Sanitation robot scheduling method, device and equipment based on sanitation task distribution

Abstract

The invention relates to the technical field of robot scheduling, and discloses an environmental sanitation robot scheduling method, device and equipment based on environmental sanitation task distribution, and the method comprises the steps: collecting the garbage accumulation information and people flow density data of a plurality of urban environmental sanitation regions, and carrying out the task distribution prediction through a space-time convolution network model, an environmental sanitation demand prediction result is obtained; according to the environmental sanitation demand prediction result, creating an urban region division scheme of a plurality of urban environmental sanitation regions and robot initial grouping configuration information; according to the urban area division scheme and the initial grouping configuration information of the robots, dynamic scheduling grouping of the sanitation robots is calculated in combination with the real-time garbage accumulation event data; the path planning and cleaning operation instruction of each environmental sanitation robot is output in groups based on dynamic scheduling of the environmental sanitation robots, intelligent planning of the paths of the environmental sanitation robots is realized, and the collaborative operation efficiency and stability of a multi-environmental sanitation robot system are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of robot scheduling, and particularly to a scheduling method, device and equipment for sanitation robots based on the distribution of sanitation tasks. Background Art

[0002] Existing sanitation robot scheduling systems generally have problems such as inaccurate prediction of sanitation task distribution, unreasonable resource allocation, and insufficient ability to respond to emergencies. These systems often adopt static scheduling strategies and cannot be adjusted in real time according to the dynamic changes in urban sanitation needs, resulting in untimely cleaning in some areas while there is resource waste in other areas.

[0003] Existing sanitation robot scheduling methods usually lack in-depth analysis of the spatio-temporal distribution characteristics of urban sanitation tasks and cannot accurately predict the time pattern and spatial distribution of garbage accumulation, thus affecting the scientific nature of scheduling decisions. Traditional scheduling algorithms are mostly based on simple rules or empirical models and are difficult to handle uncertain factors in complex and changeable urban environments, such as sudden garbage accumulation events and changes in pedestrian flow density. In addition, the existing cooperation among robots is insufficient and there is a lack of a unified coordination mechanism. In the face of large-scale cleaning tasks or emergencies, the rapid integration and allocation of resources cannot be achieved, reducing the overall cleaning efficiency. Summary of the Invention

[0004] The present invention provides a scheduling method, device and equipment for sanitation robots based on the distribution of sanitation tasks. The present invention realizes the intelligent planning of the paths of sanitation robots and ensures the cooperation efficiency and stability of the multi-sanitation robot system.

[0005] In a first aspect, the present invention provides a scheduling method for sanitation robots based on the distribution of sanitation tasks. The scheduling method for sanitation robots based on the distribution of sanitation tasks includes: Collecting garbage accumulation information and pedestrian flow density data of multiple urban sanitation areas, and performing task distribution prediction through a spatio-temporal convolutional network model to obtain a sanitation demand prediction result; Creating a city area division plan and initial robot grouping configuration information for the multiple urban sanitation areas according to the sanitation demand prediction result; Calculating the dynamic scheduling grouping of sanitation robots according to the city area division plan and the initial robot grouping configuration information, in combination with real-time garbage accumulation event data; Outputting the path planning and cleaning operation instructions for each sanitation robot based on the dynamic scheduling grouping of the sanitation robots.

[0006] In a second aspect, the present invention provides a scheduling device for sanitation robots based on the distribution of sanitation tasks. The scheduling device for sanitation robots based on the distribution of sanitation tasks includes: A collection module, configured to collect garbage accumulation information and pedestrian flow density data of sanitation areas in multiple cities, and perform task distribution prediction through a spatio-temporal convolutional network model to obtain a sanitation demand prediction result; A creation module, configured to create an urban area division plan and initial robot grouping configuration information for the multiple urban sanitation areas according to the sanitation demand prediction result; A calculation module, configured to calculate a dynamic scheduling grouping of sanitation robots according to the urban area division plan and the initial robot grouping configuration information, in combination with real-time garbage accumulation event data; An output module, configured to output a path planning and cleaning operation instruction for each sanitation robot based on the dynamic scheduling grouping of the sanitation robots.

[0007] A third aspect of the present invention provides a computer device, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor invokes the instructions in the memory to enable the computer device to execute the above-mentioned sanitation robot scheduling method based on sanitation task distribution.

[0008] In the technical solution provided by the present invention, the spatio-temporal convolutional network model is used to analyze the garbage accumulation information and pedestrian flow density data in each area of the city, realizing accurate prediction of sanitation demands, providing a scientific basis for robot scheduling, and effectively solving the problem of inaccurate prediction in traditional scheduling methods. Combining the sanitation demand prediction result, an improved spectral clustering algorithm is used to perform balanced partitioning of urban areas, and the balance, connectivity, boundary simplicity, and area suitability of the area division are ensured through a comprehensive evaluation function, realizing the reasonable allocation of sanitation resources. Based on real-time garbage accumulation event data, the robot grouping is dynamically adjusted through weighted distance calculation and priority sorting mechanisms, ensuring that events with a high degree of urgency and a large amount of garbage accumulation are processed first, while ensuring that the original area sanitation operations are not overly affected. The robust Lyapunov model predictive control method is used to comprehensively consider multiple factors such as path length, task time window, battery energy consumption, and cleaning efficiency, generating an optimal path sequence and cleaning operation instructions, realizing intelligent planning of the robot path. Through a master-slave distributed architecture and a sliding time domain control strategy, real-time status monitoring and instruction optimization of sanitation robots are realized, and a complete closed-loop control mechanism is established, ensuring the collaborative operation efficiency and stability of the multi-robot system. Description of the Drawings

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0010] Figure 1 It is a schematic diagram of the steps of the sanitation robot scheduling method based on the distribution of sanitation tasks in the embodiments of the present invention; Figure 2 It is a schematic structural diagram of the sanitation robot scheduling device based on the distribution of sanitation tasks in the embodiments of the present invention; Figure 3 It is a schematic block diagram of the structure of the computer device in the embodiments of the present invention. Detailed implementation manners

[0011] The embodiments of the present invention provide a sanitation robot scheduling method, device and equipment based on the distribution of sanitation tasks. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and drawings of the present invention are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order different from that shown or described here. In addition, the term "comprising" or "having" and any variation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0012] For ease of understanding, the specific process of the embodiments of the present invention will be described below. Please refer to Figure 1 One embodiment of the sanitation robot scheduling method based on the distribution of sanitation tasks in the embodiments of the present invention includes: Step S1: Collect the garbage accumulation information and pedestrian flow density data of multiple urban sanitation areas, and perform task distribution prediction through a spatio-temporal convolutional network model to obtain the sanitation demand prediction result; It can be understood that the execution subject of the present invention can be a sanitation robot scheduling device based on the distribution of sanitation tasks, or a terminal or a server. Specifically, it is not limited here. The embodiments of the present invention will be described by taking the server as the execution subject as an example.

[0013] Specifically, by deploying a sensor network within multiple urban sanitation areas, information on garbage accumulation and data on pedestrian flow density are collected. The sensor network includes an image acquisition unit, a weight detection unit, and an infrared detection unit, and the three units work together to form a multi-dimensional perception system. The image acquisition unit is installed at major urban intersections, squares, and areas with dense pedestrian flow, and it acquires environmental images at a fixed frequency (for example, once every 15 minutes), and performs preliminary processing through an edge computing unit to extract feature information such as the area and height of garbage accumulation from the images. The weight detection unit is embedded in ground garbage collection points to record in real time the weight and growth rate data of garbage growth, so as to ensure accurate monitoring of the garbage accumulation volume. At the same time, the infrared detection unit is used to detect the density and activity patterns of mobile populations, with a sampling interval set at once every 5 minutes to achieve high-frequency perception of the dynamic changes in pedestrian flow. These different types of data are aggregated to a central data processing center through a low-power wireless transmission network and stored in a time series database. Data correction and time series interpolation processing are performed on the original data to obtain a standardized sanitation data sequence. The data correction process includes five steps: outlier detection and elimination, data completion, time alignment, feature extraction, and data standardization. Outlier detection uses a modified Z-Score algorithm to identify data points that deviate from the normal range, and by statistically analyzing the distribution characteristics of historical data, outliers are eliminated to ensure data quality. Data completion is for data missing due to sensor failure or communication faults. An time series interpolation algorithm is applied, and missing values are filled through interpolation prediction technology to maintain the continuity and integrity of the data. During the time alignment process, since the sampling frequencies of different sensors are inconsistent, all data are unified to a standard time point to ensure the consistency of data dimensions during subsequent model processing. After completing data alignment, in the feature extraction step, by calculating key indicators such as the garbage generation rate in each area, the correlation between pedestrian flow density and garbage volume, and the characteristics of regional garbage, multi-source data are converted into meaningful feature information. The data standardization step converts the features of each dimension to the [0,1] interval to eliminate the differences in different feature dimensions, so that multi-dimensional feature data are comparable in subsequent analysis. By performing interval conversion processing on the processed multi-dimensional feature sanitation data, a standardized sanitation task distribution data set containing timestamps, geographical location coordinates, standardized garbage volume, pedestrian flow density, and garbage growth rate is obtained. Timestamp information ensures the temporal continuity of the data, geographical location coordinates are used to clarify the spatial distribution of the data, and the standardized garbage volume eliminates the influence of the absolute dimension through the standardization process and reflects the relative level of garbage accumulation in different areas. Pedestrian flow density data provides information on population activities related to the garbage generation rate, while the garbage growth rate reveals the dynamic change trend of garbage accumulation in a specific area. The standardized sanitation task distribution data set is input into a spatio-temporal convolutional network model for task distribution prediction to obtain the prediction results of the sanitation needs in each urban sanitation area within a specific future time window.The spatio-temporal convolutional network model includes a temporal convolutional layer, a spatial convolutional layer, an attention mechanism layer, and a fully-connected output layer. It extracts temporal features through one-dimensional convolutional operations, captures spatial correlations between regions using two-dimensional convolutional operations, and dynamically evaluates the importance of different features and regions using the attention mechanism. During the training process of the prediction model, historical data is combined, and the time slicing method is used to divide the dataset into a training set and a validation set to ensure the generalization ability of the model. In the design of the loss function of the model, the mean squared error loss and the spatio-temporal regularization term are combined, and the Adam optimization algorithm is used to optimize the model parameters. The learning rate is dynamically adjusted through the cosine annealing strategy to balance the convergence speed and accuracy of the model. The fully-trained prediction model maps the input multi-dimensional feature data into specific sanitation demand prediction values, including the garbage volume prediction for each grid cell in the future time and the uncertainty estimation of the prediction results.

[0014] Construct a spatio-temporal data tensor based on a standardized dataset of sanitation task distributions. The standardized dataset contains multi-dimensional feature information such as timestamps, geographical location coordinates, standardized garbage amounts, population flow densities, and garbage growth rates. The urban area is divided into grid cells of m×n, and each grid cell corresponds to a specific geographical location. The spatio-temporal data tensor is constructed in the form of [l, m, n, f], where l represents the historical time step, m and n represent the number of rows and columns of the urban area grid respectively, and f represents the feature dimension of each grid cell. This tensor construction method can effectively organize multi-dimensional feature data, maintaining both the coherence of the time series and the structured information of the spatial distribution. Input the spatio-temporal data tensor into a spatio-temporal convolutional network model for task distribution prediction. The model performs one-dimensional convolutional operations to extract temporal pattern features from the data. During the one-dimensional convolution process, the convolution kernel size is set to 24 to capture daily cycle patterns, which are suitable for the daily periodic changes in garbage accumulation amounts and population flow densities in sanitation tasks. By sliding the convolution kernel in the time dimension, the model identifies short-term and long-term trends in garbage growth rates and population flow density changes and maps these temporal patterns into a set of high-dimensional feature representations. Input the feature maps representing temporal patterns into the spatial convolutional layer of the spatio-temporal convolutional network model, and extract spatial correlation features through two-dimensional convolutional operations. The spatial convolutional layer uses a 3×3 convolution kernel and slides over the urban area grid. By performing convolution operations on the data of each grid and its surrounding neighborhoods, the model captures the spatial correlation characteristics between regions. Spatial convolution not only focuses on changes in absolute garbage amounts but also can analyze the spatial distribution patterns of population flow densities, thereby introducing more spatial semantic information into the prediction model and improving the accuracy and generalization ability of the prediction. Perform self-attention analysis and cross-attention analysis on the spatial correlation feature representations to enhance the effectiveness of the feature representations and obtain attention-enhanced feature representations. Self-attention analysis calculates the importance weights of different feature dimensions within a single region. By introducing learnable weight parameters, the model automatically focuses on the key features affecting sanitation demand prediction. For example, during peak garbage generation periods, the model pays more attention to the garbage growth rate feature, while during peak population flow periods, the model tends to amplify the influence of the population flow density feature. At the same time, cross-attention analysis is used to establish dynamic associations between different urban areas. By calculating the mutual influence between different grid cells, the model can more accurately predict demand changes caused by population flow movement or garbage diffusion. Input the attention-enhanced feature representations into the fully connected output layer of the spatio-temporal convolutional network model for non-linear mapping and regression operations to obtain the predicted values of sanitation demand. In the fully connected layer, the model maps high-dimensional features into specific prediction results through a multi-layer perceptron structure, and non-linear activation functions (such as ReLU or LeakyReLU) enhance the model's expressive ability to handle complex task distribution patterns.During the regression operation, the model outputs the predicted garbage volume values for each grid cell within a specific future time window and provides a confidence interval for the prediction results through an uncertainty estimation method. This estimated value can help the scheduling system better evaluate the reliability of the prediction results when allocating resources. For example, in areas with high prediction uncertainty, the number of sanitation robots is appropriately increased to cope with demand fluctuations. Based on the predicted values of sanitation demand, sanitation demand prediction results are generated, including the predicted garbage volume values and uncertainty estimates. These prediction results are visually presented in the form of a heat map of the urban sanitation task distribution. The color depth of each grid cell reflects the size of the garbage demand in that area, and the prediction uncertainty is shown through error bands or confidence intervals.

[0015] Step S2: Create urban area division plans and initial robot grouping configuration information for multiple urban sanitation areas according to the sanitation demand prediction results; Specifically, the city is divided into multiple grid units, each grid unit corresponding to a geographical area, and a feature matrix is constructed using the sanitation demand data provided by the prediction model. This feature matrix includes the predicted garbage volume of each area and also incorporates information on the topological characteristics of the urban road network, such as road connectivity, traffic flow, and distances between regions. Through the spectral clustering algorithm, the feature matrix is transformed into a graph structure, where the nodes represent urban area units and the weights of the edges represent the similarity or connection strength between different regions. Then, through the Laplacian eigenvalue decomposition and clustering operation of the graph, multiple sets of candidate division schemes for urban areas are generated. By calculating the variance of the predicted sanitation task volume within each candidate sub-region, a first index characterizing the balance of sanitation demand within the region is obtained. The higher the balance of demand within the region, the more it means that the sanitation robots assigned to this region can maintain a stable workload for a longer time, avoiding resource waste or reduced cleaning efficiency caused by overly concentrated or overly dispersed demand. The variance calculation method is to statistically calculate the predicted garbage volume of each grid unit, calculate the mean value, and then obtain the degree of deviation of each grid value from the mean value. The smaller the variance, the more balanced the demand within the region. To evaluate the spatial characteristics of the urban area division scheme, for each candidate sub-region, the statistical characteristics of the shortest path distance between any two points within the region are calculated to obtain a second index characterizing the connectivity of the region. In actual calculations, by constructing the shortest path graph of the urban road network, the path distances between all nodes are statistically calculated, and the convenience of internal traffic in the region is evaluated using statistical quantities such as the mean value and variance of the distances. If the internal connectivity of a region is good, the sanitation robots can move efficiently within this region, shortening the cleaning path and improving the operation efficiency. For each candidate sub-region, the ratio of the boundary length to the area of the region is calculated, and this ratio is used as a third index characterizing the simplicity of the region boundary, reflecting the regularity of the region shape and the complexity of the boundary. A region with a simple boundary is not only convenient for path planning and robot scheduling but also can reduce the risk of dead corners or omissions during the cleaning process. At the same time, the matching degree between the area of the region and the operation ability of the robot is calculated to obtain a fourth index characterizing the suitability of the region area. The suitability of the region area is not only related to the cleaning efficiency of a single robot but also needs to consider factors such as battery endurance and garbage collection capacity, ensuring that the assigned region area can enable the robot to fully exert its operation ability and will not cause the task to be unable to be completed within the effective time due to an overly large region. The first index, the second index, the third index, and the fourth index are weighted and fused for calculation to form a comprehensive evaluation function. The comprehensive evaluation function is designed in the form of F = w1·I1 + w2·I2 + w3·I3 + w4·I4, where I1 to I4 respectively represent the four evaluation indicators, and w1 to w4 are the weight coefficients of each indicator. The setting of the weight coefficients is automatically optimized through genetic algorithms or particle swarm optimization algorithms to ensure that the final urban area division scheme can achieve a balance in multiple dimensions and realize the optimal allocation of sanitation resources.During the calculation process, all candidate solutions are scored through an evaluation function, and the solution with the highest score is selected as the final urban area division plan. According to the predicted results of the environmental sanitation needs of each sub-region, the basic environmental sanitation demand of each sub-region is calculated, and the number of robot allocations is determined. The basic environmental sanitation demand is obtained by spatially integrating the predicted demand values within each sub-region, and its calculation formula is Bk = ∫∫Rk P(x, y, t)dxdy, where P(x, y, t) represents the predicted value of the environmental sanitation demand in the region (x, y) at time t, and Rk represents the spatial range of sub-region k. According to the total demand of all sub-regions and the number of available environmental sanitation robots N, a proportional allocation method is used to determine the number of robot allocations Nk for each region. The specific calculation formula is Nk = ⌈N · (Bk / ∑Bi)⌉. Through the ceiling operation, it is ensured that the number of robots allocated to each region is an integer, and the sum of all allocation results is equal to N. For the N environmental sanitation robots allocated within each sub-region, the initial deployment position coordinates of these N robots are determined through a reasonable deployment strategy to complete the initial grouping configuration of the robots. The K-means++ algorithm is used to determine the initial deployment position through cluster analysis of the environmental sanitation demand distribution within the sub-region, so that the demand within the area served by each robot is as balanced as possible. The K-means++ algorithm can effectively avoid the instability of the clustering results caused by the randomness of the initial center selection, and can also fully consider the demand intensity at each location when allocating robots. The finally obtained initial grouping configuration information includes the number of robots allocated to each sub-region and the initial coordinate positions of each robot.

[0016] Step S3: According to the urban area division plan and the initial grouping configuration information of the robots, combined with the real-time garbage accumulation event data, calculate the dynamic scheduling groups of the environmental sanitation robots; Specifically, real-time garbage accumulation event data from the sensor network and monitoring system are received. Each garbage accumulation event includes key information such as event location coordinates, garbage accumulation volume, and urgency level. By analyzing the event data, the amount of resources required for each event is calculated to form a set of event resource demand quantities. This process is achieved through a resource demand mapping function, which not only considers the garbage accumulation volume but also combines the urgency level of the event to perform a weighted calculation of the resource demand. For example, for events with a large amount of garbage, more cleaning resources are preferentially allocated, while for events with a high urgency level, the response speed is accelerated to ensure that emergencies can be handled promptly and effectively. Based on the set of event resource demand quantities, an event priority queue is sorted to form an ordered event priority queue. During the sorting process, the priority score of each event is calculated, comprehensively considering the urgency level and garbage accumulation volume of the event to ensure that resources are preferentially allocated to the events that most need them. The priority calculation formula is Pi = ui·vi / max(vi), where ui represents the urgency level of the event and vi represents the garbage accumulation volume. Through this method, events that are both urgent and require a large amount of resources to handle are automatically identified and these high-priority events are preferentially processed during the scheduling process. For each event in the ordered event priority queue, the weighted distance from all sanitation robots to the event location is calculated. When calculating the weighted distance, the geographical distance between the robot and the event is considered, and various influencing factors are introduced, such as the current load factor of the robot, the task completion rate, the battery consumption factor, and the remaining battery power. The formula for the weighted distance is Di,j = d((xi,yi),(xj,yj))·(1 + αj·Lj + βj·Ej), where d((xi,yi),(xj,yj)) represents the geographical distance between robot j and event i, αj is the current load factor of robot j, Lj is the current task completion rate of robot j, βj is the battery consumption factor, and Ej is the percentage of the remaining battery power. Through the multi-dimensional weighted distance calculation method, when selecting the responding robots, those robots that are closer to the event location, have a lighter operation burden, and have sufficient battery power are preferentially considered. According to the robot weighted distance matrix, the top k sanitation robots with the smallest weighted distance are selected for each event to form a set of temporary response groups. The value of k is determined in relation to the event resource demand quantity, and the value of k is dynamically adjusted according to the amount of resources required for event processing to ensure that the number of selected robots is sufficient to meet the needs of event processing. The selection process focuses on minimizing the weighted distance and checks the current status of each robot to avoid selecting robots that are already close to the operation limit or have too low a battery power, thus ensuring the stability and safety of the response process. For the set of temporary response groups, the coverage range of the operation areas of the remaining sanitation robots within each original group is recalculated to complete the adjustment process of the dynamic scheduling grouping of sanitation robots.During the adjustment process, analyze the remaining number of robots in each original group and the task load in their current working areas. By calculating the task coverage rate of the remaining robots within the new working range, dynamically adjust the task areas of each robot to ensure that the cleaning tasks in each sub-area can still be efficiently executed. For example, when some robots in a certain area are transferred to the temporary response group, the system automatically expands the working radius of the remaining robots or changes the cleaning path so that the areas not affected by the response event can still maintain the original cleaning effect. At the same time, set a limiting condition, that is, when making dynamic grouping adjustments, the number of robots transferred from each original group to the temporary response group shall not exceed 30% of the total number of robots in that group, to prevent a significant decline in the cleaning ability of a certain area due to excessive transfer of robots. If the transfer quantity exceeds the limit, the system will give priority to selecting the next robot with a smaller weighted distance until the allocation condition is met. Through the above calculations and dynamic adjustments, the dynamic scheduling grouping results of the sanitation robots are output in real time.

[0017] Step S4: Based on the dynamic scheduling grouping of the sanitation robots, output the path planning and cleaning operation instructions for each sanitation robot.

[0018] Specifically, each packet in the dynamic scheduling group is analyzed to determine the set of sanitation task points within the responsible area of each packet. These task point sets consist of garbage accumulation points, garbage collection stations, and other cleaning task points within the urban area, and are dynamically updated through the garbage volume data provided by the prediction model and the real-time sensor feedback information. The task point set contains the geographical location coordinates of each task point, as well as information such as the predicted garbage volume, emergency handling priority, and optimal cleaning time window. Based on the complete task allocation scheme, a robust Lyapunov model predictive control (RLMPC) problem is constructed to achieve optimal path planning and the output of cleaning operation instructions. In the RLMPC model, the state variable is designed as x(t), which includes key information such as the position information, speed, battery power, and load information of the sanitation robot. These state variables reflect the current working state and executable capabilities of the robot. The control input u(t) consists of a speed command and a cleaning operation command. The speed command is used to control the direction and speed of the robot's movement, while the cleaning operation command determines the type of cleaning action of the robot at a specific position, including starting garbage collection, obstacle avoidance actions, and the standby mode after task completion. By establishing the association between these state variables and control inputs, the predictive control model can dynamically simulate the working performance of the robot in different environments. To ensure the effectiveness and safety of the predictive control model in actual operation, a comprehensive objective function is constructed for the model and a series of constraint conditions are introduced. In the design of the objective function, four core objectives are comprehensively considered: minimizing the path length, matching the task completion time window, balancing the battery energy consumption, and maximizing the cleaning efficiency. These objectives are weighted and summed through different weight coefficients to form a multi-objective optimization problem. For example, in the objective of minimizing the path length, the system not only hopes that the robot can complete the task with the shortest path, but also needs to avoid congested roads and high-risk areas as much as possible; in the objective of matching the task completion time window, through the output of the prediction model, the task points within the optimal cleaning time window are preferentially planned to ensure that the areas with a high garbage volume growth rate can be cleaned in time. The objective of balancing the battery energy consumption mainly focuses on the battery consumption speed of the robot. By predicting the power consumption of the robot on different paths, the allocation of cleaning tasks is dynamically adjusted so that all robots can complete the operation within the safe battery power range, while the objective of maximizing the cleaning efficiency requires the system to clean as much garbage as possible in the shortest possible time to improve the overall cleaning efficiency. While constructing the objective function, a series of physical and operation constraint conditions are introduced for the predictive control model to ensure the executability of the path planning and operation instructions in the actual environment. The robot dynamics constraints ensure the continuity of the robot's movement path and the smoothness of the speed change, avoiding the robot getting out of control or mechanical failures caused by sudden acceleration or sharp turns.The battery capacity constraint sets the minimum safety threshold for the battery power during the robot's task execution. Once the prediction model calculates that the robot will run out of power during task execution, the system automatically adjusts the task allocation or plans the nearest charging path. The path increment constraint restricts the amplitude of the path change in each control cycle, improves the stability of path planning, and can also effectively reduce the operation of the robot frequently adjusting its direction during execution, thereby reducing mechanical wear and energy consumption. The task time window constraint avoids the robot arriving at the task point at a non-optimal time, resulting in an unsatisfactory cleaning effect, by setting the service time period for specific task points. The obstacle avoidance constraint ensures that the planned path can avoid static and dynamic obstacles, including fixed facilities in the city, temporarily parked vehicles, and pedestrian flow areas. By introducing an obstacle avoidance model, the robot's driving path is dynamically adjusted to ensure operation safety to the greatest extent. The predictive control model, the comprehensive objective function, and the set of constraint conditions are input into a sequential quadratic programming solver to solve the optimal control sequence for each sanitation robot separately. The sequential quadratic programming method is an iterative optimization algorithm that approximates the original non-linear optimization problem as a quadratic programming problem in each iteration and gradually approaches the global optimal solution by solving the quadratic programming sub-problem. Through this process, the system calculates the optimal control input sequence for each robot in the next period of time while maintaining all constraint conditions, including the specific movement path and operation instructions. According to the optimal control sequence, a detailed path point sequence and the corresponding cleaning operation sequence are generated for each sanitation robot. These sequences provide operation guidance for each robot in a fine-grained format. For example, the path point sequence Path(r) = {p0, p1,..., pq} defines all intermediate nodes for the robot to move from the current position to the target task point, while the cleaning operation sequence Op(r) = {o0, o1,..., oq} assigns specific operation instructions to each path node, such as accelerating, decelerating, turning, starting cleaning, avoiding pedestrians, etc. These path planning and cleaning operation instructions are output to the control module of each sanitation robot, enabling the robot to reach the target task position at a specific time point according to the planned optimal path and perform the corresponding cleaning task.

[0019] Build a master-slave distributed architecture to achieve information synchronization and control instruction distribution between the central scheduling unit and each sanitation robot. In the entire system architecture, the central scheduling unit acts as the master node, responsible for global task scheduling, data analysis, and path planning, while each sanitation robot acts as a slave node, specifically executing the cleaning operation instructions issued by the central scheduling unit. After the initial path planning and cleaning operation instructions are generated, the central scheduling unit distributes these instructions to each sanitation robot to form an initial execution plan. After each sanitation robot receives the instructions, it stores the path planning information in the local controller and starts executing the cleaning task according to the preset path points and operation sequences. To ensure the stability and operation efficiency of the entire system, a fixed control period τ is set, for example, one cycle every 5 minutes. In each cycle, each sanitation robot collects its own status information and surrounding environmental data to form a real-time sanitation robot status and environmental data set. These status information include the robot's current position, speed, battery level, current load, operation progress, and the working status of the cleaning equipment, while the environmental data include the surrounding garbage accumulation situation, road traffic conditions, obstacle distribution, and emergency event information detected by sensors. Through the collection and aggregation of these data, the system comprehensively grasps the specific performance of each robot during the current task execution and the dynamic changes in the environment it is in. After each robot completes the collection of status data, these data are uploaded to the central scheduling unit in real time through a low-latency wireless network. After receiving these data, the central scheduling unit performs fusion processing with the real-time updated data in the task distribution prediction model to construct a global state representation of the current system. The global state representation includes the sanitation robot status and environmental data set, and also combines the prediction information of the task prediction model on the change trend of garbage accumulation in the next period of time. Through the comprehensive analysis of multi-dimensional data, the progress and potential risks of the current task execution are dynamically evaluated in both the spatial and temporal dimensions. After obtaining the global state representation of the system, the central scheduling unit detects whether the current task allocation needs to be dynamically grouped and adjusted through the global coordination layer. The global coordination layer, based on the global state matrix G, evaluates whether the current grouping structure still adapts to the existing task environment by analyzing the sanitation demand load in each sub-region, the task completion rate of each group of robots, and the environmental change information. If it is detected that the garbage volume surges or an emergency occurs in a certain area, and the existing robot grouping cannot meet the task requirements, the global coordination layer will trigger the dynamic grouping adjustment process, re-execute the dynamic scheduling grouping, re-allocate the robot tasks by calculating the weighted distance between the robot and the new task point and evaluating the mobilizability of the robot resources, to achieve the optimal utilization of resources. If the system determines that the current task grouping is still reasonable, instead of making large-scale grouping adjustments, it enters the local optimization stage to fine-tune the path planning of each sanitation robot.During the path fine-tuning process, the rolling horizon predictive control method is adopted, which only focuses on the optimal control solution in the near-term time domain. By calculating the possible obstacles, road congestion conditions, and the changing trends of garbage accumulation points that the robot may encounter in the next control cycle, the path point sequence and cleaning operation instructions of the robot are dynamically adjusted. For example, when the garbage volume at a certain task point has dropped below the cleaning standard, the system automatically skips this task point and schedules the robot to the next task point with high demand, thus saving resources and improving the cleaning efficiency. At the same time, the local optimization algorithm, based on the current battery power and load conditions of each robot, preferentially selects a path with lower energy consumption, prolongs the operation time of the robot, and avoids task interruption caused by battery depletion. The obstacle avoidance algorithm is introduced during the path fine-tuning process. When the sensor detects dynamic obstacles (such as pedestrians and vehicles) ahead, a safe path is automatically generated to ensure that the robot can maintain the cleaning task progress without collisions or safety accidents. After generating the locally optimized control instructions, the central scheduling unit sends these updated instructions to each sanitation robot via wireless network, and the on-board controller of the robot executes the closed-loop control to achieve real-time collaborative operation control. In the closed-loop control mode, the robot continuously monitors its own state and environmental data during the execution process. When it detects a deviation between the instruction and the actual execution situation, the controller automatically makes fine adjustments, such as adjusting the speed, changing the cleaning mode, or re-planning a short path, to achieve refined execution of the instruction. In special cases, such as equipment failure, too low battery power, or impassable obstacles, the on-board controller automatically triggers an emergency handling mechanism, including staying in place, returning to the charging station, or sending a distress signal to the central scheduling unit. Through closed-loop control, an adaptive adjustment mechanism during the task execution process is established, enabling each robot to not only execute tasks according to the plan but also flexibly respond to changes according to the actual situation, ensuring that the entire cleaning system still maintains a highly efficient and stable operating state in a dynamic environment. Through this master-slave distributed architecture and real-time collaborative operation control strategy between the central scheduling unit and the sanitation robots, the efficient scheduling, dynamic path adjustment, and autonomous decision-making ability during the operation of the sanitation robots are realized.

[0020] In the embodiments of the present invention, by analyzing the garbage accumulation information and pedestrian flow density data of each urban area through a spatio-temporal convolutional network model, accurate prediction of sanitation needs is achieved, providing a scientific basis for robot scheduling, and effectively solving the problem of inaccurate prediction in traditional scheduling methods. Combining the sanitation needs prediction results, an improved spectral clustering algorithm is used to evenly partition urban areas, and through a comprehensive evaluation function, the balance, connectivity, simplicity of boundaries, and suitability of areas of the area division are ensured, realizing the rational allocation of sanitation resources. Based on real-time garbage accumulation event data, through weighted distance calculation and priority sorting mechanism, the robot grouping is dynamically adjusted to ensure that events with high urgency and large garbage accumulation are processed first, while ensuring that the original area sanitation operations are not overly affected. By using the robust Lyapunov model predictive control method, considering multiple factors such as path length, task time window, battery energy consumption, and cleaning efficiency, an optimal path sequence and cleaning operation instructions are generated, realizing the intelligent planning of the robot path. Through the master-slave distributed architecture and the sliding time domain control strategy, real-time status monitoring and instruction optimization of sanitation robots are achieved, and a complete closed-loop control mechanism is established to ensure the collaborative operation efficiency and stability of the multi-robot system.

[0021] In a specific embodiment, the process of executing step S1 may specifically include the following steps: Collect garbage accumulation information and pedestrian flow density data through a sensor network deployed in multiple urban sanitation areas; Perform data correction and time series interpolation on the garbage accumulation information and pedestrian flow density data to obtain a standard sanitation data sequence; Based on the standard sanitation data sequence, calculate the garbage generation rate, the correlation between pedestrian flow density and garbage volume, and the regional garbage characteristics of each urban sanitation area to obtain multi-dimensional characteristic sanitation data; Perform interval conversion processing on the multi-dimensional characteristic sanitation data to obtain a standardized sanitation task distribution data set including time stamps, geographical location coordinates, standardized garbage volume, pedestrian flow density, and garbage growth rate; Input the standardized sanitation task distribution data set into the spatio-temporal convolutional network model for task distribution prediction to obtain the sanitation needs prediction result.

[0022] Specifically, by deploying various types of sensor networks in different urban areas, including image acquisition units, weight detection units, and infrared detection units, etc., the accumulation of urban garbage and the change of pedestrian flow density are monitored in real time. For example, in the main intersections and crowded square areas, high-resolution cameras are installed to capture image data of garbage accumulation. The image data is preprocessed by an edge computing unit to extract features such as the area and height of garbage accumulation; weight sensors are embedded in the ground of garbage collection stations, and by monitoring the load change of garbage collection devices, the garbage growth rate per unit time is calculated; at the same time, infrared sensors are used to analyze the pedestrian flow density and flow direction, and these data are uploaded to the central data processing center in real time through a low-latency wireless network to form an initial multi-source data set. Data correction and time series interpolation operations are performed on the original data to generate a standardized sanitation data sequence. During the data correction process, noise data is removed through outlier detection methods. For example, using an improved Z-Score algorithm, outliers are identified by calculating the standard deviation of data points deviating from the mean, and data exceeding the set threshold is removed. For missing values in sensor data, time series interpolation methods are used to complete data filling. The interpolation methods include linear interpolation, spline interpolation, or Kalman filtering methods, so that the data remains smooth and continuous on the time axis. After completing the data correction, data from different sources are uniformly aligned to the same time granularity. For example, the sampling frequencies of image data, weight data, and infrared data are aligned to the standard of one data point every 5 minutes. The time alignment operation ensures the consistency and synchronization of subsequent model training data. Based on the standardized sanitation data sequence, by calculating the garbage generation rate of each urban sanitation area, the correlation between pedestrian flow density and garbage volume, and the characteristics of regional garbage, multi-dimensional feature sanitation data is generated. In this process, through the formula:

[0023] Calculate the garbage generation rate, where represents the location at time the garbage generation rate, is the change in garbage weight per unit time, is the time interval. Through the garbage generation rate, the peak areas of garbage growth are identified, and these areas are given priority when scheduling sanitation resources. At the same time, calculate the correlation between pedestrian flow density and garbage volume, using the Pearson correlation coefficient calculation formula:

[0024] where, is the location pedestrian flow density and the garbage volume the correlation coefficient, and They are the means of population density and garbage volume respectively. If the correlation coefficient is close to 1, it indicates that the change in population density has a significant impact on the garbage volume, and the population flow changes in these areas need to be focused on when predicting the garbage accumulation trend. Perform interval conversion processing on multi-dimensional feature data to generate a standardized dataset of sanitation tasks distribution. The standardization process includes normalizing all feature data to the interval [0, 1], for example, through the min-max normalization formula:

[0025] where, is the standardized feature value, is the original feature value, and are the minimum and maximum values of this feature in the dataset respectively. The standardized dataset contains key features such as timestamps, geographical location coordinates, standardized garbage volume, population density, and garbage growth rate. When inputting the standardized dataset of sanitation tasks distribution into the spatio-temporal convolutional network model for task distribution prediction, the data is converted into the form of spatio-temporal tensors, and the tensor is constructed:

[0026] where, represents the historical time steps, and represent the number of rows and columns of the urban grid, represents the feature dimension. The spatio-temporal convolutional network model includes a time convolutional layer, a spatial convolutional layer, an attention mechanism layer, and a fully connected output layer. The model extracts time features through one-dimensional convolutional operations, and the size of the time convolutional kernel is 24 to capture daily cycle features, such as the law of garbage volume growth in densely populated areas during morning and evening rush hours. After obtaining the time pattern features, the model extracts spatial correlation features through two-dimensional convolutional operations, and the size of the spatial convolutional kernel is 3×3. The spatial propagation characteristics of garbage volume between adjacent areas in the urban grid are analyzed through the sliding window method. For example, when the garbage volume in a certain area increases significantly, its surrounding areas are also affected. In order to enhance the model's focusing ability on key features, an attention mechanism is introduced after the convolutional operation. The importance weights of each feature are calculated through self-attention analysis, and dynamic mutual correlations are established between different areas through cross-attention analysis. The model performs non-linear mapping and regression operations through the fully connected output layer, and outputs the predicted value of the sanitation demand for each grid cell in the future time window. This prediction result contains the specific predicted value of the garbage volume, and the confidence interval of the prediction result is output through the uncertainty estimation method. For example, the standard deviation of the model prediction value is calculated through the Monte Carlo dropout method, providing a reference basis for the reliability of the prediction result for the scheduling system.

[0027] In a specific embodiment, the process of inputting the standardized sanitation task distribution dataset into the spatio-temporal convolutional network model for task distribution prediction to obtain the sanitation demand prediction result may specifically include the following steps: Construct a spatio-temporal data tensor based on the standardized sanitation task distribution dataset; Input the spatio-temporal data tensor into the spatio-temporal convolutional network model to perform a one-dimensional convolutional operation to obtain a feature map representing the time pattern; Input the feature map representing the time pattern into the spatial convolutional layer of the spatio-temporal convolutional network model to perform a two-dimensional convolutional operation to obtain a spatial correlation feature representation; Perform self-attention analysis and cross-attention analysis on the spatial correlation feature representation to obtain an attention-enhanced feature representation; Input the attention-enhanced feature representation into the fully connected output layer of the spatio-temporal convolutional network model for non-linear mapping and regression operations to obtain the predicted value of the sanitation demand; Generate a sanitation demand prediction result based on the predicted value of the sanitation demand. The sanitation demand prediction result includes the predicted garbage volume value and the uncertainty estimation value.

[0028] Specifically, convert the standardized sanitation task distribution dataset into a spatio-temporal data tensor. This dataset contains multi-dimensional feature information such as timestamps, geographical location coordinates, standardized garbage volume, population flow density, and garbage growth rate. When constructing the spatio-temporal data tensor, divide the urban area into grid cells, each grid cell represents a specific geographical area, and the feature data within each grid cell is organized in a time series form. The constructed spatio-temporal data tensor is represented as:

[0029] where, represents the historical time step, and are the number of rows and columns of the urban area grid respectively, represents the feature dimension of each grid cell. represents the time at the grid cell with coordinates , and the standardized value of the th feature. Input the spatio-temporal data tensor into the spatio-temporal convolutional network model, and perform a one-dimensional convolutional operation through the time convolutional layer to extract the time pattern features in the data. During the time convolution process, the convolution kernel size is set to 24 to adapt to the daily cycle pattern. For example, by analyzing the periodic law of garbage generation and population flow density changes within a day, automatically learn feature patterns such as rapid garbage volume growth in the early morning and evening from the data. The core of the time convolution operation lies in the formula:

[0030] Among them, is the feature representation after temporal convolution, is the weight of the temporal convolution kernel, is the size of the convolution kernel (i.e., the length of the time window), represents the eigenvalue at time step in the tensor. Through one-dimensional convolution operation, the model extracts the changing trends of each grid cell in the time dimension, including information such as the periodic changes in the garbage generation rate and the peak hours of the pedestrian flow density, so as to map the time characteristics into a set of high-dimensional feature representations. The feature maps representing the time patterns are input into the spatial convolution layer of the spatio-temporal convolution network model, and the spatial correlation features are extracted through two-dimensional convolution operation. During the spatial convolution process, the model uses a two-dimensional convolution kernel of size 3×3. The convolution operation slides the convolution kernel on the urban area grid to analyze the spatial relationship between each grid cell and its surrounding neighborhoods. The mathematical expression of spatial convolution is:

[0031] Among them, is the feature representation after spatial convolution, is the weight of the two-dimensional convolution kernel, represents the eigenvalue of the adjacent grid cells covered by the convolution window. This spatial convolution operation can effectively capture the mutual influence between different regions in the city. For example, when the garbage volume in a certain region increases significantly, the potential changing trend of the garbage volume in its surrounding regions can also be reflected through convolution calculation, thereby providing the model with richer spatial correlation characteristics. After obtaining the spatial correlation feature representation, the spatio-temporal convolution network model introduces an attention mechanism, including self-attention analysis and cross-attention analysis, to enhance the effectiveness of the feature representation. In self-attention analysis, the model focuses on key features by calculating the importance weights of each feature within a specific region, and the calculation is specifically carried out through the following formula:

[0032] Among them, represents the output result of self-attention, , , are the query, key, and value matrices of the feature representation respectively, is the scaling factor, which is used to stabilize the training process. During this process, the model can automatically learn which features are more critical in the current prediction task. For example, in areas where the amount of garbage is growing rapidly, more emphasis is placed on the garbage generation rate feature, while in areas with dense pedestrian flow, more attention is paid to the pedestrian density feature. At the same time, cross-attention analysis is used to establish the dynamic relationship between different regions. Especially when predicting the garbage distribution, the model can identify the implicit connections between different grid cells caused by the movement of pedestrians or the diffusion of garbage. After completing the processing of the attention mechanism, the model passes the attention-enhanced feature representation into the fully connected output layer. Through non-linear mapping and regression operations, the predicted value of the sanitation demand is finally obtained. This output process includes the predicted garbage amount value for each grid cell and also provides the confidence interval of the prediction result based on the uncertainty estimation method. For example, the standard deviation of the predicted value is calculated through Bayesian inference or Monte Carlo dropout method to obtain the uncertainty estimation value of the prediction.

[0033] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Perform spectral clustering analysis by combining the urban road network topology structure and the sanitation demand prediction result to obtain multiple groups of candidate urban area division schemes; Based on multiple groups of candidate urban area division schemes, calculate the variance of the predicted sanitation task volume within each candidate sub-region to obtain the first index characterizing the balance of sanitation demand within the region; For each candidate sub-region in multiple groups of candidate urban area division schemes, calculate the statistical characteristics of the shortest path distance between any two points within the region, the ratio of the boundary length to the region area, and the matching degree of the region area to the robot operation ability to obtain the second index characterizing the regional connectivity, the third index characterizing the simplicity of the regional boundary, and the fourth index characterizing the suitability of the regional area; Perform weighted fusion calculation on the first index, the second index, the third index, and the fourth index to obtain the urban area division scheme; According to the sanitation demand prediction result of each sub-region in the urban area division scheme, calculate the basic sanitation demand of each sub-region and determine the number of robots allocated to each sub-region; For the N sanitation robots allocated in each sub-region, determine the N initial deployment position coordinates to obtain the initial grouping configuration information of the robots.

[0034] Specifically, by combining the topology structure of the urban road network and the sanitation demand prediction result, multiple groups of candidate urban area division schemes are generated through the spectral clustering analysis method. The spectral clustering algorithm is a graph theory-based method that represents the urban area as a weighted undirected graph , where represents all grid nodes in the city, represents the connection between nodes (such as road passages), Represents the weight matrix of the edges, and the weight values are jointly determined by the spatial distance between nodes and the similarity of sanitation requirements. For example, for two adjacent urban areas and , its edge weight is calculated by the Gaussian kernel function:

[0035] Among them, represents the geographical distance between area and area , is the scale parameter that controls the influence of distance, is the similarity of sanitation requirements between regions, which is calculated by predicting the correlation between the amount of garbage and the density of pedestrian flow. When two regions are both geographically close and have similar sanitation requirements, their weight values are higher, meaning they are more suitable to be divided into the same partition. After obtaining the weight matrix, the spectral clustering algorithm performs eigenvalue decomposition through the Laplacian matrix (where is the degree matrix, and the diagonal elements are the connection degrees of the nodes) to obtain multiple candidate partitioning schemes for urban areas through clustering of eigenvectors. After obtaining multiple urban area partitioning schemes, evaluate the characteristics of each candidate sub-region, and select the optimal urban area partitioning scheme by calculating various evaluation indicators. Calculate the variance of the predicted sanitation task volume within each candidate sub-region to obtain the first indicator representing the balance of sanitation requirements within the region. The calculation formula of this indicator is:

[0036] Among them, represents the balance of sanitation requirements within area , is the predicted sanitation task volume at the coordinate within the region, is the mean value of the task volume within the region, is the number of grids within the region. By calculating the variance, identify those regions with a relatively uniform distribution of task volumes to avoid problems of uneven resource allocation caused by excessive differences in task volumes within the region. To comprehensively evaluate the rationality of the regional division, calculate the statistical characteristics of the shortest path distances between any two points within the region to obtain the second indicator representing regional connectivity. This calculation process is based on the shortest path algorithm of the urban road network (such as the Dijkstra algorithm or the Floyd-Warshall algorithm). By statistically analyzing the shortest path distances between all grid nodes, calculate their mean and variance to evaluate the traffic convenience within the region. The calculation formula of the connectivity indicator is:

[0037] Among them, is the connectivity index of the area , is the shortest path distance from node to node . The smaller this index is, the smoother the passage between task points in the area is, which helps improve the efficiency of the sanitation robot when performing tasks. At the same time, calculate the ratio of the boundary length to the area of each candidate sub-area to obtain the third index characterizing the simplicity of the area boundary. The boundary simplicity index is calculated by the formula:

[0038] where represents the boundary simplicity of the area , is the boundary length of the area, is the area of the area. When the ratio of the boundary length to the area is small, the area shape is more regular, and the robot path planning is simpler, which can reduce the complexity of repeated cleaning and path planning. At the same time, evaluate the matching degree between the area and the operation ability of the robot to obtain the fourth index characterizing the suitability of the area area, through the formula:

[0039] where represents the area suitability of the area , is the number of sanitation robots assigned to this area, is the standard cleaning area ability of a single robot (for example, the number of square meters that can be cleaned per hour). This index is used to ensure that the number of robots assigned to each area matches the actual cleaning task volume of the area, avoiding waste of resources caused by too many robots or untimely cleaning caused by too few robots. After calculating the above four evaluation indexes, through the weighted fusion method, these indexes are comprehensively calculated to determine the optimal urban area division plan. The comprehensive evaluation function is:

[0040] where is the comprehensive evaluation score, are the weight coefficients of each evaluation index. These weight coefficients are optimized through historical data analysis or expert experience to ensure that the comprehensive evaluation method has good adaptability in different scenarios. After calculating the evaluation score, select the area division plan with the highest score as the final implementation plan. According to the urban area division plan, calculate the basic sanitation demand of each sub-area and determine the number of sanitation robots to be assigned. The basic sanitation demand of the area is calculated by the spatial integral of the predicted task volume in the area, and the formula is:

[0041] Among them, represents the basic sanitation demand of the area , is the coordinate of the predicted task volume. After determining the basic demand, according to the total number of sanitation robots for robot allocation. The number of robots in each area is calculated by the formula:

[0042] Among them, is the number of robots allocated to area , is the total number of areas, represents the ceiling operation to ensure that the allocated quantity is an integer and the total does not exceed the available number of robots. For each sanitation robot allocated within each sub-area, the K-means++ algorithm is used to determine the initial deployment positions of these robots. During the initial position selection process, the task volume balance of the service area of each robot is maximized to achieve the optimal initial grouping configuration.

[0043] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Receive real-time garbage accumulation event data, and calculate the required resource quantity for each event in the real-time garbage accumulation event data to obtain an event resource demand quantity set; Based on the event resource demand quantity set, perform event priority queue sorting to obtain an ordered event priority queue; For each event in the ordered event priority queue, calculate the weighted distance from all sanitation robots to the event location to obtain a robot weighted distance matrix; According to the robot weighted distance matrix, select the first k sanitation robots with the smallest weighted distance for each event to obtain a set of temporary response groups; For the set of temporary response groups, recalculate the operation area coverage range of the remaining sanitation robots in each original group to obtain a dynamic scheduling grouping of sanitation robots.

[0044] Specifically, through the sensor network and monitoring system deployed in each area of the city, real-time garbage accumulation event data is received. This data includes the geographical location coordinates of the event , the current garbage accumulation volume and the urgency of the event. . To effectively evaluate the resource quantity required for each event, a resource demand mapping function is introduced. By calculating the comprehensive impact of the garbage volume and urgency, the event resource demand quantity is obtained. Its calculation formula is:

[0045] Among them, represents the resource demand of the event , is the garbage accumulation amount, is the urgency level of the event, is the weight coefficient of the urgency level. By adjusting , the different impacts of the garbage amount and the urgency level on the resource demand are balanced. For example, when is relatively large, the system will give priority to responding to urgent events, even if the garbage accumulation amount is relatively small. Based on the set of event resource demands, a priority ranking is performed to form an ordered event priority queue. The priority ranking is achieved by calculating the priority score of the event, and its calculation formula is:

[0046] Among them, represents the priority score of the event , is the maximum garbage accumulation amount among all current events. Through normalization processing, the resource demand differences between different events are balanced, so that events with high urgency and large garbage amounts are processed first. After obtaining the ordered event priority queue, the weighted distance from each sanitation robot to the event location is calculated for each event in the queue to obtain the robot weighted distance matrix. The calculation of the weighted distance not only considers the geographical distance but also needs to introduce the influence of the current state of the robot (such as load, task progress, battery power, etc.). The calculation formula of the weighted distance is:

[0047] Among them, represents the weighted distance from the robot to the event , is the geographical distance, is the current task completion rate of the robot , is the battery consumption coefficient of the robot , and are the weight coefficients of the load and the battery state respectively. Based on the robot weighted distance matrix, the first sanitation robots with the smallest weighted distance are selected for each event to construct a set of temporary response groups. The selection process is achieved by sorting the weighted distances and selecting the first values, The value of

[0048] Among them, For an event the number of robots required, represents the cleaning capacity of a single robot per unit time (such as the amount of garbage that can be cleaned per hour). The dynamic allocation method enables the system to mobilize sufficient robot resources in the face of large-scale garbage accumulation events, while maintaining the simplicity of resource allocation in small-scale events and avoiding resource waste. After the formation of the temporary response group, recalculate the coverage range of the operating areas of the remaining sanitation robots in each original group to adjust the dynamic scheduling groups of sanitation robots. During the process of adjusting the area coverage, analyze the distribution of the remaining robots in the original group, and by calculating the task density and remaining task volume in the current operating areas of these robots, dynamically expand or contract the task coverage range of each robot. For example, for an area with a large task volume, if some robots in this area are transferred to the temporary response group, the system will reallocate the cleaning paths of the remaining robots so that the average coverage distance between task points remains within a reasonable range. To this end, introduce an operating area coverage model and recalculate the objective function of the coverage area of each robot:

[0049] wherein, is the task volume of the robot after reallocation of the coverage area and represents the predicted task volume at the position . Adjust the area covered by each robot through an iterative algorithm so that the task density of all areas tends to be balanced, avoiding the situation where the cleaning tasks in some areas lag due to insufficient robot resources.

[0050] In a specific embodiment, the process of executing step S4 may specifically include the following steps: Determine the set of sanitation task points within the responsible area for each group in the dynamic scheduling groups of sanitation robots to obtain a complete task allocation plan; Based on the complete task allocation plan, construct a robust Lyapunov model predictive control problem, set the state variables to include the position, speed, battery power, and load information of the sanitation robots, and the control inputs to include speed commands and operation commands to obtain a predictive control model; Construct a comprehensive objective function for the predictive control model, introduce robot dynamics constraints, battery capacity constraints, path increment constraints, task time window constraints, and obstacle avoidance constraints for the predictive control model, and construct a contraction constraint based on the Lyapunov function to obtain a set of constraint conditions; Input the predictive control model, the comprehensive objective function, and the set of constraint conditions into a sequential quadratic programming solver to solve the optimal control sequence for each sanitation robot; ​Generate a detailed sequence of path points and the corresponding sweeping operation sequence for each sanitation robot according to the optimal control sequence, and output the path planning and sweeping operation instructions for each sanitation robot.

[0051] Specifically, according to each group in the dynamic scheduling grouping of sanitation robots, determine the set of sanitation task points within its responsible area to obtain a task allocation plan. The set of task points includes the geographical location coordinates of each task point , and also includes the predicted garbage volume as well as the optimal sweeping time window . These task points are obtained by analyzing the garbage accumulation prediction data in different regions of the city, the real-time sensor feedback information, and the task priorities of emergencies. After completing the task allocation plan, construct a robust Lyapunov model predictive control problem based on the set of task points, and realize the dynamic path planning and operation instruction generation of the sanitation robot by setting state variables and control inputs. In the predictive control model, the state variables mainly include the position of the sanitation robot , speed , battery power and payload information . The control input includes the speed instruction and the sweeping operation instruction . Through the dynamic changes of these variables, the system simulates the operation performance of the robot in different environments during the predictive control process. In order to achieve the optimal path planning in actual operation for the predictive control model, construct a comprehensive objective function, which comprehensively considers four optimization objectives: path length, task completion time window matching degree, battery energy consumption balance, and maximization of sweeping efficiency. The comprehensive objective function is expressed as:

[0052] where, is the comprehensive objective function, are the weight coefficients of different objectives, represents the minimization of the path length, represents the task completion time window matching degree, is the battery energy consumption balance, is the maximization of sweeping efficiency. In the part of minimizing the path length, calculate the distance from the robot's current task point to the next task point:

[0053] where, is the distance from task point to task point , is the total number of task points. The time window matching degree target evaluates whether the time when the robot completes the task is within the optimal service time window of the task, and realizes the optimal path planning under time constraints. The formula is:

[0054] In this formula, tasks that deviate from the time window will increase the penalty value of the objective function. During the optimization process, the system will automatically select those task points that can not only keep the path shortest but also be completed within the specified time as the priority execution targets. To achieve the balance of battery energy consumption, the objective function introduces a battery consumption model, calculates the battery power change of the robot under different path selections, balances the battery usage rates of all robots, and avoids the situation that some robots stop working halfway due to battery exhaustion. The part of maximizing the cleaning efficiency analyzes the relationship between the garbage accumulation at the task point and the robot payload to maximize the garbage cleaning efficiency. For example, at task points with a large amount of garbage, robots with a large payload are preferentially assigned to improve the operation efficiency. While constructing the comprehensive objective function, a series of constraint conditions are introduced for the predictive control model to ensure the executability of the planned path and operation instructions in the actual environment. These constraint conditions include robot dynamics constraints, battery capacity constraints, path increment constraints, task time window constraints, and obstacle avoidance constraints. For example, in the dynamics constraints, it is ensured that the robot's moving speed does not exceed the maximum allowable speed , and the acceleration change is also within the safe range, which is achieved through the following formula:

[0055] At the same time, in the battery capacity constraint, the battery consumption in each step is calculated through the formula:

[0056] to ensure that the battery power ​​Always remain above the safety threshold. The path increment constraint avoids the safety risks brought by sharp turns and sudden stops by restricting the moving distance of the robot within each control cycle. In the task time window constraint, a contraction constraint is constructed through the Lyapunov function to ensure that the path planning gradually converges to the optimal solution at each prediction time step. The predictive control model, the comprehensive objective function, and the set of constraint conditions are input into the sequential quadratic programming solver, and the optimal control sequence is solved separately for each sanitation robot. Sequential quadratic programming is an iterative optimization algorithm that solves the nonlinear optimization problem by decomposing it into multiple quadratic programming subproblems, thereby gradually approaching the global optimal solution in each iteration. Through this process, the system calculates the optimal control input sequence of the robot in the future time domain under the condition of satisfying all constraint conditions, including specific path points and cleaning operation instructions. Convert the optimal control sequence into a detailed path point sequence and the corresponding cleaning operation sequence. The path point sequence specifies the movement trajectory of the robot from the current position to the target task point, while the cleaning operation sequence defines the cleaning operations that the robot needs to perform at each path point, including instructions such as starting garbage collection, obstacle avoidance operations, and returning to the standby position after completing the task. Output these path planning and operation instructions to the control module of each sanitation robot, so that the robot can reach the target task position within the specified time window according to the planned optimal path and perform the corresponding cleaning tasks, ensuring the efficiency and accuracy of urban cleaning operations.

[0057] In a specific embodiment, the method for scheduling sanitation robots based on the distribution of sanitation tasks further includes the following steps: Adopt a master-slave distributed architecture with the central scheduling unit as the master node and each sanitation robot as the slave node, and distribute the path planning and cleaning operation instructions to each sanitation robot to obtain an initial execution plan; Set the control cycle so that each sanitation robot collects its own state information within each cycle to obtain the sanitation robot state and environment data set; Upload the sanitation robot state and environment data set to the central scheduling unit, and combine the real-time updated data in the sanitation demand prediction result to construct the current global state representation of the system; Based on the current global state representation of the system, detect whether it is necessary to trigger dynamic grouping adjustment through the global coordination layer. If adjustment is required, re-execute the dynamic scheduling grouping. If not, perform path fine-tuning on each sanitation robot to obtain locally optimized control instructions; Send the locally optimized control instructions to each sanitation robot, and the on-board controller of the sanitation robot executes closed-loop control to obtain a real-time collaborative operation control strategy.

[0058] Specifically, the central scheduling unit acts as the master node, responsible for the overall task planning, scheduling, optimization decision-making, and instruction issuance. Each sanitation robot acts as a slave node, specifically executing the cleaning tasks and feeding back its own status and environmental data. The design of the master-slave architecture enables the system to achieve both global unified coordination and fine-grained real-time control at the level of each robot, thus maintaining efficient and stable cleaning capabilities in the complex and ever-changing urban environment. When the system starts, the central scheduling unit generates initial path planning and cleaning operation instructions for each sanitation robot according to the task allocation and path planning algorithms, and distributes these instructions to each robot through the wireless network to obtain the initial execution plan. The initial path planning is usually based on the spatio-temporal task prediction results. By constructing a path planning model, the moving trajectory and specific cleaning operations of each robot on the shortest path or the lowest energy consumption path are calculated. For example, for each robot , the path planning model generates a sequence of path points and the corresponding sequence of operation instructions , where represents the coordinates of the th path point, and represents the operation instruction at this path point, such as "start cleaning", "avoid obstacles", or "return to the standby area". These initial instructions ensure that each robot can execute the cleaning tasks according to the established route and operation mode during the first control cycle after the system starts, covering all high-demand sanitation areas in the city. To achieve real-time control and dynamic scheduling of the system, a fixed control cycle is set, for example, every 5 minutes as a control cycle. During each cycle, each sanitation robot automatically collects its own status information and surrounding environmental data to form a sanitation robot status and environmental data set. The robot status information includes the current position , the current speed , the battery power , and the payload information . The environmental data is collected in real time by the sensors equipped on the robot, including the garbage accumulation amount , the positions of surrounding dynamic obstacles, and the road traffic conditions, etc. These data are preprocessed by the edge computing module inside the robot, for example, eliminating sensor noise through filtering algorithms and filling in missing data using interpolation methods, so as to ensure the accuracy and continuity of the data uploaded to the central scheduling unit. After collecting the status and environmental data, the robot uploads these data to the central scheduling unit through a low-latency wireless network. After receiving the status data of all robots, the central scheduling unit combines the data updated in real time in the sanitation demand prediction model to construct a global state representation of the current system. The global state representation is formalized as a state matrix , including the task requirements of each area, the positions of each robot, the task progress, the battery status, and the environmental changes. The state matrix is calculated by the following formula:

[0059] where is the set of all robot states, is the current environmental state, including the predicted value of garbage distribution and the position distribution of dynamic obstacles. By analyzing the global state matrix, the sanitation requirements and resource allocation of each area in the city are dynamically evaluated for matching. For example, when the garbage volume in a certain area increases significantly beyond expectations and the robot resources in this area are insufficient, the system will give an early warning of the cleaning delay and calculate the resource scheduling plan in advance. Based on the global state representation, the global coordination layer detects whether it is necessary to trigger dynamic grouping adjustment. The global coordination layer introduces a dynamic adjustment algorithm based on resource requirements and task completion. When the system detects that the predicted value of the garbage volume in a certain area exceeds the preset threshold or the task completion of the robots in the original group is lower than the set lower limit , the system will judge that there may be a bottleneck in the current task allocation, and then trigger dynamic grouping adjustment. The dynamic grouping adjustment algorithm calculates the weighted distance of all robots to the new task area:

[0060] where represents the geographical distance from the robot to the event location, and are the weight coefficients of load and battery consumption, and represent the current task load and battery status of the robot respectively. By calculating the weighted distance matrix, the robot with the fastest response speed and the best resource status is selected for the area that needs dynamic adjustment, and the tasks are reallocated to obtain the updated dynamic scheduling grouping. If the system judges that the current task grouping is still reasonable and does not require large-scale grouping adjustment, it enters the path fine-tuning stage, and new control instructions are calculated for each sanitation robot through local optimization methods. During the path fine-tuning process, the rolling horizon predictive control method is adopted, only focusing on the optimal path within the next one or several control cycles, and a local optimization objective function is constructed:

[0061] where is the distance between path points, is the target speed, is the battery power, are the weight coefficients for different optimization objectives, is the time domain length of the predictive control. By calculating this objective function, the robot tries to maintain a balanced state of the shortest path, optimal speed, and lowest energy consumption when performing tasks. After generating the locally optimized control instructions, the central scheduling unit will send these instructions to each sanitation robot through the wireless network, and the robot will execute the closed-loop control through the on-board controller to implement the real-time collaborative operation control strategy. In the closed-loop control mode, each robot will continuously monitor its own state and external environment data during the execution process. When it detects a deviation between the instruction and the actual execution situation, the on-board controller will automatically perform path micro-adjustment and operation adjustment, such as adjusting the speed, changing the traveling direction, dynamically avoiding obstacles, or switching the cleaning mode to adapt to the changes in the on-site environment. This real-time closed-loop control mechanism ensures that the system can quickly respond when encountering emergencies (such as obstacle movement, road blockage, or equipment failure), and maintain the coherence and efficiency of the cleaning task.

[0062] The above described the sanitation robot scheduling method based on the sanitation task distribution in the embodiments of the present invention. Next, the sanitation robot scheduling device based on the sanitation task distribution in the embodiments of the present invention will be described. Please refer to Figure 2 One embodiment of the sanitation robot scheduling device based on the sanitation task distribution in the embodiments of the present invention includes: An acquisition module, configured to acquire the garbage accumulation information and pedestrian flow density data of multiple urban sanitation areas, and perform task distribution prediction through a spatio-temporal convolutional network model to obtain the sanitation demand prediction result; A creation module, configured to create the urban area division plan and the initial robot grouping configuration information of multiple urban sanitation areas according to the sanitation demand prediction result; A calculation module, configured to calculate the dynamic scheduling grouping of the sanitation robots according to the urban area division plan and the initial robot grouping configuration information, in combination with the real-time garbage accumulation event data; An output module, configured to output the path planning and cleaning operation instructions of each sanitation robot based on the dynamic scheduling grouping of the sanitation robots.

[0063] Through the collaborative cooperation of the above-mentioned various components, the spatio-temporal convolutional network model analyzes the garbage accumulation information and pedestrian flow density data in each area of the city, realizes the accurate prediction of the sanitation demand, provides a scientific basis for robot scheduling, and effectively solves the problem of inaccurate prediction in traditional scheduling methods. Combining the sanitation demand prediction results, an improved spectral clustering algorithm is used to evenly partition the urban areas, and the balance, connectivity, simplicity of the boundary, and suitability of the area of the regional division are ensured through a comprehensive evaluation function, realizing the rational allocation of sanitation resources. Based on the real-time garbage accumulation event data, the robot grouping is dynamically adjusted through weighted distance calculation and priority sorting mechanism to ensure that events with high urgency and large garbage accumulation are processed first, while ensuring that the original area sanitation operations are not overly affected. The robust Lyapunov model predictive control method is adopted to comprehensively consider multiple factors such as path length, task time window, battery energy consumption, and cleaning efficiency, generate the optimal path sequence and cleaning operation instructions, and realize the intelligent planning of the robot path. Through the master-slave distributed architecture and the sliding time-domain control strategy, the real-time status monitoring and instruction optimization of the sanitation robots are realized, a complete closed-loop control mechanism is established, and the collaborative operation efficiency and stability of the multi-robot system are ensured.

[0064] Referring to Figure 3 , an embodiment of the present invention also provides a computer device, which can be a server, and its internal structure can be as Figure 3 shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with external terminals through a network connection. The computer program, when executed by the processor, implements the above method.

[0065] Those skilled in the art can understand that Figure 3 the structure shown in

[0066] is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0067] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this 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 for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0068] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of the present invention.

Claims

1. A scheduling method for sanitation robots based on the distribution of sanitation tasks, characterized in that, Including: Collecting garbage accumulation information and pedestrian flow density data in multiple urban sanitation areas, and performing task distribution prediction through a spatio-temporal convolutional network model to obtain a sanitation demand prediction result; Creating an urban area division plan and initial robot grouping configuration information for the multiple urban sanitation areas according to the sanitation demand prediction result; Calculating the dynamic scheduling grouping of sanitation robots according to the urban area division plan and the initial robot grouping configuration information, in combination with real-time garbage accumulation event data; Outputting the path planning and cleaning operation instructions for each sanitation robot based on the dynamic scheduling grouping of the sanitation robots.

2. The method for scheduling a sanitation robot based on the distribution of sanitation tasks according to claim 1, wherein, The collecting garbage accumulation information and pedestrian flow density data in multiple urban sanitation areas, and performing task distribution prediction through a spatio-temporal convolutional network model to obtain a sanitation demand prediction result includes: Collecting garbage accumulation information and pedestrian flow density data through a sensor network deployed in multiple urban sanitation areas; Performing data correction and time series interpolation on the garbage accumulation information and the pedestrian flow density data to obtain a standard sanitation data sequence; Calculating the garbage generation rate, the correlation between pedestrian flow density and garbage volume, and the regional garbage characteristics of each urban sanitation area based on the standard sanitation data sequence to obtain multi-dimensional feature sanitation data; Performing interval conversion processing on the multi-dimensional feature sanitation data to obtain a standardized sanitation task distribution data set including timestamp, geographical location coordinates, standardized garbage volume, pedestrian flow density, and garbage growth rate; Inputting the standardized sanitation task distribution data set into a spatio-temporal convolutional network model for task distribution prediction to obtain a sanitation demand prediction result.

3. The method for scheduling a sanitation robot based on the distribution of sanitation tasks according to claim 2, characterized in that, The inputting the standardized sanitation task distribution data set into a spatio-temporal convolutional network model for task distribution prediction to obtain a sanitation demand prediction result includes: Constructing a spatio-temporal data tensor based on the standardized sanitation task distribution data set; Inputting the spatio-temporal data tensor into the spatio-temporal convolutional network model to perform a one-dimensional convolutional operation to obtain a feature map representing the time pattern; Inputting the feature map representing the time pattern into the spatial convolutional layer of the spatio-temporal convolutional network model to perform a two-dimensional convolutional operation to obtain a spatial correlation feature representation; Performing self-attention analysis and cross-attention analysis on the spatial correlation feature representation to obtain an attention-enhanced feature representation; Passing the attention-enhanced feature representation into the fully connected output layer of the spatio-temporal convolutional network model for non-linear mapping and regression operations to obtain a predicted value of the sanitation demand; Generating a sanitation demand prediction result based on the predicted value of the sanitation demand, where the sanitation demand prediction result includes a predicted garbage volume value and an uncertainty estimation value.

4. The method for scheduling a sanitation robot based on the distribution of sanitation tasks according to claim 1, wherein The creating an urban area division plan and initial robot grouping configuration information for the multiple urban sanitation areas according to the sanitation demand prediction result includes: Performing spectral clustering analysis by combining the urban road network topology structure and the sanitation demand prediction result to obtain multiple groups of candidate urban area division plans; Calculating the variance of the predicted sanitation task volume in each candidate sub-region based on the multiple groups of candidate urban area division plans to obtain a first index representing the balance of sanitation demand within the region; For each candidate sub-region in the multiple groups of candidate urban area division schemes, calculate the statistical characteristics of the shortest path distance between any two points within the region, the ratio of the boundary length to the region area, and the matching degree between the region area and the robot operation ability, and obtain the second index representing the regional connectivity, the third index representing the simplicity of the regional boundary, and the fourth index representing the suitability of the regional area; Perform weighted fusion calculation on the first index, the second index, the third index, and the fourth index to obtain an urban area division scheme; According to the predicted results of the sanitation requirements for each sub-region in the urban area division scheme, calculate the basic sanitation demand for each sub-region and determine the number of robots allocated to each sub-region; For the N sanitation robots allocated in each sub-region, determine the coordinates of N initial deployment positions to obtain the initial grouping configuration information of the robots.

5. The method for scheduling a sanitation robot based on the distribution of sanitation tasks according to claim 1, wherein The calculation of the dynamic scheduling grouping of sanitation robots according to the urban area division scheme and the initial grouping configuration information of the robots, including: Receive real-time garbage accumulation event data, and calculate the amount of resources required for processing each event in the real-time garbage accumulation event data to obtain a set of event resource requirements; Perform event priority queue sorting based on the set of event resource requirements to obtain an ordered event priority queue; For each event in the ordered event priority queue, calculate the weighted distance of all sanitation robots to the event location to obtain a robot weighted distance matrix; According to the robot weighted distance matrix, select the first k sanitation robots with the smallest weighted distance for each event to obtain a set of temporary response groups; For the set of temporary response groups, recalculate the operation area coverage range of the remaining sanitation robots in each original group to obtain the dynamic scheduling grouping of sanitation robots.

6. The method for scheduling a sanitation robot based on the distribution of sanitation tasks according to claim 1, wherein, The output of the path planning and cleaning operation instructions for each sanitation robot based on the dynamic scheduling grouping of sanitation robots, including: Determine the set of sanitation task points within the responsible area for each group in the dynamic scheduling grouping of sanitation robots to obtain a complete task allocation plan; Based on the complete task allocation plan, construct a robust Lyapunov model predictive control problem, set the state variables to include the position, speed, battery power, and load information of the sanitation robot, and the control input to include the speed command and the operation command to obtain a predictive control model; Construct a comprehensive objective function for the predictive control model, introduce robot dynamics constraints, battery capacity constraints, path increment constraints, task time window constraints, and obstacle avoidance constraints for the predictive control model, and construct a contraction constraint based on the Lyapunov function to obtain a set of constraint conditions; Input the predictive control model, the comprehensive objective function, and the set of constraint conditions into a sequential quadratic programming solver to solve the optimal control sequence for each sanitation robot; According to the optimal control sequence, generate a detailed path point sequence and the corresponding cleaning operation sequence for each sanitation robot, and output the path planning and cleaning operation instructions for each sanitation robot.

7. The method for scheduling a sanitation robot based on the distribution of sanitation tasks according to claim 1, wherein The sanitation robot scheduling method based on the sanitation task distribution further includes: Adopt a master-slave distributed architecture with a central scheduling unit as the master node and each sanitation robot as a slave node, and distribute the path planning and cleaning operation instructions to each sanitation robot to obtain an initial execution plan; Set a control period so that each sanitation robot collects its own status information within each period to obtain a sanitation robot status and environment data set; Upload the sanitation robot status and environment data set to the central scheduling unit, and combine the real-time updated data in the sanitation demand prediction result to construct a current system global status representation; Based on the current system global status representation, detect through the global coordination layer whether dynamic grouping adjustment needs to be triggered. If adjustment is required, re-execute dynamic scheduling grouping. If not, perform path fine-tuning on each sanitation robot to obtain a locally optimized control instruction; Send the locally optimized control instruction to each sanitation robot, and the on-board controller of the sanitation robot executes closed-loop control to obtain a real-time collaborative operation control strategy.

8. A sanitation robot scheduling device based on the distribution of sanitation tasks, characterized in that, For executing the sanitation robot scheduling method based on sanitation task distribution according to any one of claims 1-7, the sanitation robot scheduling device based on sanitation task distribution includes: An acquisition module for acquiring garbage accumulation information and pedestrian flow density data of multiple urban sanitation areas, and performing task distribution prediction through a spatio-temporal convolutional network model to obtain a sanitation demand prediction result; A creation module for creating a city area division plan and robot initial grouping configuration information of the multiple urban sanitation areas according to the sanitation demand prediction result; A calculation module for calculating the dynamic scheduling grouping of sanitation robots according to the city area division plan and the robot initial grouping configuration information, in combination with real-time garbage accumulation event data; An output module for outputting the path planning and cleaning operation instructions of each sanitation robot based on the dynamic scheduling grouping of sanitation robots.

9. A computer device, characterized in that, Comprising a memory and a processor, the memory stores a computer program that can run on the processor, and is characterized in that when the processor executes the computer program, it implements the sanitation robot scheduling method based on sanitation task distribution according to any one of claims 1 to 7.

Images