Intelligent Scheduling Method and System for Medical Waste Robots Based on the Internet of Things

Abstract

This invention discloses an intelligent scheduling method and system for medical waste robots based on the Internet of Things (IoT), relating to the field of IoT technology. The method includes: installing an IoT sensing device group on the medical waste robot; starting the robot and processing data to obtain a multi-dimensional data stream of standard medical waste; obtaining medical waste production prediction information and simultaneously collecting the robot's working status information; determining a robot work scheduling scheme; controlling the robot to perform medical waste processing and full-process recording, obtaining a processing data chain, and tracing and managing the process through the data chain. This invention solves the technical problems of low efficiency, exposure risks, and difficulty in rationally scheduling waste processing tasks based on actual conditions in existing medical waste processing processes. It achieves intelligent and automated medical waste processing, improves processing efficiency, reduces occupational exposure risks for processing personnel, and enables rational scheduling based on actual waste conditions, achieving the technical effect of refined management throughout the entire process.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, and in particular to an intelligent scheduling method and system for IoT-based medical waste robots. Background Technology

[0002] Traditional medical waste disposal methods rely heavily on manual labor. Manual sorting, weighing, and transporting of medical waste is not only inefficient and costly, but also exposes waste handlers to direct contact with the waste, posing a high risk of occupational exposure.

[0003] Currently, while some simple auxiliary equipment is used for medical waste disposal, intelligent management and scheduling are lacking. Existing waste disposal methods struggle to accurately predict the volume, type distribution, and peak periods of medical waste generation, making it impossible to rationally allocate disposal tasks based on actual conditions. Furthermore, traditional methods cannot achieve full-process recording and traceability control of medical waste disposal, resulting in a lack of transparency, management loopholes, and an inability to meet the stringent requirements of the modern medical industry for efficient, safe, and environmentally friendly waste disposal. Summary of the Invention

[0004] This application addresses the technical problems of low efficiency, exposure risks, and difficulty in rationally scheduling waste disposal tasks based on actual conditions during existing medical waste disposal processes. This application solves these problems by installing IoT sensing devices on medical waste robots to collect multi-dimensional data streams of medical waste. After data cleaning and waste production prediction, key information is obtained. Combined with the robot's operating status, an ant colony algorithm is used to determine a scheduling scheme, control the robot to execute tasks, and record the entire process to form a data chain for traceability and control. This achieves intelligent and efficient scheduling and refined management of the entire medical waste disposal process, making it more scientific, safe, and environmentally friendly.

[0005] To address the aforementioned technical issues, this application proposes a technical solution for an intelligent scheduling method and system for medical waste robots based on the Internet of Things.

[0006] Firstly, this application provides an intelligent scheduling method for medical waste robots based on the Internet of Things (IoT). The method includes: installing an IoT sensing device group on the medical waste robot, the IoT sensing device group including a weight sensor, a type identification sensor, a GPS positioning device, an environmental sensor, and a video monitoring device; starting the medical waste robot and acquiring multi-dimensional data streams of medical waste in real time through the IoT sensing device group; cleaning and processing the multi-dimensional data streams of medical waste to obtain standard multi-dimensional data streams of medical waste; predicting waste production from the standard multi-dimensional data streams of medical waste to obtain medical waste production prediction information, while simultaneously acquiring the working status information of the medical waste robot; analyzing scheduling schemes based on the medical waste production prediction information and the working status information of the medical waste robot to determine a robot work scheduling scheme; controlling the medical waste robot to perform medical waste processing and full-process recording according to the robot work scheduling scheme to obtain a robot medical waste processing data chain, and performing full-process traceability and control through the robot medical waste processing data chain.

[0007] Secondly, this application provides an intelligent scheduling system for medical waste robots based on the Internet of Things (IoT). The system includes: a sensing device installation module for installing an IoT sensing device group on the medical waste robot, the IoT sensing device group including a weight sensor, a type identification sensor, a GPS positioning device, an environmental sensor, and a video monitoring device; a data stream acquisition module for starting the medical waste robot and acquiring multi-dimensional data streams of medical waste in real time through the IoT sensing device group, performing data cleaning processing on the multi-dimensional data streams of medical waste to obtain standard multi-dimensional data streams of medical waste; a work information acquisition module for predicting waste production from the standard multi-dimensional data streams of medical waste to obtain medical waste production prediction information, and simultaneously acquiring the working status information of the medical waste robot; a scheduling scheme determination module for analyzing scheduling schemes based on the medical waste production prediction information and the working status information of the medical waste robot to determine the robot's work scheduling scheme; and a full-process control module for controlling the medical waste robot to perform medical waste processing and full-process recording according to the robot's work scheduling scheme, obtaining a robot medical waste processing data chain, and performing full-process traceability control through the robot medical waste processing data chain.

[0008] This application proposes one or more technical solutions, which have at least the following technical effects:

[0009] This application identifies data sources by installing an IoT sensing device group on a medical waste robot. Then, it collects multi-dimensional data streams of medical waste in real time, employing techniques such as diversion identification, preprocessing, and anomaly detection to obtain a standard multi-dimensional data stream of medical waste. Next, it trains a time-series neural network using historical medical waste datasets to generate a medical waste prediction model and obtain predictive information on medical waste production. Subsequently, it combines the robot's working status information to define processing objectives and construct functions, using an ant colony algorithm to determine the robot's work scheduling scheme. While controlling the robot to execute tasks, it monitors road conditions in real time and adjusts the scheme accordingly, recording the entire process to form a data chain for traceability and control. This achieves intelligent scheduling and management of medical waste disposal, realizing intelligent and automated medical waste disposal, improving processing efficiency, reducing occupational exposure risks for processing personnel, and enabling reasonable scheduling based on actual waste conditions, achieving the technical effect of refined management throughout the entire process.

[0010] The foregoing outlines the intelligent scheduling method and system for IoT-based medical waste robots in this application. The following detailed embodiments will describe the steps of the technical solution in detail to enable those skilled in the art to have a clear and complete understanding of this application. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating the intelligent scheduling method for medical waste robots based on the Internet of Things provided in this application embodiment.

[0013] Figure 2 This is a schematic diagram of the structure of the intelligent scheduling system for medical waste robots based on the Internet of Things provided in this application embodiment.

[0014] Explanation of reference numerals in the attached diagram: 1. Sensing device installation module; 2. Data stream acquisition module; 3. Work information collection module; 4. Scheduling scheme determination module; 5. Full-process control module. Detailed Implementation

[0015] This application collects multi-dimensional data streams by installing IoT sensing devices on medical waste robots, and obtains standardized data after cleaning. Historical data is used to train a model to predict waste production information, and combined with the robot's working status, an algorithm is used to determine a scheduling scheme. The robot is controlled to execute tasks, and the system monitors road conditions in real time to correct the scheme. The entire process data chain is recorded for traceability and control, achieving intelligent scheduling and full-process management of medical waste disposal. This results in intelligent and automated medical waste disposal, improved processing efficiency, reduced occupational exposure risks for processing personnel, and the ability to rationally schedule waste based on actual conditions, achieving the technical effect of refined management throughout the entire process.

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0017] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0018] Example 1, as Figure 1 As shown, an intelligent scheduling method for medical waste robots based on the Internet of Things (IoT) includes:

[0019] Step A100: Install an IoT sensing device group on the medical waste robot. The IoT sensing device group includes a weight sensor, a type identification sensor, a GPS positioning device, an environmental sensor, and a video monitoring device.

[0020] Specifically, the process begins with equipment selection and hardware integration. Addressing the need for real-time acquisition of weight, type, location, environmental conditions, and visualized images in medical waste disposal scenarios, sensor equipment meeting medical environment standards was selected. The weight sensor employs a high-precision pressure sensing module with a range of 0-50kg and an accuracy error controlled within ±0.5%, ensuring the accuracy of medical waste weighing data and providing a data foundation for subsequent automatic billing functions. The type recognition sensor integrates a multispectral imaging module and a recognition module, capable of identifying 6 major categories and 23 subcategories of waste in the "Medical Waste Classification Catalog," achieving automatic classification through differences in spectral reflectance (e.g., infectious waste with a reflectance characteristic value >0.6 in the 450-550nm band). The GPS positioning device utilizes a centimeter-level differential positioning module, combined with an indoor UWB positioning system, ensuring accurate positioning of the robot in the complex hospital environment. With a depth of ≤10cm, it meets the precise navigation needs of waste collection points in different departments; the environmental sensor group includes temperature and humidity, and harmful gas (such as formaldehyde and ozone) monitoring modules, with temperature and humidity measurement ranges of 0-100%RH and -20℃-60℃ respectively, and accuracy of ±2%RH and ±0.5℃. The sensitivity of harmful gas detection reaches the ppb level, providing real-time early warning of abnormal changes in the medical waste storage environment; the video monitoring equipment is equipped with a 1080P high-definition camera, infrared night vision and a wide-angle lens, supporting 360° surround scanning, a video frame rate of 25fps, and a storage resolution of 1920×1080, ensuring visual traceability of the waste collection site. To facilitate the robot's recognition of the surrounding environment, lidar, depth cameras and ultrasonic sensors are also installed.

[0021] Next, the hardware installation and layout followed the principle of "functional zoning - anti-interference design." The weight sensor was embedded in the bottom load-bearing structure of the robot's storage compartment, rigidly connected to the compartment body, and sensed pressure changes through strain gauges. The type recognition sensor was installed above the storage compartment entrance, with its lens pointing vertically downwards towards the waste disposal opening to ensure frontal imaging and recognition during waste disposal. The GPS antenna was fixed to an unobstructed area on the top of the robot, and the environmental sensor group was integrated into the ventilation area on the side of the body to avoid heat interference from mechanical components. The video monitoring camera was mounted on the front of the body using an adjustable bracket, supporting remote angle adjustment. All sensors connected to the robot's main control chip via RS-485, USB-C, and other interfaces, using shielded cables to reduce electromagnetic interference, and an independent power supply module was configured to ensure stable power supply to the sensors.

[0022] Through the above steps, the IoT sensing device group and the medical waste robot achieve physical hardware integration, providing a reliable data foundation for subsequent data cleaning, waste production prediction and intelligent scheduling, while ensuring the realization of the self-weighing function.

[0023] Step A200: Start the medical waste robot and collect multi-dimensional data streams of medical waste in real time through the IoT sensing device group. Perform data cleaning processing on the multi-dimensional data streams of medical waste to obtain standard multi-dimensional data streams of medical waste.

[0024] In this embodiment of the application, the multidimensional data stream of medical waste refers to a collection of multidimensional and multi-type data collected in real time by an IoT sensing device group installed on a medical waste robot.

[0025] Optionally, after the medical waste robot is activated, the IoT sensing device group first collects multi-dimensional data in real time at a preset frequency (10Hz), forming a raw data stream containing weight, type, location, environmental parameters, and video images. When all sensors work synchronously, the weight sensor provides real-time feedback on the storage compartment load data with an accuracy of ±0.5%, the type identification sensor generates 25 frames of waste image data per second through a multispectral imaging module, the GPS positioning module, combined with the indoor UWB system, outputs centimeter-level coordinates (positioning error ≤10cm), the environmental sensor group simultaneously collects temperature and humidity (accuracy ±2%RH, ±0.5℃) and harmful gas concentration (sensitivity up to ppb level), and the video monitoring equipment records 1080P high-definition video at 25fps. This raw data is transmitted to the robot's main control chip via an RS-485 / USB-C interface and initially packaged into a timestamped binary data stream (approximately 1.2MB per cycle).

[0026] Then, the collected multi-dimensional medical waste data streams are sorted and identified by the IoT sensing device group to obtain multi-source medical waste data streams with weight, type, location, etc. Based on the characteristic information of each data stream, the corresponding multi-source medical waste data preprocessing procedure is determined. The multi-source data streams are processed through the preprocessing procedure to remove initial noise and format differences, resulting in usable multi-source medical waste data streams. Finally, the usable data streams are anomaly identified and cleaned to obtain standard medical waste multi-dimensional data streams. The specific steps are explained in detail in A210-A240.

[0027] Step A300: Perform waste production prediction on the standard medical waste multidimensional data stream to obtain medical waste production prediction information, and at the same time collect the working status information of the medical waste robot.

[0028] In this embodiment of the application, waste production prediction is achieved by training a medical waste prediction model using historical medical waste datasets. Based on this model, the multidimensional data stream of cleaned standard medical waste is analyzed to output prediction information such as the amount of medical waste generated by each department, the proportion of different types, and peak periods in the future.

[0029] In one embodiment of this application, firstly, a historical medical waste dataset is collected and time-series segmented and identified to obtain a historical medical waste sequence dataset. Then, a medical waste prediction model is generated by training a time-series neural network. Based on the prediction model, the multidimensional data stream of standard medical waste is analyzed to obtain medical waste production prediction information. The specific steps are described in detail in A310-A330.

[0030] While predicting waste production, the robot's built-in sensors collect real-time operational status information. On the hardware level, a power sensor monitors the remaining lithium battery power (accuracy ±1.5%), triggering a low-battery warning when the power level drops below 20%; an odometer records the robot's movement distance (error ≤0.5%), and calculates the current speed (resolution 0.1m / s) using GPS positioning data; a gyroscope and accelerometer sense the robot's posture (tilt angle error ≤2°) to determine if a collision or tipping has occurred; and a task status register provides real-time feedback on the current task type (e.g., "go to the 3rd floor laboratory to collect chemical waste"), progress (percentage), and fault codes (e.g., 0x01 indicates the sorting compartment door is stuck).

[0031] At the software level, status data is packaged (including timestamp, robot ID, and status parameter array) by the edge computing module (a distributed computing unit deployed locally on the medical waste robot or near its communication network, possessing real-time data processing and edge intelligent decision-making capabilities, and undertaking the core function of data preprocessing in the intelligent scheduling system, detailed in A210-A240). This data is then uploaded to the cloud via the MQTT protocol at a frequency of 5Hz (transmission latency ≤150ms), forming a working status information stream containing four dimensions: "battery level - location - task - fault" (each data entry is approximately 500B). The system performs real-time verification of the status data. When it detects three consecutive instances of no task progress updates or abnormal speed (>1.5m / s), it automatically triggers the robot's self-check program to ensure the real-time nature and accuracy of the working status information.

[0032] Through the above steps, a closed loop of "historical data training - real-time production prediction - multi-dimensional status monitoring" is achieved, providing dual data support for intelligent scheduling. Specifically, medical waste prediction information is used to proactively plan collection routes and frequencies, while robot working status information ensures the feasibility of dynamically adjusting scheduling schemes.

[0033] Step A400: Based on the medical waste production prediction information and the working status information of the medical waste robot, analyze the scheduling scheme and determine the robot's work scheduling scheme.

[0034] Specifically, the robot waste disposal objective is defined and an objective function is constructed. With medical waste production prediction information and robot working status information as constraints, the scheduling scheme space is obtained through ant colony algorithm analysis. Then, the robot work scheduling scheme is determined by global optimization within this space using the objective function. The specific steps are explained in detail in A410-A430.

[0035] Step A500: Control the medical waste robot to perform medical waste treatment and full-process recording according to the robot work scheduling scheme, obtain the robot medical waste treatment data chain, and perform full-process traceability and control through the robot medical waste treatment data chain.

[0036] Specifically, firstly, the medical waste robot is controlled to perform medical waste disposal according to the scheduling plan, the working road conditions are monitored in real time and obstacle information is identified, and the scheduling plan is corrected based on this information to obtain the robot work scheduling correction plan. The specific steps are explained in detail in A510-A530.

[0037] Throughout the entire medical waste disposal process, the robot records the entire process. Equipped with omnidirectional monitoring cameras and a data recording module, the robot records real-time video of each stage of waste collection, transportation, and storage in high-definition (1080P) at a frame rate of at least 25 frames per second. Simultaneously, the data recording module automatically records the timestamp, location information (accurate to the specific department and coordinates), and relevant task parameters (such as the type and weight of the collected waste) for each operation step. This data is stored in encrypted form on the robot's local storage device, with a storage capacity sufficient for at least 30 days of continuous recording. It is also uploaded in real-time to a cloud server via wireless communication technology for backup and further processing.

[0038] Based on the data recorded throughout the entire process, a robotic medical waste management data chain is generated. This data chain integrates all key information from the source of waste generation (department information) to the final processing endpoint (waste treatment center), forming a complete and traceable information chain. Each data node in the data chain contains detailed metadata, such as time, location, operation type, and responsible person, ensuring the integrity and accuracy of the data.

[0039] Finally, a robotic medical waste disposal data chain is used for end-to-end traceability and control. Those skilled in the art can use a dedicated management platform to input specific query criteria (such as time range, department name, waste type, etc.) to quickly locate and view relevant medical waste disposal records. For example, when an anomaly is detected in a batch of medical waste, the data chain can trace back to detailed information such as the department that generated the waste, the collection time, the transportation route, and the disposal method, allowing for timely problem identification and appropriate measures. Simultaneously, the data chain can provide strong data support for quality assessment, cost accounting, and compliance checks of medical waste disposal, enabling refined management and effective control of the entire medical waste disposal process.

[0040] Furthermore, step A200 in the method provided in this application embodiment includes:

[0041] A210: The multi-dimensional data stream of medical waste is divided and identified according to the IoT sensing device group to obtain a multi-source medical waste data stream.

[0042] A220: Determine the multi-source medical waste data preprocessing procedure based on the characteristic information of the multi-source medical waste data stream.

[0043] A230: Based on the multi-source medical waste data preprocessing program, the multi-source medical waste data stream is preprocessed to obtain a usable multi-source medical waste data stream.

[0044] A240: Perform anomaly identification and data cleaning on the available multi-source medical waste data stream to obtain a standard multi-dimensional medical waste data stream.

[0045] Specifically, after the medical waste robot is started, the IoT sensing device group collects multi-dimensional data streams of medical waste in real time at a frequency of 10Hz. After these data are transmitted to the robot's main control chip via RS-485 / USB-C interface, they are first sorted and identified: each data is tagged according to the unique ID of the sensor (such as weight sensor ID-W01, type identification sensor ID-T02), and split into 5 independent sub-data streams of weight, type, location, environment, and image.

[0046] Simultaneously, a CRC-16 checksum algorithm is employed to eliminate erroneous data during transmission: First, the sensor adds a 16-bit checksum to the original data frame (such as sensor output signals for weight, type, etc.). This checksum is generated by performing a modulo-2 division operation on the data frame using a preset generator polynomial (such as CRC-16-CCITT). When the main control chip receives the data frame, it extracts the data portion, recalculates the CRC checksum value, and compares it bit by bit with the checksum carried at the end of the frame. If the two do not match, the data frame is determined to have a transmission error (i.e., checksum failure). Frames with consecutive checksum failures are counted. When three consecutive frames fail checksums, these abnormal data frames are marked in red and placed in a pending state, triggering a retransmission mechanism or manual intervention for verification. This ensures that the pass rate of data frame checksums entering the subsequent processing flow is ≥99.9%, effectively eliminating erroneous data caused by electromagnetic interference and transmission errors, thereby forming a preliminary classified multi-source medical waste data stream.

[0047] Next, based on the differences in characteristics among the above five types of sub-data streams, the edge computing module built into the main control chip automatically matches the preprocessing program:

[0048] Step a: The weight data is fed into a Kalman filter algorithm (iteration period 50ms) to filter out high-frequency noise generated by mechanical vibration. Specifically, the weight estimate for the current moment is predicted based on the weight value of the previous moment and the robot's motion state (velocity, acceleration). Then, the sensor measured values ​​and the predicted values ​​are weighted and fused, and the weights are adaptively adjusted using the noise covariance matrix to filter out vibration noise with a frequency >10Hz (such as interference caused by the trolley's bumps). After processing, the noise amplitude of the weight data is reduced, providing stable input data for waste weighing and load balancing calculations.

[0049] Step b: The type image data is processed using a combination of histogram equalization and median filtering to improve the contrast of waste surface features and is then segmented into a standard input size of 224×224 pixels. Specifically, histogram equalization expands the dynamic range of image grayscale values ​​(e.g., expanding the grayscale distribution from [50,150] to [0,255]) to improve the contrast of waste surface textures (e.g., reflective markings on IV bags, metallic luster of needles); the median filtering algorithm targets salt-and-pepper noise (e.g., sensor noise) in the image by replacing the center pixel with the neighborhood median using a 3×3 pixel sliding window to remove discrete noise points. The processed image is further segmented into a standard size of 224×224 pixels to match the input requirements of subsequent classification models, ensuring that each image contains complete waste feature regions (e.g., waste samples in the center of the disposal port), thus improving the confidence of the type recognition algorithm.

[0050] Step c: The location data undergoes coordinate transformation based on the hospital's electronic map (WGS84 geographic coordinate system to local coordinate system, error ≤ 5cm), and is associated with floor and department information (e.g., the coordinates X, Y corresponding to "3rd floor laboratory"). The coordinate transformation process first converts latitude and longitude to Cartesian coordinates using a seven-parameter model (translation, rotation, and scaling parameters), and then performs secondary calibration by combining the coordinate offset of the hospital's electronic map (e.g., the coordinates of the 1st floor reference point X = 0cm, Y = 0cm), ultimately controlling the error within 5cm (meeting the navigation accuracy requirements of narrow corridor areas). Simultaneously, the system pre-generates a floor-department coordinate mapping table (e.g., the coordinate range X = 12000-13000cm, Y = 8000-9000cm corresponding to the 3rd floor laboratory), and automatically associates the department name corresponding to the current coordinates using spatial geometric algorithms (e.g., point-polygon inclusion detection), transforming the location data from simple numerical coordinates into meaningful information such as "3rd floor laboratory," supporting subsequent waste production statistics and task allocation by department.

[0051] Step d: Environmental data and video streams are synchronized via a hardware clock (due to different sampling frequencies, the former is 10Hz and the latter is 25fps, requiring a hardware clock synchronization mechanism to ensure the timing consistency of the two types of data). Specifically, the edge computing unit has a built-in high-precision crystal oscillator (clock error ±5ppm) or connects to a GPS timing signal, assigning a unified timestamp reference to each sensor and camera. When the environmental sensor collects data (e.g., 14:00:00.000), the video stream synchronously records frame data within 40ms before and after that moment (corresponding to 1 frame at 25fps). Through timestamp matching (allowing ±50ms error), the environmental parameters of "25℃, 60%RH" are bound to the waste disposal image at the same time (e.g., showing medical staff disposing of infectious waste at 14:00:00.030), forming spatiotemporally aligned correlated data. This synchronization ensures that subsequent tracing can quickly locate the corresponding time period's image records through environmental anomalies (e.g., a sudden temperature rise), providing accurate temporal data support for analyzing the correlation between the medical waste storage environment and safety risks.

[0052] The preprocessed data is uniformly packaged into JSON format (approximately 2KB per data entry), containing fields such as "timestamp, sensor ID, data value, and device status," forming a structured and usable multi-source medical waste data stream.

[0053] Finally, anomaly threshold analysis is performed on the available multi-source medical waste data stream to determine the anomaly judgment threshold. Based on this threshold, the abnormal multi-source medical waste dataset is identified and cleaned to finally obtain the standard medical waste multidimensional data stream. The specific steps are explained in detail in A241-A243.

[0054] Through the above steps, high-quality input data with controllable error and uniform format is provided for subsequent waste production prediction, solving the problem of fusion of multi-source heterogeneous data and supporting the accurate realization of automatic recording and full-process traceability functions.

[0055] Furthermore, step A240 in the method provided in this application embodiment includes:

[0056] A241: Perform anomaly threshold analysis on the available multi-source medical waste data streams respectively to determine the anomaly judgment threshold for multi-source medical waste data.

[0057] A242: Based on the anomaly judgment threshold of the multi-source medical waste data, perform anomaly judgment on the available multi-source medical waste data stream to obtain an abnormal multi-source medical waste dataset.

[0058] A243: Perform data cleaning processing on the abnormal multi-source medical waste dataset to obtain a standard medical waste multidimensional data stream.

[0059] In this embodiment of the application, the abnormal threshold analysis is a step of setting reasonable abnormal judgment criteria for data of each dimension by analyzing data characteristics and combining historical statistical patterns and industry standards for preprocessed multi-source medical waste data streams.

[0060] Optionally, firstly, anomaly threshold analysis is performed on data from different dimensions such as weight, type, location, and environment. For weight data, a physical range threshold is set (0-50kg, matching the maximum load-bearing capacity of the storage compartment), and a dynamic fluctuation threshold is calculated based on historical data, that is, the 95th percentile (usually 12%-15%) of the weight change rate at adjacent moments in the past 7 days is taken. When the real-time change rate is >15%, an anomaly warning is triggered.

[0061] The type identification data is based on the confidence distribution during the training of the classification model, with a confidence threshold of 0.9. Data below this value is labeled as ambiguous. The specific training process of the classification model is as follows:

[0062] Step e: Obtain multispectral image data of a large amount of medical waste after preprocessing and cutting (detailed in step b), and label these preprocessed image data with the corresponding waste type labels to form a training dataset.

[0063] Step f: Select a suitable deep learning architecture, such as ResNet-18, as the base model. Input the training dataset into the model and continuously adjust the model's parameters using the backpropagation algorithm to minimize the error between the predicted results and the labeled values. During training, use methods such as cross-validation to evaluate and optimize the model to prevent overfitting. The training process will continue for multiple rounds until the model's performance stabilizes.

[0064] Step g: Finally, a stable medical waste classification model is obtained. The input of this model is a preprocessed 224×224 pixel multispectral image of medical waste, and the output is the type of medical waste corresponding to the image, as well as the confidence level of the classification result. For example, the output might be "Infectious waste, confidence level 95%".

[0065] Location data, combined with the hospital's floor plan, was used to delineate 20 restricted areas (such as operating rooms and sterile pharmacies, corresponding to a set of coordinates on an electronic map), and a movement speed threshold of 1.5 m / s was set (exceeding this value may lead to collision risks). Environmental data was used to set safe zones according to the "Regulations on the Management of Medical Waste," such as temperature 2-40℃, humidity 30%-80% RH, and formaldehyde concentration ≤0.08 mg / m³. 3 Any value exceeding the specified range is considered an environmental anomaly. Integrating these thresholds yields the comprehensive anomaly assessment thresholds for multi-source medical waste data.

[0066] Next, based on the aforementioned thresholds, the edge computing module scans the available multi-source data streams point-by-point at a frequency of 10Hz. An abnormal data labeling process is executed through a logic judgment engine. For example, if the weight data shows 55kg (out of range) or a sudden increase from 10kg to 25kg within 10 seconds (a change rate of 150% > 15%), a weight anomaly is triggered. If the type identification result is "unclassified" with a confidence level of 0.85 (< 0.9), it is labeled as a vague type. If the location coordinates fall into a restricted area (e.g., the operating room coordinate range X = 5000-6000cm, Y = 3000-4000cm) or the speed is 1.8m / s (> 1.5m / s), it is judged as a location anomaly. If the environmental data shows a temperature of 45℃ (> 40℃) or formaldehyde of 0.1mg / m³, it is considered an abnormal location. 3 (>0.08mg / m 3 Data points meeting any of these conditions are recorded as exceeding environmental limits. Data points meeting any of these conditions are extracted into an anomalous multi-source medical waste dataset, which typically comprises 1.5%-3% of the preprocessed data and includes four main anomaly types: numerical out-of-bounds, classification ambiguity, location conflict, and environmental risk.

[0067] Finally, for abnormal datasets, the system initiates a hierarchical cleaning mechanism. For numerical anomalies such as weight, temperature, and humidity, the sliding window interpolation method is used first. If there is a valid value within 5 cycles (500ms) before and after the anomaly point, it is repaired by linear interpolation (e.g., if weight data is missing, the average of 12kg at the previous moment and 13kg at the next moment is taken as 12.5kg), with a repair success rate of 96.2%. If there is no valid value, it is marked as "N / A" and the sensor self-test is triggered (fault diagnosis is completed within 2 minutes).

[0068] Data with ambiguous types enters the secondary recognition process. The edge computing module uploads the image data to the cloud-based enhancement model (processing latency ≤300ms with computing power support). For samples whose confidence level is still <0.9 after review, the corresponding video clips (1 second before and after, 25 frames of video) are automatically associated, and a manual review work order is generated.

[0069] Anomaly data is repaired using a trajectory fitting algorithm. A Bézier curve is constructed based on at least three valid coordinates from the last 30 seconds to calculate reasonable coordinates for anomaly points (error ≤ 20cm), ensuring path continuity (e.g., coordinate jumps caused by lost elevator signals are smoothed by 40% after fitting). During the cleaning process, format standardization is performed simultaneously: weight data is uniformly retained to two decimal places, type data is mapped to the "Medical Waste Classification Catalog" code (e.g., HW01 represents infectious waste), location coordinates are converted to the hospital's local coordinate system (unit: cm), and environmental parameters are output according to national standard units (e.g., ℃, %RH).

[0070] The final generated standard medical waste multidimensional data stream is uploaded to the cloud via the MQTT protocol. Each data entry contains multiple standardized fields (time, department, type code, weight, temperature and humidity, coordinates, etc.).

[0071] The above steps lay a high-quality data foundation for the subsequent training of waste production prediction models and the operation of intelligent scheduling algorithms, realizing the key transformation from raw heterogeneous data to standardized and usable data, and supporting the realization of core functions such as full-process traceability and control, intelligent scheduling optimization, and automatic classification.

[0072] Furthermore, step A300 in the method provided in this application embodiment includes:

[0073] A310: Collect and acquire historical medical waste datasets, perform time-series segmentation and labeling on the historical medical waste datasets, and obtain historical medical waste sequence datasets.

[0074] A320: Use a time-series neural network to train a model on the historical medical waste sequence dataset to generate a medical waste prediction model.

[0075] A330: Based on the medical waste prediction model, perform waste production prediction on the standard medical waste multidimensional data stream to obtain the medical waste production prediction information.

[0076] In this embodiment of the application, the time series neural network is a neural network model used to process data with time dependencies.

[0077] Specifically, when obtaining information on predicting medical waste production, the first step is to build a historical data foundation. A historical medical waste dataset covering the past 12 months is extracted from a cloud database (fields include collection time, generating department, waste weight, type distribution, collection period, etc.). The data is then segmented into a time-series partitioning algorithm based on a granularity of "year-quarter-month-day-hour," forming a historical medical waste sequence dataset with one hour as the smallest time unit.

[0078] Based on the above dataset, an LSTM time series neural network is used to train the model, generating a medical waste prediction model with multi-task prediction capabilities. The specific steps are explained in detail in A321-A323.

[0079] After model deployment, rolling predictions are performed on the real-time input of standard medical waste multidimensional data streams. Using the current time as the base point, a sliding window of data from the previous 24 hours is extracted. Numerical features (such as weight, temperature, and humidity) are scaled to the [-1,1] interval using the Z-score normalization method. Categorical features (such as department and type code) are converted into one-hot encoding (for a categorical variable with n different categories, one-hot encoding creates a binary vector of length n, where only one element is 1 and the rest are 0, to represent the category to which the variable belongs) and then input into the prediction model.

[0080] The model outputs predictions for the next 4 hours, including: the predicted hourly waste generation for each department (e.g., the emergency department is expected to generate 18.5kg ± 1.5kg of hazardous waste between 15:00 and 16:00); dynamic changes in the distribution of waste types (e.g., the proportion of chemical waste during the afternoon surgical peak increases from 18% to 25%); and peak period warnings (three peak periods are identified within the next 24 hours, with probabilities of 93%, 89%, and 91%, respectively).

[0081] The prediction results are cross-validated with the patient volume data and operating room schedule of the hospital's HIS system. When the prediction confidence is less than 85%, a secondary prediction is triggered (by calling the data of the same time period in the past 7 days to augment the model input) to ensure that the error rate of the output medical waste production prediction information is reduced.

[0082] The above steps provide high-precision data support for the forward-looking planning of subsequent robot scheduling schemes.

[0083] Furthermore, step A320 in the method provided in this application embodiment includes:

[0084] A321: Use a time series neural network to train attribute identification on the historical medical waste sequence dataset to obtain a waste generation prediction model, a waste type distribution prediction model, and a waste generation peak period prediction model.

[0085] A322: The waste generation prediction model, waste type distribution prediction model, and waste generation peak period prediction model are merged to obtain the basic waste prediction model.

[0086] A323: Perform performance verification and optimization on the basic waste prediction model to generate the medical waste prediction model.

[0087] In this embodiment of the application, attribute labeling training is a training method for datasets with multiple attribute features, which allows the model to learn the correlation between different attributes and target prediction.

[0088] Specifically, firstly, a multi-task sub-model system is constructed using attribute labels for training. For the historical medical waste sequence dataset (1-hour time resolution), a multi-output time series neural network architecture based on LSTM is adopted. The input features are divided into three attribute spaces: time attributes (hour encoding 0-23, weekday / weekend identifier, seasonal factor), department attributes (60 departments are one-hot encoded and then reduced to 5 dimensions by PCA), and time-series attributes (mean output of the first 3 hours, variance of type distribution, and fluctuation coefficient of collection frequency). The network is designed with three independent output branches: the output quantity prediction branch outputs continuous values ​​(unit: kg / h) using the mean squared error (MSE) loss function; the type distribution branch outputs the probability distribution of 6 types of waste (summing up to 1) using classification cross-entropy loss; and the peak period branch outputs binary classification results (peak / off-peak, threshold 0.8) using binary cross-entropy loss.

[0089] During training, general temporal features are extracted by sharing the first two LSTM hidden layers (128 memory units each, Dropout rate 0.2), while the third hidden layer performs specific feature mapping for different task branches. After 100 rounds of iterative training (batch size 32, Adam optimizer learning rate 0.001), the prediction error of garbage generation is stabilized at ±8%, the prediction accuracy of type distribution reaches over 90%, and the F1 score for peak period identification is over 0.9, thus forming three independent prediction sub-models for garbage generation, type distribution, and peak period.

[0090] Next, the model merging stage is entered, where the network structures of the three sub-models are integrated in a lightweight manner: the shared first two feature extraction layers are retained, and the output layers of each task branch are merged into a unified output interface through a concatenation operation, enabling a single input to simultaneously generate three types of prediction results (output quantity value, type probability vector, and peak probability value). The merged basic waste prediction model maintains a 15-dimensional input dimension (5 dimensions of time + 5 dimensions of department + 5 dimensions of time series), and expands the output dimension to 8 dimensions (1 dimension of output + 6 dimensions of type probability + 1 dimension of peak probability). A confidence calibration layer is added to the model output to calibrate the peak period prediction probability to the [0.7, 0.95] confidence interval, ensuring the numerical stability of the multi-task output. Through testing on a cross-validation set (accounting for 20% of the dataset), the inference speed of the merged model (basic waste prediction model) is improved compared to the independent sub-models, and the performance degradation of each sub-task is controlled within 3%, which can meet the computing power constraints of edge computing nodes.

[0091] Finally, performance verification and optimization were performed using a dual process of "historical data backtesting + real-time scenario stress testing": First, historical data from the past three months was used for backtesting, comparing the model's predicted values ​​with the actual values. The mean absolute error (MAE) of the predicted waste generation was calculated to be 1.2 kg / h, the Kullback-Leibler divergence (KL divergence) of the type distribution was stable below 0.15, and the false positive rate (actual peak periods not identified) was ≤5%, while the false positive rate (off-peak periods misjudged) was ≤3%. For the weaker performance in the early morning period (low waste generation, large fluctuations), data augmentation techniques were used to expand the training samples (copying the data for this period and adding ±10% random noise), while the LSTM forget gate parameter was adjusted (increasing from 0.8 to 0.85) to enhance long-term memory capabilities. In a real-world hospital deployment environment, three departments were selected for 72-hour real-time validation. When there was a sudden surge in patient volume (e.g., the emergency department saw over 500 patients in a single day, a significant change from the usual level), the model's prediction error temporarily increased from the usual 8% to 12%, triggering a dynamic optimization mechanism—automatically adding data from similar abnormal scenarios from the past 7 days to the training set. After 5 rounds of rapid incremental training, the error returned to below 9%. The final medical waste prediction model possesses multi-dimensional prediction capabilities (simultaneously outputting output, type, and peak information), with over 90% of the data having a prediction confidence level ≥85%, providing high-precision forward-looking data support for intelligent robot scheduling.

[0092] Furthermore, step A400 in the method provided in this application embodiment includes:

[0093] A410: Define the robot waste disposal target, and construct the robot waste disposal target function based on the robot waste disposal target.

[0094] A420: Using the medical waste production prediction information and the working status information of the medical waste robot as constraints, the ant colony algorithm is used to analyze the scheduling scheme of the constraints to obtain the robot scheduling scheme space.

[0095] A430: The robot's waste disposal objective function is used to perform global optimization within the robot scheduling scheme space to determine the robot's work scheduling scheme.

[0096] In this embodiment, the waste disposal objective function is used to quantify the optimization objective in the intelligent scheduling of IoT-based medical waste robots. The robot scheduling scheme space refers to the set of all possible robot scheduling schemes under given constraints.

[0097] Specifically, firstly, the processing objectives are clarified and a mathematical model is constructed. Based on the actual needs of hospital medical waste disposal, the objectives of robotic waste disposal are defined, such as minimizing the total waste collection time (ensuring timeliness and avoiding infection risks), maximizing resource utilization (reducing robot idle mileage and lowering energy consumption), and balancing the load on each robot (preventing any robot from overworking and increasing the failure rate). Based on these objectives, the objective function for robotic waste disposal is constructed. Taking minimizing the total waste collection time as an example, its objective function can be expressed as: Where n is the number of medical waste collection points (e.g., n is 60 if a hospital has 60 departments), m is the number of robots, and t ij Let x represent the travel time of robot j from its current position to collection point i. ij The variables are 0-1 (1 indicates that robot j goes to collection point i, and 0 indicates that it does not go). Meanwhile, those skilled in the art can integrate other objectives (resource utilization, load balancing) into the objective function through weighted coefficients to form a comprehensive optimization objective.

[0098] Next, using medical waste production forecast information and robot working status information as constraints, an ant colony algorithm is employed for analysis. The medical waste production forecast information includes the amount and type distribution of waste generated by each department over the next 4 hours, as well as peak periods (e.g., the emergency department is expected to generate 18.5 kg of explosive waste between 15:00 and 16:00). The robot working status information includes current location coordinates, remaining storage compartment capacity, and battery level. The ant colony algorithm simulates the foraging process of ants, treating each robot's scheduling path as the walking path of an ant, and searching for feasible solutions in the robot scheduling scheme space. The algorithm iteratively calculates different scheduling schemes (e.g., robot 1 responsible for departments 1-20, robot 2 responsible for departments 21-40, etc.) using a pheromone update mechanism (initial pheromone concentration set to 1) and a state transition probability formula (calculated in conjunction with the objective function and constraints). After 50 iterations (convergence speed typically ≤10 minutes), a robot scheduling scheme space containing more than 1000 feasible schemes is generated.

[0099] Finally, a global optimization is performed within the scheme space using the robot waste disposal objective function. Each scheme in the robot scheduling scheme space is traversed and substituted into the objective function for calculation and evaluation. For example, Scheme 1 has a total waste collection time of 60 minutes, a resource utilization rate of 85%, and a load balancing coefficient of 0.9, resulting in a comprehensive score of 88 points. Scheme 2 has corresponding indicators of 55 minutes, 90%, and 0.85, resulting in a comprehensive score of 92 points. The scores of all schemes are compared, and the scheme with the highest score is selected as the final robot work scheduling scheme.

[0100] The robot work scheduling scheme with the highest score obtained through the above steps effectively achieves efficient, balanced, and energy-saving medical waste disposal.

[0101] Furthermore, step A500 in the method provided in this application embodiment includes:

[0102] A510: Controls the medical waste robot to perform medical waste disposal according to the robot work scheduling plan, and monitors and obtains the robot's working conditions in real time.

[0103] A520: Identify obstacles in the robot's working environment and determine obstacle information.

[0104] A530: Based on the road condition and obstacle information, the robot work scheduling scheme is modified to obtain a modified robot work scheduling scheme.

[0105] In one embodiment, when controlling a medical waste robot to perform medical waste disposal according to a work schedule, the robot first begins to move according to the predetermined work schedule while its working conditions are monitored in real time. The medical waste robot is equipped with various sensors, such as lidar, cameras, and ultrasonic sensors, to collect road condition information. Lidar scans the surrounding environment at a frequency of 100,000 to 200,000 data points per second, enabling precise distance measurement with centimeter-level accuracy; cameras capture images at a rate of 30-60 frames per second, providing data for subsequent visual analysis; and ultrasonic sensors assist in detecting nearby obstacles. The data from these sensors is transmitted to the robot's control system in real time. The control system integrates and processes this data to form a preliminary understanding of the current working conditions.

[0106] Next, obstacle identification is performed on the acquired robot's working path data to determine obstacle information. For LiDAR data, a point cloud-based obstacle identification algorithm is used. This algorithm first filters the raw point cloud data to remove noise and outliers to improve data quality. Then, a clustering algorithm divides the point cloud data into different clusters, each representing a possible obstacle. Commonly used clustering algorithms include DBSCAN (density-based spatial clustering), which divides clusters based on point cloud density and can adapt to obstacles of different shapes and sizes. For images captured by the camera, deep learning object detection algorithms, such as the YOLO (You Only Look Once) series, are used. The YOLO algorithm has high real-time performance, capable of detecting obstacles in images in a short time and providing their location and category information. The identification results from LiDAR and camera are fused to improve the accuracy and reliability of obstacle identification.

[0107] Then, the robot scheduling scheme is revised based on the determined road condition and obstacle information. The A* algorithm is used for path planning revision. The A* algorithm comprehensively considers the actual cost and estimated cost from the current position to the target position. First, the robot's current position and target position are used as input, and the road condition and obstacle information is transformed into obstacle areas on a map. Then, the A* algorithm searches the map for an optimal path that avoids the obstacles. During the search process, the algorithm continuously calculates the cost of each node, selecting the node with the lowest cost for expansion, until the target position is found or all possible nodes have been traversed. Based on the newly planned path, the original robot scheduling scheme is adjusted to form a revised robot scheduling scheme. For example, if the original scheme requires the robot to pass through an area blocked by obstacles, the revised scheme will allow the robot to choose a path to bypass that area, ensuring that the robot can safely and efficiently complete the medical waste disposal task.

[0108] Finally, after the robot completes the medical waste disposal task, the edge computing module initiates the automatic billing process. First, the robot collects and uploads key data via its built-in sensors, such as the weight and volume of various types of medical waste collected, the mileage traveled, and the working time. Upon receiving this data, the module calculates the cost according to pre-set billing rules. For waste collection, different unit prices are set for different types of waste, and fees are calculated based on the weight or volume collected; corresponding charging standards are also set for the robot's mileage and working time. Next, the module summarizes all costs to arrive at the total cost of the medical waste disposal. Finally, a detailed cost list is generated, including details of each cost and the total amount, and sent to the relevant management department or billing party, completing the automatic billing process.

[0109] By monitoring road conditions in real time and identifying obstacle information, the robot's work scheduling scheme is dynamically modified accordingly. This achieves intelligent obstacle avoidance of the robot's path and adaptive adjustment of the scheduling scheme during the medical waste disposal process, ensuring the realization of automatic billing function and effectively improving the safety of robot operations and the efficiency of task execution.

[0110] In summary, the intelligent scheduling method for medical waste robots based on the Internet of Things provided in this application has the following technical effects:

[0111] This application establishes a data transmission channel between the medical waste data preprocessing module and the intelligent scheduling decision-making module. It utilizes an anomaly threshold analysis algorithm to identify data anomalies, and then performs Kalman filtering for noise reduction and histogram equalization enhancement to obtain a standardized multidimensional data stream of medical waste. In the prediction model training module, time-series segmentation and attribute labeling training are performed. Through LSTM network iterative optimization and multi-model merging verification, combined with ant colony optimization and dynamic path correction mechanisms, a robot work scheduling scheme is generated and adjusted based on real-time road conditions and task requirements. This ensures efficient and orderly medical waste treatment, achieving intelligent and automated medical waste management, improving processing efficiency, reducing occupational exposure risks for processing personnel, and enabling reasonable scheduling based on actual waste conditions, thus achieving the technical effect of refined management throughout the entire process.

[0112] Example 2, as Figure 2 As shown, based on the same inventive concept as Embodiment 1 above, this application provides an intelligent scheduling system for medical waste robots based on the Internet of Things, the system comprising:

[0113] The sensing device installation module 1 is used to install an Internet of Things (IoT) sensing device group on the medical waste robot. The IoT sensing device group includes a weight sensor, a type recognition sensor, a GPS positioning device, an environmental sensor, and a video monitoring device.

[0114] The data stream acquisition module 2 is used to start the medical waste robot and collect multi-dimensional data streams of medical waste in real time through the Internet of Things sensing device group, and perform data cleaning processing on the multi-dimensional data streams of medical waste to obtain standard multi-dimensional data streams of medical waste.

[0115] The work information acquisition module 3 is used to predict the production of medical waste from the multi-dimensional data stream of the standard medical waste, obtain medical waste production prediction information, and at the same time collect the working status information of the medical waste robot.

[0116] The scheduling scheme determination module 4 is used to analyze the scheduling scheme based on the medical waste production prediction information and the working status information of the medical waste robot, and determine the robot's work scheduling scheme.

[0117] The full-process control module 5 is used to control the medical waste robot to perform medical waste treatment and full-process recording according to the robot work scheduling plan, obtain the robot medical waste treatment data chain, and perform full-process traceability and control through the robot medical waste treatment data chain.

[0118] Furthermore, the data stream acquisition module 2 is used to perform the following steps:

[0119] The IoT sensing device group performs diversion and identification on the multi-dimensional medical waste data stream to obtain a multi-source medical waste data stream; based on the characteristic information of the multi-source medical waste data stream, a multi-source medical waste data preprocessing procedure is determined; the multi-source medical waste data stream is preprocessed based on the multi-source medical waste data preprocessing procedure to obtain a usable multi-source medical waste data stream; anomaly identification and data cleaning are performed on the usable multi-source medical waste data stream to obtain a standard medical waste multi-dimensional data stream.

[0120] Furthermore, the data stream acquisition module 2 is used to perform the following steps:

[0121] The available multi-source medical waste data stream is analyzed for anomaly thresholds to determine the anomaly judgment thresholds for multi-source medical waste data. Based on the anomaly judgment thresholds for multi-source medical waste data, the available multi-source medical waste data stream is judged for anomalies to obtain an abnormal multi-source medical waste dataset. The abnormal multi-source medical waste dataset is then cleaned to obtain a standard multi-dimensional medical waste data stream.

[0122] Furthermore, the work information acquisition module 3 is used to perform the following steps:

[0123] A historical medical waste dataset is collected and time-series segmented to obtain a historical medical waste sequence dataset. A time-series neural network is used to train a model on the historical medical waste sequence dataset to generate a medical waste prediction model. Based on the medical waste prediction model, waste production prediction is performed on the standard medical waste multidimensional data stream to obtain the medical waste production prediction information.

[0124] Furthermore, the work information acquisition module 3 is used to perform the following steps:

[0125] The historical medical waste sequence dataset is trained using a time series neural network to identify attributes, thereby obtaining a waste generation volume prediction model, a waste type distribution prediction model, and a waste generation peak period prediction model. The waste generation volume prediction model, waste type distribution prediction model, and waste generation peak period prediction model are then merged to obtain a basic waste prediction model. The performance of the basic waste prediction model is then validated and optimized to generate the medical waste prediction model.

[0126] Furthermore, the scheduling scheme determination module 4 is used to perform the following steps:

[0127] Define the robot waste disposal objective, and construct a robot waste disposal objective function based on the robot waste disposal objective. Use the medical waste production prediction information and the working status information of the medical waste robot as constraints, and use the ant colony algorithm to analyze the scheduling schemes of the constraints to obtain the robot scheduling scheme space. Use the robot waste disposal objective function to perform global optimization in the robot scheduling scheme space to determine the robot work scheduling scheme.

[0128] Furthermore, the full-process control module 5 is used to perform the following steps:

[0129] The system controls the medical waste robot to perform medical waste processing according to the robot work scheduling plan, and monitors and acquires the robot's working road conditions in real time; it identifies obstacles in the robot's working road conditions and determines the road obstacle information; based on the road obstacle information, it corrects the robot work scheduling plan to obtain a corrected robot work scheduling plan.

[0130] The intelligent scheduling system for IoT-based medical waste robots provided in this embodiment of the invention can execute the intelligent scheduling method for IoT-based medical waste robots provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0131] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0132] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. An intelligent scheduling method for a medical waste robot based on the Internet of Things, characterized in that, The method comprises: installing an Internet of Things sensing device group on a medical waste robot, the Internet of Things sensing device group comprising a weight sensor, a type identification sensor, a GPS positioning device, an environmental sensor, and a video monitoring device; starting the medical waste robot and collecting medical waste multi-dimensional data streams in real time through the Internet of Things sensing device group, performing data cleaning processing on the medical waste multi-dimensional data streams to obtain standard medical waste multi-dimensional data streams; performing waste production prediction on the standard medical waste multi-dimensional data streams to obtain medical waste production prediction information, and collecting working state information of the medical waste robot; performing scheduling scheme analysis based on the medical waste production prediction information and the working state information of the medical waste robot to determine a robot working scheduling scheme; controlling the medical waste robot to perform medical waste treatment and whole-process recording according to the robot working scheduling scheme to obtain a robot medical waste treatment data chain, and performing whole-process traceability management through the robot medical waste treatment data chain; the standard medical waste multi-dimensional data streams are obtained by: performing shunt identification on the medical waste multi-dimensional data streams according to the Internet of Things sensing device group to obtain multi-source medical waste data streams; determining a multi-source medical waste data preprocessing program according to characteristic information of the multi-source medical waste data streams; preprocessing the multi-source medical waste data streams based on the multi-source medical waste data preprocessing program to obtain available multi-source medical waste data streams; performing abnormality identification and data cleaning on the available multi-source medical waste data streams to obtain standard medical waste multi-dimensional data streams; the medical waste production prediction information is obtained by: collecting a historical medical waste data set, performing time sequence division identification on the historical medical waste data set to obtain a historical medical waste sequence data set; using a time sequence neural network to perform model training on the historical medical waste sequence data set to generate a medical waste prediction model; performing waste production prediction on the standard medical waste multi-dimensional data streams based on the medical waste prediction model to obtain the medical waste production prediction information; the medical waste prediction model is generated by: performing attribute identification training on the historical medical waste sequence data set using a time sequence neural network, constructing a multi-task sub-model system using attribute identification training, for the historical medical waste sequence data set, using a multi-output time sequence neural network architecture based on LSTM, dividing input features into three attribute spaces of time attributes, department attributes, and time sequence attributes, designing three independent output branches of the network, the output branch of the quantity prediction branch outputs continuous values using a mean square error loss function, the output branch of the type distribution branch outputs probability distributions of six types of waste using a classification cross-entropy loss, the output branch of the peak period branch outputs a binary classification result using a binary cross-entropy loss, in the training process, common time sequence features are extracted through the shared first two layers of LSTM hidden layers, the third layer of hidden layers maps specific features for different task branches, and a waste quantity prediction model, a waste type distribution prediction model, and a waste production peak period prediction model are obtained. The garbage generation prediction model, the garbage type distribution prediction model and the garbage generation peak period prediction model are combined to obtain a basic garbage prediction model; The performance of the basic garbage prediction model is verified and optimized to generate the medical waste prediction model; The performance of the basic garbage prediction model is verified and optimized, including: The performance verification optimization is performed, and a double process of historical data backtracking verification and real-time scene stress testing is adopted. The historical data of the past three months are backtracked, the model prediction value is compared with the actual value, the average absolute error of the generation prediction, the Kullback-Leibler divergence of the type distribution, the missed detection rate and the false alarm rate of the peak period are calculated, and for the weak morning period, the data enhancement technology is used to expand the training samples, and the LSTM forgetting gate parameter is adjusted to strengthen the long-term memory ability.

2. The intelligent scheduling method of the medical waste robot based on the Internet of Things according to claim 1, characterized in that, The standard medical waste multi-dimensional data stream is obtained, including: The available multi-source medical waste data stream is respectively subjected to abnormal threshold analysis to determine a multi-source medical waste data abnormality judgment threshold; Based on the multi-source medical waste data abnormality judgment threshold, the available multi-source medical waste data stream is subjected to abnormal data judgment to obtain an abnormal multi-source medical waste data set; The abnormal multi-source medical waste data set is subjected to data cleaning processing to obtain a standard medical waste multi-dimensional data stream. 3.The intelligent scheduling method of the medical waste robot based on the Internet of Things according to claim 1, wherein, The robot working scheduling scheme is determined, including: A robot garbage processing target is defined, and a robot garbage processing target function is constructed according to the robot garbage processing target; The medical waste production prediction information and the working state information of the medical waste robot are used as constraint conditions, an ant colony algorithm is used to analyze the scheduling scheme of the constraint conditions to obtain a robot scheduling scheme space; Global optimization is performed in the robot scheduling scheme space by using the robot garbage processing target function to determine the robot working scheduling scheme.

4. The intelligent scheduling method for the medical waste robot based on the Internet of Things according to claim 3, characterized in that, The medical waste robot is controlled to perform medical waste processing according to the robot working scheduling scheme, including: The medical waste robot is controlled to perform medical waste processing according to the robot working scheduling scheme, and real-time monitoring is performed to obtain the robot working road condition; Obstacle identification is performed on the robot working road condition to determine road obstacle information; Based on the road obstacle information, the robot working scheduling scheme is modified to obtain a robot working scheduling modification scheme.

5. The intelligent scheduling system of medical waste robots based on the Internet of Things, characterized in that, The system for implementing the intelligent scheduling method of the medical waste robot based on the Internet of Things according to any one of claims 1-4, the system comprising: A perception device installation module is used to install an Internet of Things perception device group on the medical waste robot, and the Internet of Things perception device group includes a weight sensor, a type identification sensor, a GPS positioning device, an environment sensor and a video monitoring device; A data stream obtaining module is used to start the medical waste robot, and real-time collection of medical waste multi-dimensional data streams is performed through the Internet of Things perception device group. The medical waste multi-dimensional data streams are subjected to data cleaning processing to obtain standard medical waste multi-dimensional data streams. The work information collection module is configured to perform garbage production prediction on the standard medical garbage multidimensional data flow, obtain medical garbage production prediction information, and collect work state information of the medical garbage robot; The scheduling scheme determination module is configured to perform scheduling scheme analysis based on the medical garbage production prediction information and the work state information of the medical garbage robot, and determine a robot work scheduling scheme; The whole-process management and control module is configured to control the medical garbage robot to perform medical garbage treatment and whole-process recording according to the robot work scheduling scheme, obtain a robot medical garbage treatment data chain, and perform whole-process traceability management and control through the robot medical garbage treatment data chain.

Images