A household garbage transfer route planning method and system based on multi-source data

Abstract

The application discloses a household garbage transfer route planning method and system based on multi-source data, and relates to the technical field of route planning for household garbage transfer. In the method, first, the multi-source basic data is time-aligned and fused, the spatio-temporal correlation features are extracted, and the preset network is used to predict the garbage yield of each region. When the predicted yield abnormally increases, the local surge state is determined, and the spatial position matching degree and the transport capacity load are combined to optimize the generation of the first transfer route under the constraints of the service time window and the empty driving proportion. If the time consumption exceeds the limit, the second route is generated through shift allocation. Finally, the route is smoothed and the blind end is removed in combination with the road network topology and the real-time traffic flow, and the target transfer route is output. The technical scheme provided by the application can improve the adaptability of the household garbage transfer route planning.

Description

Technical Field

[0001] This application belongs to the field of route planning technology for municipal solid waste transfer, specifically a method and system for municipal solid waste transfer route planning based on multi-source data. Background Technology

[0002] With the accelerating pace of urbanization and the continuous improvement of residents' living standards, the amount of urban domestic waste is growing rapidly. Timely collection and transfer of domestic waste is a crucial component of urban environmental sanitation management, directly impacting the overall environmental quality and residents' quality of life. Utilizing big data analytics to integrate multi-source data such as meteorological, activity, and traffic data to uncover the spatiotemporal evolution patterns of waste production provides technical support for improving waste transfer efficiency, reducing operating costs, and developing scientifically sound, data-driven transfer route planning schemes. This has become a significant challenge for urban management departments.

[0003] Currently, the planning of municipal solid waste transfer routes mainly relies on historical statistical data and experience-based judgment. A common practice is to pre-determine the routes and service sequences of transfer vehicles based on the historical average waste production of each region and a fixed transfer schedule. This static planning method based on historical averages can, to some extent, meet the basic needs of daily waste collection and transportation, providing transfer vehicles with relatively stable operating paths, thereby achieving the planning of municipal solid waste transfer routes.

[0004] However, urban waste production exhibits significant spatiotemporal dynamics. During holidays, large-scale events, extreme weather, and other special periods, waste production in certain areas may experience short-term surges. Static planning methods based on historical averages struggle to accurately predict and respond promptly to such localized, sudden spikes in waste production when developing waste transfer routes. Due to a lack of accurate understanding of the spatiotemporal fluctuations in waste production, pre-planned transfer routes often cannot be dynamically adjusted to adapt to changes in actual demand, leading to waste backlogs or idle transfer resources in some areas, thus reducing the adaptability of waste transfer route planning.

[0005] Application content

[0006] To address the above issues, this application provides a method and system for planning municipal solid waste transfer routes based on multi-source data. This method solves the problem that existing static route planning methods based on historical averages are difficult to dynamically adjust to adapt to fluctuations in waste production, thereby improving the adaptability of municipal solid waste transfer route planning.

[0007] To achieve the above objectives, the technical solution adopted in this application is as follows:

[0008] Acquire multi-source basic data of the target area, and perform time step alignment processing on the multi-source basic data to obtain multi-source fused data;

[0009] A feature matrix is ​​generated based on the multi-source fusion data, and spatiotemporal correlation features are determined based on the feature matrix;

[0010] The spatiotemporal correlation features are calculated by using a preset network layer structure to obtain the production fluctuation sensitivity coefficient between adjacent grids, and the predicted waste production of each region is calculated based on the production fluctuation sensitivity coefficient.

[0011] If the ratio of the predicted waste production to the historical average is greater than the preset production threshold, then the spatial grid number of the corresponding area and the start and end times of the surge period are marked to obtain the local surge state;

[0012] Based on the local surge state, calculate the spatial location matching degree between the surge grid and the existing route and the current vehicle capacity load, determine the node service time window based on the spatial location matching degree, and generate the vehicle empty driving distance ratio based on the capacity load.

[0013] The spatial location matching degree and the transportation capacity load are extracted as initial parameters for optimization. Under the constraints of the node service time window and the ratio of vehicle empty driving distance, the first transfer route is calculated.

[0014] If the estimated time for the first transfer route is greater than the preset time threshold, then the first transfer route is allocated shifts to obtain a second transfer route.

[0015] Based on the preset road network topology and real-time traffic flow data of the target area, the second transfer route is smoothed and blind end sections are removed to output the target transfer route.

[0016] By adopting the above technical solution, multi-source fusion data is obtained by acquiring multi-source basic data of the target area and aligning it with time steps. This integrates data information from different sources and at different time granularities, providing a unified data foundation for subsequent analysis. Based on this, a feature matrix is ​​generated from the multi-source fusion data, and spatiotemporal correlation characteristics are determined. The production fluctuation sensitivity coefficient between adjacent grids is calculated using a preset network layer structure, capturing the spatial transmission pattern and temporal evolution trend of waste production. This allows for accurate calculation of predicted waste production in each region based on the production fluctuation sensitivity coefficient. When the ratio of predicted waste production to the historical average exceeds a preset production threshold, a local surge state is obtained by marking the spatial grid number of the corresponding region and the start and end times of the surge period. This enables timely identification of localized and sudden surges in waste production. Based on the local surge state, the spatial location matching degree between the surge grid and existing routes, as well as the current vehicle capacity load, are calculated. The node service time window is determined based on the spatial location matching degree, and the vehicle empty-run distance ratio is generated based on the capacity load. This comprehensively considers the spatial relationship between the surge area and existing routes, as well as the actual carrying capacity of vehicles. Using spatial location matching degree and transport capacity load as initial parameters for optimization, a first transfer route is calculated under the constraints of node service time window and vehicle empty running distance ratio. This optimizes route arrangement while meeting time window requirements and avoiding excessive empty running. When the estimated time of the first transfer route exceeds a preset time threshold, a second transfer route is obtained through shift allocation, avoiding efficiency decline caused by excessively long single transfer times. Finally, combining the preset road network topology of the target area and real-time traffic flow data, the second transfer route is smoothed and blind end sections are removed before outputting the target transfer route, ensuring the feasibility and traffic efficiency of the planned route in the actual road network. Compared with the static planning method based on historical averages in the prior art, this application can dynamically respond to the spatiotemporal fluctuations of waste production, adjust transfer routes in a timely manner to adapt to changes in actual demand, avoid waste accumulation or idle transfer resources in some areas, and thus improve the adaptability of municipal solid waste transfer route planning.

[0017] Secondly, embodiments of this application provide a municipal solid waste transfer route planning system based on multi-source data. The system includes one or more processors and a memory. The memory is coupled to the processors and stores computer program code, which includes computer instructions. The processors invoke the computer instructions to cause the system to perform the method described in the first aspect and any possible implementation thereof.

[0018] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages:

[0019] By acquiring multi-source basic data of the target area and performing time-step alignment processing to obtain multi-source fused data, a feature matrix is ​​generated and spatiotemporal correlation features are determined. This fully integrates multi-dimensional information to accurately capture the inherent correlation patterns of waste production in time and space. By calculating the production fluctuation sensitivity coefficient between adjacent grids through a preset network layer structure, and using this to calculate the predicted waste production of each region, and then comparing historical averages with preset production thresholds, the spatial grid numbers and start and end times of surge periods for corresponding regions are marked to obtain the local surge state. This allows for accurate prediction and timely response to localized and sudden surges in waste production, overcoming the shortcomings of existing static planning methods based on historical averages in capturing spatiotemporal dynamic fluctuations. Furthermore, based on the local surge state calculation... The system calculates spatial location matching degree and transport capacity load, using these as initial parameters for optimization. Under the constraints of node service time window and vehicle empty driving distance ratio, a first transfer route is calculated. This allows for targeted dynamic optimization based on existing routes and actual transport capacity during sudden surges, effectively avoiding localized garbage backlogs or idle transfer resources. Subsequently, a second transfer route is obtained by allocating shifts to the first transfer route whose estimated time exceeds a preset time threshold. Path smoothing and blind spot removal are then performed using a preset road network topology and real-time traffic flow data, ultimately outputting the target transfer route. This ensures the timeliness of garbage collection and effectively solves the problem of existing technologies struggling to dynamically adjust to fluctuations in garbage production, improving the adaptability of household waste transfer route planning. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a method for planning municipal solid waste transfer routes based on multi-source data, as disclosed in an embodiment of this application.

[0021] Figure 2 This is another flowchart illustrating a method for planning municipal solid waste transfer routes based on multi-source data disclosed in an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of a municipal solid waste transfer route planning system based on multi-source data provided in an embodiment of this application.

[0023] In the diagram: 301, Central Processing Unit; 302, Read-Only Memory; 303, Random Access Memory; 304, Bus; 305, Input / Output Interface; 306, Input Section; 307, Output Section; 308, Storage Section; 309, Communication Section; 310, Driver; 311, Removable Media. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solution, the present application will be described in detail below with reference to the embodiments. The description in this section is only exemplary and explanatory, and should not be used to limit the scope of protection of the present application in any way.

[0025] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.

[0026] This application provides a method for planning municipal solid waste transfer routes based on multi-source data, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a method for planning municipal solid waste transfer routes based on multi-source data, provided in an embodiment of this application. The method is applied to a system, which refers to a hardware and software integrated platform capable of executing a municipal solid waste transfer route planning program based on multi-source data. The system can execute a municipal solid waste transfer route planning program based on multi-source data. The method includes steps 101 to 106, as follows:

[0027] Step 101: Obtain multi-source basic data of the target area, perform time step alignment processing on the multi-source basic data, and obtain multi-source fused data.

[0028] In this embodiment, the target area represents the geographical region where waste transfer route planning is required. Multi-source basic data refers to the raw data set from different collection channels and sensors, including meteorological rainfall intensity data, assembly date data, consecutive holiday data, and hourly waste production records. Time step alignment processing refers to unifying data with different time resolutions to the same time interval, such as aggregating minute-level data into hour-level data or decomposing daily-level data into hour-level data. Multi-source fused data is used to represent a unified format dataset formed after spatiotemporal standardization, where data in each dimension maintains consistency in spatial grid and time scale.

[0029] Specifically, rainfall intensity data is first obtained from meteorological monitoring stations. This data is typically stored in a latitude-longitude grid format, requiring a projection transformation to convert the geographic coordinate system to a planar coordinate system. Then, the grid is resampled according to the spatial resolution of the target area, ensuring the grid size is uniformly set to a preset value, such as 500 meters by 500 meters. For event date data, the year, month, day, hour, minute, and second of the event are extracted and mapped to the hour on the standard time axis using timestamp alignment. Data on consecutive holidays needs to be normalized according to the time zone of the target area, converting holiday information into consecutive days based on midnight. Hourly waste production records may contain missing values ​​due to sensor malfunctions or data transmission interruptions; these are filled using linear interpolation of values ​​from previous and subsequent times. Finally, the processed meteorological data, event data, holiday data, and production data are concatenated according to a unified spatial grid number and hourly time step to form a multi-dimensional feature vector for each grid at each time point, resulting in multi-source fused data.

[0030] In one possible implementation, multi-source basic data of the target area is acquired, and time-step alignment processing is performed on the multi-source basic data to obtain multi-source fused data. Specifically, this includes steps 1011-1012, as follows:

[0031] Step 1011: Obtain meteorological rainfall intensity data, meeting date data, number of consecutive holidays data, and hourly garbage production records for the target area.

[0032] In this embodiment, the target area refers to the geographical area requiring waste collection and transfer services, such as an administrative district of a city or a combination of several streets. Meteorological rainfall intensity data represents the measured rainfall within the target area at any given time, typically recorded in millimeters per hour. Event date data refers to the specific dates and event types of large-scale events held within the target area, such as the dates of concerts, sporting events, or festivals. Consecutive holiday days data represents the number of consecutive days off for statutory holidays or weekends, such as 3 days for New Year's Day or 7 days for National Day. Hourly waste production records refer to the hourly waste generation weight or volume data for each waste collection point.

[0033] Specifically, hourly rainfall intensity measurements from various meteorological monitoring stations within the target area are obtained from the meteorological department's server via a meteorological data interface. This data includes the latitude and longitude coordinates of the monitoring stations, timestamps, and corresponding rainfall intensity values. Large-scale gathering schedules for the target area are obtained from event management departments or public information platforms, extracting information such as event dates, event types, expected number of participants, and event location coordinates to generate gathering date data. National statutory holidays and weekend calendars are queried to calculate the number of consecutive rest periods in the target area within the analysis period, including preceding and following rest days, forming consecutive holiday data. Historical waste production records from all waste collection points within the target area are retrieved from the waste collection management system, extracting the waste weight measurement or bin filling rate data for each collection point at each hour, and arranging them chronologically to generate hourly waste production records. The acquired meteorological rainfall intensity data, gathering date data, consecutive holiday data, and hourly waste production records are stored in corresponding data tables, providing raw data sources for subsequent data fusion processing.

[0034] Step 1012: Perform projection transformation and grid resampling on the meteorological rainfall intensity data, perform timestamp alignment on the assembly date data, perform time zone normalization on the number of consecutive holidays data, perform linear interpolation to fill missing values ​​on the hourly garbage production records, and unify the processed data to the preset spatial resolution and preset time step to obtain multi-source fused data.

[0035] In this embodiment, projection transformation refers to the computational operation of converting meteorological rainfall intensity data from the original geographic coordinate system to a unified Cartesian coordinate system. Grid resampling processing represents the process of converting irregularly distributed monitoring point data into regular grid data using a spatial interpolation algorithm. Timestamp alignment processing represents the conversion operation of unifying the time stamps of assembly date data to the same format as timestamps from other data sources. Time zone normalization processing refers to the operation of uniformly converting data on consecutive holidays recorded in different time zones to the time zone of the target area. Missing value linear interpolation imputation processing represents the method of linearly estimating and filling missing data points in hourly waste production records using values ​​from adjacent times. Preset spatial resolution refers to the smallest unit size of the unified grid data in the spatial dimension, such as 500 meters by 500 meters. Preset time step represents the smallest sampling interval of the unified data in the time dimension, such as 1 hour. Multi-source fusion data refers to a comprehensive dataset containing information from all data sources after unified processing.

[0036] Specifically, the meteorological rainfall intensity data is converted from a geographic coordinate system to a planar coordinate system. Within the target area, a regular grid is established with a preset spatial resolution, such as 500 meters. The distance between the center point of each grid and surrounding meteorological monitoring stations is calculated. Inverse distance weighted interpolation is used to map the rainfall intensity values ​​of the monitoring points onto the grid points, completing grid resampling. The activity date string is extracted from the meeting date data, and the date format is uniformly converted to the standard timestamp format of year, month, day, hour, minute, and second, ensuring consistency with the timestamp format of waste production records, completing timestamp alignment. The time zone information used when recording consecutive holiday days is identified, and the hour difference between this time zone and the target area's time zone is calculated. The hour difference is added to or subtracted from the recorded time, converting all data to the target area's time zone, completing time zone normalization. The hourly waste production records are traversed, identifying missing times with empty values. The production values ​​of the preceding and following non-empty times are found, and their arithmetic mean is calculated as the filling value for the missing time, completing linear interpolation filling of missing values. The processed meteorological rainfall intensity data, meeting date data, consecutive holiday data, and hourly garbage production records are linked by a unified grid number and timestamp, so that the data of the same grid at the same time are merged into a single record, generating multi-source fusion data.

[0037] Step 102: Generate a feature matrix based on the multi-source fusion data, and determine the spatiotemporal correlation features based on the feature matrix; calculate the spatiotemporal correlation features through a preset network layer structure to obtain the production fluctuation sensitivity coefficient between adjacent grids, and calculate the predicted waste production of each region based on the production fluctuation sensitivity coefficient.

[0038] In this embodiment, the feature matrix refers to a two-dimensional array structure formed by organizing the multidimensional features of each grid in the multi-source fusion data at each time point according to rows and columns. Rows represent spatiotemporal sample points, and columns represent feature dimensions. Spatiotemporal correlation features are used to represent feature subsets that simultaneously contain temporal evolution patterns and spatial distribution patterns, such as the production sequence of a grid over the past 24 hours and the current production of neighboring grids within a 1-kilometer radius. The preset network layer structure refers to a pre-constructed multi-layer neural network architecture, including an input layer, hidden layers, and an output layer. The production fluctuation sensitivity coefficient represents the degree of mutual influence between changes in waste production between adjacent grids, and reflects spatial correlation by quantifying the adjustment of the predicted values ​​of neighboring grids by the increase or decrease in the production of a certain grid. Predicted waste production refers to the amount of waste generated by each grid in the future time period calculated based on historical patterns and current features.

[0039] Specifically, the rainfall intensity, activity identifiers, holiday days, historical yield, and other dimensions of the multi-source fusion data are arranged in chronological order and spatial location to construct a feature matrix. The number of rows equals the total number of times across all grids, and the number of columns equals the number of feature dimensions. The difference between the initial Gini impurity value and the weighted average value after splitting at each feature column in the decision tree node is calculated. This difference represents the decrease in Gini impurity. All features are sorted from largest to smallest by their decrease to obtain a feature importance sequence. A predetermined number of features, such as 30, are selected from the front of this sequence to form a target feature subset. For this target feature subset, historical yield data from each grid over a predetermined backtracking period (e.g., 72 hours) is extracted as the temporal dimension feature, and current yield data from all adjacent grids within a predetermined buffer distance (e.g., 2 kilometers) are extracted as the spatial dimension feature. These are combined to form the initial spatiotemporal features. The Pearson correlation coefficient between any two feature columns in the initial spatiotemporal features is calculated. Among feature pairs with an absolute correlation coefficient greater than a predetermined threshold (e.g., 0.85), the one with the larger decrease in Gini impurity is retained, and the other is discarded, resulting in the spatiotemporal correlation features. Standardized feature vectors are obtained by subtracting the mean and dividing the standard deviation from the spatiotemporal correlation features. These standardized feature vectors are then input into the hidden layers of a predefined network structure. Nonlinear feature mapping is performed using predefined activation functions such as hyperbolic tangent or modified linear units, outputting the hidden layer state features. The Euclidean or cosine distance between the hidden layer state feature vectors of each grid cell in the target region and its neighboring grid cells is calculated. This normalized distance value becomes the production fluctuation sensitivity coefficient. The hidden layer state features are then input into the output layer for linear regression or fully connected layer calculations to obtain the baseline predicted production for each region. Finally, the baseline predicted production for each grid cell is multiplied by 1 and then weighted and summed with the product of the production fluctuation sensitivity coefficients of its neighboring grid cells and the ratios of the deviations of the baseline predicted production of its neighboring grid cells from the mean, yielding the corrected predicted waste production for each region.

[0040] In one possible implementation, a feature matrix is ​​generated based on the multi-source fusion data, and spatiotemporal correlation features are determined based on the feature matrix, specifically including steps 1021-1023, as follows:

[0041] Step 1021: Extract the data of each dimension from the multi-source fusion data and splice them to generate a feature matrix; calculate the Gini impurity decrease of each dimension feature in the feature matrix when the node splits, and sort the features of each dimension according to the Gini impurity decrease to obtain the feature importance sequence.

[0042] In this embodiment, multi-source fusion data refers to a comprehensive dataset containing information such as meteorological rainfall intensity, meeting dates, number of consecutive holidays, and waste production after unified processing. Each dimension of the data represents a data column in the multi-source fusion data divided according to different measurement indicators; for example, rainfall intensity is one dimension, and meeting dates are another. Concatenation refers to the operation of horizontally combining multiple dimensions of data according to sample order to form a unified data structure. The feature matrix represents the two-dimensional array structure formed after concatenating the data from each dimension, where rows represent different samples and columns represent different dimensional features. Node splitting refers to the operation in a decision tree algorithm of dividing a parent node into left and right child nodes according to a certain feature threshold. The Gini impurity reduction represents the reduction in Gini impurity before and after node splitting using a certain dimensional feature; a larger value indicates a stronger discriminative ability of the feature. Dimensional features refer to the measurement indicators corresponding to each column of data in the feature matrix. The feature importance sequence represents the list of dimensional features arranged in descending order of Gini impurity reduction.

[0043] Specifically, meteorological rainfall intensity data columns, meeting date data columns, consecutive holiday day data columns, and hourly waste production data columns are extracted from multi-source fusion data. The sample rows of each dimension are aligned according to timestamps and grid numbers. The extracted data columns are horizontally concatenated in a fixed order, with the first dimension data as the first column of the feature matrix, the second dimension data as the second column, and so on, forming a feature matrix where the number of samples is multiplied by the number of dimensions. A decision tree is constructed, using waste production as the target variable, and iterating through each dimension feature in the feature matrix. For each dimension feature, all possible splitting thresholds are calculated, and the sample set is divided into two subsets based on whether the feature value is greater than the threshold. The Gini impurity of the parent node before splitting is calculated by subtracting the sum of the squares of the sample proportions of each category from 1. The Gini impurity of the left and right child nodes after splitting is calculated separately, and the weighted sum of the Gini impurities of the two child nodes is obtained by weighting the sum of the sample proportions of the left and right child nodes. Subtracting the weighted Gini impurity after splitting from the Gini impurity before splitting yields the Gini impurity reduction for that dimension at the given splitting threshold. Calculate the Gini impurity reduction for all splitting thresholds for that dimension, and take the maximum value as the Gini impurity reduction for that dimension. Calculate the Gini impurity reduction for each dimension feature in the feature matrix, and sort the dimensions features in descending order of Gini impurity reduction value to generate a feature importance sequence.

[0044] Step 1022: Extract a preset number of dimensional features from the feature importance sequence and combine them to generate a target feature subset; for the target feature subset, extract time dimension features based on a preset backtracking period and extract spatial dimension features based on a preset buffer distance, and combine the time dimension features and spatial dimension features to obtain the initial spatiotemporal features.

[0045] In this embodiment, the feature importance sequence refers to a list of dimensional features arranged in descending order of Gini impurity decrease. The preset quantity indicates the number of dimensional features selected from the feature importance sequence, such as 10 or 15. A dimensional feature refers to a measurement index corresponding to a column in the feature matrix. The target feature subset represents a combination of the top few important dimensional features extracted from the feature importance sequence. The preset backtracking period refers to the length of time to trace back to history on the time axis, such as 24 hours or 72 hours. The time dimension feature represents time-series statistical features extracted based on historical time windows, such as the average value or trend change of the past day. The preset buffer distance represents the spatial radius range extending from the current grid to the surrounding area, such as 1000 meters or 2000 meters. The spatial dimension feature refers to spatial statistical features extracted based on surrounding neighboring grids, such as the neighborhood average or spatial gradient. The initial spatiotemporal feature represents a feature set containing spatiotemporal information formed by combining the time dimension feature and the spatial dimension feature.

[0046] Specifically, a predetermined number of dimensional features are extracted from the feature importance sequence in sorted order, for example, the top 10 dimensional features. The names of these dimensional features and their column indices in the feature matrix are recorded and combined to generate a target feature subset. For each dimensional feature in the target feature subset, a predetermined backtracking period (e.g., 24 hours) is performed along the time axis to extract all historical values ​​of that dimensional feature within the backtracking time window. Statistical measures within the time window are calculated: the sum of historical values ​​divided by the number of values ​​yields the mean feature; the difference between the last and first values ​​yields the change feature; and the average of the differences between adjacent values ​​after sorting the values ​​by time yields the trend feature. These statistical measures are used as the time dimension features. For each dimensional feature in the target feature subset, a predetermined buffer distance (e.g., 1000 meters) is extended outward from the current grid to identify all neighboring grids within the buffer area. Extract the current value of the feature in that dimension from the neighboring grid, calculate the neighborhood statistics, sum the neighborhood values ​​and divide by the number of neighboring grids to obtain the neighborhood mean feature, subtract the neighborhood mean from the current grid value to obtain the spatial contrast feature, and calculate the standard deviation of the difference between the current grid value and the values ​​of each neighboring grid to obtain the spatial dispersion feature. Use these statistics as the spatial dimension features. Concatenate the temporal and spatial dimension features corresponding to the same dimension feature column by column. Repeat the above operation for all dimension features in the target feature subset, summarize all temporal and spatial dimension features, and combine them to obtain the initial spatiotemporal features.

[0047] Step 1023: Calculate the correlation coefficients between the features of each dimension in the initial spatiotemporal features, and remove the dimensional features whose absolute values ​​of the correlation coefficients in the initial spatiotemporal features are greater than the preset correlation coefficient threshold to obtain the spatiotemporal correlation features.

[0048] In this embodiment, the initial spatiotemporal features refer to a comprehensive feature set including time-dimensional and spatial-dimensional features. Each dimensional feature represents each feature column in the initial spatiotemporal features. The correlation coefficient is a numerical indicator used to represent the degree of linear correlation between two dimensional features, ranging from -1 to +1, with a larger absolute value indicating a stronger correlation. The preset correlation coefficient threshold is a critical value for determining whether two features are highly correlated, such as 0.8 or 0.9. Removal refers to the operation of removing dimensional features that meet the conditions from the initial spatiotemporal features. The spatiotemporal correlation features represent the set of low-redundancy features retained after removing highly correlated features.

[0049] Specifically, all dimensional features are extracted from the initial spatiotemporal features, and a feature pair combination list is established. Any two different dimensional features are combined into a feature pair, generating all possible feature pair combinations. For each feature pair, the numerical sequences of the two dimensional features are extracted across all samples. The mean of each feature is calculated, and the sum of the numerical sequences of each feature is divided by the number of samples. The covariance of the two features is calculated. For each sample, the deviation is obtained by subtracting the mean of the first feature from the value of the first feature, and the deviation is obtained by subtracting the mean of the second feature from the value of the second feature. The two deviations are multiplied, and the sum of the products is divided by the number of samples minus 1. The standard deviation of each feature is calculated. For each feature, the mean is subtracted from the value, the squares are summed, and the sum of the squares is divided by the number of samples minus 1, then the square root is taken. The covariance is divided by the product of the standard deviations of the two features to obtain the correlation coefficient of the feature pair. The correlation coefficient is calculated for all feature pairs, and a correlation coefficient matrix is ​​established. The correlation coefficient matrix is ​​traversed, and feature pairs with an absolute correlation coefficient greater than a preset correlation coefficient threshold, such as 0.8, are identified. For the identified highly correlated feature pairs, compare the ranking of the two features in the feature importance sequence, retain the feature with the higher ranking, and mark the feature with the lower ranking as a feature to be removed. Remove all dimensional features marked as to be removed from the initial spatiotemporal features, and retain the unmarked dimensional features to obtain the spatiotemporal correlated features.

[0050] In one possible implementation, the spatiotemporal correlation features are calculated using a preset network layer structure to obtain the production fluctuation sensitivity coefficient between adjacent grids. Based on the production fluctuation sensitivity coefficient, the predicted waste production of each region is calculated. Specifically, this includes steps 1024-1026, as follows:

[0051] Step 1024: Standardize the spatiotemporal correlation features to obtain standardized feature vectors; input the standardized feature vectors into the hidden layers of the preset network layer structure and use the preset activation function to perform feature mapping, and output the hidden layer state features.

[0052] In this embodiment, spatiotemporal correlation features refer to a set of low-correlation features containing time-series features and spatial neighborhood features retained after feature filtering. Standardization calculation refers to the normalization operation of performing a zero-mean, unit-variance transformation on the feature data. Standardized feature vectors refer to vector data where the feature values ​​of each dimension are distributed within a similar numerical range after standardization calculation. A preset network layer structure is used to represent a neural network architecture composed of an input layer, hidden layers, and an output layer, such as a fully connected network containing two hidden layers. A hidden layer is an intermediate layer in the preset network layer structure located between the input and output layers, used to extract the nonlinear expression of features. A preset activation function represents a function in the hidden layer that performs a nonlinear transformation on the linear calculation result, such as the ReLU function or the Sigmoid function. Feature mapping refers to the computational process of converting input features into a high-dimensional nonlinear expression through a preset activation function. Hidden layer state features are used to represent the intermediate feature vector output after the spatiotemporal correlation features are calculated by the hidden layer.

[0053] Specifically, the arithmetic mean and standard deviation of each dimension of the spatiotemporal correlation features are calculated across all samples. For each element in the feature vector of each dimension, the mean of that dimension is subtracted, and then divided by the standard deviation of that dimension, resulting in a standardized value with a mean of 0 and a standard deviation of 1. These standardized values ​​from all dimensions are combined to form a standardized feature vector. This standardized feature vector is then fed into the input layer of a predefined network structure. Matrix multiplication is performed between the input layer and the first hidden layer using the weight matrix, mapping the input features from their original dimensions to the neuron dimensions of the first hidden layer. For each element in the matrix multiplication result vector of the first hidden layer, a predefined activation function, such as ReLU, is applied. If the element value is less than 0, 0 is output; otherwise, the original value is output, completing the nonlinear feature mapping of the first hidden layer. The output of the first hidden layer is then fed into the second hidden layer, where matrix multiplication is performed using the weight matrix, and a predefined activation function is applied again for nonlinear transformation, yielding the output of the second hidden layer. The output vector of the last hidden layer is saved as the hidden layer state feature, containing a high-order semantic expression of the spatiotemporal correlation features after multiple nonlinear transformations.

[0054] Step 1025: Calculate the difference between the hidden layer state characteristics of adjacent grids in the target area to obtain the yield fluctuation sensitivity coefficient.

[0055] In this embodiment, the target area refers to the geographical range where waste production prediction needs to be performed. Adjacent grids represent surrounding grid cells that share a common boundary or vertex with the current grid in space. Hidden layer state features refer to the intermediate feature vectors obtained after processing spatiotemporal correlation features through the hidden layer of a neural network. The difference value represents the numerical gap or distance metric between the hidden layer state features corresponding to two adjacent grids. The production fluctuation sensitivity coefficient is a numerical index reflecting the degree of influence of waste production changes between adjacent grids; a larger value indicates a stronger production correlation between adjacent grids.

[0056] Specifically, all grid cells in the target area are traversed, and for each grid cell, its spatially adjacent surrounding grid cells are identified, including those in the four cardinal directions (north, south, east, and west) and the four diagonal directions. The hidden layer state feature vectors corresponding to the current grid cell and each adjacent grid cell are extracted. The Euclidean distance between the hidden layer state feature vectors of the current grid cell and those of the adjacent grid cells is calculated. The elements at corresponding positions of the two vectors are subtracted to obtain a difference vector. The sum of the squares of all elements in the difference vector is calculated, and the square root is taken to obtain the difference value for that adjacent grid pair. After calculating the difference value for all adjacent grid pairs, the reciprocal of the difference value is taken and normalized to map the value to the range of 0 to 1. The smaller the difference value, the larger the normalized value of the adjacent grid pair, indicating more similar features and a stronger production correlation. The normalized value is used as the production fluctuation sensitivity coefficient for that adjacent grid pair. The above calculation is repeated for all adjacent grid pairs in the target area to generate a complete production fluctuation sensitivity coefficient matrix, where each element corresponds to the production fluctuation sensitivity coefficient value of a pair of adjacent grid cells.

[0057] Step 1026: Input the hidden layer state features into the output layer of the preset network layer structure for regression calculation to obtain the basic predicted output of each region; adjust the basic predicted output according to the output fluctuation sensitivity coefficient to obtain the predicted garbage output of each region.

[0058] In this embodiment, the hidden layer state features refer to the intermediate feature vectors output after the spatiotemporal correlation features are calculated by the hidden layer of the neural network. The output layer represents the last neural network layer in the preset network layer structure, used to convert the hidden layer state features into prediction results. Regression calculation refers to the computational operation of mapping the hidden layer state features to the real-number domain prediction value through linear transformation. The basic predicted output represents the preliminary waste output prediction value obtained directly through the output layer based solely on the hidden layer state features. The output fluctuation sensitivity coefficient represents a numerical index of the degree of influence of waste output changes between adjacent grids. Weighted correction refers to the computational process of adjusting the basic predicted output for neighborhood influence using the output fluctuation sensitivity coefficient. The predicted waste output represents the final waste output prediction result for each grid after weighted correction.

[0059] Specifically, the hidden layer state feature vectors of each grid are input into the output layer of a predefined network structure. Matrix multiplication is performed between the weight matrix of the output layer and the hidden layer state feature vectors, mapping the hidden layer state features from a high-dimensional feature space to one-dimensional real values. The bias parameters of the output layer are added to the matrix multiplication result to obtain the basic predicted output value for that grid. For each grid within the target region, the output fluctuation sensitivity coefficient between that grid and all its neighboring grids is extracted, and the corresponding element value is retrieved from the output fluctuation sensitivity coefficient matrix. The difference between the basic predicted output of all neighboring grids and the basic predicted output of that grid is calculated, and each difference is multiplied by the corresponding output fluctuation sensitivity coefficient to obtain the output influence of each neighboring grid on the current grid. The output influence of all neighboring grids is summed to obtain the total neighborhood influence. The basic predicted output of that grid is added to the total neighborhood influence to complete the weighted correction of the basic predicted output of that grid, resulting in the predicted garbage output of that grid. The above weighted correction operation is repeated for all grids within the target region to generate the predicted garbage output dataset for each region.

[0060] Step 103: If the ratio of the predicted waste production to the historical average is greater than the preset production threshold, mark the spatial grid number of the corresponding area and the start and end times of the surge period to obtain the local surge state.

[0061] In this embodiment, the historical average refers to the average waste production of each grid in the target area within a historical statistical period, such as the past 30 or 90 days. The preset production threshold is used to represent the critical point for determining whether production has abnormally surged; for example, 1.5 indicates that a surge indicator is triggered when the predicted production exceeds the historical average by 50%. The spatial grid number is a unique identifier assigned to each grid after the target area is regularly divided, typically using row and column indexes or hash codes. The start and end times of the surge period represent the start and end times of the continuous time period during which waste production exceeds the threshold, recorded in year-month-day-hour-minute format. The local surge state refers to a set of state descriptions containing information on the spatial location of the surge area and the duration of the surge.

[0062] Specifically, the process iterates through all spatial grids in the target area, extracting the hourly predicted waste production sequence for each grid within a future prediction period (e.g., the next 24 hours). Simultaneously, it queries the waste production records for the corresponding time periods within the historical statistical period for that grid, calculating the arithmetic mean of the historical records to obtain the historical mean for that grid. The ratio of the predicted waste production to the historical mean is calculated hour by hour. If the ratio at any given moment exceeds a preset production threshold (e.g., 1.5), the grid is determined to have entered a surge state. The row and column indices or codes corresponding to this grid are marked as spatial grid numbers. The moment when the ratio condition is first met is recorded as the start time of the surge period. The process continues to check whether the ratio at subsequent moments remains greater than the threshold until the ratio first falls below the threshold or reaches the end of the prediction period. The last moment that meets the condition is recorded as the end time of the surge period. The spatial grid numbers and the start and end times of the surge period are combined to form data records containing location identifiers and time ranges. Corresponding records are generated for all grids in the target area that meet the conditions, and the results are summarized to obtain the local surge state.

[0063] Step 104: Based on the local surge state, calculate the spatial location matching degree between the surge grid and the existing route and the current vehicle capacity load, determine the node service time window based on the spatial location matching degree, and generate the vehicle empty driving distance ratio based on the capacity load.

[0064] In this embodiment, the surge grid refers to a spatial grid marked as indicating an abnormal increase in waste production during a local surge. The existing route refers to the currently executing or planned waste transfer vehicle travel path and its sequence of service nodes. Spatial location matching degree represents the spatial proximity between the surge grid and nodes on the existing route; a smaller value indicates a closer proximity and higher matching degree. Capacity load refers to the ratio between the amount of waste currently carried by the transfer vehicle and the vehicle's maximum loading capacity. The node service time window represents the allowed time range within which a transfer vehicle can arrive at and serve a node, including the earliest arrival time and the latest departure time. The vehicle empty driving distance ratio refers to the proportion of the transfer vehicle's empty driving distance (without waste) to the total driving distance during the entire journey.

[0065] Specifically, the system obtains the maximum vehicle loading capacity of the current transfer shift (e.g., 10 tons), the departure interval (e.g., 30 minutes), and the geographic coordinate sequence of each node in the initial transfer route, as well as the travel time matrix of the road segments between nodes. For each surge grid in a local surge state, the coordinates of the center point corresponding to its spatial grid number are extracted, and the Euclidean distance between this center point and the coordinates of all nodes on the initial transfer route is calculated. The minimum distance value is taken as the spatial location matching degree of the surge grid. The predicted waste output of all nodes in the initial transfer route is accumulated, and the sum is divided by the maximum vehicle loading capacity to obtain the current vehicle's transport capacity load value. Based on the spatial location matching degree, the route node closest to the surge grid is determined, and the cumulative travel time from the vehicle's origin to this node is extracted from the road segment travel time matrix as the target travel time. Extract the start and end times of the surge period. Subtract a preset time margin, such as 15 minutes, before the start time to obtain the earliest service time. Add the preset time margin after the end time to obtain the latest service time. Add the target travel time to the departure time to obtain the estimated arrival time. Fine-tune the earliest and latest service times based on the relationship between the estimated arrival time and the range of the surge period to determine the node's service time window. Multiply the capacity load value by a preset empty-run ratio upper limit, such as 0.2, to obtain the allowable vehicle empty-run distance ratio.

[0066] Step 105: Extract spatial location matching degree and transport capacity load as initial parameters for optimization. Under the constraints of node service time window and vehicle empty driving distance ratio, calculate the first transfer route. If the estimated time of the first transfer route is greater than the preset time threshold, then allocate shifts to the first transfer route to obtain the second transfer route.

[0067] In this embodiment, the initial optimization parameters refer to the initial adjustable variable values ​​set before the route optimization algorithm begins iteration, including spatial location matching degree and transport capacity load. Constraints represent the restrictions that must be met during route planning, including node service time windows and vehicle empty-run distance ratios. The first transfer route represents the transfer path scheme, including node access order and road segment connection relationships, obtained through preliminary optimization calculations under constraints. Estimated time consumption refers to the total time required for a vehicle to complete all services, calculated by accumulating the node order and road segment travel time according to the first transfer route. The preset time consumption threshold represents the maximum allowable operating time for a single transfer shift, such as 4 hours. Shift allocation refers to the scheduling operation of splitting overdue transfer routes into multiple shifts for separate execution. The second transfer route refers to the multi-shift route scheme that meets time constraints after shift allocation.

[0068] Specifically, the spatial location matching degree value is extracted as the initial weight for node insertion position, and the transport capacity load value is extracted as the initial reference value for vehicle capacity constraints. A route optimization objective function is constructed, aiming to minimize the sum of the total travel distance and the time window violation penalty. During iteration, after each adjustment of the node access order, it is checked whether the arrival time of vehicles at each node under the new order falls within the corresponding node service time window. If the time window is violated, a penalty term is added to the objective function. The total travel distance under the new order is calculated, and the sum of the distances of road segments where vehicles are not loaded with garbage is extracted and divided by the total travel distance. If the resulting ratio exceeds the proportion of empty vehicle distances, the adjustment scheme is rejected. The optimal node order is searched in the feasible solution space that satisfies the constraints to obtain the first transfer route. The travel time of all road segments in the first transfer route is added together with the loading and unloading operation time of each node to obtain the estimated time. If the estimated travel time exceeds a preset time threshold (e.g., 4 hours), the travel time for each segment is accumulated in the node sequence of the first transfer route, starting from the origin. When the accumulated time first reaches or exceeds the preset single-shift travel time limit (e.g., 3.5 hours), the current node is marked as a cutoff point. The node sequence is divided into a first-shift node sequence and a second-shift node sequence, using the cutoff point as the boundary. Departure start times are set for the two shifts based on the departure interval. The first shift starts from the initial departure time, and the second shift adds the departure interval to the first shift's departure time, forming a departure timetable for each shift. For each shift's node sequence, starting from the current node, the unvisited node with the shortest travel time is selected as the next node, and this process is repeated until all nodes for that shift have been traversed. The node access order for each shift is then updated, and a second transfer route containing multiple shift departure times and corresponding routes is generated.

[0069] In one possible implementation, if the estimated time for the first transfer route is greater than a preset time threshold, then the first transfer route is allocated shifts to obtain a second transfer route, specifically including steps 1051-1053, as follows:

[0070] Step 1051: When the estimated travel time is greater than the preset travel time threshold, find the target node in the first transfer route that has reached the preset single-shift travel time limit as the cutoff point and split it to obtain the first shift node sequence and the second shift node sequence.

[0071] In this embodiment, the estimated time consumption represents the total time required to execute the first transfer route, calculated by summing the travel time of all segments and the dwell time of nodes in the route. A preset time consumption threshold represents the maximum allowed working time for a single shift, such as 8 or 10 hours. The first transfer route refers to the complete transfer path sequence generated in the initial planning, including the starting point, intermediate nodes, and the destination. The cumulative travel time represents the total accumulated time taken from the route starting point to a certain node. The preset single-shift time limit refers to the maximum working time for a single shift set after considering driver working hours and vehicle continuous operation time limitations, such as 480 minutes. The target node represents the route node where the cumulative travel time first reaches or exceeds the preset single-shift time limit. The cutoff point refers to the boundary node position used to divide the first transfer route into two segments. Splitting represents the operation of dividing the complete route into two independent routes before and after the cutoff point. The first shift node sequence refers to the node sequence before and after the cutoff point, including the first half of the node sequence. The second shift node sequence represents the node sequence after the cutoff point, including the second half of the node sequence.

[0072] Specifically, the process iterates through all segments and nodes of the first transfer route, accumulating the estimated travel time for each segment and the estimated dwell time for each node to calculate the estimated time required to complete the entire route. This estimated time is then compared to a preset time threshold, such as 480 minutes, to determine if it exceeds the threshold. If the estimated time exceeds the threshold, starting from the starting node of the first transfer route, the travel time to each node is accumulated sequentially to calculate the cumulative travel time for each node. The route nodes are then sequentially searched, and nodes whose cumulative travel time first reaches or just exceeds the preset single-shift time limit are marked as target nodes. The target node is set as the cutoff point, and the first transfer route is split along this cutoff point. All nodes from the route start point to the cutoff point are extracted and arranged in the original route order to form the first shift node sequence. All nodes from the cutoff point to the route end point are extracted and arranged in the original route order to form the second shift node sequence, with the starting point of the second shift node sequence being the next node immediately following the cutoff point.

[0073] Step 1052: Extract the node sequence of the first and second trains, and allocate the departure start and end times according to the preset departure interval to obtain the departure timetable for each train.

[0074] In this embodiment, the first shift node sequence refers to the node access sequence for the first half of the route after splitting. The second shift node sequence represents the node access sequence for the second half of the route after splitting. The preset departure interval is used to indicate the minimum interval required between the departure of the first shift vehicle and the departure of the second shift vehicle, such as 30 minutes or 60 minutes. The departure start and end times refer to the start time of each shift vehicle's departure from the transfer station and the end time of its return to the transfer station. The departure timetable for each shift represents a detailed time arrangement table including the departure time, estimated return time, and arrival times of the nodes passed through.

[0075] Specifically, the estimated total operating time for the first shift is calculated by summing the travel time of all road segments and the dwell time of all nodes in the first shift's node sequence. The departure start time for the first shift is set, for example, 6:00 AM on the current day. The estimated return time for the first shift is calculated by adding the departure start time to the estimated total operating time of the first shift. Starting from the beginning of the first shift's node sequence, the travel time and dwell time to each node are accumulated sequentially, and the estimated arrival time of each node is obtained by accumulating these amounts based on the departure start time of the first shift, forming the detailed timetable for the first shift. The estimated total operating time for the second shift is calculated by summing the travel time of all road segments and the dwell time of all nodes in the second shift's node sequence. The estimated total operating time for the second shift is obtained by adding a preset departure interval, such as 60 minutes, to the departure start time of the first shift, for example, 7:00 AM. The estimated return time for the second shift is calculated by adding the departure start time to the estimated total operating time of the second shift. Starting from the beginning of the second trip's node sequence, the travel time and dwell time to each node are accumulated sequentially. These are then added to the estimated arrival time of each node based on the second trip's departure start time, forming the detailed timetable for the second trip. The departure and return times of the first and second trips, along with the arrival times of the nodes along the way, are then combined to generate the departure timetable for each trip.

[0076] Step 1053: Based on the departure timetable of each bus, connect the unvisited nodes with the shortest travel time from the current node in sequence to update the node access order and generate the second transfer route.

[0077] In this embodiment, the departure timetable for each bus trip refers to a schedule that includes the departure time of each trip and the arrival time of the nodes along the route. The current node represents the node location being processed during route optimization. Travel time represents the travel time required to reach a candidate node from the current node. Unvisited nodes refer to the set of nodes in the route that have not yet been assigned a visit order. Node visit order represents the order in which vehicles visit each node. The second transfer route represents the new transfer path generated after optimizing the node visit order based on the departure timetable.

[0078] Specifically, the departure start time and node sequence of the first bus are extracted from the departure timetables of each bus service. The starting point of the first bus node sequence is set as the current node, and all other nodes in the first bus node sequence except the starting point are added to the unvisited node set. Starting from the current node, the travel time from the current node to each node in the unvisited node set is calculated. The distance and speed limit information of each road segment are obtained by querying the road network database, and the travel time is obtained by dividing the distance by the speed limit. All travel time values ​​are compared, and the unvisited node with the shortest travel time is selected as the next visited node. The selected node is removed from the unvisited node set and added to the end of the visited node sequence, updating the node visit order. The selected node is set as the new current node, and the operation of finding the unvisited node with the shortest travel time is repeated until the unvisited node set of the first bus is empty. According to the updated node visit order, all nodes of the first bus are connected sequentially to form the new route segment of the first bus. Perform the same operation on the second shift's node sequence. Starting from the second shift's starting point, sequentially select the unvisited node with the shortest travel time from the current node, update the node access order of the second shift, and form a new route segment for the second shift. Merge the new route segments of the first and second shifts in chronological order to generate a complete second transfer route.

[0079] Step 106: Combining the preset road network topology of the target area with real-time traffic flow data, perform path smoothing and blind end segment removal on the second transfer route, and output the target transfer route.

[0080] In this embodiment, the preset road network topology refers to the pre-stored graph structure data of road connections and intersection node locations in the target area, typically represented as an adjacency matrix or adjacency list. Real-time traffic flow data is used to represent dynamic traffic status information such as vehicle density, average speed, and congestion level of each road segment at the current moment. Path smoothing refers to curve fitting processing of node connections with excessively large turning angles in the transfer route to make the vehicle trajectory smoother. Blind ends represent road sections that are impassable due to traffic control, road construction, or severe congestion. The target transfer route refers to the final output route scheme that meets all actual traffic conditions after path refinement processing.

[0081] Specifically, an adjacency matrix of a preset road network topology is obtained. Elements in the matrix represent the connectivity weights of road segments, with the weight of normally passable road segments determined by their length or travel time. Real-time traffic flow data is acquired, and road segments with average speeds below a preset congestion speed threshold (e.g., 15 km / h) are marked as congested segments. Road segments prohibited from passage in traffic control information are marked as restricted segments. The connectivity weights of congested and restricted segments in the adjacency matrix are updated to preset values, such as positive infinity, indicating impassability, resulting in a real-time available road network matrix. The geographical coordinates of each node in the second transfer route are mapped to the node indices of the real-time available road network matrix. The connectivity weights of adjacent nodes in the matrix are queried according to the node access order. If a weight is positive infinity, the corresponding road segment is determined to be a blind end segment, which is then deleted from the route. An alternative path between the start and end points is found in the available road network using a shortest path algorithm to reconnect them, resulting in a blind-end-free transfer route. Extract the coordinates of three adjacent nodes in the blind-endless transfer route, calculate the angle between the direction vectors of the preceding and following segments, and subtract this angle from 180 degrees to obtain the turning angle. When the turning angle is greater than a preset turning angle threshold, such as 120 degrees, the turning node is used as a control point. Extract the coordinates of the starting control point at a preset cutoff length, such as 20 meters, before the end of the preceding segment, and extract the coordinates of the ending control point at a preset cutoff length after the starting point of the following segment. Use Bézier curve or spline interpolation methods with a preset smoothing parameter, such as a curve order of 3, for discrete sampling to generate several intermediate trajectory points between the starting control point, turning control point, and ending control point, forming a smooth trajectory point sequence. Replace the original right-angle turning nodes in the blind-endless transfer route with the smooth trajectory point sequence to obtain a smooth transfer route. Calculate whether each segment in the smooth transfer route is loaded with garbage, and divide the sum of the lengths of unloaded segments by the total route length to obtain the vehicle empty driving distance ratio. If the ratio is less than the preset empty load threshold, such as 0.15, the route is determined to meet the operational efficiency requirements, and the smooth transfer route is output as the target transfer route.

[0082] In one possible implementation, by combining the preset road network topology of the target area with real-time traffic flow data, the second transfer route is smoothed and blind-end road segments are removed to output the target transfer route. Specifically, this includes steps 1061-1064, as follows:

[0083] Step 1061: Identify congested and restricted road segments in the preset road network topology by combining real-time traffic flow data, and update the connectivity weight of the corresponding road segments in the adjacency matrix to the preset value to obtain the real-time available road network matrix.

[0084] In this embodiment, real-time traffic flow data refers to the current traffic flow, speed, and congestion status information of each road segment obtained from the traffic management system. A preset road network topology represents the connection graph structure of road nodes and segments within the target area, including road intersection nodes and connecting road segments. Congested segments are used to represent segments where the real-time speed is below the normal speed threshold or the traffic flow exceeds the road capacity. Restricted segments refer to road segments temporarily closed due to construction, traffic control, or emergencies. An adjacency matrix represents a two-dimensional matrix of connectivity relationships between nodes in the road network topology, with matrix element values ​​representing the connectivity weight of the corresponding road segment. The connectivity weight represents the drivability or cost of a road segment between two nodes, typically using segment length or travel time as the weight value. Preset values ​​are specific values ​​indicating that a road segment is impassable, such as positive infinity or negative 1. A real-time available road network matrix represents an adjacency matrix reflecting the current actual road traffic status after updating congestion and restricted information.

[0085] Specifically, real-time traffic flow data for all road segments within the target area is obtained from the traffic management system interface, and the current vehicle speed, traffic density, and congestion level information for each segment are extracted. The real-time traffic flow data is traversed to identify road segments with speeds below 50% of the speed limit or those marked as severely congested, and these segments are marked as congested. Road construction announcements, traffic control notices, and emergency warnings for the current moment are obtained from the traffic control information platform, and the road segment numbers that are temporarily closed or prohibited are extracted and marked as prohibited. An adjacency matrix with a preset road network topology is loaded. The rows and columns of the adjacency matrix correspond to the intersection nodes in the road network, and the matrix elements store the passage time between corresponding node pairs as connectivity weights. The list of road segments marked as congested and prohibited is traversed, and the row and column indices of the start and end nodes for each segment in the adjacency matrix are found. The connectivity weight element value at the corresponding row and column position in the adjacency matrix is ​​updated to a preset value, such as positive infinity, indicating that the road segment is currently impassable. After updating the weights of all congested and restricted road sections, a real-time available road network matrix reflecting the current actual road traffic status is generated.

[0086] Step 1062: Map the second transfer route to the real-time available road network matrix, find and remove blind end road segments with a preset connectivity weight, and obtain a transfer route without blind ends.

[0087] In this embodiment, the second transfer route refers to the transfer path generated after optimizing the node access order based on the departure timetable. The real-time available road network matrix represents the adjacency matrix reflecting the current actual road traffic status after updating congestion and traffic restrictions. The connectivity weight is used to represent the drivability or travel cost of road segments between two nodes in the adjacency matrix. The preset value refers to a specific value indicating that a road segment is impassable, typically positive infinity or negative 1. Blind-end road segments represent road segments in the second transfer route that are actually impassable due to congestion or traffic restrictions. A transfer route without blind ends represents a transfer path that can be actually executed after removing impassable road segments.

[0088] Specifically, the sequence of adjacent node pairs in the second transfer route is extracted, and the node number of each pair of adjacent nodes is obtained sequentially according to the route access order. For each pair of adjacent nodes, the corresponding row and column index position is found in the real-time available road network matrix, and the connectivity weight value stored at that position is read. It is determined whether the connectivity weight value is equal to a preset value, such as positive infinity. If it is equal to the preset value, it indicates that the road segment is a blind end segment, and the start and end node information of the road segment is recorded. All adjacent node pairs of the second transfer route are traversed to complete the drivability check of all road segments and generate a blind end segment list. Starting from the starting point of the second transfer route, each road segment in the route is checked sequentially to see if it is in the blind end segment list. When a blind end segment is encountered, the segment is removed from the route, and a drivable alternative path from the start node to the end node of the segment is searched in the real-time available road network matrix. The shortest path algorithm is used to calculate the detour path. If a drivable alternative path exists, the alternative path is inserted into the original route to replace the blind end segment. If no alternative path exists, the end node of the blind end segment and its subsequent nodes are removed from the route as a whole. After completing the removal or replacement of all blind-end road segments, retain the passable road segments and nodes to generate blind-end-free transfer routes.

[0089] Step 1063: Extract the turning angles of adjacent road segments in the blind-end-free transfer route, and perform curve interpolation calculations on the nodes with turning angles greater than the preset turning angle threshold to obtain a smooth transfer route.

[0090] In this embodiment, a blind-end-free transfer route refers to a transfer path that can be actually executed after removing impassable road sections. Adjacent road sections represent two consecutive road sections in a blind-end-free transfer route. Turning angle represents the angle of directional change when a vehicle enters a subsequent road section from the previous one. A preset turning angle threshold is a critical angle value used to determine whether a turn is too sharp, such as 90 degrees or 120 degrees. Node connection represents the connection relationship between adjacent road sections at intersection nodes. Curve interpolation calculation refers to a mathematical calculation method for generating a smooth curve trajectory at sharp turn nodes. A smooth transfer route represents a transfer path that is more suitable for vehicle travel after optimizing the turning trajectory.

[0091] Specifically, all nodes in the blind-endless transfer route are traversed, and information about the preceding and following road segments is extracted for each intermediate node. The inbound direction vector of the preceding segment is calculated by subtracting the starting coordinates from the end coordinates of the preceding segment. The outbound direction vector of the following segment is calculated by subtracting the starting coordinates from the end coordinates of the following segment. The angle between the inbound and outbound direction vectors is calculated by dividing the dot product of the vectors by the product of their magnitudes to obtain the cosine value of the angle, and then performing an inverse cosine operation on the cosine value to obtain the angle. The supplementary angle is calculated as the turning angle, and the turning angle value is obtained by subtracting the angle from 180 degrees. The turning angle is compared with a preset turning angle threshold, such as 90 degrees. When the turning angle is greater than the preset turning angle threshold, the intermediate node is marked as a turning node requiring smoothing. On the inbound segment of the turning node, a preset intercept length, such as 20 meters, is moved backward along the road segment direction from the turning node, and the coordinates at this position are extracted as the starting control point. On the exit section of the turning node, a preset intercept length, such as 20 meters, is moved forward from the turning node towards the leading section. The coordinates at this location are extracted as the endpoint control point. Using the turning node coordinates as intermediate control points, and the starting control point, intermediate control point, and endpoint control point as the three control points of the Bézier curve, equal-interval sampling is performed based on preset smoothing parameters between 0 and 1. The coordinates of the curve points corresponding to each sampling parameter are calculated to generate a smooth trajectory point sequence. In the blind-end-free transfer route, the right-angle connections of the original turning nodes are replaced with the smooth trajectory point sequence, while keeping other road sections unchanged, resulting in a smooth transfer route.

[0092] In one possible implementation, the turning angles of adjacent road segments in the blind-end-free transfer route are extracted, and curve interpolation calculations are performed on the node connections where the turning angle is greater than a preset turning angle threshold to obtain a smooth transfer route. Specifically, this includes steps 10631-10633, as follows:

[0093] Step 10631: Extract the inbound and outbound direction vectors of adjacent node connecting segments in the blind-end transfer route; calculate the supplementary angle of the angle between the inbound and outbound direction vectors to obtain the turning angle.

[0094] In this embodiment, a blind-end-free transfer route refers to a transfer path that can be actually executed after removing impassable road sections. Adjacent node connection segments represent road connection sections between two adjacent nodes in the route. The entering direction vector represents the direction of movement of a vehicle when it travels along the previous road segment to reach the current node. The exit direction vector represents the direction of movement of a vehicle when it leaves the current node along the next road segment. The supplementary angle refers to the difference between the angle between two direction vectors and 180 degrees. The turning angle represents the angle at which a vehicle changes its direction of travel at a node; a larger value indicates a sharper turn.

[0095] Specifically, starting from the beginning of the blind-endless transfer route, each intermediate node in the route is traversed sequentially, skipping the starting and ending nodes. For each intermediate node, the coordinates of its preceding and following nodes are extracted. The inbound segment from the preceding node to the current node is calculated by subtracting the preceding node's x-coordinate from the current node's x-coordinate to obtain the x-coordinate component of the inbound direction vector, and subtracting the preceding node's y-coordinate from the current node's y-coordinate to obtain the y-coordinate component of the inbound direction vector. These components are then combined to obtain the inbound direction vector. The outbound segment from the current node to the next node is calculated by subtracting the current node's x-coordinate from the next node's x-coordinate to obtain the x-coordinate component of the outbound direction vector, and subtracting the current node's y-coordinate from the next node's y-coordinate to obtain the y-coordinate component of the outbound direction vector. These components are then combined to obtain the outbound direction vector. The dot product of the inbound and outbound direction vectors is calculated, and the sum of the products of the x-coordinate components and y-coordinate components of the two vectors is taken as the dot product value. Calculate the magnitudes of the approach and exit direction vectors respectively. Add the squares of the x-coordinate components of the vectors and take the square root. Divide the dot product by the product of the magnitudes of the two vectors to obtain the cosine of the angle between them. Perform an inverse cosine function on the cosine to obtain the angle between the two vectors, in degrees. Subtract the angle from 180 degrees to obtain the supplementary angle, which is the turning angle at that node.

[0096] Step 10632: When the turning angle is greater than the preset turning angle threshold, the corresponding turning node is used as a control point, and the coordinates are extracted on the corresponding entry and exit road segments according to the preset intercept length to obtain the starting control point and the ending control point.

[0097] In this embodiment, the turning angle refers to the angle at which a vehicle changes its direction of travel at a node. A preset turning angle threshold represents the critical angle value for determining whether a turn requires smoothing, such as 90 degrees. A turning node represents a route node whose turning angle exceeds the preset turning angle threshold. A control point is a key location point used to generate a smooth curve, typically the coordinates of the turning node itself. An entry segment represents the section of road before the vehicle reaches the turning node. An exit segment represents the section of road after the vehicle leaves the turning node. A preset intercept length refers to the distance between the entry and exit segments from which control points are extracted, such as 20 meters or 30 meters. A starting control point represents the first control point intercepted on the entry segment for curve generation. An ending control point represents the third control point intercepted on the exit segment for curve generation.

[0098] Specifically, obtain the turning angle values ​​of each intermediate node calculated in the previous step. Compare each turning angle value with a preset turning angle threshold, such as 90 degrees, to determine if the turning angle is greater than the preset turning angle threshold. When the turning angle of a node is greater than the preset turning angle threshold, mark the node as a turning node that needs smoothing, and record the coordinates of the turning node as an intermediate control point. Extract the entry segment of the turning node, and obtain the coordinates of the starting point of the entry segment and the coordinates of the turning node. Calculate the length of the entry segment by adding the square root of the sum of the square of the difference between the starting x-coordinate and the ending x-coordinate, and the square root of the sum of the square of the square of the difference between the starting y-coordinate and the ending y-coordinate. Calculate the unit direction vector of the entry segment by dividing each component of the entry direction vector by the length of the entry segment. Move backward along the entry segment from the turning node coordinates by a preset intercept length, such as 20 meters. Subtract the x-coordinate component of the unit direction vector from the turning node x-coordinate and multiply by the preset intercept length to obtain the x-coordinate of the starting control point. Subtract the y-coordinate component of the unit direction vector from the turning node y-coordinate and multiply by the preset intercept length to obtain the y-coordinate of the starting control point. Extract the exit segment of the turning node and calculate the unit direction vector of the exit segment. Move the turning node coordinates along the exit segment in the positive direction by a preset intercept length. Add the x-coordinate of the turning node to the x-coordinate component of the unit direction vector and multiply by the preset intercept length to obtain the x-coordinate of the endpoint control point. Add the y-coordinate of the turning node to the y-coordinate component of the unit direction vector and multiply by the preset intercept length to obtain the y-coordinate of the endpoint control point. Record the coordinates of the starting control point, intermediate control point, and endpoint control point corresponding to this turning node.

[0099] Step 10633: Discretely sample the starting control point, control point and ending control point based on the preset smoothing parameters to generate a smooth trajectory point sequence; replace the turning nodes in the blind-end transfer route with the smooth trajectory point sequence to obtain the smooth transfer route.

[0100] In this embodiment, the preset smoothing parameter refers to a parameter used to control the smoothness of the curve, typically an interpolation parameter for a Bézier curve with a value range of 0 to 1. The starting control point represents the coordinates of the first control point generated by the curve. The control point represents the coordinates of the second control point generated by the curve itself, acting as the turning node. The ending control point represents the coordinates of the third control point generated by the curve. Discrete sampling refers to the operation of extracting parameter values ​​at fixed intervals within the preset smoothing parameter range. The smoothed trajectory point sequence represents a series of trajectory point coordinate sequences generated through curve interpolation. A blind-end-free transfer route refers to the transfer path after removing impassable road sections. A turning node connection represents the straight-line connection between road sections before and after a turning node in a blind-end-free transfer route. A smoothed transfer route represents the optimized transfer path after replacing sharp-turn connections with smooth curves.

[0101] Specifically, for each turning node requiring smoothing, the coordinates of its starting control point, intermediate control point, and ending control point are extracted. A preset sampling interval for the smoothing parameters is set, for example, 0.1. Within the range of 0 to 1, a sampling parameter sequence of 0, 0.1, 0.2, up to 1.0 is generated with a step size of 0.1. For each parameter value in the sampling parameter sequence, the coordinates of the corresponding point on the quadratic Bézier curve are calculated. The current parameter value is subtracted from 1 to obtain the complement parameter value. The square of the complement parameter value is multiplied by the x-coordinate of the starting control point, plus twice the complement parameter value multiplied by the current parameter value multiplied by the x-coordinate of the intermediate control point, and then the square of the current parameter value multiplied by the x-coordinate of the ending control point is added to obtain the x-coordinate of the curve point. The y-coordinate of the curve point is calculated in the same way: the square of the complement parameter value multiplied by the y-coordinate of the starting control point, plus twice the complement parameter value multiplied by the current parameter value multiplied by the y-coordinate of the intermediate control point, and then the square of the current parameter value multiplied by the y-coordinate of the ending control point is added. The coordinates of all curve points corresponding to the sampling parameters are arranged in ascending order of parameter value to generate the smoothed trajectory point sequence for that turning node. In a blind-end-free transfer route, locate the turning node and the connecting segments of its preceding and following adjacent nodes. Delete the original path that connects the preceding node to the turning node in a straight line and then to the following node in a straight line, and replace this path segment with points from the smoothed trajectory point sequence. Repeat the above operation for all turning nodes that require smoothing to complete the smoothing of all sharp turns, resulting in a smooth transfer route.

[0102] Step 1064: Calculate the vehicle empty driving distance ratio of the smooth transfer route. If the vehicle empty driving distance ratio is less than the preset empty load threshold, the smooth transfer route will be output as the target transfer route.

[0103] In this embodiment, a smooth transfer route refers to an optimized transfer path after cornering smoothing. The vehicle empty-run distance ratio represents the proportion of the distance a vehicle travels without loading garbage within the transfer route, relative to the total travel distance; a smaller value indicates higher route efficiency. A preset empty-run threshold represents the critical value for determining whether a route meets efficiency requirements, such as 0.3 or 0.35. The target transfer route represents the final determined transfer path output to the vehicle for execution.

[0104] Specifically, the process iterates through all segments of the smooth transfer route, identifying the start and end nodes of each segment. It queries the node type of the start and end nodes to determine if they are garbage collection points. If neither the start nor end of a segment is a garbage collection point, or if the vehicle did not perform a garbage loading operation while traveling on that segment, the segment is marked as an empty segment. The total length of all empty segments is calculated. For each empty segment, the start and end coordinates are extracted. The square root of the sum of the squares of the start and end coordinates is calculated to obtain the segment length. The total empty distance is then summed to obtain the total empty distance traveled by the vehicle. Finally, the total empty distance is divided by the total vehicle travel distance to obtain the empty distance ratio. This empty distance ratio is then compared with a preset empty threshold, such as 0.3. If the proportion of empty driving distance is less than the preset empty load threshold, it means that the route efficiency meets the requirements. The smooth transfer route is then determined as the target transfer route and output to the vehicle dispatching system or navigation system through the data interface to guide the vehicles to perform transfer operations.

[0105] In the above embodiments, by acquiring multi-source fusion data and constructing spatiotemporal correlation features, accurate calculation of predicted waste production in each region and generation of preliminary transfer routes were achieved. To further improve the dynamic response capability of transfer routes to sudden surges in waste production and the spatiotemporal allocation accuracy of vehicle transport capacity resources, this application also provides a method for planning municipal solid waste transfer routes based on multi-source data. This method obtains the maximum vehicle loading capacity and departure interval of the current transfer shift, as well as the node access order and road segment travel time matrix of the initial transfer route. It calculates the Euclidean distance between the spatial grid of the local surge state and the nodes of the initial transfer route to obtain the spatial location matching degree. It accumulates the ratio of the predicted waste output of nodes in the initial transfer route to the maximum vehicle loading capacity to obtain the transport capacity load. Based on the spatial location matching degree, it extracts the target travel time of corresponding nodes from the road segment travel time matrix and uses a preset time margin to perform interval expansion calculations on the start and end times of the surge period to obtain the baseline time interval. It also performs offset calculations on the boundary of the baseline time interval based on the target travel time to determine the node service time window. Finally, it generates the vehicle empty-run distance ratio based on the product of the transport capacity load and the preset empty-run ratio upper limit. This accurately quantifies the spatial coupling relationship between the surge area and the existing route and solves the problem of dynamic adjustment of the time window under vehicle loading capacity constraints. It provides subsequent dynamic route optimization and vehicle scheduling decisions with node service time windows considering spatial location matching degree and vehicle empty-run distance ratio constraints reflecting the transport capacity load status. The following section combines... Figure 2 Another method for planning municipal solid waste transfer routes based on multi-source data is described in the embodiments of this application:

[0106] Please see Figure 2 This is another flowchart illustrating a method for planning municipal solid waste transfer routes based on multi-source data in an embodiment of this application.

[0107] Step 201: Obtain the vehicle loading capacity limit and departure interval of the current transfer shift, as well as the node access order and road segment travel time matrix of the initial transfer route. The initial transfer route is a static transfer path pre-planned based on historical average waste production.

[0108] In this embodiment, the current transfer shift refers to the time batch of a waste transfer operation that is in progress or about to be performed. The maximum vehicle loading capacity represents the maximum weight or volume of waste that a transfer vehicle can carry, such as 8 tons or 12 cubic meters. The departure interval represents the time interval between two adjacent transfer shifts departing from the transfer station, such as 60 minutes or 90 minutes. The initial transfer route refers to a transfer path scheme pre-planned based on historical average waste production, including a starting point, intermediate nodes, and an ending point. The node access order represents the sequential order in which waste collection points and transfer nodes are accessed in the initial transfer route. The road segment travel time matrix is ​​a two-dimensional matrix structure representing the estimated travel time data of road segments between any two nodes in the initial transfer route. The rows and columns of the matrix correspond to each node, and the matrix elements store the travel time values ​​between corresponding node pairs.

[0109] Specifically, the system queries the vehicle configuration information for the current transfer shift from the vehicle dispatch management system, extracts the rated loading capacity parameter corresponding to the model of the transfer vehicle used in this shift, and uses this capacity parameter as the upper limit of the vehicle loading capacity. It retrieves the departure time of the current transfer shift and the departure time of the next transfer shift from the transfer operation schedule, calculates the time difference between the two times, and obtains the departure interval. It reads the initial transfer route data generated for the current area from the route planning module. This route data includes the starting transfer station node, several garbage collection point nodes, and the ending transfer station node. According to the order of the nodes in the initial transfer route, it extracts the number and location information of each node and generates a node access order list. It loads the road segment information from the road network database, and for any two nodes in the initial transfer route, queries the estimated travel time of the shortest path or planned path between these two nodes. It constructs a road segment travel time matrix, with the matrix dimension equal to the total number of nodes in the initial transfer route. The element in the i-th row and j-th column of the matrix stores the travel time value from the i-th node to the j-th node. After filling in the travel time for all node pairs, a complete road segment travel time matrix is ​​generated.

[0110] Step 202: Calculate the Euclidean distance between the spatial grid of the local surge state and the nodes of the initial transport route to obtain the spatial location matching degree.

[0111] In this embodiment, a local surge state refers to an abnormal state in which waste production at certain spatial locations within a target area is significantly higher than the average level during a specific time period. The spatial grid represents the geographic grid unit corresponding to the local surge state, typically a rectangular area divided by a fixed size. Euclidean distance is used to represent the straight-line distance between the center point of the spatial grid and the route node, obtained by calculating the square root of the sum of the squares of the coordinate differences between the two points. Spatial location matching degree represents the spatial proximity index between the spatial grid of the local surge state and the initial transfer route node; the smaller the distance, the higher the matching degree.

[0112] Specifically, the process involves identifying spatial grids experiencing localized surges in waste production. Grids whose predicted production exceeds the historical average by a predetermined multiple (e.g., 1.5 times the average) are selected from the waste production prediction results. Geographical location information for these grids is extracted, obtaining the coordinates of their center point, including both x and y coordinates. Geographical coordinates for all nodes along the initial transfer route are extracted, including the x and y coordinates of each waste collection point and transfer station. For each node along the initial transfer route, the Euclidean distance between that node and the center point of the localized surge grid is calculated. The x-coordinate of the node is subtracted from the x-coordinate of the center point to obtain the horizontal difference, which is then squared. Similarly, the y-coordinate of the node is subtracted from the y-coordinate of the center point to obtain the vertical difference, which is also squared. The squared horizontal and vertical differences are then added together, and the result is squared to obtain the Euclidean distance between the node and the localized surge grid. This calculation is repeated for all nodes along the initial transfer route to generate a list of Euclidean distances from each node to the localized surge grid. Take the reciprocal of the Euclidean distance and normalize it. Map all reciprocal values ​​to the range of 0 to 1. The smaller the distance, the larger the reciprocal value. The normalized value is used as the spatial location matching degree.

[0113] Step 203: Accumulate the ratio of the predicted waste output of nodes in the initial transfer route to the upper limit of vehicle loading capacity to obtain the transport capacity load.

[0114] In this embodiment, nodes represent waste collection points and transfer stations in the initial transfer route. Predicted waste production is used to represent the expected amount of waste generated or to be collected at each node location within the predicted time period, typically in tons or cubic meters. Vehicle loading capacity limit refers to the maximum weight or volume of waste that a transfer vehicle can carry. The ratio represents the dimensionless value obtained by dividing the predicted waste production of a node by the vehicle loading capacity limit, reflecting the proportion of that node's production to the vehicle's capacity. Transport capacity load represents the cumulative proportion of production to vehicle capacity for all nodes in the initial transfer route; a larger value indicates a higher overall load on the route.

[0115] Specifically, the process involves extracting the node numbers from the node access order list of the initial transfer route. For each node, the waste production prediction database is queried to obtain the predicted waste production value for the current time period. The maximum vehicle loading capacity value for the current transfer shift is extracted. For the first node in the initial transfer route, its predicted waste production is divided by the maximum vehicle loading capacity to obtain the production-capacity ratio. The same process is repeated for the second node. The production-capacity ratios for all nodes in the initial transfer route are calculated sequentially. An accumulation variable is created and initialized to 0. The production-capacity ratios for each node are iterated over and added sequentially to the accumulation variable. After accumulating the production-capacity ratios for all nodes, the final value of the accumulation variable represents the transport capacity load. If the transport capacity load value is greater than 1, it indicates that the total predicted waste production of all nodes exceeds the maximum vehicle loading capacity. If the transport capacity load value is less than or equal to 1, it indicates that a single transfer can complete waste collection for all nodes. The capacity load values ​​are saved for subsequent route optimization and resource allocation calculations.

[0116] Step 204: Based on the road segment travel time matrix corresponding to the spatial location matching degree, generate the node service time window by combining the start and end times of the surge period with the preset time margin; generate the vehicle empty driving distance ratio by multiplying the capacity load with the preset empty driving ratio upper limit.

[0117] In this embodiment, spatial location matching degree refers to the spatial proximity index between the spatial grid of the local surge state and the nodes of the initial transfer route. The road segment travel time matrix represents a two-dimensional matrix structure of the estimated travel time of road segments between any two nodes in the initial transfer route. The surge period start and end times are used to represent the start and end times of the local surge state. The preset time margin refers to the additional buffer time length added based on the surge period start and end times, such as 30 minutes or 60 minutes. The node service time window represents the time range within which each node is allowed to provide waste collection services, including the earliest and latest service times. The transport capacity load is used to represent the cumulative ratio of output to vehicle capacity of all nodes in the initial transfer route. The preset empty-run ratio upper limit refers to the maximum allowable percentage threshold of vehicle empty-run distance to total travel distance, such as 0.3 or 0.35. The vehicle empty-run distance ratio represents the estimated empty-run distance percentage index calculated based on the transport capacity load.

[0118] Specifically, the node with the highest matching degree with the spatial grid of the local surge state is identified based on the spatial location matching degree value. The travel time data of the row or column containing this node is extracted from the road segment travel time matrix. The start and end times of the surge period of the local surge state are extracted, including the start and end times of the surge. A preset time margin, such as 30 minutes, is subtracted before the start time of the surge to obtain the earliest time of the node service time window. A preset time margin, such as 30 minutes, is added after the end time of the surge to obtain the latest time of the node service time window. The earliest and latest times are combined to generate the node service time window, which restricts vehicles to arriving at the node within this time range for service. The capacity load value and the preset empty-run ratio upper limit value are extracted. The capacity load is multiplied by the preset empty-run ratio upper limit to obtain a preliminary estimate of the vehicle empty-run distance ratio. If the capacity load is high and close to 1, the vehicle load rate is high, and the empty-run distance ratio is relatively small. If the capacity load is low and much less than 1, the vehicle load rate is low, and the empty-run distance ratio is relatively large. The product result is used as the vehicle empty-run distance ratio for the empty-run control constraint in subsequent route optimization.

[0119] In one possible implementation, a node service time window is generated based on the road segment travel time matrix corresponding to the spatial location matching degree, combined with the start and end times of the surge period plus a preset time margin. Specifically, this includes steps 2041-2043, as follows:

[0120] Step 2041: Based on the spatial location matching degree, extract the target travel time of the corresponding node from the road segment travel time matrix.

[0121] In this embodiment, spatial location matching degree refers to the spatial proximity index between the local surge state spatial grid and the nodes of the initial transfer route; a larger value indicates that the node is closer to the surge location. The road segment travel time matrix represents a two-dimensional matrix structure of the estimated travel time of road segments between any two nodes in the initial transfer route, with matrix elements storing the travel time values ​​between node pairs. The corresponding node refers to the route node identified based on the spatial location matching degree that is closest to the local surge state spatial grid. The target travel time is used to represent the estimated travel time required to reach the corresponding node from the transfer station or other key nodes.

[0122] Specifically, extract a list of spatial location matching scores for all nodes in the initial transfer route. Sort the list in descending order, identify the node with the highest matching score, and record its number and index position in the node access order. Locate the row or column of the corresponding node from the road segment travel time matrix. Extract the travel time from the transfer station's starting node to the corresponding node from the road segment travel time matrix; this value is stored at the intersection of the starting node's row and the corresponding node's column. If the initial transfer route contains multiple key nodes, extract the travel time from each key node to its corresponding node. Summarize the extracted travel time values; if multiple paths lead to the corresponding node, select the path with the shortest travel time. Save the selected travel time value as the target travel time, reflecting the time required to reach the corresponding node from the starting point, for subsequent node service time window calculations. Record the target travel time value and its corresponding starting and corresponding node information to provide basic data for the next step's time window boundary offset calculation.

[0123] Step 2042: Calculate the interval expansion of the start and end times of the surge period using the preset time margin to obtain the base time interval.

[0124] In this embodiment, the preset time margin refers to the additional buffer time length added based on the start and end times of the surge period, used to cope with traffic uncertainties and service delays, such as 30 minutes or 45 minutes. The start and end times of the surge period represent the start and end times of the local surge, identifying the time range of peak waste production. Interval expansion calculation refers to the operation of extending the preset time margin outward from the start and end boundaries of the original time interval. The baseline time interval is used to represent the more relaxed time range obtained after interval expansion, including the earliest and latest times after expansion.

[0125] Specifically, the start and end times of the surge period are extracted from the local surge data. The start time indicates the first occurrence of the surge, and the end time indicates the subsidence of the surge. A preset time margin is configured, for example, 30 minutes. The earliest extended time is calculated by subtracting the preset time margin from the timestamp of the start time. For example, if the surge starts at 10:00 AM and the preset time margin is 30 minutes, the earliest extended time is 9:30 AM. The latest extended time is calculated by adding the preset time margin to the timestamp of the end time. For example, if the surge ends at 11:00 AM and the preset time margin is 30 minutes, the latest extended time is 11:30 AM. The earliest and latest extended times are combined to form a baseline time interval. The baseline time interval is extended by a preset time margin before and after the original surge period, providing more flexible service time scheduling space, and recording the earliest and latest time values ​​of the baseline time interval.

[0126] Step 2043: Calculate the offset of the boundary of the reference time interval based on the target passage time to determine the node service time window.

[0127] In this embodiment, the target travel time refers to the estimated travel time required to reach the corresponding node from a transfer station or key node. The baseline time interval represents the time range obtained after interval expansion, including the earliest and latest times after expansion. The boundary refers to the starting and ending boundaries of the baseline time interval, i.e., the earliest and latest times. Offset calculation is used to represent the operation of adjusting the time boundaries based on the target travel time. The node service time window represents the time range within which the corresponding node is allowed to provide garbage collection services, including the earliest and latest service times.

[0128] Specifically, the baseline time interval calculated in the previous step is extracted, and the earliest and latest times within the baseline time interval are obtained. The target travel time value is extracted, representing the travel time required for a vehicle to travel from the transfer station to the corresponding node. The earliest time in the baseline time interval is shifted forward, and the target travel time is subtracted from the earliest time to obtain the earliest departure time of the vehicle, ensuring that the vehicle arrives at the corresponding node at the earliest time in the baseline time interval after departing at this time. The earliest departure time is added to the target travel time, and the earliest service time of the corresponding node is calculated in reverse, which is equal to or close to the earliest time in the baseline time interval. The latest time in the baseline time interval is shifted, and the target travel time is subtracted from the latest time to obtain the latest departure time of the vehicle, ensuring that a vehicle departing before this time arrives at the corresponding node before the latest time in the baseline time interval. The latest departure time is added to the target travel time to calculate the latest service time of the corresponding node, which is equal to or close to the latest time in the baseline time interval. The earliest and latest service times of the corresponding node are combined to determine the node's service time window.

[0129] The following describes a municipal solid waste transfer route planning system based on multi-source data from the perspective of hardware processing in this application embodiment. Please refer to [link to relevant documentation]. Figure 3 This is a schematic diagram of a municipal solid waste transfer route planning system based on multi-source data in an embodiment of this application.

[0130] It should be noted that, Figure 3 The structure of a municipal solid waste transfer route planning system based on multi-source data shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0131] like Figure 3 As shown, a municipal solid waste transfer route planning system based on multi-source data includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes according to a program stored in a Read-Only Memory (ROM) 302 or a program loaded from a storage section 308 into a Random Access Memory (RAM) 303, such as executing the method described in the above embodiment. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.

[0132] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0133] The above description is merely a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for planning municipal solid waste transfer routes based on multi-source data, characterized in that, The method includes: Acquire multi-source basic data of the target area, and perform time step alignment processing on the multi-source basic data to obtain multi-source fused data; A feature matrix is ​​generated based on the multi-source fusion data, and spatiotemporal correlation features are determined based on the feature matrix; The spatiotemporal correlation features are calculated by using a preset network layer structure to obtain the production fluctuation sensitivity coefficient between adjacent grids, and the predicted waste production of each region is calculated based on the production fluctuation sensitivity coefficient. If the ratio of the predicted waste production to the historical average is greater than the preset production threshold, then the spatial grid number of the corresponding area and the start and end times of the surge period are marked to obtain the local surge state; Based on the local surge state, calculate the spatial location matching degree between the surge grid and the existing route and the current vehicle capacity load, determine the node service time window based on the spatial location matching degree, and generate the vehicle empty driving distance ratio based on the capacity load. The spatial location matching degree and the transportation capacity load are extracted as initial parameters for optimization. Under the constraints of the node service time window and the ratio of vehicle empty driving distance, the first transfer route is calculated. If the estimated time for the first transfer route is greater than the preset time threshold, then the first transfer route is allocated shifts to obtain a second transfer route. Based on the preset road network topology and real-time traffic flow data of the target area, the second transfer route is smoothed and blind end sections are removed to output the target transfer route.

2. The method according to claim 1, characterized in that, The process of acquiring multi-source basic data of the target area and performing time-step alignment processing on the multi-source basic data to obtain multi-source fused data includes: Acquire meteorological rainfall intensity data, assembly date data, consecutive holiday data, and hourly garbage production records for the target area; The meteorological rainfall intensity data is subjected to projection transformation and grid resampling processing; the meeting date data is subjected to timestamp alignment processing; the number of consecutive holiday days data is subjected to time zone normalization processing; and the hourly garbage production records are subjected to missing value linear interpolation filling processing. All processed data are unified to a preset spatial resolution and a preset time step to obtain the multi-source fused data.

3. The method according to claim 1, characterized in that, The step of generating a feature matrix based on the multi-source fusion data and determining spatiotemporal correlation features based on the feature matrix includes: Extract data from each dimension of the multi-source fusion data and concatenate them to generate the feature matrix; Calculate the decrease in Gini impurity of each dimension feature in the feature matrix when the node splits, and sort the features of each dimension according to the decrease in Gini impurity to obtain the feature importance sequence; A predetermined number of dimensional features are extracted from the feature importance sequence and combined to generate a target feature subset; For the target feature subset, time dimension features are extracted based on a preset backtracking period, spatial dimension features are extracted based on a preset buffer distance, and the time dimension features and spatial dimension features are combined to obtain initial spatiotemporal features; Calculate the correlation coefficients between the features of each dimension in the initial spatiotemporal features, and remove the features of dimensions whose absolute values ​​of the correlation coefficients are greater than a preset correlation coefficient threshold to obtain the spatiotemporal correlation features.

4. The method according to claim 1, characterized in that, The process involves calculating the spatiotemporal correlation features using a preset network layer structure to obtain the production fluctuation sensitivity coefficient between adjacent grids, and then calculating the predicted waste production for each region based on the production fluctuation sensitivity coefficient, including: The spatiotemporal correlation features are standardized to obtain a standardized feature vector; The standardized feature vector is input into the hidden layer of the preset network layer structure and a preset activation function is used for feature mapping to output the hidden layer state features. The difference between the hidden layer state features corresponding to adjacent grids in the target region is calculated to obtain the production fluctuation sensitivity coefficient; The hidden layer state features are input into the output layer of the preset network layer structure for regression calculation to obtain the basic predicted output of each region. The predicted waste output for each region is obtained by weighting and correcting the basic predicted output based on the output fluctuation sensitivity coefficient.

5. The method according to claim 1, characterized in that, The process of calculating the spatial location matching degree between the surge grid and existing routes and the current vehicle capacity load based on the local surge state, determining the node service time window based on the spatial location matching degree, and generating the vehicle empty driving distance ratio based on the capacity load includes: Obtain the vehicle loading capacity limit and departure interval of the current transfer shift, as well as the node access order and road segment passage time matrix of the initial transfer route. The initial transfer route is a static transfer path pre-planned based on historical average waste production. Calculate the Euclidean distance between the spatial grid of the local surge state and the nodes of the initial transport route to obtain the spatial location matching degree; The transport capacity load is obtained by summing the ratio of the predicted waste output of the nodes in the initial transfer route to the upper limit of the vehicle loading capacity; Based on the road segment travel time matrix corresponding to the spatial location matching degree, the node service time window is generated by combining the start and end times of the surge period with a preset time margin. The vehicle empty driving distance ratio is generated by multiplying the capacity load with the preset empty driving ratio upper limit.

6. The method according to claim 5, characterized in that, The process of generating the node service time window based on the road segment travel time matrix corresponding to the spatial location matching degree, combined with the start and end times of the surge period plus a preset time margin, includes: Based on the spatial location matching degree, the target travel time of the corresponding node is extracted from the road segment travel time matrix; Using the preset time margin, the start and end times of the surge period are calculated by interval expansion to obtain the base time interval; The boundary of the reference time interval is offset based on the target passage time to determine the node service time window.

7. The method according to claim 1, characterized in that, If the estimated travel time of the first transfer route is greater than a preset time threshold, then the first transfer route is allocated shifts to obtain a second transfer route, including: When the estimated travel time is greater than the preset travel time threshold, the target node in the first transfer route that has reached the preset single-shift travel time limit is found as the cutoff point and split to obtain the first shift node sequence and the second shift node sequence. Extract the first and second shift node sequences, and allocate the start and end times of departures based on the preset departure interval to obtain the departure timetable for each shift. Based on the departure timetables for each bus, the unvisited nodes with the shortest travel time from the current node are sequentially connected to update the node access order, thereby generating the second transfer route.

8. The method according to claim 1, characterized in that, The process involves combining the preset road network topology of the target area with real-time traffic flow data to perform path smoothing and blind spot removal on the second transfer route, and outputting the target transfer route, including: By combining the real-time traffic flow data, congested road segments and restricted road segments in the preset road network topology are identified, and the connectivity weights of the corresponding road segments in the adjacency matrix are updated to preset values ​​to obtain a real-time available road network matrix. The second transfer route is mapped to the real-time available road network matrix, and blind end road segments with a preset connectivity weight are found and eliminated to obtain a transfer route without blind ends. The turning angles of adjacent road segments in the blind-end-free transfer route are extracted, and curve interpolation calculations are performed on the nodes with turning angles greater than a preset turning angle threshold to obtain a smooth transfer route. Calculate the vehicle empty driving distance ratio of the smooth transfer route. If the vehicle empty driving distance ratio is less than a preset empty load threshold, then output the smooth transfer route as the target transfer route.

9. The method according to claim 8, characterized in that, The step of extracting the turning angles of adjacent road segments in the blind-end-free transfer route, and performing interpolation calculations on the node connections where the turning angle is greater than a preset turning angle threshold to obtain a smooth transfer route includes: Extract the inbound and outbound direction vectors of the adjacent node connection segments in the blind-endless transfer route; Calculate the supplementary angle of the angle between the entering direction vector and the exit direction vector to obtain the turning angle; When the turning angle is greater than the preset turning angle threshold, the corresponding turning node is used as a control point, and the coordinates are extracted from the corresponding entry and exit road segments according to the preset intercept length to obtain the starting control point and the ending control point. Based on preset smoothing parameters, the starting control point, the control point and the ending control point are discretely sampled to generate a smooth trajectory point sequence; The smooth transfer route is obtained by replacing the turning nodes in the blind-endless transfer route with the smooth trajectory point sequence.

10. A system for planning municipal solid waste transfer routes based on multi-source data, characterized in that, The multi-source data-based municipal solid waste transfer route planning system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the multi-source data-based municipal solid waste transfer route planning system to perform the method as described in any one of claims 1-9.

Images