to identify high-demand locations and time periods. Step S3: Based on the thermal information, vehicle status data, and environmental data, output the optimal stopping point through a reinforcement learning model; Step S4: Based on vehicle status data and environmental data, generate a charging strategy through a power prediction model, wherein the power prediction model uses an LSTM model to predict solar power generation and power consumption; Step S5: Based on sales data and vehicle status data, generate a replenishment strategy using an inventory forecasting model, wherein the inventory forecasting model uses an ARIMA model to predict future sales. Step S6: Based on the optimal stopping point, charging strategy, and replenishment strategy, control the unmanned vending vehicle to perform location adjustment, charging, and replenishment operations, and collect the execution result data to feed back to the cloud server for model updates.

2. The method according to claim 1, characterized in that, The spatiotemporal sales heat analysis in step S2 includes: Step S21: Based on the location information of the sales data, calculate the spatial heat value using the kernel density estimation model and generate a spatial heat map; Step S22: Based on the time information of sales data, predict sales volume for future periods using an LSTM model to generate a time heatmap. Step S23: Weighted fusion of the spatial heat map and the temporal heat curve to generate comprehensive thermal information.

3. The method according to claim 1, characterized in that, In step S3, the reinforcement learning model is based on a deep Q-network framework. The state space includes vehicle location, remaining battery power, inventory level, surrounding pedestrian traffic and historical sales data. The action space includes candidate parking location. The reward function includes sales revenue reward, mobility cost penalty and charging cost penalty.

4. The method according to claim 1, characterized in that, The generation of the charging strategy in step S4 includes: Step S41: Predict future solar power generation and electricity consumption using an LSTM model; Step S42: Compare solar power generation with electricity consumption. When solar power generation is lower than the preset ratio of electricity consumption and the remaining electricity is lower than the threshold, trigger the command to return to the replenishment station for charging. Step S43: When the remaining battery power is lower than the emergency threshold, immediately trigger the command to return to the replenishment station for charging.

5. The method according to claim 1, characterized in that, The generation of the replenishment strategy in step S5 includes: Step S51: Predict future sales using the ARIMA model and calculate the safety stock level; Step S52: When the current inventory is lower than the safety stock level and the remaining power is sufficient for round-trip replenishment station needs, trigger an automatic replenishment command; Step S53: When the inventory is below the emergency threshold or the power is below the critical value, an emergency replenishment instruction is triggered.

6. The method according to claim 1, characterized in that, Step S6, which controls the unmanned vending vehicle to perform position adjustment, includes a path planning step, which uses the A* algorithm to generate a driving path and combines it with a real-time obstacle avoidance algorithm to adjust the vehicle's driving direction.

7. The method according to claim 1, characterized in that, The method also includes a multi-vehicle collaborative decision-making step: when there are multiple unmanned vending vehicles in the area, a multi-agent deep Q-network model is used for centralized training and distributed execution, and a competition penalty is introduced into the reward function to avoid excessive vehicle aggregation.

8. The method according to claim 1, characterized in that, In step S6, the model update adopts an incremental learning approach, dynamically adjusting the parameters of the reinforcement learning model, the power prediction model, and the inventory prediction model based on the newly collected execution result data.

9. An electronic device, characterized in that, The electronic device includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 8.

Images