Automatic charging method, device and equipment of auxiliary robot and storage medium

Abstract

The invention provides an automatic charging method and device for an auxiliary robot, equipment and a storage medium, and the method comprises the steps: obtaining a state data set of a charging station, and generating a charging station state matrix according to the state data set; performing long-short-term memory network model training and prediction processing on the charging station state matrix to obtain expected utilization rate data of the charging station in a preset time period; attribute data of the auxiliary robots are obtained, weighted evaluation is carried out on the attribute data, and a dynamic priority matrix of each auxiliary robot is obtained; performing deep learning network training and optimization processing on the charging station state matrix and the dynamic priority matrix to obtain an optimal charging scheduling strategy; determining a target charging path of the auxiliary robot according to preset three-dimensional map information, the expected utilization rate data and the optimal charging scheduling strategy, wherein the preset three-dimensional map information comprises three-dimensional position information of the charging station; and controlling the auxiliary robot to perform automatic charging according to the target charging path.

Description

Technical Field

[0001] The present application relates to the field of automatic charging control technology, and in particular to an automatic charging method, device, equipment and storage medium for an auxiliary robot. Background Art

[0002] At present, with the widespread application of automated equipment such as unmanned delivery vehicles and delivery robots, how to realize the automatic charging of these devices has become an important issue. Existing methods for automatic charging of auxiliary machines usually involve technologies such as charging base positioning, contact charging, and charging time scheduling. However, in practical applications, when multiple charging sites need to be managed at the same time, existing technologies often find it difficult to effectively coordinate the charging of multiple devices and allocate charging resources. Specifically, existing technologies face the following challenges when dealing with multiple charging sites: first, there is a lack of an effective multi-device charging coordination mechanism, making it difficult to reasonably allocate devices to be charged among multiple charging sites; second, the charging priority management and processing are not perfect, and the charging order cannot be flexibly adjusted according to the actual needs and importance of the equipment; finally, there are deficiencies in charging time scheduling, making it difficult to achieve optimal utilization of charging resources while ensuring charging efficiency. Summary of the invention

[0003] The present application provides an automatic charging method, device, equipment and storage medium for assisting a robot, which are used to improve the utilization rate of charging resources while ensuring charging efficiency.

[0004] In a first aspect, an embodiment of the present application provides an automatic charging method for an auxiliary robot, the method comprising: Acquire a state data set of a charging site, and generate a charging site state matrix according to the state data set; Performing long short-term memory network model training and prediction processing on the charging site state matrix to obtain expected usage rate data of the charging site within a preset time period; Acquiring attribute data of the assistive robots, performing weighted evaluation on the attribute data, and obtaining a dynamic priority matrix of each assistive robot; Performing deep learning network training and optimization processing on the charging station state matrix and the dynamic priority matrix to obtain an optimal charging scheduling strategy; Determining a target charging path of the auxiliary robot according to preset three-dimensional map information, the expected usage rate data and the optimal charging scheduling strategy, wherein the preset three-dimensional map information includes three-dimensional location information of the charging station; The auxiliary robot is controlled to perform automatic charging according to the target charging path.

[0005] In a second aspect, an embodiment of the present application provides an automatic charging device for an auxiliary robot, the device comprising: A data processing module, used to obtain a status data set of a charging site, and generate a charging site status matrix according to the status data set; A data prediction module, used to perform long short-term memory network model training and prediction processing on the charging site state matrix to obtain expected usage rate data of the charging site within a preset time period; A data evaluation module, used for acquiring attribute data of the auxiliary robot, performing weighted evaluation on the attribute data, and obtaining a dynamic priority matrix of each auxiliary robot; A strategy generation module, used to perform deep learning network training and optimization processing on the charging station state matrix and the dynamic priority matrix to obtain an optimal charging scheduling strategy; a path planning module, used to determine a target charging path of the auxiliary robot according to preset three-dimensional map information, the expected usage rate data and the optimal charging scheduling strategy, wherein the preset three-dimensional map information includes three-dimensional location information of the charging station; A charging control module is used to control the auxiliary robot to automatically charge according to the target charging path.

[0006] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the automatic charging method of the auxiliary robot as described in any one of the embodiments of the present application when executing the computer program.

[0007] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor enables the processor to implement the automatic charging method for an auxiliary robot as described in any one of the embodiments of the present application.

[0008] An embodiment of the present application provides an automatic charging method for an auxiliary robot, the method comprising: obtaining a status data set of a charging site, and generating a charging site status matrix according to the status data set; performing long short-term memory network model training and prediction processing on the charging site status matrix to obtain expected usage rate data of the charging site within a preset time period; obtaining attribute data of the auxiliary robot, and performing weighted evaluation on the attribute data to obtain a dynamic priority matrix for each auxiliary robot; performing deep learning network training and optimization processing on the charging site status matrix and the dynamic priority matrix to obtain an optimal charging scheduling strategy; determining a target charging path for the auxiliary robot according to preset three-dimensional map information, expected usage rate data and the optimal charging scheduling strategy, the preset three-dimensional map information including three-dimensional position information of the charging site; and controlling the auxiliary robot to automatically charge according to the target charging path. Through the above method, the status data set of the charging site is obtained and the charging site status matrix is ​​generated. The state matrix is ​​trained and predicted through the long short-term memory network (LSTM) model to generate expected usage rate data, realize the prediction of future charging demand, obtain the attribute data of the auxiliary robot, and perform weighted evaluation on it to generate a dynamic priority matrix. By comprehensively considering factors such as the remaining battery power and the task type, it is ensured that each auxiliary robot is charged on demand to improve the charging efficiency. The charging site status matrix and the dynamic priority matrix are trained and optimized through deep learning networks to obtain the optimal charging scheduling strategy, so that the allocation of charging resources is more reasonable and the excessive crowding or idleness of charging sites is effectively avoided. Combined with the preset three-dimensional map information, the charging path of each auxiliary robot can be intelligently planned to ensure that it reaches the charging site quickly and accurately, reducing waiting time and path conflicts. In this way, the utilization rate of charging resources is improved while ensuring the charging efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0010] Figure 1 A schematic flow chart of an automatic charging method for an auxiliary robot provided in an embodiment of the present application; Figure 2 A schematic block diagram of an automatic charging device for an auxiliary robot provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0012] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0013] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0014] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0015] See also Figure 1 , Figure 1 FIG. 1 is a schematic flow chart of an automatic charging method for an auxiliary robot provided in an embodiment of the present application. Figure 1 As shown, the specific steps of the method include: S101-S106.

[0016] S101. Obtain a status data set of a charging site, and generate a charging site status matrix according to the status data set.

[0017] For example, the status data set of a charging station includes: waiting time, charging speed and idle time. In the charging station status matrix, rows represent different charging stations; columns represent various attributes, such as the number of charging piles, the number of currently available charging piles, the average waiting time, etc.; cell values ​​represent the specific values ​​of the corresponding attributes of each charging station at a specific time point.

[0018] S102: Perform long short-term memory network model training and prediction processing on the charging site state matrix to obtain expected usage rate data of the charging sites within a preset time period.

[0019] Exemplarily, a long short-term memory network (LSTM) model is used to predict the expected utilization data of each charging station in the next 24 hours. Based on these prediction results, the availability of each charging station is dynamically adjusted. For example, for charging stations with higher predicted utilization rates, a certain number of charging interfaces are reserved to cope with sudden demands. So that when multiple auxiliary robots request charging at the same time, the expected utilization data of each charging station can be combined to assign the optimal charging station to each auxiliary robot. This allocation process uses a heuristic algorithm to strike a balance between minimizing charging waiting time and balancing station loads. In this way, it is possible to effectively avoid overcrowding of some charging stations while other stations are idle, thereby improving the overall charging efficiency.

[0020] S103. Acquire the attribute data of the auxiliary robot, perform weighted evaluation on the attribute data, and obtain a dynamic priority matrix for each auxiliary robot.

[0021] Exemplarily, a dynamic priority index is established for each auxiliary robot. This index is calculated through a complex multi-factor evaluation model, which takes into account the following attribute data: (1) the remaining battery power, which is mapped by a nonlinear function so that the lower the power, the faster the priority increases; (2) the task importance of the device, which is weighted according to the preset task type and urgency; (3) historical charging behavior, which rewards devices that charge regularly and punishes devices that frequently charge temporarily; (4) expected working time, which takes into account the future work plan of the device to ensure that devices that work for a long time have sufficient power; (5) charging efficiency, which prioritizes devices with fast charging speed to improve the overall charging station utilization. Based on these attribute data, the priority index of each auxiliary robot is calculated. When multiple auxiliary robots request charging at the same time, they are sorted according to their priority indexes, and charging resources are dynamically allocated in combination with the availability of the current charging station. In addition, the priority of the device is quickly adjusted according to emergencies (such as emergency task allocation) to ensure that critical devices can get charging opportunities in time.

[0022] S104: Perform deep learning network training and optimization on the charging station status matrix and the dynamic priority matrix to obtain an optimal charging scheduling strategy.

[0023] Exemplarily, the core of the deep learning network is a dynamic programming model that aims to maximize charging efficiency while balancing the load of charging stations. In the deep learning network, a day is divided into multiple time periods (for example, 15 minutes as a unit), and a deep state matrix is ​​established for each time period and each charging station. This deep state matrix contains the charging station state matrix and the dynamic priority matrix. Based on the deep state matrix, reinforcement learning technology, especially the deep Q learning network (DQN), is used to learn the optimal charging scheduling strategy. The input of the deep Q learning network (DQN) includes the deep state matrix and possible scheduling actions, and the output is the expected reward value of each action. The reward function is designed as a weighted sum of charging efficiency, load balance and user satisfaction. By continuously interacting and learning with the environment, the deep Q learning network (DQN) can gradually optimize its decision-making strategy. In addition, the deep Q learning network (DQN) also introduces a dynamic price mechanism to appropriately increase the charging price during peak hours and guide some non-urgent charging demands to shift to low-peak hours, thereby further optimizing the utilization of charging stations.

[0024] S105. Determine a target charging path for the auxiliary robot according to preset three-dimensional map information, expected usage rate data, and an optimal charging scheduling strategy, wherein the preset three-dimensional map information includes three-dimensional location information of charging stations.

[0025] Exemplarily, a three-dimensional map information of a charging station is constructed. The three-dimensional map information not only shows the three-dimensional location of each charging station, but also updates the availability, current queue situation and expected waiting time of the station in real time. The A* algorithm is combined with real-time traffic data to plan the optimal path for each auxiliary machine that needs to be charged. This path planning not only takes into account the distance factor, but also takes into account the current status and predicted utilization rate of the charging station to avoid directing the machine to a station that may be congested. At the same time, the auxiliary machine can make an appointment for the charging period in advance according to its own work plan and power status. By using a heuristic algorithm, while meeting user needs, efforts are made to achieve load balancing of charging stations. For example, if a station in a certain period of time has too many appointments, other nearby time periods or stations are automatically recommended. In addition, the present application also implements a dynamic adjustment mechanism that can automatically adjust the appointment according to the real-time situation, such as when a machine completes a task ahead of schedule, automatically advance its charging appointment, thereby improving the overall charging efficiency.

[0026] S106. Control the auxiliary robot to automatically charge according to the target charging path.

[0027] An embodiment of the present application provides an automatic charging method for an auxiliary robot, the method comprising: obtaining a status data set of a charging site, and generating a charging site status matrix based on the status data set; performing long short-term memory network model training and prediction processing on the charging site status matrix to obtain expected usage rate data of the charging site within a preset time period; obtaining attribute data of the auxiliary robot, and performing weighted evaluation on the attribute data to obtain a dynamic priority matrix for each auxiliary robot; performing deep learning network training and optimization processing on the charging site status matrix and the dynamic priority matrix to obtain an optimal charging scheduling strategy; determining a target charging path for the auxiliary robot based on preset three-dimensional map information, expected usage rate data, and the optimal charging scheduling strategy, the preset three-dimensional map information including three-dimensional position information of the charging site; and controlling the auxiliary robot to automatically charge according to the target charging path. Through the above method, the status data set of the charging site is obtained and the charging site status matrix is ​​generated. The state matrix is ​​trained and predicted through the long short-term memory network (LSTM) model to generate expected usage rate data, realize the prediction of future charging demand, obtain the attribute data of the auxiliary robot, and perform weighted evaluation on it to generate a dynamic priority matrix. By comprehensively considering factors such as the remaining battery power and the task type, it is ensured that each auxiliary robot is charged on demand to improve the charging efficiency. The charging site status matrix and the dynamic priority matrix are trained and optimized through deep learning networks to obtain the optimal charging scheduling strategy, so that the allocation of charging resources is more reasonable and the excessive crowding or idleness of charging sites is effectively avoided. Combined with the preset three-dimensional map information, the charging path of each auxiliary robot can be intelligently planned to ensure that it reaches the charging site quickly and accurately, reducing waiting time and path conflicts. In this way, the utilization rate of charging resources is improved while ensuring the charging efficiency.

[0028] In order to more clearly introduce the technical solution of the present application, the technical solution of the present application will be introduced through specific embodiments below. It should be noted that the specific embodiments are used to expand the technical solution of the present application, but are not intended to limit the present application.

[0029] In some embodiments, a status data set of a charging site is obtained, and a charging site status matrix is ​​generated based on the status data set, including: preprocessing the status data set to obtain a processed status data set; dividing the processed status data set according to preset time intervals to form time series data; extracting features from the time series data to obtain a data feature set; constructing a multidimensional feature vector based on the data feature set and preset static information of the charging site; and using a principal component analysis (PCA) algorithm to perform dimensionality reduction processing on the multidimensional feature vector to obtain a charging site status matrix.

[0030] This series of processing steps not only improves the usability and analysis efficiency of the status data set, but also provides high-quality input data for subsequent prediction and optimization algorithms, thereby improving the prediction accuracy and scheduling optimization effect of the entire charging system.

[0031] In some embodiments, a long short-term memory network model is trained and predicted on the charging site state matrix to obtain expected usage rate data of the charging sites within a preset time period, including: S1021-S1027.

[0032] S1021. Extract time series features from the charging site state matrix to obtain a multidimensional time series data set including a charging power value sequence P(t), a charging duration value sequence D(t), and a waiting time value sequence W(t), where t represents a time index.

[0033] It should be noted that the charging power value P(t) reflects the power usage of the charging station at different time points. The charging duration value D(t) indicates the time required for each charging. The waiting time value W(t) indicates the time the robot waits for charging. These features constitute multi-dimensional time series data, providing basic data for subsequent predictions.

[0034] When extracting the time series features of the charging station state matrix, for each charging station i, its charging power value P is collected i (t), charging time D i (t) and waiting time W i (t), construct a multidimensional time series data set S(t). For time point t, the data set is expressed as: S(t)=[P i (t), D i (t), W i (t)]; where i represents the charging station number and t represents the time index value.

[0035] S1022. Segment the multidimensional time series data set into sliding windows, set the window length to 48 time steps, and set the sliding step to 6 time steps to obtain a training sample sequence X(t), and perform data enhancement processing on the training sample sequence X(t) to generate an extended sample set Xe(t) containing Gaussian noise.

[0036] For example, when a sliding window is used to segment a multidimensional time series data set S(t), the window length L is set to 48 time steps to cover the complete daily cycle characteristics; the sliding step s is set to 6 time steps to ensure that the data overlap is moderate. For window k, the generation formula of the training sample sequence X(t) is: X(t) k=[S(t-L+1), S(t-L+2), ..., S(t)] Generate augmented samples by adding Gaussian noise N(0, σ²): Xe(t)=X(t)+N(0, σ²); where σ² is the noise variance, set to 0.1 times the standard deviation of the original data.

[0037] By adding Gaussian noise to expand the training samples, the model's robustness to noise is improved, the amount of training data is increased, and overfitting is prevented.

[0038] S1023. Input the expanded sample set Xe(t) into the wavelet transform unit for multi-scale decomposition, set the decomposition layer number to 3, select the db4 wavelet basis function, obtain the multi-scale feature set F(t), and perform principal component analysis on the multi-scale feature set F(t), select the feature subset with a cumulative contribution rate of 95%, and obtain the reduced-dimensional feature matrix R(t).

[0039] When the extended sample set Xe(t) is input into the wavelet transform unit, the db4 wavelet basis function is used for 3-layer decomposition to obtain the approximate coefficient a3 and detail coefficients {d1, d2, d3}. The multi-scale feature set F(t) includes: F(t)={a3, d3, d2, d1}. The principal component analysis of F(t) is performed, and the eigenvalue λ i Arrange in descending order and select k eigenvectors that meet the following conditions: ∑(λ i ) / ∑(λ total )≥95%, i ranges from 1 to k to obtain the reduced dimension feature matrix R(t).

[0040] S1024. Construct a four-layer long short-term memory network structure based on the reduced dimension feature matrix R(t). The number of neurons in each layer is 256, 128, 64, and 32 respectively. Add a Dropout layer between each layer, set the inactivation rate to 0.2, and select the ReLU function as the activation function to obtain the initial network model M0.

[0041] For example, when constructing a four-layer LSTM network based on the reduced-dimensional feature matrix R(t), the forward propagation calculation of the lth layer is: l (t)=LSTM(h l (t-1), x l (t)x l (t)=Dropout(ReLU(h (l-1) (t)), p=0.2); where h l (t) is the hidden state of the lth layer at time t, x l (t) is the input.

[0042] S1025. Perform cross-validation training on the initial network model M0, using a 5-fold cross-validation method, 800 training rounds per fold, the mean square logarithmic error loss is selected as the loss function, the AdaBelief is selected as the optimizer, and the initial learning rate is set to 0.0005 to obtain the verified model set Mv.

[0043] In the cross-validation training process, the data set D is randomly divided into 5 subsets {D1, D2, D3, D4, D5}, and 4 subsets are selected as training sets each time, and the remaining 1 is used as the validation set. The loss function is defined as: L = 1 / n∑(log(y pred -y true )²); where y pred is the predicted value, y true is the true value.

[0044] S1026. Build an integrated learning model based on the verified model set Mv, use the Bagging strategy to perform weighted averaging on the prediction results of each sub-model, and determine the weight coefficient through grid search to obtain the optimal integrated model Me.

[0045] When constructing an integrated model for the verified model set Mv={M1, M2, M3, M4, M5}, the prediction value calculation formula is: ensemble =∑(w i *y i ), where w i is the weight of the i-th model, which is optimized in the range [0, 1] through grid search and satisfies ∑w i =1.

[0046] S1027. Input the optimal integrated model Me into the status data of the most recent 72 time steps for rolling prediction, predict the status of the charging site for the next 24 time steps, and calculate the utilization rate indicator U(t) based on the prediction results, where U(t) is equal to the predicted charging time D(t) divided by the sum of the total time T and the predicted waiting time W(t), to obtain the expected utilization rate data of the charging site within the preset time period.

[0047] When using the optimal integrated model Me for prediction, the predicted value at time t+k is calculated as: U(t+k)=D pred (t+k) / (T+W pred (t+k)); where k∈[1, 24] represents the prediction step length, T is the total time constant, and D pred and W pred They are the predicted charging time and waiting time respectively.

[0048] Through the above technical solutions, the time series dynamic change rules of charging station usage characteristics can be fully captured. At the same time, the robustness of the prediction model is improved through data enhancement and multi-model integration, and the future usage rate of charging stations is accurately predicted, providing reliable data support for the subsequent formulation of the optimal charging scheduling strategy. Each technical step of the solution has been carefully designed and coordinated with each other, which not only ensures the prediction accuracy, but also has strong generalization and anti-interference capabilities.

[0049] In some embodiments, the attribute data of the assistive robot is obtained, and a weighted evaluation is performed on the attribute data, including: real-time collection of the attribute data of the assistive robot, including the remaining battery power, the current task type, the historical charging records, the expected working time and the charging efficiency; a nonlinear transformation is performed on the remaining battery power, and the power is mapped to the interval [0, 1] using the sigmoid function to obtain a power evaluation score; according to a predefined task importance level table, an importance weight is assigned to the current task type to obtain a task type score; the historical charging records are analyzed to calculate the charging regularity score; based on the expected working time, a time urgency score is determined according to a preset exponential decay function; a preliminary priority score is obtained based on the power evaluation score, the task type score, the charging regularity score and the time urgency score; and a softmax function is applied to convert the preliminary priority score into a probability distribution to obtain a dynamic priority matrix.

[0050] Exemplarily, by acquiring and weighted evaluating the attribute data of the auxiliary robot, this method can achieve accurate priority sorting of the robot's charging needs. Specifically, key attribute data such as the remaining battery power, current task type, historical charging records, and expected working time collected in real time are subjected to nonlinear transformation and weighted processing to generate a comprehensive evaluation score. This multi-dimensional evaluation method not only takes into account the urgency of the battery power, but also combines the importance and time urgency of the task to ensure a comprehensive evaluation of the charging needs. The preliminary priority score is converted into a dynamic priority matrix through the softmax function, making the charging scheduling more scientific and flexible, thereby improving the utilization efficiency of the charging station and the continuity of the robot's task execution.

[0051] In some embodiments, the charging site status matrix and the dynamic priority matrix are subjected to deep learning network training and optimization processing, including: S1041-S1047.

[0052] S1041. Concatenate the charging station state matrix and the dynamic priority matrix into an input state matrix of the deep Q learning network.

[0053] The charging station state matrix S and the dynamic priority matrix P are concatenated to form the input state matrix X of the deep Q learning network. Assuming that the dimension of the charging station state matrix is ​​m×n and the dimension of the dynamic priority matrix is ​​k×l, the dimension of the concatenated state matrix X is (m+k)×max(n,l). The concatenation operation can be expressed as: X=[S;P] S1042. Define an action space of a deep Q-learning network, where the action space includes: all possible combinations of each auxiliary robot and charging station.

[0054] For example, for the action space A of the deep Q-learning network, all possible combinations of each auxiliary robot ri and charging station sj need to be considered. Assuming there are N auxiliary robots and M charging stations, the size of the action space is N×M. Each action can be represented as a two-tuple a(i, j), where i represents the index of the auxiliary robot and j represents the index of the charging station: A={a(i, j)|i∈[1, N], j∈[1, M]}.

[0055] S1043: Determine a reward function according to the charging efficiency, the preset load balance and the preset waiting time.

[0056] For example, the design of the reward function R needs to consider multiple factors. For any action a(i, j), its reward value consists of the following three parts: R(a)=w1×E(a)+w2×B(a)+w3×T(a); where E(a) represents the charging efficiency term, which is determined by the charging power and charging time; B(a) represents the load balancing term, which reflects the degree of balance in the use of charging stations; T(a) represents the waiting time term, which represents the waiting time for the robot to arrive at the charging station; w1, w2, and w3 are the weight coefficients of each item.

[0057] S1044. At each time step of the deep Q learning network, an experimental action is determined from the action space according to the current strategy, next state information and reward information are generated according to the experimental action, and the next state information and reward information are stored in the experience replay buffer.

[0058] Exemplarily, at each time step t, the deep Q learning network selects an action at from the action space A according to the ε-greedy strategy. The specific selection method is: randomly select an action with probability ε, and select the action with the largest current Q value with probability 1-ε. After selecting the action, the environment is transferred to the new state st+1 and receives the reward rt. The transfer quadruple (st, at, rt, st+1) is stored in the experience replay buffer D: D←D∪{(st, at, rt, st+1)}.

[0059] S1045. Randomly sample batch data from the experience replay buffer and calculate the timing difference error between the target Q value and the current Q value.

[0060] Exemplarily, batches of data of size B are randomly sampled from the experience replay buffer D for training. For each sample (s, a, r, s'), the target Q value is calculated: y = r + γ × max (Q (s', a'; θ-)); where γ is the discount factor and θ- is the target network parameter; the current Q value is calculated at the same time: q = Q (s, a; θ); the difference between the two is the temporal difference error: δ = yq.

[0061] S1046. Use the gradient descent method to update the deep Q learning network to minimize the mean square error loss of the deep Q learning network.

[0062] Exemplarily, the gradient descent method is used to update the parameters θ of the deep Q learning network. The loss function of the network is defined as the mean square error: L(θ)=E[(yQ(s, a; θ))²]; the parameter update method is: θ←θ-α×∇L(θ); where α is the learning rate.

[0063] S1047, until the deep Q learning network converges or reaches a preset number of iterations, the optimal charging scheduling strategy is obtained.

[0064] For example, when the value of the loss function changes less than the preset threshold ε for K consecutive epochs, or reaches the maximum number of iterations Nmax, the network training is considered to have converged. The network parameters at this time correspond to the optimal charging scheduling strategy π*: π*(s)=argmax(Q(s, a; θ*)); where θ* is the network parameter after convergence. This strategy can be used to select the optimal charging site for each auxiliary robot to achieve optimization of charging scheduling.

[0065] The above technical solution significantly improves the utilization efficiency of charging resources and the system response speed, reduces the charging waiting time, and improves the overall charging efficiency and user satisfaction.

[0066] In some embodiments, a target charging path of an auxiliary robot is determined based on preset three-dimensional map information, expected usage rate data, and an optimal charging scheduling strategy, including: converting the preset three-dimensional map information into a graph structure, the graph structure including: nodes and edges, the nodes representing passable points, and the edges representing the connection between two adjacent nodes; preprocessing the graph structure, including: obstacle expansion and path smoothing, to obtain a processed graph structure; integrating the expected usage rate data of the charging station into the processed graph structure to obtain an augmented graph structure; according to the optimal charging scheduling strategy and the augmented graph structure, assigning a target charging station to each auxiliary robot; based on a preset A* algorithm, determining an initial charging path according to the real-time position information of the target charging station and the auxiliary robot; applying a dynamic programming method to optimize the initial path to obtain an optimized path; generating a velocity curve and an acceleration curve according to the kinematic constraints of the auxiliary robot; and determining the target charging path according to a combination of the optimized path, velocity curve, and acceleration curve.

[0067] By converting the preset three-dimensional map information into a graph structure, and performing obstacle expansion and path smoothing processing on the graph structure, the feasibility and safety of path planning are ensured. The expected utilization rate data of the charging station is integrated into the processed graph structure, which can dynamically reflect the real-time load situation of the charging station. Based on the optimal charging scheduling strategy and the augmented graph structure, the target charging station is assigned to each auxiliary robot, and the A* algorithm is used to calculate the initial path, and then the dynamic programming method is applied to optimize the path to ensure the optimality and effectiveness of the path. Combined with the kinematic constraints of the auxiliary robot, the velocity curve and acceleration curve are generated to determine the target charging path. This method effectively reduces path conflicts and congestion, and improves the efficiency and stability of the charging process.

[0068] In some embodiments, the assistive robot is controlled to automatically charge according to a target charging path, including: discretizing the target charging path into multiple groups of path points and timestamps; using a model predictive control (MPC) algorithm to optimize the control input of the assistive robot in a rolling time domain; constructing a dynamic model of the assistive robot, including state equations of position, velocity, and acceleration; defining an objective function of the MPC, including path tracking error, energy consumption, and smoothness terms; setting constraints for the MPC, including velocity and acceleration limits of the assistive robot; solving the optimal control problem in a finite time domain in each control cycle; applying the first control input obtained by optimization to the assistive robot; using a Kalman filter to fuse multi-sensor data to estimate the position and posture of the assistive robot in real time; calculating the tracking error and performing real-time path adjustments based on the estimated state and target path; and when the assistive robot arrives at the target charging site, executing a precise docking procedure, including visual servoing and force control, to achieve automatic charging.

[0069] By discretizing the target charging path into multiple groups of path points and timestamps, and using the model predictive control (MPC) algorithm to optimize the control input of the auxiliary robot in the rolling time domain, the real-time and accuracy of the charging path are ensured. The dynamic model of the auxiliary robot, including the state equations of position, velocity and acceleration, is constructed to accurately simulate the motion characteristics of the robot. The objective function of MPC is defined, and the path tracking error, energy consumption and smoothness are comprehensively considered to ensure the optimality of the control input. The constraints of velocity and acceleration are set to ensure the safety of the robot's movement. By solving the optimal control problem in a finite time domain and applying the optimized control input to the auxiliary robot, accurate path tracking is achieved. The accuracy of path tracking is further improved by combining the Kalman filter for state estimation and real-time path adjustment. When the auxiliary robot arrives at the target charging station, it executes a precise docking procedure, including visual servoing and force control, to achieve efficient and reliable automatic charging process. This solution effectively improves the intelligence level and operation efficiency of the automatic charging system.

[0070] See also Figure 2 , Figure 2 2 is a schematic block diagram of an automatic charging device for an auxiliary robot provided in an embodiment of the present application, wherein the automatic charging device 200 for the auxiliary robot is used to execute the automatic charging method for the auxiliary robot described above. The automatic charging device 200 for the auxiliary robot can be configured in a server.

[0071] Among them, the server can be an independent server or a server cluster, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN) and big data and artificial intelligence platforms.

[0072] like Figure 2 As shown, the automatic charging device 200 of the auxiliary robot includes: a data processing module 201, a data prediction module 202, a data evaluation module 203, a strategy generation module 204, a path planning module 205 and a charging control module 206.

[0073] The data processing module 201 is used to obtain a status data set of a charging site and generate a charging site status matrix according to the status data set.

[0074] The data prediction module 202 is used to perform long short-term memory network model training and prediction processing on the charging site state matrix to obtain the expected usage rate data of the charging site within a preset time period.

[0075] The data evaluation module 203 is used to obtain the attribute data of the auxiliary robot, perform weighted evaluation on the attribute data, and obtain a dynamic priority matrix for each auxiliary robot.

[0076] The strategy generation module 204 is used to perform deep learning network training and optimization processing on the charging station status matrix and the dynamic priority matrix to obtain the optimal charging scheduling strategy.

[0077] The path planning module 205 is used to determine the target charging path of the auxiliary robot according to preset three-dimensional map information, expected usage rate data and optimal charging scheduling strategy. The preset three-dimensional map information includes three-dimensional location information of the charging station.

[0078] The charging control module 206 is used to control the auxiliary robot to perform automatic charging according to the target charging path.

[0079] An embodiment of the present application provides an electronic device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement an automatic charging method for an auxiliary robot such as any one of the embodiments of the present application when executing the computer program.

[0080] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements an automatic charging method for an auxiliary robot such as any one of the embodiments of the present application.

[0081] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An automatic charging method for an auxiliary robot, characterized in that: The method comprises: Acquire a state data set of a charging site, and generate a charging site state matrix according to the state data set; Performing long short-term memory network model training and prediction processing on the charging site state matrix to obtain expected usage rate data of the charging site within a preset time period; Acquiring attribute data of the assistive robots, performing weighted evaluation on the attribute data, and obtaining a dynamic priority matrix of each assistive robot; Performing deep learning network training and optimization processing on the charging station state matrix and the dynamic priority matrix to obtain an optimal charging scheduling strategy; Determining a target charging path of the auxiliary robot according to preset three-dimensional map information, the expected usage rate data and the optimal charging scheduling strategy, wherein the preset three-dimensional map information includes three-dimensional location information of the charging station; The auxiliary robot is controlled to perform automatic charging according to the target charging path.

2. The automatic charging method of the assistive robot according to claim 1, characterized in that: The step of acquiring a status data set of a charging site and generating a charging site status matrix according to the status data set includes: Preprocessing the state data set to obtain a processed state data set; Dividing the processed state data set according to preset time intervals to form time series data; Extracting features from the time series data to obtain a data feature set; Constructing a multidimensional feature vector according to the data feature set and preset static information of the charging station; The multi-dimensional feature vector is subjected to dimensionality reduction processing to obtain the charging station state matrix.

3. The automatic charging method of the assistive robot according to claim 1, characterized in that: The acquiring the attribute data of the auxiliary robot and performing weighted evaluation on the attribute data comprises: Collecting the attribute data of the auxiliary robot in real time, including the remaining battery power, current task type, historical charging records, expected working time and charging efficiency; Performing a nonlinear transformation on the remaining battery power and mapping it using a sigmoid function to obtain a power evaluation score; According to a predefined task importance level table, an importance weight is assigned to the current task type to obtain a task type score; Analyzing the historical charging records and calculating a charging regularity score; Based on the estimated working time, determining a time urgency score according to a preset exponential decay function; Obtaining a preliminary priority score according to the power evaluation score, the task type score, the charging regularity score, and the time urgency score; The preliminary priority scores are converted into probability distributions using a softmax function to obtain the dynamic priority matrix.

4. The automatic charging method of the assistive robot according to claim 3, characterized in that: The performing deep learning network training and optimization processing on the charging station state matrix and the dynamic priority matrix includes: Concatenating the charging station state matrix and the dynamic priority matrix into an input state matrix of a deep Q learning network; Defining an action space of the deep Q-learning network, the action space comprising: all possible combinations of each of the assistive robots and the charging stations; Determining a reward function according to the charging efficiency, the preset load balance and the preset waiting time; At each time step of the deep Q learning network, determining an experimental action from the action space according to a current strategy, generating next state information and reward information according to the experimental action, and storing the next state information and the reward information in an experience replay buffer; Randomly sampling batch data from the experience replay buffer, and calculating the timing difference error between the target Q value and the current Q value; Using a gradient descent method to update the deep Q-learning network, the mean square error loss of the deep Q-learning network is minimized; Until the deep Q learning network converges or reaches a preset number of iterations, the optimal charging scheduling strategy is obtained.

5. The automatic charging method of the assistive robot according to claim 1, characterized in that: The determining the target charging path of the auxiliary robot according to the preset three-dimensional map information, the expected usage rate data and the optimal charging scheduling strategy includes: Converting the preset three-dimensional map information into a graph structure, wherein the graph structure includes nodes and edges, wherein a node represents a traversable point, and an edge represents a connection between two adjacent nodes; Integrating the expected usage rate data of the charging station into the graph structure to obtain an augmented graph structure; According to the optimal charging scheduling strategy and the augmented graph structure, assigning a target charging site to each of the auxiliary robots; Determining an initial charging path according to the real-time position information of the target charging site and the auxiliary robot; Applying a dynamic programming method to optimize the initial path to obtain an optimized path; generating a velocity curve and an acceleration curve according to the kinematic constraints of the assistive robot; The target charging path is determined according to a combination of the optimized path, the speed curve and the acceleration curve.

6. The automatic charging method of the assistive robot according to claim 1, characterized in that: The step of controlling the auxiliary robot to automatically charge according to the target charging path includes: Discretizing the target charging path into multiple groups of path points and timestamps; Use a model predictive control algorithm to optimize the control input of the assistive robot in the receding horizon; Constructing a dynamic model of the assistive robot, including state equations of position, velocity and acceleration; defining an objective function of the model predictive control algorithm, including path tracking error, energy consumption, and smoothness terms; Setting constraints for the model predictive control algorithm, including speed and acceleration limits for the assistive robot; In each control cycle, solve the optimal control problem in a finite time domain; Applying the optimized first control input to the assisting robot; Use Kalman filter to fuse multi-sensor data to estimate the position and posture of the assistive robot in real time; Calculating tracking errors and making real-time path adjustments based on the estimated state and target path; When the assistive robot reaches the target charging site, it executes a precise docking procedure, including visual servoing and force control, to achieve autonomous charging.

7. An automatic charging device for an auxiliary robot, characterized in that: The automatic charging device of the auxiliary robot comprises: A data processing module, used to obtain a status data set of a charging site, and generate a charging site status matrix according to the status data set; A data prediction module, used to perform long short-term memory network model training and prediction processing on the charging site state matrix to obtain expected usage rate data of the charging site within a preset time period; A data evaluation module, used for acquiring attribute data of the auxiliary robot, performing weighted evaluation on the attribute data, and obtaining a dynamic priority matrix of each auxiliary robot; A strategy generation module, used to perform deep learning network training and optimization processing on the charging station state matrix and the dynamic priority matrix to obtain an optimal charging scheduling strategy; a path planning module, used to determine a target charging path of the auxiliary robot according to preset three-dimensional map information, the expected usage rate data and the optimal charging scheduling strategy, wherein the preset three-dimensional map information includes three-dimensional location information of the charging station; A charging control module is used to control the auxiliary robot to automatically charge according to the target charging path.

8. An electronic device, characterized in that: The electronic device comprises a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the automatic charging method for the assistive robot as described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the automatic charging method for the assistive robot as described in any one of claims 1 to 6.

Images