Intelligent early warning method of intelligent environmental sanitation platform

Abstract

This invention relates to the field of smart city management technology and discloses an intelligent early warning method for a smart sanitation platform. The method acquires multi-source data from sanitation facilities, including sensor data, pedestrian flow data, meteorological data, and traffic data. It constructs an urban sanitation topology network, defining downstream nodes using pedestrian movement direction vectors and dynamically assigning edge weights. A spatiotemporal prediction model is used to aggregate features of the dynamic topology network to predict the time of garbage overflow. When the garbage enters the warning range, a vision-sensor mutual verification mechanism is activated, using a semantic segmentation algorithm to verify the accuracy of sensor data to generate a definitive warning. Finally, meteorological data is combined to calculate the garbage decomposition rate factor, and traffic fluctuations are considered to calculate the optimal interception time window to generate dynamic dispatch instructions. This invention can predict overflow risks in advance based on pedestrian flow trends, reduce false alarm rates, and achieve precise garbage collection and dispatch based on environmental conditions.

Description

Technical Field

[0001] This invention relates to the field of smart city management technology, and in particular to an intelligent early warning method for a smart sanitation platform. Background Technology

[0002] With the advancement of smart city construction, urban environmental sanitation management is gradually transforming from manual inspection to intelligent supervision. Existing smart sanitation systems mainly rely on single sensors (such as ultrasonic overflow detectors or weighing sensors) deployed inside trash cans for status monitoring, triggering an alarm when the detected value exceeds a fixed threshold (such as 80%).

[0003] However, existing technologies have significant drawbacks in practical applications: First, the early warning mechanism is lagging. Most existing solutions are "post-event responses," unable to predict overflow times based on changes in regional pedestrian traffic, often resulting in garbage overflow by the time collection vehicles arrive, severely impacting the city's appearance. Second, single-source data has a high false alarm rate. Ultrasonic sensors are easily obstructed by hollow objects (such as cardboard boxes), leading to false overflow alarms, or malfunction due to dirt covering the probe; relying solely on camera monitoring is significantly affected by light and obstruction, lacking an effective multimodal data verification mechanism. Finally, scheduling strategies lack refined consideration. Existing route planning typically only aims for the shortest distance, ignoring the impact of temperature and humidity on garbage decomposition speed and the uncertainty of road traffic fluctuations, making it difficult to find the optimal balance between preventing garbage odor and reducing transportation costs.

[0004] Therefore, there is an urgent need for a smart sanitation early warning method that can integrate spatiotemporal multidimensional data, has proactive prediction capabilities, and provides precise scheduling. Summary of the Invention

[0005] This invention provides an intelligent early warning method for a smart sanitation platform. By constructing a spatiotemporal topology prediction model driven by human flow and a visual-sensor mutual verification mechanism, it achieves proactive and accurate early warning of garbage overflow. Furthermore, by combining environmental decomposition factors to optimize the operation time window, it effectively solves the problems of delayed early warning, high false alarm rate, and poor scheduling timeliness in traditional sanitation systems.

[0006] This invention provides an intelligent early warning method for a smart sanitation platform, comprising:

[0007] S1. Acquire real-time sensor data of sanitation facilities, regional pedestrian flow heat data, real-time meteorological data, and related road traffic flow data within the target area; wherein, the target area contains multiple sanitation facilities;

[0008] S2. Construct an urban sanitation topology network based on the geographical location of the sanitation facilities within the target area, map each sanitation facility to a network node of the urban sanitation topology network, calculate the direction vector of pedestrian movement based on the temporal changes in the location of the regional pedestrian flow heat data, and define the node it points to as the downstream node, so as to dynamically assign values ​​to the edge weights pointing to the downstream nodes.

[0009] S3. Input the real-time sensor data of the sanitation facilities, the regional human flow heat data and the real-time meteorological data into the pre-trained spatiotemporal prediction model, use the dynamically assigned urban sanitation topology network to perform feature aggregation, and output the garbage saturation change curve of each sanitation facility in the future preset time period to extract the predicted value of the overflow time.

[0010] S4. When the predicted value of the overflow moment enters the warning range, the visual acquisition device associated with the sanitation facility is invoked to obtain real-time images and the visual saturation value is determined using a semantic segmentation algorithm. If the difference between the visual saturation value and the real-time sensor data of the sanitation facility is within the preset tolerance range, a certainty warning signal is generated.

[0011] S5. In response to the confirmed early warning signal, calculate the garbage decomposition rate factor based on the real-time meteorological data and calculate the optimal interception time window by combining the associated road traffic flow data and the garbage saturation change curve. It generates dispatch instructions and plans the travel routes of the work vehicles, ensuring that the expected arrival time of the work vehicles at the target sanitation facility falls within the optimal interception time window. Inside, and It is less than the predicted value at the overflow time.

[0012] Furthermore, S1 specifically includes:

[0013] S101. Traverse all sanitation facilities within the target area, assign a unique equipment identification code to each sanitation facility, and record its static attribute data.

[0014] S102. Periodically collect data using IoT sensor terminals deployed inside the sanitation facility and send it to the cloud server via a wireless communication network to obtain real-time sensor data of the sanitation facility.

[0015] S103. Delineate a radiation zone with the sanitation facility as the center, and obtain the real-time population density value within the radiation zone as the population thermal data of the area.

[0016] S104. Obtain the current atmospheric temperature, relative humidity, and precipitation type of the target area as the real-time meteorological data.

[0017] S105. Identify the cleaning operation roads adjacent to each of the sanitation facilities, and obtain the real-time average vehicle speed and congestion index of the cleaning operation roads as traffic flow data of the associated roads.

[0018] S106. The real-time sensor data of the sanitation facilities, the regional population flow heat data, the real-time meteorological data and the associated road traffic flow data are subjected to outlier removal, and spatiotemporal alignment processing is performed based on the preset time granularity to form a standardized multidimensional feature tensor.

[0019] Furthermore, S2 specifically includes:

[0020] S201. Construct a static basic topology graph structure G=(V,E), mapping each sanitation facility to a vertex V in the graph, and construct connecting edges E based on the principle of geographical proximity, assigning initial weights to the connecting edges; wherein, the specific method for constructing connecting edges and assigning initial weights is as follows: calculate any two nodes and Euclidean distance between ;like Less than the preset distance threshold Then at node and Establish connecting edges between them; set the initial weights of the connecting edges. and Inversely proportional;

[0021] S202. Obtain the regional pedestrian flow heat map data at the current time t and the previous time t-1, calculate the pedestrian flow movement direction vector in the grid cell using the image optical flow method or gradient calculation method, and map the pedestrian flow movement direction vector to the corresponding vertex.

[0022] S203. For the starting node and the target node connected in the topology network, calculate the cosine value of the angle between the physical path vector and the direction vector of the flow of people at the starting node. If the cosine value of the angle is greater than zero, then determine that the target node is the downstream node of the starting node.

[0023] S204. Based on the magnitude of the pedestrian movement direction vector and the cosine of the included angle, calculate the pedestrian impulse gain, and use the pedestrian impulse gain to correct the initial weight of the connecting edge pointing to the downstream node to obtain the dynamic weight.

[0024] Furthermore, in S204, the formula for calculating the dynamic weight is:

[0025]

[0026] in, Represents a node Pointing to node The dynamic weights of the connecting edges; Represents a node and nodes Initial weights between them; This represents the preset influence adjustment coefficient, used to adjust the proportion of pedestrian flow factors in the weighting calculation; Represents a node The magnitude of the vector representing the direction of pedestrian movement at a given location characterizes the intensity of pedestrian movement. Represents a node Pointing to node physical path vector With nodes The direction vector of pedestrian movement at the location The cosine of the angle between them.

[0027] Furthermore, S3 specifically includes:

[0028] S301. Construct a spatiotemporal prediction model, which includes a graph spatial feature extraction layer, a temporal evolution layer, and an output layer connected in sequence.

[0029] S302. Integrate the real-time sensor data of sanitation facilities, regional pedestrian flow heat data, real-time meteorological data and related road traffic flow data that have undergone spatiotemporal alignment processing into a node feature matrix, and construct a multi-dimensional feature input tensor.

[0030] S303. Input the multidimensional feature input tensor into the graph space feature extraction layer, construct a dynamic adjacency matrix using the dynamic weights, and perform spatial feature aggregation on the node feature matrix. The aggregation operation formula is as follows:

[0031]

[0032] in, Indicates the feature state of the node in the l-th layer; This represents the feature state of a node in the (l+1)th layer; This indicates a dynamic adjacency matrix with self-loops, whose element values ​​are determined based on the dynamic weights and are dynamically updated as the time step changes. express The degree matrix; This represents the learnable weight parameter matrix of the l-th layer; Indicates the activation function;

[0033] S304. Input the feature vector sequence after aggregating spatial features into the temporal evolution layer to capture the temporal evolution pattern of garbage generation, and generate a prediction sequence for a future preset time period through the output layer.

[0034] S305. Smooth the predicted sequence using spline interpolation to generate a continuous function. As the curve showing the change in waste saturation; an overflow threshold is set. Solve the equation The smallest positive real root is used as the predicted value at the overflow time.

[0035] Furthermore, S4 specifically includes:

[0036] S401. Monitor the predicted value of the overflow moment in real time, calculate the time difference between the current time and the predicted value of the overflow moment, and if the time difference is less than the preset warning response threshold, send a capture command to the visual acquisition device to obtain real-time RGB image frames.

[0037] S402. Input the real-time RGB image frame into a pre-trained semantic segmentation model, classify and statistically analyze the image pixels, and calculate the visual saturation value based on the statistical results. The calculation formula is as follows:

[0038]

[0039] in, This represents the visual saturation value; This indicates the number of pixels in the real-time RGB image frame that are identified as garbage accumulation categories; This represents the number of pixels in the real-time RGB image frame that are identified as belonging to the trash can container region category; if the semantic segmentation model identifies that the accumulated trash extends beyond the boundary of the trash can container region, then the visual saturation value is directly increased. Set to 100%;

[0040] S403. Read the real-time sensor data of the sanitation facilities, and normalize the real-time sensor data of the sanitation facilities according to the physical attribute parameters of the sanitation facilities to generate the sensor saturation value in percentage form; wherein, the sensor saturation value The calculation method includes: if the real-time sensing data of the sanitation facilities is an ultrasonic ranging value And the total depth of the sanitation facilities is ,but If the real-time sensor data of the sanitation facilities is a weight value And the maximum load-bearing capacity of the sanitation facilities is ,but ;

[0041] S404. Calculate the absolute difference between the visual saturation value and the sensor saturation value. If the absolute difference is less than or equal to a preset tolerance threshold, determine that the visual recognition result is consistent with the sensor data and generate a confirmation warning signal.

[0042] Furthermore, S5 specifically includes:

[0043] S501. Set a preset collection threshold, traverse the waste saturation change curve, and determine the moment when the waste saturation change curve first reaches the preset collection threshold as the starting point of the interception time window.

[0044] S502. Based on the real-time meteorological data, construct an environmental impact model and calculate the waste decomposition rate factor to characterize the risk of waste decomposition; wherein, the calculation formula is:

[0045]

[0046] Wherein, Temp and Hum represent the atmospheric temperature and relative humidity in the real-time meteorological data, respectively; and These represent the preset reference temperature and reference humidity, respectively. and These represent the temperature weighting coefficient and the humidity weighting coefficient, respectively. Indicates the activation function;

[0047] S503. Calculate the traffic fluctuation variance based on the associated road traffic flow data, calculate the safety buffer time in conjunction with the garbage decomposition rate factor, and subtract the safety buffer time from the predicted overflow time to obtain the endpoint of the interception time window, thereby determining the optimal interception time window; wherein, the endpoint of the interception time window... The calculation formula is:

[0048]

[0049] in, The value at the overflow moment is represented by K; K represents the preset adjustment constant. This represents the waste decomposition rate factor; This represents the traffic fluctuation variance calculated based on the associated road traffic flow data, used to characterize the instability of road conditions;

[0050] S504. Construct a path planning objective function with time window constraints, introduce an empty load penalty term for arrival earlier than the start of the interception time window and an overflow penalty term for arrival later than the end of the interception time window into the path planning objective function, and solve for the optimal driving path sequence.

[0051] S505. Generate a scheduling instruction containing a mandatory arrival time period based on the optimal driving route sequence, and send it to the work vehicle terminal.

[0052] Furthermore, in S504, the path planning objective function is constructed as follows:

[0053] The optimization objective is to minimize the total transportation cost, which includes fuel consumption cost and time cost.

[0054] When the scheduled arrival time of the work vehicle At that time, the empty-load penalty item is added to the total transportation cost, wherein This is the starting point of the interception time window;

[0055] When the scheduled arrival time of the work vehicle At that time, the overflow penalty item is added to the total transportation cost, wherein This is the end point of the interception time window;

[0056] The objective function of the path planning is solved using either the ant colony algorithm or the genetic algorithm.

[0057] The present invention also provides an intelligent early warning device for a smart sanitation platform, based on the intelligent early warning method for a smart sanitation platform as described above, the device comprising:

[0058] The acquisition module is used to acquire real-time sensor data of sanitation facilities, regional pedestrian flow heat data, real-time meteorological data, and related road traffic flow data within the target area; wherein, the target area contains multiple sanitation facilities;

[0059] The construction module is used to construct an urban sanitation topology network based on the geographical location of the sanitation facilities within the target area, map each sanitation facility to a network node of the urban sanitation topology network, calculate the direction vector of pedestrian movement based on the temporal location change of the area's pedestrian flow heat data, and define the node it points to as the downstream node, so as to dynamically assign values ​​to the edge weights pointing to the downstream nodes.

[0060] The aggregation module is used to input the real-time sensor data of the sanitation facilities, the regional human flow heat data and the real-time meteorological data into the pre-trained spatiotemporal prediction model, and use the dynamically assigned urban sanitation topology network to perform feature aggregation, and output the garbage saturation change curve of each sanitation facility in the future preset time period to extract the predicted value of the overflow time.

[0061] The early warning module is used to call the visual acquisition device associated with the sanitation facility to acquire real-time images and use a semantic segmentation algorithm to determine the visual saturation value when the predicted value of the overflow time enters the early warning range. If the difference between the visual saturation value and the real-time sensor data of the sanitation facility is within the preset tolerance range, a certainty warning signal is generated.

[0062] The indicator module is used to respond to the confirmed warning signal by calculating the garbage decomposition rate factor based on the real-time meteorological data and combining it with the associated road traffic flow data and garbage saturation change curve to calculate the optimal interception time window. It generates dispatch instructions and plans the travel routes of the work vehicles, ensuring that the expected arrival time of the work vehicles at the target sanitation facility falls within the optimal interception time window. Inside, and It is less than the predicted value at the overflow time.

[0063] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0064] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0065] The beneficial effects of this invention are as follows:

[0066] This invention constructs an urban sanitation topology network based on pedestrian movement direction vectors and utilizes graph neural networks to capture the potential impact of upstream pedestrian flow on downstream nodes, achieving a leap from passive alarm to proactive prediction and significantly improving the foresight of overflow warnings. Simultaneously, this invention introduces a vision-sensor mutual verification mechanism, using semantic segmentation algorithms to perform secondary verification of sensor data, effectively filtering false alarms caused by sensor obstruction or malfunction. Furthermore, this invention establishes a garbage decomposition rate model based on meteorological data and combines it with traffic fluctuation variance to calculate the optimal interception time window, ensuring that cleaning vehicles arrive at the optimal time when garbage is still valuable and has not yet fermented and overflowed, achieving dual optimization of sanitation resource scheduling in terms of both economy and environmental sanitation quality. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0068] Figure 2 This is a schematic diagram of the device structure according to an embodiment of the present invention.

[0069] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.

[0070] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0071] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0072] like Figure 1 As shown, the present invention provides an intelligent early warning method for a smart sanitation platform, comprising:

[0073] S1. Acquire real-time sensor data of sanitation facilities, regional pedestrian flow heat data, real-time meteorological data, and related road traffic flow data within the target area; wherein, the target area contains multiple sanitation facilities.

[0074] In one embodiment, step S1 specifically includes the following sub-steps S101 to S106:

[0075] S101. Delineate the electronic fence of the target area using a GIS (Geographic Information System). Traverse all physical sanitation facilities (such as smart trash cans, garbage rooms, and transfer stations) within the target area, assign a unique device ID to each facility, and record its static attribute data. The static attribute data includes at least: geographic coordinates (longitude, latitude), facility capacity (volume / L), facility type (recyclable / kitchen waste / other), and the grid cell number to which it belongs.

[0076] S102. Collect real-time sensor data of sanitation facilities.

[0077] Periodic data collection is performed using IoT sensor terminals deployed inside the sanitation facility. Specifically, ultrasonic ranging sensors or lidar sensors are used to collect the garbage filling height, and the ratio of the filling height to the total height of the bin is calculated to obtain the filling percentage data; bottom weighing sensors are used to collect the current weight data of the garbage; and gas sensors (such as NH3 or H2S sensors) are used to collect the odor concentration data inside the bin. The filling percentage data, weight data, and odor concentration data are combined as real-time sensing data for the sanitation facility.

[0078] The real-time sensor data of the aforementioned sanitation facilities are sent to the cloud server via NB-IoT (Narrowband Internet of Things) or 4G / 5G communication modules at a preset sampling frequency (e.g., once every 5 minutes).

[0079] S103. Obtain regional pedestrian flow heat map data.

[0080] Pedestrian flow data is obtained by connecting to the API interface of a third-party location service provider (LBS) or by calling the SDK of urban public security cameras. Specifically, a circular radiation zone or corresponding grid zone with a preset radius (e.g., 50 meters) is delineated centered on the sanitation facility. The number of mobile terminal signaling accesses within this area or the human detection count based on video streams is obtained to generate a real-time pedestrian density value (unit: people / square meter) as regional pedestrian flow heat map data. This regional pedestrian flow heat map data has been anonymized and does not contain specific individual privacy information. This data reflects the intensity of social activity in the area at the current moment and is used in step S2 to calculate the direction of pedestrian movement and potential waste generation pressure.

[0081] S104. Obtain real-time meteorological data.

[0082] By calling the meteorological bureau's open API interface or reading data from micro-weather stations deployed in the target area, the current atmospheric temperature, relative humidity, and precipitation type of the target area are obtained as real-time meteorological data. Atmospheric temperature and relative humidity will serve as the basic parameters for calculating the waste decomposition rate factor in step S5 (e.g., the decomposition factor increases under high temperature and high humidity conditions). Precipitation type (such as heavy rain or strong winds) is used to assist in determining whether an emergency response plan needs to be activated.

[0083] S105. Obtain traffic flow data for associated roads.

[0084] Identify the cleaning operation roads adjacent to each sanitation facility, and obtain the real-time average vehicle speed and congestion index (TPI) of the road segment as related road traffic flow data by connecting to the traffic status API of online map service providers (such as Gaode / Baidu Maps).

[0085] The associated road traffic flow data is not only used to characterize the ease with which operating vehicles can reach their targets, but also serves as a feature input to the spatiotemporal prediction model to help determine the potential impact of road congestion on the amount of roadside garbage dumping.

[0086] S106. Since the data obtained in S102 to S105 come from different sources, their timestamps and data formats differ, therefore standardization is required:

[0087] Outlier removal: Filter the real-time sensor data to remove noise caused by instantaneous sensor jitter (e.g., the fill percentage jumps from 20% to 100% and then immediately recovers).

[0088] Spatiotemporal alignment: Based on a preset time granularity (e.g., 10 minutes), linear interpolation or nearest neighbor interpolation is used to uniformly align real-time sensor data of sanitation facilities, regional pedestrian flow heat data, real-time meteorological data, and related road traffic flow data onto the same time axis, forming a standardized multidimensional feature tensor, which is stored in a time series database for use in steps S2 and S3.

[0089] S2. Construct an urban sanitation topology network based on the geographical location of the sanitation facilities within the target area, map each sanitation facility to a network node of the urban sanitation topology network, calculate the direction vector of pedestrian movement based on the temporal location changes of the regional pedestrian flow heat data, and define the node it points to as the downstream node, so as to dynamically assign values ​​to the edge weights pointing to the downstream nodes.

[0090] In one embodiment, step S2 specifically includes the following sub-steps S201 to S204:

[0091] S201. Establish an urban sanitation topology network represented by graph G=(V,E).

[0092] Node definition: Each sanitation facility (such as a smart trash can or a garbage room) obtained in step S1 is mapped to a vertex in the graph, denoted as the node set. , where n is the total number of sanitation facilities within the target area. Each node Includes its latitude and longitude coordinates .

[0093] Edge definition: Construct an edge set E based on the principle of geographical proximity. For any two nodes... and Calculate the Euclidean distance between them. Set a distance threshold. (For example, 500 meters), if Then at node and Establish a bidirectional connection edge between them.

[0094] Initial weights: The weights of the edges in the initial state. It is inversely proportional to geographical distance; that is, the closer the distance, the stronger the association. The formula can be expressed as: .

[0095] S202. Calculate the direction vector of pedestrian movement based on the time-series heatmap.

[0096] The optical flow method or gradient calculation method in computer vision is used to process regional pedestrian flow heat map data. Specifically, the pedestrian flow heat map matrix at the current time t is obtained. And the heat map matrix of pedestrian traffic at the previous time t-1 The geographic space containing the heatmap is divided into a k×k grid. For each grid cell g, its position at time intervals is calculated. The gradient change in pedestrian density within the area. Constructing a vector representing the direction of pedestrian movement. This vector contains two components: direction (indicating the primary direction of pedestrian movement, such as from the subway station to the shopping street) and modulus (indicating the intensity or speed of pedestrian movement). The pedestrian flow vector of the grid is mapped to the nodes of the topology network. For each node... Find the grid cell where it is located, and assign the vector of that grid cell to that node, denoted as . .

[0097] S203. Determine the relationship between upstream and downstream nodes in the topology network.

[0098] For any two connected nodes in the topology network (Starting node) and (Target nodes), determine their geometric relationship with the direction of pedestrian movement, specifically including:

[0099] Calculate from node Pointing to node physical path vector This vector is determined by the difference in geographic coordinates between the two. Calculate the pedestrian flow vector at the node. With physical path vector cosine value of the angle between .

[0100] like (i.e., the included angle is acute), indicating that the physical path direction is roughly the same as the direction of pedestrian movement, and nodes are defined. For nodes The downstream node. At this point, garbage is more likely to be carried by people from... Generated and directed to Migration (e.g., a pedestrian carrying a beverage bottle from...) walk to Throw it away).

[0101] like This indicates that the two directions are opposite or perpendicular, and the nodes It is not a downstream node.

[0102] S204. Dynamically assign edge weights based on pedestrian flow volume.

[0103] The weights of the edges pointing to the downstream nodes are dynamically adjusted by introducing a flow momentum gain. The adjusted dynamic weights are... The calculation formula is as follows:

[0104]

[0105] in, The initial weights are based on geographical distance; The magnitude of the vector representing the direction of human movement (representing the flow rate); This is the directional similarity coefficient (representing the consistency of flow direction). This is a preset influence adjustment coefficient (used to balance the influence ratio of distance weight and flow weight).

[0106] The dynamic topology network constructed in step S2 not only reflects the physical location relationship of sanitation facilities, but also includes the trend characteristics of potential garbage transfer with the flow of people, providing a spatiotemporal semantic-rich input structure for the graph neural network prediction in step S3.

[0107] S3. Input the real-time sensor data of the sanitation facilities, the regional population flow heat data and the real-time meteorological data into the pre-trained spatiotemporal prediction model, use the dynamically assigned urban sanitation topology network to perform feature aggregation, and output the garbage saturation change curve of each sanitation facility in the future preset time period to extract the predicted value of the overflow time.

[0108] In one embodiment, the spatiotemporal prediction model employs a hybrid deep learning architecture based on graph convolutional networks (GCN) and long short-term memory networks (LSTM), and step S3 specifically includes the following sub-steps S301 to S305:

[0109] S301. Constructing the network architecture for spatiotemporal prediction models.

[0110] Construct a deep neural network model that includes a graph space feature extraction layer and a temporal evolution layer, specifically including:

[0111] Graph spatial feature extraction layer: A multi-layer graph convolutional network (GCN) is used to capture the spatial dependencies between sanitation facilities and the impact of pedestrian flow at each time step.

[0112] Temporal evolution layer: Long Short-Term Memory (LSTM) network or Gated Recurrent Unit (GRU) is used to capture the periodic patterns and long-term trends of waste generation.

[0113] Output layer: Employs a fully connected layer to output the future. The saturation value at each time step.

[0114] S302. Integrate the spatiotemporally aligned data from step S1 into a node feature matrix. , specifically,

[0115] For each node in the topology network Its eigenvectors It includes the current sensor saturation value, the current pedestrian density value, the current meteorological parameters (normalized values ​​of temperature and humidity), and the traffic congestion index of the associated roads. The input features of the entire network are represented as tensors. , where k is the time step of historical observation (e.g., the past 12 time steps).

[0116] S303, the node feature matrix Input the graph space feature extraction layer. In this step, the dynamic weights calculated in step S2 are introduced. To construct the adjacency matrix .

[0117] The core calculation formula for the GCN layer is:

[0118]

[0119] in: It is the node feature state of the l-th layer; It introduces a dynamic adjacency matrix with self-loops, where the element values ​​are determined by the crowd flow impulse gain in step S2. It is a learnable weight parameter matrix; express The degree matrix; This represents the learnable weight parameter matrix of the l-th layer; It is an activation function (such as ReLU). By using... The garbage generation characteristics of upstream nodes in the flow of people will be propagated and aggregated to downstream nodes with greater weight, thus enabling the model to sense the upcoming increase in garbage.

[0120] S304. The feature vector sequence output from S303, which aggregates spatial information, is input into the temporal evolution layer (LSTM). The LSTM unit updates its internal cell state through forget gate, input gate, and output gate mechanisms. This process memorizes historical garbage growth patterns (such as the growth slope during morning and evening rush hours). After multi-layer LSTM processing, the output layer decodes the data to generate a prediction sequence for a predetermined time period in the future (e.g., one point every 10 minutes for the next 2 hours). .

[0121] S305, Based on Predicted Sequence A smooth curve of waste saturation variation was generated by fitting the data using spline interpolation. Set an overflow threshold. (Usually set to 100% or 95%). Analytical solution of the equation. The smallest positive real root obtained is the predicted value at the overflow time. If the curve does not reach the threshold within the preset time period, it will be marked. It can be infinity or the end point of a preset time period.

[0122] S4. When the predicted overflow time value enters the warning range, the visual acquisition device associated with the sanitation facility is invoked to obtain real-time images and the visual saturation value is determined using a semantic segmentation algorithm. If the difference between the visual saturation value and the real-time sensor data of the sanitation facility is within the preset tolerance range, a confirmation warning signal is generated.

[0123] In one embodiment, step S4 specifically includes the following sub-steps S401 to S404:

[0124] S401, Real-time monitoring of the overflow time prediction value output by S3 Calculate the current time Compared with the predicted value of the overflow time Time difference .like Less than the preset warning response threshold If the timeframe is 30 minutes (e.g., 30 minutes), the system is considered to be within the warning range, triggering the verification process. Based on the device association mapping table, a capture command is sent to the visual acquisition device (such as a pole-mounted camera or a vehicle-mounted camera) aimed at the target sanitation facility to obtain real-time RGB image frames containing the opening area of ​​the sanitation facility.

[0125] S402. Input the real-time RGB image frame into a pre-trained semantic segmentation model (e.g., DeepLabV3+ or U-Net). The model classifies each pixel in the image into three categories: background, trash can container area, and garbage accumulation. The total number of pixels belonging to the trash can container area in the image is counted. And the total number of pixels belonging to the garbage accumulation located inside the container area and above the opening. Calculate visual saturation values. The calculation formula is: If it is detected that waste has exceeded the container area boundary and fallen to the ground, then directly... Set it to 100% and mark it as overflowing.

[0126] S403. Read the real-time sensor data of sanitation facilities obtained in S1, and convert it into a percentage value consistent with the visual numerical dimension, which is recorded as the sensor saturation value. .

[0127] If the sensor data is an ultrasonic ranging value Given the depth of the bucket ,but .

[0128] If the sensor data is a weight value Known maximum load ,but .

[0129] If multiple sensors exist simultaneously, the maximum value after normalization of each sensor is taken as the final value. To maintain sensitivity to overflow risks.

[0130] S404, Calculate the visual saturation value. With the aforementioned sensor saturation value absolute difference between Set tolerance threshold (For example, 15%).

[0131] Scenario 1 (Confirmation): If This indicates that the visual observation and physical sensing results are consistent, and the risk of overflow is determined to be real. A confidence warning signal is generated, and the process proceeds to step S5.

[0132] Scenario 2 (Sensor False Alarm): If and (For example, if the sensor shows 90% full, but the visual display shows only 20%), it is determined that there may be a hollow object blocking the sensor probe. Instead of generating a confirmation warning signal, a device fault verification work order is generated.

[0133] Scenario 3 (Visual Occlusion): If and (For example, if the visual display is overflowing and the sensor display is empty), it is determined that there may be dirt on the camera lens or abnormal lighting, and a manual review request is generated.

[0134] S5. In response to the confirmed early warning signal, calculate the garbage decomposition rate factor based on the real-time meteorological data and calculate the optimal interception time window by combining the associated road traffic flow data and the garbage saturation change curve. It generates dispatch instructions and plans the travel routes of the work vehicles, ensuring that the expected arrival time of the work vehicles at the target sanitation facility falls within the optimal interception time window. Inside, and It is less than the predicted value at the overflow time.

[0135] In one embodiment, step S5 specifically includes the following sub-steps S501 to S505:

[0136] S501, Extraction time window start point ( ).

[0137] Call the garbage saturation change curve generated in S3 Set a preset collection threshold. (For example, 75%). This threshold represents the minimum economic loading capacity for waste removal operations; below this percentage, waste removal is economically unprofitable. Iteration function. The search makes The earliest point in time when it was established is marked as the starting point of the interception time window. This means that the earliest the work vehicles can arrive is when the amount of garbage has accumulated to a level that is worth clearing.

[0138] S502, Calculate the garbage decomposition rate factor ( ).

[0139] Based on real-time meteorological data (atmospheric temperature (Temp) and relative humidity (Hum) acquired by S1, an environmental impact model is constructed to calculate the waste decomposition rate factor. Considering that high temperature and high humidity environments accelerate the anaerobic fermentation of organic waste, producing odors and leachate, this embodiment uses the following empirical formula for calculation. :

[0140]

[0141] in, and The baseline temperature and humidity (e.g., 25°C and 50%). and These are the weighting coefficients; This is the activation function used to map temperature differences to a non-linearly increasing risk value. When the temperature is low and dry, When the temperature is hot and humid, The higher the value, the higher the risk of garbage decay.

[0142] S503, Correct and determine the end point of the time window ( The initial time window endpoint is theoretically the overflow moment predicted in step S3. However, in actual operation, a safety buffer time must be reserved, which is affected by both the risk of corruption and traffic uncertainty.

[0143] Read the associated road traffic flow data from step S1 and calculate the traffic fluctuation variance. This is used to characterize the instability of road conditions. Calculate the safe buffer time. :

[0144]

[0145] Where K is the adjustment constant. This formula indicates that the more easily garbage decomposes ( Larger roads or more unpredictable road conditions The larger the size, the longer the safety buffer time is reserved.

[0146] Calculate the final interception time window end point :

[0147]

[0148] This determined the optimal interception time window. .

[0149] S504. Model the sanitation dispatching problem as a vehicle routing problem with time windows (VRPTW).

[0150] The objective function J is defined as minimizing the total transportation cost (fuel consumption + manpower time). A penalty function P(t) is introduced: if the vehicle's estimated arrival time... This results in an "idle load penalty" (going too early); if This results in an "overflow penalty" (arriving too late, leading to complaints of overflow or foul odor).

[0151] An improved ant colony algorithm or genetic algorithm is used to solve the objective function J and search for the optimal driving path sequence.

[0152] S505: Based on the optimal path planned in S504, generate specific dispatch instructions. The instructions include the target sanitation facility ID, suggested route, and mandatory arrival time period (i.e., ...). The system displays the command and its associated warning level (conviction warning). This command is pushed to the operators via the vehicle terminal or mobile app, and the vehicle navigation system automatically locks the optimal route to guide the vehicle there.

[0153] like Figure 2 As shown, the present invention also provides an intelligent early warning device for a smart sanitation platform. Based on the intelligent early warning method for a smart sanitation platform described above, the device includes:

[0154] The acquisition module 1 is used to acquire real-time sensor data of sanitation facilities, regional pedestrian flow heat data, real-time meteorological data, and related road traffic flow data within the target area; wherein, the target area contains multiple sanitation facilities;

[0155] Module 2 is used to construct an urban sanitation topology network based on the geographical location of the sanitation facilities within the target area, map each sanitation facility to a network node of the urban sanitation topology network, calculate the direction vector of pedestrian movement based on the temporal location change of the area's pedestrian flow heat data, and define the node it points to as the downstream node, so as to dynamically assign values ​​to the edge weights pointing to the downstream nodes.

[0156] The aggregation module 3 is used to input the real-time sensor data of the sanitation facilities, the regional human flow heat data and the real-time meteorological data into the pre-trained spatiotemporal prediction model, and use the dynamically assigned urban sanitation topology network to perform feature aggregation, and output the garbage saturation change curve of each sanitation facility in the future preset time period to extract the predicted value of the overflow time.

[0157] The early warning module 4 is used to call the visual acquisition device associated with the sanitation facility to acquire real-time images and use a semantic segmentation algorithm to determine the visual saturation value when the predicted value of the overflow time enters the early warning range. If the difference between the visual saturation value and the real-time sensor data of the sanitation facility is within the preset tolerance range, a certainty warning signal is generated.

[0158] Indication module 5 is used to respond to the confirmed warning signal by calculating the garbage decomposition rate factor based on the real-time meteorological data and combining it with the associated road traffic flow data and garbage saturation change curve to calculate the optimal interception time window. It generates dispatch instructions and plans the travel routes of the work vehicles, ensuring that the expected arrival time of the work vehicles at the target sanitation facility falls within the optimal interception time window. Inside, and It is less than the predicted value at the overflow time.

[0159] Each of the above modules is used to perform the corresponding steps in the intelligent early warning method of the smart sanitation platform. The specific implementation method is as described in the above method embodiment, and will not be repeated here.

[0160] like Figure 3 As shown, the present invention also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores all data required for the intelligent early warning method of the smart sanitation platform. The network interface is used for communication with external terminals via a network connection. The computer program is executed by the processor to implement the intelligent early warning method of the smart sanitation platform.

[0161] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0162] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described intelligent sanitation platform intelligent early warning methods.

[0163] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by hardware related to computer program instructions. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM), such as dynamic RAM (used as main storage) or static RAM (commonly used as cache memory). By way of illustration and not limitation, RAM has various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), and Rambus DRAM (RDRAM).

[0164] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0165] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A smart early warning method for a smart sanitation platform, characterized in that, include: S1. Acquire real-time sensor data of sanitation facilities, regional pedestrian flow heat data, real-time meteorological data, and related road traffic flow data within the target area; wherein, the target area contains multiple sanitation facilities; S2. Construct an urban sanitation topology network based on the geographical location of the sanitation facilities within the target area, map each sanitation facility to a network node of the urban sanitation topology network, calculate the direction vector of pedestrian movement based on the temporal changes in the location of the regional pedestrian flow heat data, and define the node it points to as the downstream node, so as to dynamically assign values ​​to the edge weights pointing to the downstream nodes. S3. Input the real-time sensor data of the sanitation facilities, the regional human flow heat data and the real-time meteorological data into the pre-trained spatiotemporal prediction model, use the dynamically assigned urban sanitation topology network to perform feature aggregation, and output the garbage saturation change curve of each sanitation facility in the future preset time period to extract the predicted value of the overflow time. S4. When the predicted value of the overflow moment enters the warning range, the visual acquisition device associated with the sanitation facility is invoked to obtain real-time images and the visual saturation value is determined using a semantic segmentation algorithm. If the difference between the visual saturation value and the real-time sensor data of the sanitation facility is within the preset tolerance range, a certainty warning signal is generated. S5. In response to the confirmed early warning signal, calculate the garbage decomposition rate factor based on the real-time meteorological data and calculate the optimal interception time window by combining the associated road traffic flow data and the garbage saturation change curve. It generates dispatch instructions and plans the travel routes of the work vehicles, ensuring that the expected arrival time of the work vehicles at the target sanitation facility falls within the optimal interception time window. Inside, and It is less than the predicted value at the overflow time.

2. The intelligent early warning method for the smart sanitation platform according to claim 1, characterized in that, S1 specifically includes: S101. Traverse all sanitation facilities within the target area, assign a unique equipment identification code to each sanitation facility, and record its static attribute data. S102. Periodically collect data using IoT sensor terminals deployed inside the sanitation facility and send it to the cloud server via a wireless communication network to obtain real-time sensor data of the sanitation facility. S103. Delineate a radiation zone with the sanitation facility as the center, and obtain the real-time population density value within the radiation zone as the population thermal data of the area. S104. Obtain the current atmospheric temperature, relative humidity, and precipitation type of the target area as the real-time meteorological data. S105. Identify the cleaning operation roads adjacent to each of the sanitation facilities, and obtain the real-time average vehicle speed and congestion index of the cleaning operation roads as traffic flow data of the associated roads. S106. The real-time sensor data of the sanitation facilities, the regional population flow heat data, the real-time meteorological data and the associated road traffic flow data are subjected to outlier removal, and spatiotemporal alignment processing is performed based on the preset time granularity to form a standardized multidimensional feature tensor.

3. The intelligent early warning method for the smart sanitation platform according to claim 1, characterized in that, S2 specifically includes: S201. Construct a static basic topology graph structure G=(V,E), mapping each sanitation facility to a vertex V in the graph, and construct connecting edges E based on the principle of geographical proximity, assigning initial weights to the connecting edges; wherein, the specific method for constructing connecting edges and assigning initial weights is as follows: calculate any two nodes and Euclidean distance between ;like Less than the preset distance threshold Then at node and Establish connecting edges between them; set the initial weights of the connecting edges. and Inversely proportional; S202. Obtain the regional pedestrian flow heat map data at the current time t and the previous time t-1, calculate the pedestrian flow movement direction vector in the grid cell using the image optical flow method or gradient calculation method, and map the pedestrian flow movement direction vector to the corresponding vertex. S203. For the starting node and the target node connected in the topology network, calculate the cosine value of the angle between the physical path vector and the direction vector of the flow of people at the starting node. If the cosine value of the angle is greater than zero, then determine that the target node is the downstream node of the starting node. S204. Based on the magnitude of the pedestrian movement direction vector and the cosine of the included angle, calculate the pedestrian impulse gain, and use the pedestrian impulse gain to correct the initial weight of the connecting edge pointing to the downstream node to obtain the dynamic weight.

4. The intelligent early warning method for the smart sanitation platform according to claim 3, characterized in that, In S204, the formula for calculating the dynamic weight is: in, Represents a node Pointing to node The dynamic weights of the connecting edges; Represents a node and nodes Initial weights between them; This represents the preset influence adjustment coefficient, used to adjust the proportion of pedestrian flow factors in the weighting calculation; Represents a node The magnitude of the vector representing the direction of pedestrian movement at a given location characterizes the intensity of pedestrian movement. Represents a node Pointing to node physical path vector With nodes The direction vector of pedestrian movement at the location The cosine of the angle between them.

5. The intelligent early warning method for the smart sanitation platform according to claim 1, characterized in that, S3 specifically includes: S301. Construct a spatiotemporal prediction model, which includes a graph spatial feature extraction layer, a temporal evolution layer, and an output layer connected in sequence. S302. Integrate the real-time sensor data of sanitation facilities, regional pedestrian flow heat data, real-time meteorological data and related road traffic flow data that have undergone spatiotemporal alignment processing into a node feature matrix, and construct a multi-dimensional feature input tensor. S303. Input the multidimensional feature input tensor into the graph space feature extraction layer, construct a dynamic adjacency matrix using the dynamic weights, and perform spatial feature aggregation on the node feature matrix. The aggregation operation formula is as follows: in, Indicates the feature state of the node in the l-th layer; This represents the feature state of a node in the (l+1)th layer; This indicates a dynamic adjacency matrix with self-loops, whose element values ​​are determined based on the dynamic weights and are dynamically updated as the time step changes. express The degree matrix; This represents the learnable weight parameter matrix of the l-th layer; Indicates the activation function; S304. Input the feature vector sequence after aggregating spatial features into the temporal evolution layer to capture the temporal evolution pattern of garbage generation, and generate a prediction sequence for a future preset time period through the output layer. S305. Smooth the predicted sequence using spline interpolation to generate a continuous function. As the curve showing the change in waste saturation; an overflow threshold is set. Solve the equation The smallest positive real root is used as the predicted value at the overflow time.

6. The intelligent early warning method for the smart sanitation platform according to claim 1, characterized in that, S4 specifically includes: S401. Monitor the predicted value of the overflow moment in real time, calculate the time difference between the current time and the predicted value of the overflow moment, and if the time difference is less than the preset warning response threshold, send a capture command to the visual acquisition device to obtain real-time RGB image frames. S402. Input the real-time RGB image frame into a pre-trained semantic segmentation model, classify and statistically analyze the image pixels, and calculate the visual saturation value based on the statistical results. The calculation formula is as follows: in, This represents the visual saturation value; This indicates the number of pixels in the real-time RGB image frame that are identified as garbage accumulation categories; This represents the number of pixels in the real-time RGB image frame that are identified as belonging to the trash can container region category; if the semantic segmentation model identifies that the accumulated trash extends beyond the boundary of the trash can container region, then the visual saturation value is directly increased. Set to 100%; S403. Read the real-time sensor data of the sanitation facilities, and normalize the real-time sensor data of the sanitation facilities according to the physical attribute parameters of the sanitation facilities to generate the sensor saturation value in percentage form; wherein, the sensor saturation value The calculation method includes: if the real-time sensing data of the sanitation facilities is an ultrasonic ranging value And the total depth of the sanitation facilities is ,but If the real-time sensor data of the sanitation facilities is a weight value And the maximum load-bearing capacity of the sanitation facilities is ,but ; S404. Calculate the absolute difference between the visual saturation value and the sensor saturation value. If the absolute difference is less than or equal to a preset tolerance threshold, determine that the visual recognition result is consistent with the sensor data and generate a confirmation warning signal.

7. The intelligent early warning method for the smart sanitation platform according to claim 1, characterized in that, S5 specifically includes: S501. Set a preset collection threshold, traverse the waste saturation change curve, and determine the moment when the waste saturation change curve first reaches the preset collection threshold as the starting point of the interception time window. S502. Based on the real-time meteorological data, construct an environmental impact model and calculate the waste decomposition rate factor to characterize the risk of waste decomposition; wherein, the calculation formula is: Wherein, Temp and Hum represent the atmospheric temperature and relative humidity in the real-time meteorological data, respectively; and These represent the preset reference temperature and reference humidity, respectively. and These represent the temperature weighting coefficient and the humidity weighting coefficient, respectively. Indicates the activation function; S503. Calculate the traffic fluctuation variance based on the associated road traffic flow data, calculate the safety buffer time in conjunction with the garbage decomposition rate factor, and subtract the safety buffer time from the predicted overflow time to obtain the endpoint of the interception time window, thereby determining the optimal interception time window; wherein, the endpoint of the interception time window... The calculation formula is: in, The value at the overflow moment is represented by K; K represents the preset adjustment constant. This represents the waste decomposition rate factor; This represents the traffic fluctuation variance calculated based on the associated road traffic flow data, used to characterize the instability of road conditions; S504. Construct a path planning objective function with time window constraints, introduce an empty load penalty term for arrival earlier than the start of the interception time window and an overflow penalty term for arrival later than the end of the interception time window into the path planning objective function, and solve for the optimal driving path sequence. S505. Generate a scheduling instruction containing a mandatory arrival time period based on the optimal driving route sequence, and send it to the work vehicle terminal.

8. The intelligent early warning method for the smart sanitation platform according to claim 7, characterized in that, In S504, the objective function for path planning is constructed as follows: The optimization objective is to minimize the total transportation cost, which includes fuel consumption cost and time cost. When the scheduled arrival time of the work vehicle At that time, the empty-load penalty item is added to the total transportation cost, wherein This is the starting point of the interception time window; When the scheduled arrival time of the work vehicle At that time, the overflow penalty item is added to the total transportation cost, wherein This is the end point of the interception time window; The objective function of the path planning is solved using either the ant colony algorithm or the genetic algorithm.

9. An intelligent early warning device for a smart sanitation platform, based on the intelligent early warning method for a smart sanitation platform according to any one of claims 1 to 8, characterized in that, The device includes: The acquisition module is used to acquire real-time sensor data of sanitation facilities, regional pedestrian flow heat data, real-time meteorological data, and related road traffic flow data within the target area; wherein, the target area contains multiple sanitation facilities; The construction module is used to construct an urban sanitation topology network based on the geographical location of the sanitation facilities within the target area, map each sanitation facility to a network node of the urban sanitation topology network, calculate the direction vector of pedestrian movement based on the temporal location change of the area's pedestrian flow heat data, and define the node it points to as the downstream node, so as to dynamically assign values ​​to the edge weights pointing to the downstream nodes. The aggregation module is used to input the real-time sensor data of the sanitation facilities, the regional human flow heat data and the real-time meteorological data into the pre-trained spatiotemporal prediction model, and use the dynamically assigned urban sanitation topology network to perform feature aggregation, and output the garbage saturation change curve of each sanitation facility in the future preset time period to extract the predicted value of the overflow time. The early warning module is used to call the visual acquisition device associated with the sanitation facility to acquire real-time images and use a semantic segmentation algorithm to determine the visual saturation value when the predicted value of the overflow time enters the early warning range. If the difference between the visual saturation value and the real-time sensor data of the sanitation facility is within the preset tolerance range, a certainty warning signal is generated. The indicator module is used to respond to the confirmed warning signal by calculating the garbage decomposition rate factor based on the real-time meteorological data and combining it with the associated road traffic flow data and garbage saturation change curve to calculate the optimal interception time window. It generates dispatch instructions and plans the travel routes of the work vehicles, ensuring that the expected arrival time of the work vehicles at the target sanitation facility falls within the optimal interception time window. Inside, and It is less than the predicted value at the overflow time.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images