Intelligent material sorting method and system

Abstract

The invention provides an intelligent material sorting method and system, and the method comprises the steps: S4, matching a corresponding runner adjustment strategy from a preset dynamic regulation and control rule library according to the occurrence position and type of an abnormal state, optimizing a sorting path and an equipment scheduling scheme through a genetic algorithm, and obtaining an optimized material sorting sequence; s7, extracting processing records of similar scenes from the historical safety accident database according to the isolation state of the dangerous goods, evaluating the level of potential threats by adopting a decision tree algorithm, and determining the stability index of the overall operation of the system; and S9, according to the new regulation and control instruction, the working states of all the sorting devices are adjusted through a distributed calculation framework, the short-term change trend of material flowing is analyzed and predicted through a time sequence, and a stable operation scheme of the system in a high-load scene is obtained. The safety and efficiency of material sorting are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent warehousing technology, and in particular to an intelligent material sorting method and system. Background Art

[0002] Warehouse logistics sorting management, a key component of modern supply chain systems, directly impacts logistics efficiency and enterprise competitiveness. Its research and optimization are particularly important in the context of the rapid development of global trade and e-commerce. Reasonable sorting management not only improves warehouse operation efficiency but also effectively reduces operating costs and ensures the stability and security of material flow.

[0003] However, existing solutions often exhibit significant limitations when addressing increasingly complex logistics demands. Many traditional approaches rely too heavily on manual intervention or simple automated equipment, lacking the ability to respond to dynamic risks in real time. This is particularly true when dealing with diverse materials and high-throughput scenarios. Inadequate security policies can easily lead to system failures or reduced efficiency. These limitations highlight core challenges in warehouse logistics and sorting management that urgently need to be addressed.

[0004] Specifically, risk prevention and control during the material flow process, the real-time monitoring capabilities of the sorting system, and the isolation and handling mechanisms for special items (such as hazardous materials) are key difficulties in current technical optimization. Due to the imperfect risk prevention and control measures, easily dumped items may cause flow channel blockages and even cause safety accidents; the lack of efficient real-time monitoring methods makes it difficult to detect and handle abnormal conditions in a timely manner; in addition, the isolation flow channel design and emergency response mechanisms for hazardous materials management have not yet been fully standardized, resulting in potential threats that cannot be completely eliminated. These unresolved technical factors directly lead to safety hazards and efficiency bottlenecks in the sorting system under high load or complex scenarios, forming technical difficulties that need to be overcome urgently.

[0005] Therefore, building a dynamic control system that balances safety and efficiency in warehouse logistics sorting management has become a key research issue. To address material flow risks, safety checkpoints and flow safety warning systems are needed to achieve real-time identification of abnormal patterns and automatic adjustment of flow path parameters. Furthermore, for hazardous materials management, dedicated isolation flow paths and emergency response mechanisms are required to ensure the stability and safety of the overall system operation. Solving these issues will be key to improving warehouse logistics sorting management. Summary of the Invention

[0006] In order to solve the technical problems raised in the above background technology, the first aspect of the present invention is an intelligent material sorting method, which includes: S1, acquires material flow data through the sensor network. When a tipping-prone object is detected, the ratio of the material stack height to the flow channel width is calculated. If the ratio exceeds the safety factor threshold, the safety detection point is triggered to record the abnormal pattern and obtain a high-risk material identification; S2, based on the high-risk material identification, extracts the historical flow channel blockage records of the corresponding materials from the database, uses the random forest algorithm to analyze the probability of blockage, determines the need for flow channel parameter adjustment, and outputs the adjusted flow channel speed and sorting interval parameters; S3, obtains the adjusted flow channel speed and sorting interval parameters, and compares the deviation between the current flow channel operation status and the target state through the real-time monitoring system. If the deviation is greater than the set value, the flow safety warning system is activated to determine the location and type of abnormal state; S4, based on the location and type of abnormal state, matches the corresponding flow channel adjustment strategy from the preset dynamic control rule library, uses genetic algorithm to optimize the sorting path and equipment scheduling plan, and obtains the optimized material sorting sequence; S5, through the optimized material sorting sequence, detects whether there is a dangerous goods mark in the system. If so, it calls the dedicated isolation channel allocation algorithm to divert the dangerous goods to the isolation channel and outputs the isolation channel operation parameters and emergency plan; S6, obtaining the operating parameters and emergency plan of the isolation flow channel, analyzing the flow stability of the hazardous materials in the isolation channel through the real-time monitoring system, and adjusting the air pressure and separation distance of the isolation flow channel if the stability is lower than the stability threshold to determine the isolation status of the hazardous materials; S7, based on the isolation status of hazardous materials, extracts the handling records of similar scenarios from the historical safety accident database, uses a decision tree algorithm to assess the level of potential threats, and determines the stability indicators of the overall system operation; S8, obtaining the stability index of the overall operation of the system, comparing the matching degree between the current operation data and the stability index through the real-time monitoring system, and if the matching degree is lower than the matching degree threshold, updating the parameter threshold in the dynamic control rule base and outputting a new control instruction; S9, based on the new control instructions, adjusts the working status of all sorting equipment through the distributed computing framework, uses time series analysis to predict the short-term change trend of material flow, and obtains a stable operation plan for the system under high-load scenarios.

[0007] Optionally, step S1 acquires material flow data through a sensor network, and when a tippable object is detected, calculates the ratio of the material stack height to the flow channel width. If the ratio exceeds a safety factor threshold, a safety detection point is triggered to record an abnormal mode, thereby obtaining a high-risk material identification, including: Step S11, collecting material flow data through a laser rangefinder and a photoelectric sensor network to obtain a first data stream; Step S12, performing denoising processing on the first data stream using Gaussian filtering to obtain a second data stream; Step S13, using an edge detection operator to extract object contour features from the second data stream, and determining an easily dumpable object identifier based on an aspect ratio threshold; Step S14, obtaining the stacking height measurement value and the flow channel width measurement value corresponding to the easily dumpable item identifier from the warehouse database, and calculating the ratio of the stacking height measurement value to the flow channel width measurement value; Step S15: If the ratio of the stack height measurement value to the flow channel width measurement value exceeds a preset safety factor threshold, an isolation forest algorithm is used to detect abnormal patterns and obtain an abnormality type code; Step S16: Add a new record to the detection log table according to the abnormality type code and generate a high-risk material identification; Step S17: using the DBSCAN algorithm to group the high-risk material identifiers according to their location coordinates to obtain a material risk heat map; Step S18: adjusting the collection interval of the corresponding sensor according to the coordinates of the high-risk area in the material risk heat map.

[0008] Optionally, the step S17 uses the DBSCAN algorithm to group the high-risk material identifiers according to their location coordinates to obtain a material risk heat map, including: Step S171, obtaining material location data with longitude and latitude coordinates from the warehouse database; Step S172, clustering the coordinates using the DBSCAN algorithm; Step S173, obtaining preliminary spatial grouping results; Step S174, calculating a risk concentration code for each of the groups; Step S175: If the risk concentration code exceeds the risk concentration code threshold, the inbound and outbound records of the corresponding material ID in the past specified time period are extracted from the warehouse database; Step S176: Calculate the coordinate adjustment amount of the abnormal flow material based on the inventory in and out records. If the daily flow exceeds two standard deviations of the average value, move the material coordinates toward the group center point by a fixed percentage of the abnormal number of times to obtain an adjusted coordinate data set. Step S177, performing secondary clustering on the adjusted coordinate data set using an improved DBSCAN algorithm, optimizing the neighborhood radius using the silhouette coefficient, and obtaining a final risk grouping result; Step S178: Map the risk concentration code of each final risk group into a thermal value, wherein the thermal value is linearly positively correlated with the risk concentration code, and output a dynamic thermal map.

[0009] Optionally, in step S172, clustering the coordinates using the DBSCAN algorithm includes: determining the neighborhood radius by the inflection point of the k-distance graph, and setting the minimum number of samples to a fixed proportion of the total data volume.

[0010] Optionally, step S174, calculating a risk concentration code for each of the groups, including: the risk concentration code is defined as the sum of the squares of the risk levels of the materials in the group divided by the area of ​​the group.

[0011] Optionally, step S3, obtaining the adjusted flow channel speed and sorting interval parameters, comparing the deviation between the current flow channel operation state and the target state through the real-time monitoring system, and activating the flow safety warning system if the deviation is greater than a set value to determine the location and type of the abnormal state, includes: Step S31, obtaining the original data stream of the flow channel speed and sorting interval through the real-time monitoring system; Step S32: for the original data stream, extract the speed mean value within the past specified time window as the first feature and the sorting interval standard deviation as the second feature; Step S33, respectively calculating the absolute difference between the first feature and a preset speed and the ratio of the second feature to a preset sorting interval to obtain a deviation vector; Step S34: If the Euclidean distance of the deviation vector exceeds the Euclidean distance threshold, an IsolationForest algorithm is used to detect abnormal points and obtain an abnormality probability; Step S35: generating a JSON alert including a device number and a timestamp for the data point whose abnormal probability is greater than the abnormal probability threshold; Step S36, obtaining the operating parameters corresponding to the device number in the JSON alarm from the real-time database; Step S37, performing Z-score normalization on the operating parameters to obtain normalized parameters; Step S38, using a K-Means clustering model to classify the standardized parameters to obtain cluster labels; Step S39, selecting preset PID control parameters according to the cluster labels; Step S310, adjusting the inverter output frequency and the sorting motor pulse interval through the PLC controller to obtain an adjusted device data stream; Step S311: Enter the next feature extraction cycle for the adjusted device data stream.

[0012] Optionally, the step S39 selects preset PID control parameters according to the cluster label, including: using a first speed gain coefficient when the cluster label is the first category, and using a first interval compensation amount when the cluster label is the second category.

[0013] Optionally, step S6, obtaining the operating parameters and emergency plan of the isolation channel, analyzing the flow stability of the hazardous materials in the isolation channel through a real-time monitoring system, and adjusting the air pressure and separation distance of the isolation channel if the stability is lower than a stability threshold, to determine the isolation status of the hazardous materials, includes: Step S61, obtaining the operating parameters of the isolation flow channel, collecting data through the pressure sensor and the flow sensor, and calculating the flow stability of the dangerous goods using a linear regression model to obtain an initial stability value; Step S62: continuously sampling the initial stability value through the real-time monitoring system, and performing time series analysis using the ARIMA model to determine the stability change trend; Step S63: If the stability change trend is lower than the stability change trend threshold, the air pressure value of the isolation flow channel is adjusted by the PID controller to obtain an adjusted air pressure parameter; Step S64, calculating the change in the separation distance based on the adjusted air pressure parameter and determining a new separation distance value; Step S65: collecting flow data at the new separation distance value through the monitoring system, and using a linear regression model to analyze the flow stability after adjustment to determine whether a stable state has been reached; Step S66: If the adjusted flow stability is still lower than the stability threshold, obtain the preset strategy in the emergency plan database and determine the backup isolation plan; Step S67: For the backup isolation scheme, finite element analysis is used to simulate the isolation state of the dangerous goods to determine the final isolation effect.

[0014] Optionally, the step S64, calculating the change in the separation distance based on the adjusted air pressure parameter to determine a new separation distance value, includes: The change in separation distance is calculated using the following formula: Where ΔS represents the change in separation distance, K represents the correction coefficient, Indicates standard atmospheric pressure, Indicates the current air pressure. Indicates the current temperature. represents the standard temperature, and h represents the initial distance.

[0015] A second aspect of the present invention provides an intelligent material sorting system, which uses the above method to intelligently sort materials, and the system includes: The sensor data acquisition module is used to obtain material flow data through the sensor network. When a tipping-prone object is detected, the ratio of the material stack height to the flow channel width is calculated. If the ratio exceeds the safety factor threshold, the safety detection point is triggered to record the abnormal mode and obtain a high-risk material identification; The high-risk material analysis module is used to extract the historical flow channel blockage records of the corresponding materials from the database based on the high-risk material identification, use the random forest algorithm to analyze the probability of blockage, determine the need for flow channel parameter adjustment, and output the adjusted flow channel speed and sorting interval parameters; The flow channel parameter adjustment module is used to obtain the adjusted flow channel speed and sorting interval parameters. The real-time monitoring system compares the deviation between the current flow channel operation status and the target state. If the deviation is greater than the set value, the flow safety warning system is activated to determine the location and type of abnormal state. The flow safety early warning module is used to match the corresponding flow channel adjustment strategy from the preset dynamic control rule library according to the location and type of abnormal state, and use genetic algorithms to optimize the sorting path and equipment scheduling plan to obtain the optimized material sorting sequence; The sorting path optimization module is used to detect whether there are any hazardous materials signs in the system through the optimized material sorting sequence. If so, it calls the dedicated isolation channel allocation algorithm to divert the hazardous materials to the isolation channel and output the operating parameters and emergency plan of the isolation channel; The hazardous materials isolation module is used to obtain the operating parameters and emergency plans of the isolation flow channel. It analyzes the flow stability of hazardous materials in the isolation channel through a real-time monitoring system. If the stability is lower than the stability threshold, it adjusts the air pressure and separation distance of the isolation flow channel to determine the isolation status of the hazardous materials. The flow stability analysis module is used to extract similar scenario processing records from the historical safety accident database based on the isolation status of hazardous materials, use a decision tree algorithm to evaluate the level of potential threats, and determine the stability indicators of the overall system operation; The threat assessment module is used to obtain the stability index of the overall system operation. It compares the matching degree of the current operation data with the stability index through the real-time monitoring system. If the matching degree is lower than the matching degree threshold, it updates the parameter threshold in the dynamic control rule base and outputs new control instructions. The system control module is used to adjust the working status of all sorting equipment through a distributed computing framework according to new control instructions, use time series analysis to predict short-term changes in material flow trends, and obtain a stable operation plan for the system under high-load scenarios.

[0016] The technical solution provided by the embodiment of the present invention has the following beneficial effects: The present invention provides an intelligent material sorting method and system, which monitors material flow data in real time through a sensor network. When easily tippable items are detected, the ratio of stacking height to flow channel width is calculated, triggering the recording of safety detection points. Based on the identification of high-risk materials, the system uses a random forest algorithm to analyze the blockage probability and adjust the flow channel parameters. When encountering an abnormal state, the adjustment strategy is matched from a preset rule library, and a genetic algorithm is used to optimize the sorting path. For dangerous goods, a dedicated isolation flow channel is activated and its stability is monitored in real time. Potential threats are evaluated using a decision tree algorithm to ensure the overall stable operation of the system.

[0017] The present invention can also update control rules based on real-time data, and use time series analysis to predict short-term change trends, thereby achieving stable operation under high-load scenarios and greatly improving the safety and efficiency of material sorting. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The figure is a flow chart of an intelligent material sorting method of the present invention.

[0019] Figure 2 This is a structural diagram of an intelligent material sorting system of the present invention. DETAILED DESCRIPTION

[0020] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] like Figure 1 As shown, the first aspect of the present invention provides an intelligent material sorting method, the method comprising: S1 obtains material flow data through the sensor network. When detecting items that are easy to tip over, it calculates the ratio of the material stacking height to the flow channel width. If the ratio exceeds the safety factor threshold, the safety detection point is triggered to record the abnormal mode and obtain a high-risk material identification.

[0022] Optionally, this step also includes: Step S11: Collect material flow data using a laser rangefinder and a photoelectric sensor network to obtain a first data stream. Step S12: De-noise the first data stream using a Gaussian filter to obtain a second data stream. Step S13: Use an edge detection operator to extract item contour features from the second data stream and identify easily tippable items based on an aspect ratio threshold. Step S14: Obtain the stack height and flow channel width measurements corresponding to the easily tippable item identifiers from the warehouse database and calculate the ratio of the stack height to flow channel width measurements. Step S15: If the ratio of the stack height to flow channel width measurements exceeds a preset safety factor threshold, detect abnormal patterns using an isolation forest algorithm to obtain an abnormality type code. Step S16: Add a new record to the detection log table based on the abnormality type code to generate a high-risk material identifier. Step S17: Use the DBSCAN algorithm to group the high-risk material identifiers by location coordinates to obtain a material risk heat map. Step S18: Adjust the corresponding sensor acquisition interval based on the coordinates of the high-risk areas in the material risk heat map.

[0023] Specifically, a laser rangefinder and a photoelectric sensor network are used to collect material flow data to generate the first data stream. This method, which leverages the high precision of laser ranging and the real-time performance of photoelectric sensors, is widely used in warehousing scenarios.

[0024] For example, in an automated warehouse, laser rangefinders can measure the height and position of stacked materials, while photoelectric sensors monitor the dynamic changes of materials as they pass through conveyor belts. Assuming that 10 items pass by a conveyor belt per second, the sensor network collects distance and light intensity data every 0.1 seconds, generating a first data stream containing timestamps and measurement values. This approach effectively captures the real-time status of materials and provides a foundation for subsequent analysis. Gaussian filtering is used to denoise the first data stream, generating a second data stream. Gaussian filtering smoothes data to reduce noise interference and is commonly used to address random fluctuations in sensor data.

[0025] For example, if the first data stream contains noise due to ambient light or device jitter, resulting in stack height measurements fluctuating around 10 centimeters, Gaussian filtering can smooth out these anomalies, preserve trend characteristics, and output a more stable second data stream. This improves the accuracy of subsequent feature extraction and prevents noise from misleading judgments. Edge detection operators are used to extract item outline features from the second data stream, and easily tippable items are identified based on an aspect ratio threshold.

[0026] Edge detection is often used to identify object boundaries. For example, a filtered data stream shows that a material is 20 cm high, 5 cm wide, and has an aspect ratio of 4. If the preset aspect ratio threshold is 3, the material is marked as a tippable item.

[0027] Preferably, verification can be performed from multiple angles, such as by combining data from side sensors to determine whether the contour is distorted due to viewing angle deviation. This method can accurately identify potentially risky materials and improve warehouse safety. The stack height and runner width measurements corresponding to the tippable item identification are obtained from the warehouse database, and the ratio of the stack height measurement to the runner width measurement is calculated.

[0028] For example, the database records that the stack height of a certain easily tipping item is 50 cm, the flow channel width is 20 cm, and the ratio is 2.5. This ratio reflects the stability of the material, and a larger value indicates a higher risk of tipping.

[0029] It should be noted that the database needs to be updated in real time to ensure that the data is consistent with the field, thereby ensuring the reliability of the calculation results. If the ratio of the stack height to the flow channel width exceeds the safety factor threshold, the isolation forest algorithm is used to detect abnormal patterns and obtain the abnormal type code.

[0030] Isolation Forest excels at detecting outliers in data. For example, if the safety factor threshold is 2, and the ratio of a material's stack height to its runner width is 2.5, Isolation Forest will analyze the distribution differences between this and other material data and potentially output a "highly outlier" code. This method quickly locates anomalies and improves risk identification efficiency. Based on the anomaly type code, a new record is added to the detection log table to generate a high-risk material identifier.

[0031] For example, a material is recorded as "highly abnormal" and the identifier "H001" is generated.

[0032] In one embodiment, the log table can also record information such as time and location for easy traceability. This recording method provides data support for subsequent analysis. The DBSCAN algorithm is used to group high-risk material identifiers by location coordinates to generate a material risk heat map. DBSCAN is based on density clustering and does not require a preset number of clusters.

[0033] For example, if 10 high-risk materials in a warehouse are distributed between coordinates (1,1) and (5,5), the algorithm will divide them into two high-risk areas. Heat maps visually display risk concentration areas, helping to optimize management. Based on the coordinates of the high-risk areas in the material risk heat map, the corresponding sensor collection interval is adjusted.

[0034] For example, the heat map shows that the risk in the (3,3) area is high, and the original collection interval of 0.1 seconds can be adjusted to 0.05 seconds.

[0035] For example, this dynamic adjustment can improve monitoring accuracy, provide timely warnings of potential problems, and significantly reduce the incidence of dumping accidents.

[0036] Optionally, the step S17, using the DBSCAN algorithm to group the high-risk material identifiers according to their location coordinates to obtain a material risk heat map, further includes: Step S171, obtain material location data with longitude and latitude coordinates from the warehouse database. Step S172, cluster the coordinates using the DBSCAN algorithm, where the neighborhood radius eps is determined by the inflection point of the k-distance graph, and the minimum number of samples min_samples is set to a fixed proportion of the total data volume. Step S173, obtain the preliminary spatial grouping result. Step S174, calculate the risk concentration code for each of the groups, and the risk concentration code is defined as the sum of the squares of the risk levels of the materials in the group divided by the group area. Step S175, if the risk concentration code exceeds the risk concentration code threshold, extract the entry and exit records of the corresponding material ID in the past specified time period from the warehouse database. Step S176, calculate the coordinate adjustment amount of the abnormal flow material based on the entry and exit records. If the single-day flow exceeds twice the standard deviation of the average value, the material coordinates are moved toward the center point of the group by a fixed percentage of the number of abnormal times to obtain an adjusted coordinate data set. In step S177, the adjusted coordinate dataset is subjected to secondary clustering using an improved DBSCAN algorithm, and the neighborhood radius is optimized using the silhouette coefficient to obtain the final risk grouping results. In step S178, the risk concentration code of each final risk group is mapped to a thermal value, wherein the thermal value is linearly positively correlated with the risk concentration code, and a dynamic heat map is output.

[0037] Specifically, obtaining material location data with longitude and latitude coordinates from the warehouse database is the basis of the entire process.

[0038] For example, in a large automated warehouse, each pallet of materials is equipped with a location tag that records its latitude and longitude coordinates in real time, such as Material A located at (120.5, 30.2). This data is uploaded to a database via a wireless network, providing raw input for subsequent analysis. When clustering coordinates using the DBSCAN algorithm, the neighborhood radius eps is determined by the inflection points of the k-distance graph, and the minimum number of samples min_samples is set as a fixed proportion of the total data volume.

[0039] It can be understood that the k-distance graph calculates the distance from each point to the k-th nearest neighbor and sorts them, and the inflection point reflects the dividing line of density change.

[0040] For example, if there are 1000 material points in a warehouse, the minimum sample size min_samples is set to 2%, that is, 20 points, and the neighborhood radius eps may be determined to be 5 meters. This method can adapt to material layouts with different density distributions and obtain preliminary spatial grouping results.

[0041] For example, a group may contain 50 material points near the edge of the shelf. A risk concentration code is calculated for each group, which is defined as the sum of the squares of the risk levels of the materials in the group divided by the area of ​​the group.

[0042] Specifically, suppose a group contains 10 materials with risk levels of 1, 2, and 3, respectively, with a total sum of squares of 30 and an area of ​​100 square meters. The risk concentration code is 0.3. This code quantifies the concentration of risk within the area. If the code exceeds a preset risk concentration threshold, such as 0.5, the database retrieves the inbound and outbound records for the corresponding material ID over the past 30 days.

[0043] For example, if records for material B show frequent movement, this may increase risk. When calculating coordinate adjustments for materials with abnormal movement based on inbound and outbound records, if the daily movement exceeds two standard deviations of the mean, the coordinates are adjusted.

[0044] For example, if material C has an average daily flow of 5 times and a standard deviation of 2, and on a particular day it flows 10 times, exceeding the flow threshold by 9 times, its coordinates will be shifted toward the cluster center by 10% of the number of abnormal times, for example, from (120.6, 30.1) to (120.58, 30.12). This adjustment reflects the material's actual risk distribution. A modified DBSCAN algorithm is used to perform secondary clustering on the adjusted coordinate dataset, optimizing the neighborhood radius using the silhouette coefficient.

[0045] For example, the silhouette coefficient evaluates the clustering compactness, and the neighborhood radius eps is preferably adjusted from 5 meters to 4 meters to obtain a more accurate final risk grouping result.

[0046] Preferably, this method can better distinguish materials in overlapping areas. The risk concentration codes of the final risk grouping are mapped to thermal values, and the thermal values ​​are linearly positively correlated with the codes, and a dynamic thermal map is output.

[0047] In one embodiment, a code of 0.3 is mapped to a thermal value of 30, and a code of 0.6 is mapped to 60.

[0048] For example, a heat map can show areas near exits as red, indicating high-risk areas. This visualization helps warehouse managers quickly identify problems.

[0049] It should be noted that the generation of dynamic heat maps depends on real-time updated coordinates and recorded data.

[0050] For example, if a certain area experiences elevated heat levels due to frequent inflows and outflows, managers can adjust the layout or increase monitoring accordingly. This multi-level analysis and adjustment approach ensures comprehensive and practical risk identification.

[0051] In one possible implementation, if new materials are added to the warehouse, the database can be dynamically updated and re-clustered.

[0052] For example, when a new material D is added, the coordinates (120.7, 30.3) may change the grouping structure, and the heat map will be updated accordingly. This flexibility improves the system's adaptability.

[0053] S2, based on the high-risk material identification, extracts the historical flow channel blockage records of the corresponding materials from the database, uses the random forest algorithm to analyze the probability of blockage, determines the need for flow channel parameter adjustment, and outputs the adjusted flow channel speed and sorting interval parameters.

[0054] Optionally, this step also includes: Step S21, obtain the high-risk material identification and corresponding historical flow channel data from the database, and extract the blockage record according to the rule that the flow channel speed fluctuation exceeds the speed fluctuation threshold or the stagnation time is greater than the stagnation time threshold. Step S22, use the standardization method to process the flow channel speed, material length and environmental temperature and humidity characteristics to obtain a standardized feature set. Step S23, input the standardized feature set into the random forest classifier to obtain the blockage probability. Step S24, if the blockage probability exceeds the blockage probability threshold, then according to the flow channel speed too high feature in the blockage record, reduce the current flow channel speed by a preset proportion to obtain the adjusted flow channel speed. Step S25, calculate the sorting interval based on the adjusted flow channel speed and material length, and the sorting interval is equal to the material length divided by the adjusted flow channel speed. Step S26, calculate the deviation between the sorting interval and the theoretical interval. If the deviation exceeds the deviation threshold, re-input the current flow channel parameters and the blockage record into the random forest classifier to obtain the updated blockage probability. Step S27: Adjust the flow channel speed and recalculate the sorting interval according to the updated blocking probability, and repeat the process until the deviation is less than the preset target value or the preset number of iterations is reached. Step S28: Output the final sorting interval parameter.

[0055] Specifically, when obtaining high-risk material identification and corresponding historical flow channel data from the database, key information can be extracted by querying records within a specific time period.

[0056] For example, suppose a warehouse database stores flow channel speed and stagnation time data for the past 30 days. For a high-risk material identified as "H002," it is discovered that its flow channel speed fluctuation reached 20%, exceeding the preset speed fluctuation threshold of 15%. At the same time, the stagnation time reached 5 seconds, exceeding the preset stagnation time threshold of 3 seconds. This situation is recorded as a blockage event.

[0057] Specifically, flow path velocity fluctuations may be caused by material accumulation or equipment failure, while excessive stagnation may be related to excessive conveyor loads. This extraction method can quickly identify potential problem points.

[0058] In one possible implementation, when standardizing the flow channel velocity, material length, and ambient temperature and humidity characteristics, the dimensions of these data can be unified.

[0059] For example, if the flow velocity is in meters per second, the material length is in centimeters, and the temperature and humidity are expressed as percentages and degrees Celsius, these values ​​can be converted to dimensionless values ​​by subtracting the mean and dividing by the standard deviation. For example, suppose a material has a flow velocity of 2 meters per second, a material length of 30 centimeters, an ambient temperature of 25 degrees Celsius, and a humidity of 60%. After standardization, a set of features is obtained. This method makes data of different dimensions comparable in subsequent analysis. When the standardized feature set is input into a random forest classifier, it can be understood as using multiple decision trees to predict the probability of blockage.

[0060] For example, by analyzing the characteristics of 100 material flows in historical data, Random Forest found that the probability of blockage increases significantly when the flow velocity is too high and the humidity is high. Assuming that the feature set of a certain input predicts a blockage probability of 70%, this provides a basis for subsequent adjustments.

[0061] It should be noted that the advantage of random forests is that they can integrate the weights of multiple features to avoid misjudgment based on a single factor. If the blockage probability exceeds the blockage probability threshold, such as 60%, adjustments can be made based on the characteristic of excessively high flow velocity.

[0062] Preferably, if the current flow channel speed is 3 meters per second, which is outside the normal range, it is reduced by 20% and adjusted to 2.4 meters per second.

[0063] In one embodiment, this adjustment is based on historical data analysis to ensure that the flow channel speed is not too high, causing material collisions. The adjusted flow channel speed can effectively reduce the occurrence of blockages. When calculating the sorting interval based on the adjusted flow channel speed and material length, for example, if the material length is 40 cm and the adjusted speed is 2 meters per second, the sorting interval is 0.2 seconds. This interval reflects the shortest time interval required for materials to pass through the sorting point.

[0064] For example, if the speed is too high and the sorting interval is too short, it may cause material to pile up; otherwise, it will affect efficiency. When calculating the deviation between the sorting interval and the theoretical interval, assuming the theoretical interval is 0.25 seconds and the actual interval is 0.2 seconds, the deviation is 0.05 seconds, exceeding the deviation threshold of 0.03 seconds.

[0065] Understandably, this deviation may be caused by insufficient speed adjustment. At this point, the current flow channel parameters and blockage records are re-entered into the random forest classifier, the blockage probability is updated to 65%, and further optimization is performed based on this. Based on the updated blockage probability, the flow channel speed is adjusted, for example, from 2.4 meters per second to 2.2 meters per second. The sorting interval is recalculated to 0.22 seconds, and this cycle is repeated until the deviation is less than 0.03 seconds or five iterations are performed.

[0066] In one embodiment, this iteration can gradually approach the optimal parameters. When the final sorting interval parameter is output, for example, 0.23 seconds is obtained, which can be directly applied to the sorting equipment.

[0067] For example, this dynamic adjustment can improve sorting efficiency while reducing the risk of blockage, providing reliable support for warehouse management.

[0068] S3, obtain the adjusted flow channel speed and sorting interval parameters, and compare the deviation between the current flow channel operation status and the target state through the real-time monitoring system. If the deviation is greater than the set value, the flow safety warning system is activated to determine the location and type of abnormal state.

[0069] Optionally, this step also includes: Step S31, obtain the original data stream of flow channel speed and sorting interval through the real-time monitoring system. Step S32, for the original data stream, extract the speed mean in the past specified time window as the first feature, and the sorting interval standard deviation as the second feature. Step S33, respectively calculate the absolute difference between the first feature and the preset speed and the ratio of the second feature to the preset sorting interval to obtain a deviation vector. Step S34, if the Euclidean distance of the deviation vector exceeds the Euclidean distance threshold, use the IsolationForest algorithm to detect abnormal points and obtain the abnormal probability. Step S35, for the data points whose abnormal probability is greater than the abnormal probability threshold, generate a JSON alert containing the device number and timestamp. Step S36, obtain the operating parameters corresponding to the device number in the JSON alert from the real-time database. Step S37, perform Z-score normalization on the operating parameters to obtain normalized parameters. Step S38, use the K-Means clustering model to classify the normalized parameters to obtain cluster labels. In step S39, preset PID control parameters are selected based on the cluster label. If the cluster label is the first category, the first speed gain coefficient is used; if the cluster label is the second category, the first interval compensation value is used. In step S310, the PLC controller adjusts the inverter output frequency and the sorting motor pulse interval to obtain an adjusted device data stream. In step S311, the next feature extraction cycle begins for the adjusted device data stream.

[0070] Specifically, obtaining the original data stream of flow channel speed and sorting interval through the real-time monitoring system is the basis of the entire process.

[0071] For example, suppose sensors in a warehouse record conveyor belt speed and the interval between sorting points once a second, generating a continuous data stream that provides a timely reflection of the equipment's operating status.

[0072] In one possible implementation, a 10-minute window is selected for analysis. The mean speed is extracted as the first feature, for example, 2.5 meters per second, reflecting the average flow channel operation level. The standard deviation of the sorting interval is used as the second feature, for example, 0.04 seconds, reflecting the stability of the interval. The absolute difference between the first feature and the preset speed is calculated, assuming the preset speed is 2.3 meters per second, with a difference of 0.2 meters per second. The ratio of the second feature to the preset sorting interval of 0.2 seconds is 0.04 seconds divided by 0.2 seconds, which is 0.2, forming the deviation vector.

[0073] It is understood that the size of the deviation vector reflects the degree of deviation between the actual operation and the ideal state. If the Euclidean distance of the deviation vector exceeds the preset Euclidean distance threshold of 0.3, it indicates that there may be an abnormality in the system.

[0074] Specifically, the IsolationForest algorithm detects anomalies by isolating data points. For example, an analysis of 100 data points from the past hour revealed that a particular group with a speed of 2.8 meters per second and an interval of 0.15 seconds had an 80% probability of an anomaly, far exceeding the set anomaly probability threshold of 60%. Generating JSON alerts for data points with high anomaly probabilities is very practical.

[0075] For example, an alert contains the device number "E001" and the timestamp "2025-03-31 14:23:45," making it easy to quickly locate the problematic device. When retrieving operating parameters from a real-time database, information such as a motor speed of 1800 rpm and a temperature of 30 degrees Celsius might be extracted.

[0076] In one embodiment, Z-score normalization converts these parameters into dimensionless values ​​with a mean of 0 and a standard deviation of 1, which facilitates subsequent analysis. When the normalized parameters are classified using a K-Means clustering model, the device operating status can be divided into two categories.

[0077] For example, the first category might correspond to high-speed, low-temperature scenarios, while the second category might correspond to low-speed, high-temperature scenarios. When selecting PID control parameters based on cluster labels, a speed gain factor of 0.8 might be used for the first category, while an interval compensation of 0.05 seconds might be used for the second category. This selection optimizes control performance for different conditions.

[0078] Optimally, the PLC controller adjusts the inverter frequency to 45 Hz and the motor pulse interval to 0.22 seconds, resulting in more stable equipment operation. When the adjusted equipment data stream enters the next feature extraction cycle, for example, the new mean speed becomes 2.4 meters per second and the standard deviation decreases to 0.03 seconds, indicating that the system is gradually stabilizing.

[0079] It should be noted that this circulation mechanism can continuously optimize parameters to ensure that the flow channel and sorting efficiency reach the best state.

[0080] In one possible implementation, if the probability of anomalies drops below 50%, the frequency of alarms decreases, and equipment maintenance costs are reduced. This approach has the advantages of real-time and adaptive nature, effectively improving warehouse sorting reliability.

[0081] S4, based on the location and type of abnormal state, matches the corresponding flow channel adjustment strategy from the preset dynamic control rule library, uses genetic algorithm to optimize the sorting path and equipment scheduling plan, and obtains the optimized material sorting sequence.

[0082] Optionally, this step also includes: Step S41, obtain the abnormal state code and the coordinates of the occurrence position from the sensor or database; Step S42, according to the abnormal state code, query the corresponding occurrence type in the preset rule database, and use the decision tree classifier to classify the abnormal state into equipment failure or material blockage; Step S43, according to the occurrence type, extract the flow threshold and priority parameters from the rule database, and generate a matching strategy including adjustment direction and amplitude; Step S44, according to the occurrence position coordinates, calculate the adjustment parameters of the flow channel width and diversion ratio, and obtain the flow channel solution including coordinate offset and flow distribution value; Step S45, the flow The coordinate data in the path plan is converted into path nodes, and input into the NSGA-II algorithm for iterative optimization, with path length and sorting time as the objective function, to obtain the Pareto optimal solution set; step S46, select the path plan with the shortest total time from the Pareto optimal solution set, and update the task queue priority of the sorting equipment; step S47, generate a material ID sequence according to the task queue, and calculate the difference between the material ID sequence and the expected sequence; step S48, if the difference exceeds the difference threshold, re-acquire the current flow channel state data, correct the flow distribution parameters, and generate a new material ID sequence.

[0083] Specifically, obtaining the abnormal status code and occurrence location coordinates from the sensor or database is the starting point of the entire process.

[0084] For example, the sensor may detect the code "A012", indicating that an abnormality has occurred, and record the coordinates as "X:15, Y:22". These data provide the basis for subsequent analysis.

[0085] It is understandable that the exception code is usually preset by the equipment, while the coordinates reflect the specific location where the problem occurred, such as a corner of a conveyor belt in a warehouse.

[0086] In one possible implementation, a rule database is used to query the corresponding occurrence type based on the abnormal status code. For example, suppose "A012" in the database corresponds to "equipment failure," while "A013" indicates "material blockage." A decision tree classifier, based on historical data, determines that, for example, 80 of the past 100 occurrences of "A012" were related to motor overload and are therefore classified as equipment failure. This approach can quickly identify the nature of the problem.

[0087] It should be noted that the advantage of decision trees is that they are intuitive and can automatically divide according to features.

[0088] Specifically, when extracting flow threshold and priority parameters based on the occurrence type, a device failure might correspond to a flow threshold of "80%" and a priority of "High," while a material blockage might correspond to "50%" and a priority of "Medium." When generating a matching strategy, for example, the adjustment direction for a device failure might be "Slow Down" with a magnitude of "20%." This ensures that the strategy matches the problem type.

[0089] In one example, the strategy for a material blockage might be to increase diversion by 30% to clear the blockage. When calculating the channel width and diversion ratio based on the coordinates of the location where the blockage occurred, assume the channel width at coordinates "X:15, Y:22" is 2 meters. Adjusting the parameters might yield an offset of 0.5 meters and a flow distribution of 60%.

[0090] Preferably, this can avoid local overload. The generated flow path plan will be specific, for example, to share the flow to adjacent paths to ensure overall efficiency. After converting the coordinate data into path nodes, the NSGA-II algorithm begins optimization.

[0091] For example, let's set the path length target at 50 meters and the sorting time target at 30 seconds. After iteration, multiple solutions may be generated, such as path A taking 35 seconds and path B taking 28 seconds. The Pareto optimal solution set provides multiple options.

[0092] In one embodiment, path B, which has the shortest total time, is selected and the task queue priority is updated, for example, by placing "Path B Task" at the top. When generating a material ID sequence based on the task queue, suppose the sequence "P001, P002, P003" is obtained, while the expected sequence is "P001, P003, P002." The difference is determined by comparison, for example, if two locations are inconsistent.

[0093] It should be noted that this difference reflects the deviation of the sorting order. If the difference exceeds the difference threshold "1", the flow channel status data is re-acquired.

[0094] For example, the new data may show that traffic has recovered to "90%", and the correction generates a new sequence of "P001, P003, P002", which is consistent with expectations.

[0095] For example, if the initial anomaly causes a drop in flow due to a motor failure, the flow can be restored after adjustment, which can reduce subsequent errors.

[0096] It can be understood that this dynamic correction improves the accuracy of sorting.

[0097] Preferably, by continuously optimizing queues and parameters, the system can quickly return to normal operation after an anomaly.

[0098] S5, through the optimized material sorting sequence, detects whether there are hazardous goods labels in the system. If so, calls the dedicated isolation channel allocation algorithm to divert the hazardous goods to the isolation channel and outputs the operating parameters and emergency plan of the isolation channel.

[0099] Optionally, this step also includes: Step S51, obtain material data through the sorting sequence, detect whether there is a dangerous goods mark therein, and obtain preliminary detection results. Step S52, if the preliminary detection result shows that the dangerous goods mark exists, use OpenCV for image recognition, analyze the mark type, and determine the category of dangerous goods. Step S53, call the greedy algorithm according to the category of dangerous goods, calculate the isolation flow channel allocation plan, and output the diversion path data. Step S54, use the diversion path data to drive the channel isolation operation, obtain the isolation flow channel operation status, and obtain the operation parameters. Step S55, generate an emergency plan based on the operation parameters, judge the plan matching degree, and output the plan execution plan. Step S56, extract key instructions from the plan execution plan, use Python's Pandas library to generate the final isolation flow channel operation parameters and emergency plan data. Step S57, obtain the final isolation flow channel operation parameters and emergency plan data, use Arduino for sensor data collection and processing, verify the diversion effect, and judge the operation completion status.

[0100] Specifically, after obtaining material data through the sorting sequence, detecting hazardous material identification is a key step to ensure safety.

[0101] For example, material data might include ID, weight, and packaging information, which the system scans for "flammable" or "corrosive" markings. For example, if a material has the ID "P005" and its data contains the label "Warning: Flammable Liquid," the initial detection result would be "Hazardous Material Label Present." This type of detection typically relies on a pre-set keyword library to quickly identify potential risks. If a hazardous material label is detected, image recognition technology further analyzes the label type.

[0102] In one possible implementation, OpenCV is used to process warning patterns on packaging. For example, if a package image of "P005" displays a red flame logo, the system can identify it as a "flammable liquid" through color segmentation and shape matching.

[0103] Understandably, this method can distinguish different labels such as "flammable," "toxic," or "explosive," ensuring accurate subsequent processing. After determining the category of hazardous materials, a greedy algorithm is used to calculate the isolation flow channel allocation plan.

[0104] Specifically, for flammable liquids, the algorithm prioritizes the closest dedicated isolation channel. For example, if there are three channels in a warehouse, L1, L2, and L3, a greedy strategy might select L2, as it has the shortest path, requiring only a 10-meter movement. This approach seeks a local optimum and quickly generates diversion path data, such as "L2, flow allocation 70%." After using this diversion path data, the channel isolation operation begins and its operational status is retrieved.

[0105] Preferably, after isolation channel "L2" is activated, sensors monitor flow and speed, generating operational parameters such as "flow 60%, speed 2 m / s." These parameters reflect the proper functioning of the channel and provide a basis for further action. When generating an emergency plan based on these operating parameters, the degree of compatibility is assessed based on actual conditions.

[0106] In one embodiment, if the "flow rate of 60%" falls below the expected 80%, the contingency plan might include adding a backup flow channel L3. The degree of compatibility is determined by comparing historical data. For example, if a similar situation has been successful for 90% of the time, the contingency plan can be executed. This approach allows for rapid response to anomalies. After extracting key instructions from the contingency plan, the data is organized using Pandas.

[0107] For example, the command "Start L3, flow allocation 30%" is converted into a structured table containing flow channel IDs, parameters, and timestamps, ultimately generating complete operating parameters and emergency response plan data. The efficiency of this tool lies in its clear data and ease of system call. After obtaining the final data, the Arduino is responsible for sensor data collection and verification.

[0108] For example, the sensor detects that the flow rates of "L2" and "L3" are "55%" and "25%" respectively, and the total reaches the expected value, indicating that the diversion effect is good and the operation completion status is "successful".

[0109] It should be noted that this verification ensures that actual implementation is consistent with the plan, improving safety and reliability.

[0110] S6, obtains the operating parameters and emergency plan of the isolation flow channel, and analyzes the flow stability of the hazardous materials in the isolation channel through the real-time monitoring system. If the stability is lower than the stability threshold, the air pressure and separation distance of the isolation flow channel are adjusted to determine the isolation status of the hazardous materials.

[0111] Optionally, this step also includes: Step S61: Obtain the operating parameters of the isolation flow channel. Data is collected through pressure sensors and flow sensors. A linear regression model is used to calculate the flow stability of the hazardous materials to obtain an initial stability value. Step S62: Continuously sample the initial stability value through a real-time monitoring system and perform time series analysis using an ARIMA model to determine the stability trend. Step S63: If the stability trend is lower than the stability trend threshold, adjust the air pressure in the isolation flow channel using a PID controller to obtain an adjusted air pressure parameter. Step S64: Based on the adjusted air pressure parameter, calculate the change in the separation distance and determine a new separation distance value. Step S65: Flow data at the new separation distance value is collected through the monitoring system. A linear regression model is used to analyze the adjusted flow stability to determine whether a stable state has been achieved. Step S66: If the adjusted flow stability is still lower than the stability threshold, retrieve the preset strategy from the emergency plan database and determine a backup isolation plan. Step S67: Use finite element analysis to simulate the isolation state of the hazardous materials for the backup isolation plan to determine the final isolation effect.

[0112] Specifically, collecting the operating parameters of the isolation flow channel through pressure sensors and flow sensors is an important part of ensuring the safety of hazardous materials handling.

[0113] For example, a pressure sensor can be installed at the inlet of the flow channel to monitor the change in air pressure, while a flow sensor records the volume of material passing through per unit time.

[0114] In one possible implementation, assume that when an isolated flow channel is operating, the pressure sensor displays a current value of 1.2 atmospheres, and the flow sensor records a value of 50 liters per second. The system uses this data to analyze flow stability using a linear regression model. Linear regression fits a pressure-flow curve using historical data to derive an initial stability value, such as 85%, to reflect whether the current state is stable. When the real-time monitoring system continuously samples these initial stability values, the ARIMA model can capture trends and fluctuations in the time series.

[0115] Specifically, if the sampling data for 10 consecutive minutes is 85%, 84%, and 83%, respectively, and then gradually decreases, ARIMA analysis may predict that stability will fall below 80% within the next 5 minutes. This prediction relies on the autocorrelation of historical data and is suitable for dynamic adjustment decisions.

[0116] It should be noted that the advantage of ARIMA lies in its sensitivity to short-term changes, which can provide a basis for subsequent regulation. If the stability trend falls below a preset stability trend threshold, such as 80%, the PID controller will intervene to adjust the air pressure.

[0117] Preferably, PID gradually adjusts the air pressure from 1.2 atmospheres to 1.3 atmospheres by calculating the deviation in real time to ensure pressure balance in the flow channel.

[0118] In one embodiment, after adjustment, the air pressure stabilizes at 1.25 atmospheres, and sensor feedback shows a slight increase in flow rate, indicating that the system is gradually returning to stability. After adjusting the air pressure, fluid mechanics formulas are used to calculate the change in separation distance.

[0119] Optionally, the step S64, calculating the change in the separation distance based on the adjusted air pressure parameter and determining a new separation distance value, includes: calculating the change in the separation distance using the following formula: Where ΔS represents the change in separation distance, K represents the correction coefficient, Indicates standard atmospheric pressure, Indicates the current air pressure. Indicates the current temperature. represents the standard temperature, and h represents the initial distance.

[0120] It can be understood that the separation distance affects the distribution of material in the flow channel.

[0121] For example, with an initial spacing of 2 meters and air pressure increasing from 1.2 to 1.25 atmospheres, the system calculates that the spacing needs to be increased by 0.1 meters, taking temperature into account. The new spacing of 2.1 meters reduces collision risk by optimizing material distribution. The monitoring system collects flow data at this new spacing and uses linear regression to analyze stability again.

[0122] For example, new data indicates that stability has risen to 88%, close to expectations and indicating that adjustments have been effective. This repeated verification ensures that the flow of hazardous materials remains controllable. If stability remains insufficient, for example, at only 75%, the system will retrieve a backup plan from the emergency plan database.

[0123] Specifically, the plan may recommend using an alternative flow channel to share pressure, such as diverting 30% of the flow to another channel. For the alternative plan, finite element analysis is used to simulate the isolation state.

[0124] In one example, simulations showed that hazardous materials were evenly distributed under dual-flow operation, and the temperature in the isolated area dropped by 5 degrees Celsius, indicating a reduction in risk. This approach verifies the feasibility of the solution through virtual testing, ensuring the reliability of the final results.

[0125] It should be noted that the combination of multi-level analysis and adjustment can significantly improve the accuracy and safety of hazardous materials handling.

[0126] S7, based on the isolation status of dangerous goods, extracts the processing records of similar scenarios from the historical safety accident database, uses the decision tree algorithm to evaluate the level of potential threats, and determines the stability indicators of the overall operation of the system.

[0127] Optionally, this step also includes: Step S71: Obtain similar scene records of historical accidents from the security database, and use the KNN algorithm to match the corresponding processing records of hazardous materials isolation. Step S72: Based on the matching results, use the decision tree algorithm to analyze the potential threats in similar scenes to obtain the classification results of the threat level. Step S73: Compare the threat level with the preset abnormality threshold to determine the preliminary evaluation value of the operating status. Step S74: Based on the pre-established mapping relationship between the operating status and the stability index, obtain the stability deviation of the system operation through comparison and analysis. Step S75: Based on the stability deviation, use the K-means clustering algorithm to determine the abnormal distribution characteristics of the hazardous materials isolation scene. Step S76: Based on the abnormal distribution characteristics, analyze the dynamic changes of potential threats and determine the adjustment parameters of the system operation. Step S77: Update the operating status according to the adjustment parameters to obtain the optimized value of the stability index.

[0128] Specifically, when obtaining similar scene records of historical accidents from the safety database, the KNN algorithm can be used for matching.

[0129] For example, the database stores accident data from hazardous materials isolation flow channels over the past five years, including scenarios such as temperature anomalies and sudden pressure changes. KNN calculates the distance between current operating parameters and historical data to identify the three most similar cases.

[0130] For example, if the current flow channel pressure is 1.3 atmospheres and the temperature is 45 degrees Celsius, KNN may match an accident record with a pressure of 1.28 atmospheres and a temperature of 47 degrees Celsius. This method relies on the proximity of parameters and can quickly locate similar problems.

[0131] In one possible implementation, a decision tree algorithm is used to analyze potential threats based on the matching results. The decision tree classifies the threat level based on features in historical cases, such as excessive pressure or abnormal flow.

[0132] Specifically, if a matching case shows a sudden pressure change leading to a leak, the decision tree will determine the threat level as "medium" based on criteria such as "whether the pressure exceeds 1.5 atmospheres" and "whether the flow rate drops suddenly." This classification, based on a step-by-step breakdown of rules, is clear and easy to trace.

[0133] It should be noted that when comparing the threat level with the preset abnormality threshold, the abnormality threshold may be "high-level threat corresponds to 90% abnormality probability".

[0134] In one embodiment, if the decision tree outputs a threat level of "medium," corresponding to a 70% probability of abnormality, which is lower than the abnormality threshold, the initial assessment of the operating status is "controllable but requiring attention." This comparison provides an intuitive basis for judgment and facilitates subsequent decision-making. When analyzing stability deviations based on the mapping relationship between operating status and stability indicators.

[0135] It is understandable that the mapping relationship may be a table established in advance through expert experience.

[0136] For example, the operating status of "controllable but requiring attention" corresponds to a stability index of 85%, while the current actual index is 82%, a deviation of 3%. This deviation quantifies the degree to which the system deviates from the ideal state and provides a basis for adjustment.

[0137] For stability deviation, preferably, the K-means clustering algorithm can be used to determine abnormal distribution characteristics.

[0138] Specifically, suppose 10 sets of flow channel data, including pressure, flow rate, and temperature, are collected. K-means classifies them into three categories: "normal," "slightly abnormal," and "significantly abnormal." If the majority of the data falls into the "slightly abnormal" category, it indicates potential system instability. This clustering intuitively reflects the data distribution pattern.

[0139] In one embodiment, when analyzing the dynamic changes of potential threats based on anomaly distribution characteristics, it was observed that the pressure value in the "minor anomaly" category gradually increased from 1.3 atmospheres to 1.35 atmospheres. Combined with the time series, it was inferred that the threat may have intensified due to pressure accumulation, and the adjustment parameter was determined to be "reducing the pressure to 1.25 atmospheres." This dynamic analysis captures the changing trend and ensures targeted adjustments.

[0140] For example, after updating the operating status based on the adjustment parameters, the stability index increased from 82% to 87%.

[0141] In one embodiment, sensor feedback shows that the pressure is stable at 1.25 atmospheres and the flow distribution is more uniform. This improvement in the optimization value reflects the system's ability to recover stability and provides higher reliability for hazardous materials isolation.

[0142] S8, obtains the stability index of the overall operation of the system, compares the matching degree of the current operation data with the stability index through the real-time monitoring system, and if the matching degree is lower than the matching degree threshold, updates the parameter threshold in the dynamic control rule library and outputs a new control instruction.

[0143] Optionally, this step also includes: Step S81, collect system operation data through sensors to obtain a quantitative value of the current operation state. Step S82, use the ARIMA model to perform time series analysis on the quantitative value to obtain the changing trend of the system operation. Step S83, obtain a pre-established stability index database, compare the changing trend with the stability index, and determine the degree of matching. Step S84, if the degree of matching is lower than the matching threshold, analyze the parameter threshold in the rule base through the K-means clustering algorithm to determine the adjustment range. Step S85, update the parameter threshold in the dynamic control rule base according to the adjustment range to obtain a new control instruction set. Step S86, send the control instruction set to the execution unit through the message queue to complete the system adjustment. Step S87, obtain the adjusted operation data, compare the adjusted operation data with the stability index, and determine whether the system operation state has returned to the expected state.

[0144] Specifically, by collecting system operation data through sensors, a quantitative value of the current operating status is obtained, which can be understood as the basis for real-time monitoring of the hazardous materials isolation system.

[0145] For example, in a hazardous materials flow channel, sensors might record data indicating a pressure of 1.4 atmospheres, a temperature of 50 degrees Celsius, and a flow rate of 200 liters per minute. These quantitative values ​​reflect the current operating status of the system and provide data support for subsequent analysis. Using the ARIMA model to perform time series analysis on these quantitative values ​​aims to predict changing trends in system operation.

[0146] Specifically, ARIMA analyzes the changes in pressure or temperature over time by using patterns in historical data.

[0147] In one possible implementation, if the pressure has slowly risen from 1.3 atmospheres to 1.4 atmospheres over the past hour, ARIMA might predict that the pressure will continue to rise to 1.45 atmospheres within the next 30 minutes. This trend analysis can reveal potential anomalies in advance. Obtaining a pre-established database of stability indicators, comparing the changing trends with the stability indicators, and determining the degree of match is key to assessing whether the system has deviated from normal conditions.

[0148] For example, the stability index database stipulates that the pressure should be maintained between 1.2 and 1.35 atmospheres, and the predicted value of 1.45 atmospheres is out of range, and the matching degree may be determined to be 60%, which is lower than the preset matching degree threshold of 80%.

[0149] It should be noted that this comparison provides an intuitive basis for deviation. If the matching degree is lower than the matching threshold, the parameter threshold in the rule base is analyzed by the K-means clustering algorithm to determine the adjustment range.

[0150] Preferably, K-means can classify historical data into three categories: "normal", "high", and "abnormal".

[0151] For example, if the current pressure of 1.4 atmospheres is classified as "high," the rule base recommends adjusting the pressure to 1.25 atmospheres. This approach uses data distribution patterns to determine the adjustment direction. Based on the adjustment range, the parameter thresholds in the dynamic control rule base are updated to generate a new set of control instructions, demonstrating the system's adaptability.

[0152] In one embodiment, the rule base, which originally set the upper pressure limit at 1.35 atmospheres, is now updated to 1.25 atmospheres, generating the instruction "Reduce valve opening by 10%." This update ensures that the rule matches the current state. The control instruction set is sent to the execution unit via a message queue, completing the system adjustment and ensuring efficient instruction delivery.

[0153] For example, commands are sent to the flow control valve via a queue to quickly perform pressure adjustments.

[0154] In one embodiment, sensor feedback indicates that the adjusted pressure stabilizes at 1.25 atmospheres. This approach improves real-time execution. Verifying the effectiveness of adjustments requires obtaining post-adjustment operating data and comparing it with stability indicators to determine whether the system has returned to its expected state.

[0155] Specifically, after adjustment, the pressure was 1.25 atmospheres and the temperature was 48 degrees Celsius, both falling within the stability indicator range, indicating that the system has returned to normal. This comparison provides a reliable basis for subsequent optimization.

[0156] S9, based on the new control instructions, adjusts the working status of all sorting equipment through the distributed computing framework, uses time series analysis to predict the short-term change trend of material flow, and obtains a stable operation plan for the system under high-load scenarios.

[0157] Optionally, this step also includes: In step S91, the current operating status data of the sorting equipment is acquired through the distributed computing framework Apache Spark. Real-time operating information is extracted from the equipment using the acquisition module to obtain a status dataset. In step S92, material flow-related indicators are extracted from the status dataset. The ARIMA model is used to process historical material flow data to determine short-term trends. In step S93, based on the short-term trends, it is determined whether the material flow exceeds the short-term trend threshold. If so, the sorting equipment operating status is adjusted using Apache Spark to obtain adjusted operating parameters. In step S94, the sorting equipment operating plan is updated using the adjusted operating parameters, real-time system load data under high-load scenarios is obtained, and the system stability is determined. In step S95, load fluctuation characteristics are extracted from the system stability state. A random forest algorithm is used to analyze the relationship between load fluctuations and material flow trends to obtain an optimized operating plan. In step S96, Apache Spark's task allocation is updated based on the optimized operating plan. The sorting equipment executes the adjusted operating status to obtain stable operating data for the system under high-load scenarios. Step S97: extract performance indicators from the stable operation data, use the ARIMA model to predict the material flow change trend in the short term, and determine the long-term stable operation plan of the system.

[0158] Specifically, obtaining the current working status data of the sorting equipment through the distributed computing framework Apache Spark can be understood as utilizing its powerful parallel processing capabilities to collect information from multiple device nodes in real time.

[0159] For example, in a hazardous materials sorting warehouse, the acquisition module extracts data such as speed, power, and material accumulation from the sorting equipment every second to form a status dataset. For example, a speed of 500 rpm, power of 80%, and accumulation of 300 kg per hour reflect the equipment's current operating load. When extracting material flow-related metrics from the status dataset, the focus is on material input and output rates and accumulation.

[0160] Specifically, suppose historical data shows that the material input rate over the past 24 hours was 50 kg / minute and the output was 48 kg / minute. Analyzing this data using an ARIMA model can capture the increasing trend in the difference between input and output.

[0161] In one possible implementation, the model predicts that the input rate could rise to 52 kilograms per minute over the next hour, while the output remains constant, indicating that the accumulation will increase. This trend analysis provides a basis for subsequent judgment.

[0162] When judging whether the material flow exceeds the short-term change trend threshold based on the short-term change trend, it is preferred that the preset accumulation threshold is 350 kilograms per hour, and the current predicted accumulation is 360 kilograms, which has exceeded the accumulation threshold range. At this time, Apache Spark is used to adjust the working state of the sorting equipment, for example, increasing the speed to 550 revolutions per minute and the power to 85%. The adjusted operating parameters can quickly respond to material accumulation problems and ensure a smooth process. After updating the sorting equipment operation plan with the adjusted operating parameters, it is particularly important to obtain real-time load data under high-load scenarios.

[0163] For example, during peak hours, when sorting equipment processes 500 kg of material per hour, load data may show speed fluctuations between 540 and 560 rpm. When determining system stability, if the fluctuation is less than 5%, the system is considered to be within controllable range. This judgment provides data support for optimization. After extracting load fluctuation characteristics from the system's stable state, a random forest algorithm is used to analyze the relationship between the fluctuations and material flow trends.

[0164] In one embodiment, the algorithm, using historical data, discovered that speed fluctuations are positively correlated with material input rate: for every 10 kg increase in input, the speed needs to increase by approximately 20 rpm. After obtaining an optimized operating plan, for example, it recommends dynamically adjusting the speed to 570 rpm to better adapt to high-load scenarios.

[0165] When updating Apache Spark's task allocation according to the optimized operation plan, it should be noted that the tasks will be redistributed according to the device load.

[0166] For example, offloading processing tasks from high-load equipment to backup nodes ensures overall efficiency. After the sorting equipment implements the adjusted operating state, the system's stable operating data under high-load scenarios may show that the accumulation rate has dropped to 340 kilograms per hour, which is within a safe range. After extracting performance indicators from the stable operating data, it is crucial to use the ARIMA model to predict material flow trends in the short term.

[0167] For example, current data shows a stable input rate of 50 kilograms per minute, and the model predicts a potential increase to 53 kilograms per minute within the next two hours. Based on this trend, a plan for long-term system stability can be developed, such as allowing for a 10% speed margin to effectively cope with future fluctuations. This prediction ensures long-term stable system operation.

[0168] like Figure 2 As shown, the second aspect of the present invention provides an intelligent material sorting system, which uses the above method to intelligently sort materials, and the system includes: The sensor data acquisition module is used to obtain material flow data through the sensor network. When a tipping-prone object is detected, the ratio of the material stack height to the flow channel width is calculated. If the ratio exceeds the safety factor threshold, the safety detection point is triggered to record the abnormal mode and obtain a high-risk material identification; The high-risk material analysis module is used to extract the historical flow channel blockage records of the corresponding materials from the database based on the high-risk material identification, use the random forest algorithm to analyze the probability of blockage, determine the need for flow channel parameter adjustment, and output the adjusted flow channel speed and sorting interval parameters; The flow channel parameter adjustment module is used to obtain the adjusted flow channel speed and sorting interval parameters. The real-time monitoring system compares the deviation between the current flow channel operation status and the target state. If the deviation is greater than the set value, the flow safety warning system is activated to determine the location and type of abnormal state. The flow safety early warning module is used to match the corresponding flow channel adjustment strategy from the preset dynamic control rule library according to the location and type of abnormal state, and use genetic algorithms to optimize the sorting path and equipment scheduling plan to obtain the optimized material sorting sequence; The sorting path optimization module is used to detect whether there are any hazardous materials signs in the system through the optimized material sorting sequence. If so, it calls the dedicated isolation channel allocation algorithm to divert the hazardous materials to the isolation channel and output the operating parameters and emergency plan of the isolation channel; The hazardous materials isolation module is used to obtain the operating parameters and emergency plans of the isolation flow channel. It analyzes the flow stability of hazardous materials in the isolation channel through a real-time monitoring system. If the stability is lower than the stability threshold, it adjusts the air pressure and separation distance of the isolation flow channel to determine the isolation status of the hazardous materials. The flow stability analysis module is used to extract similar scenario processing records from the historical safety accident database based on the isolation status of hazardous materials, use a decision tree algorithm to evaluate the level of potential threats, and determine the stability indicators of the overall system operation; The threat assessment module is used to obtain the stability index of the overall system operation. It compares the matching degree of the current operation data with the stability index through the real-time monitoring system. If the matching degree is lower than the matching degree threshold, it updates the parameter threshold in the dynamic control rule base and outputs new control instructions. The system control module is used to adjust the working status of all sorting equipment through a distributed computing framework according to new control instructions, use time series analysis to predict short-term changes in material flow trends, and obtain a stable operation plan for the system under high-load scenarios.

[0169] The above embodiments are intended to illustrate the technical solutions of the present invention and are not intended to limit the present invention. The present invention is described in detail with reference to the preferred embodiments only. It should be understood by those skilled in the art that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or equivalents should be included within the scope of the claims of the present invention.

Claims

1. An intelligent material sorting method, characterized in that: The method comprises: S1, acquires material flow data through the sensor network. When a tipping-prone object is detected, the ratio of the material stack height to the flow channel width is calculated. If the ratio exceeds the safety factor threshold, the safety detection point is triggered to record the abnormal pattern and obtain a high-risk material identification; S2, based on the high-risk material identification, extracts the historical flow channel blockage records of the corresponding materials from the database, uses the random forest algorithm to analyze the probability of blockage, determines the need for flow channel parameter adjustment, and outputs the adjusted flow channel speed and sorting interval parameters; S3, obtains the adjusted flow channel speed and sorting interval parameters, and compares the deviation between the current flow channel operation status and the target state through the real-time monitoring system. If the deviation is greater than the set value, the flow safety warning system is activated to determine the location and type of abnormal state; S4, based on the location and type of abnormal state, matches the corresponding flow channel adjustment strategy from the preset dynamic control rule library, uses genetic algorithm to optimize the sorting path and equipment scheduling plan, and obtains the optimized material sorting sequence; S5, through the optimized material sorting sequence, detects whether there is a dangerous goods mark in the system. If so, it calls the dedicated isolation channel allocation algorithm to divert the dangerous goods to the isolation channel and outputs the isolation channel operation parameters and emergency plan; S6, obtaining the operating parameters and emergency plan of the isolation flow channel, analyzing the flow stability of the hazardous materials in the isolation channel through the real-time monitoring system, and adjusting the air pressure and separation distance of the isolation flow channel if the stability is lower than the stability threshold to determine the isolation status of the hazardous materials; S7, based on the isolation status of hazardous materials, extracts the handling records of similar scenarios from the historical safety accident database, uses a decision tree algorithm to assess the level of potential threats, and determines the stability indicators of the overall system operation; S8, obtaining the stability index of the overall operation of the system, comparing the matching degree between the current operation data and the stability index through the real-time monitoring system, and if the matching degree is lower than the matching degree threshold, updating the parameter threshold in the dynamic control rule base and outputting a new control instruction; S9, based on the new control instructions, adjusts the working status of all sorting equipment through the distributed computing framework, uses time series analysis to predict the short-term change trend of material flow, and obtains a stable operation plan for the system under high-load scenarios.

2. The method according to claim 1, characterized in that Step S1, acquiring material flow data through a sensor network, and calculating the ratio of the material stack height to the flow channel width when a tippable object is detected. If the ratio exceeds a safety factor threshold, the safety detection point is triggered to record an abnormal mode and obtain a high-risk material identification, including: Step S11, collecting material flow data through a laser rangefinder and a photoelectric sensor network to obtain a first data stream; Step S12, performing denoising processing on the first data stream using Gaussian filtering to obtain a second data stream; Step S13, using an edge detection operator to extract object contour features from the second data stream, and determining an easily dumpable object identifier based on an aspect ratio threshold; Step S14, obtaining the stacking height measurement value and the flow channel width measurement value corresponding to the easily dumpable item identifier from the warehouse database, and calculating the ratio of the stacking height measurement value to the flow channel width measurement value; Step S15: If the ratio of the stack height measurement value to the flow channel width measurement value exceeds a preset safety factor threshold, an isolation forest algorithm is used to detect abnormal patterns and obtain an abnormality type code; Step S16: Add a new record to the detection log table according to the abnormality type code and generate a high-risk material identification; Step S17: using the DBSCAN algorithm to group the high-risk material identifiers according to their location coordinates to obtain a material risk heat map; Step S18: adjusting the collection interval of the corresponding sensor according to the coordinates of the high-risk area in the material risk heat map.

3. The method according to claim 2, characterized in that The step S17, using the DBSCAN algorithm to group the high-risk material identifiers according to their location coordinates to obtain a material risk heat map, includes: Step S171, obtaining material location data with longitude and latitude coordinates from the warehouse database; Step S172, clustering the coordinates using the DBSCAN algorithm; Step S173, obtaining preliminary spatial grouping results; Step S174, calculating a risk concentration code for each of the groups; Step S175: If the risk concentration code exceeds the risk concentration code threshold, the inbound and outbound records of the corresponding material ID in the past specified time period are extracted from the warehouse database; Step S176: Calculate the coordinate adjustment amount of the abnormal flow material based on the inventory in and out records. If the daily flow exceeds two standard deviations of the average value, move the material coordinates toward the group center point by a fixed percentage of the abnormal number of times to obtain an adjusted coordinate data set. Step S177, performing secondary clustering on the adjusted coordinate data set using an improved DBSCAN algorithm, optimizing the neighborhood radius using the silhouette coefficient, and obtaining a final risk grouping result; Step S178: Map the risk concentration code of each final risk group into a thermal value, wherein the thermal value is linearly positively correlated with the risk concentration code, and output a dynamic thermal map.

4. The method according to claim 3, characterized in that The step S172 is to cluster the coordinates using the DBSCAN algorithm, including: determining the neighborhood radius by the inflection point of the k-distance graph, and setting the minimum number of samples to a fixed proportion of the total data volume.

5. The method according to claim 4, characterized in that Step S174, calculating a risk concentration code for each of the groups, including: defining the risk concentration code as the sum of the squares of the risk levels of the materials in the group divided by the area of ​​the group.

6. The method according to claim 1, characterized in that Step S3, obtaining the adjusted flow channel speed and sorting interval parameters, and comparing the deviation between the current flow channel operation state and the target state through the real-time monitoring system. If the deviation is greater than the set value, the flow safety warning system is activated to determine the location and type of the abnormal state, including: Step S31, obtaining the original data stream of the flow channel speed and sorting interval through the real-time monitoring system; Step S32: for the original data stream, extract the speed mean value within the past specified time window as the first feature and the sorting interval standard deviation as the second feature; Step S33, respectively calculating the absolute difference between the first feature and a preset speed and the ratio of the second feature to a preset sorting interval to obtain a deviation vector; Step S34: If the Euclidean distance of the deviation vector exceeds the Euclidean distance threshold, an IsolationForest algorithm is used to detect abnormal points and obtain an abnormality probability; Step S35: generating a JSON alert including a device number and a timestamp for the data point whose abnormal probability is greater than the abnormal probability threshold; Step S36, obtaining the operating parameters corresponding to the device number in the JSON alarm from the real-time database; Step S37, performing Z-score normalization on the operating parameters to obtain normalized parameters; Step S38, using a K-Means clustering model to classify the standardized parameters to obtain cluster labels; Step S39, selecting preset PID control parameters according to the cluster labels; Step S310, adjusting the inverter output frequency and the sorting motor pulse interval through the PLC controller to obtain an adjusted device data stream; Step S311: Enter the next feature extraction cycle for the adjusted device data stream.

7. The method according to claim 6, characterized in that The step S39, selecting a preset PID control parameter according to the cluster label, includes: adopting a first speed gain coefficient when the cluster label is the first category, and adopting a first interval compensation amount when the cluster label is the second category.

8. The method according to claim 1, characterized in that Step S6, obtaining the operating parameters and emergency plan of the isolation channel, analyzing the flow stability of the hazardous materials in the isolation channel through the real-time monitoring system, and adjusting the air pressure and separation distance of the isolation channel if the stability is lower than the stability threshold to determine the isolation status of the hazardous materials, includes: Step S61, obtaining the operating parameters of the isolation flow channel, collecting data through the pressure sensor and the flow sensor, and calculating the flow stability of the dangerous goods using a linear regression model to obtain an initial stability value; Step S62: continuously sampling the initial stability value through the real-time monitoring system, and performing time series analysis using the ARIMA model to determine the stability change trend; Step S63: If the stability change trend is lower than the stability change trend threshold, the air pressure value of the isolation flow channel is adjusted by the PID controller to obtain an adjusted air pressure parameter; Step S64, calculating the change in the separation distance based on the adjusted air pressure parameter and determining a new separation distance value; Step S65: collecting flow data at the new separation distance value through the monitoring system, and using a linear regression model to analyze the flow stability after adjustment to determine whether a stable state has been reached; Step S66: If the adjusted flow stability is still lower than the stability threshold, obtain the preset strategy in the emergency plan database and determine the backup isolation plan; Step S67: For the backup isolation scheme, finite element analysis is used to simulate the isolation state of the dangerous goods to determine the final isolation effect.

9. The method according to claim 8, characterized in that The step S64, calculating the change in the separation distance based on the adjusted air pressure parameter and determining a new separation distance value, includes: The change in separation distance is calculated using the following formula: Where ΔS represents the change in separation distance, K represents the correction coefficient, Indicates standard atmospheric pressure, Indicates the current air pressure. Indicates the current temperature. represents the standard temperature, and h represents the initial distance.

10. An intelligent material sorting system, characterized in that: The method according to any one of claims 1 to 9 is used to perform intelligent sorting of materials, wherein the system comprises: The sensor data acquisition module is used to obtain material flow data through the sensor network. When a tipping-prone object is detected, the ratio of the material stack height to the flow channel width is calculated. If the ratio exceeds the safety factor threshold, the safety detection point is triggered to record the abnormal mode and obtain a high-risk material identification; The high-risk material analysis module is used to extract the historical flow channel blockage records of the corresponding materials from the database based on the high-risk material identification, use the random forest algorithm to analyze the probability of blockage, determine the need for flow channel parameter adjustment, and output the adjusted flow channel speed and sorting interval parameters; The flow channel parameter adjustment module is used to obtain the adjusted flow channel speed and sorting interval parameters. The real-time monitoring system compares the deviation between the current flow channel operation status and the target state. If the deviation is greater than the set value, the flow safety warning system is activated to determine the location and type of abnormal state. The flow safety early warning module is used to match the corresponding flow channel adjustment strategy from the preset dynamic control rule library according to the location and type of abnormal state, and use genetic algorithms to optimize the sorting path and equipment scheduling plan to obtain the optimized material sorting sequence; The sorting path optimization module is used to detect whether there are any hazardous materials signs in the system through the optimized material sorting sequence. If so, it calls the dedicated isolation channel allocation algorithm to divert the hazardous materials to the isolation channel and output the operating parameters and emergency plan of the isolation channel; The hazardous materials isolation module is used to obtain the operating parameters and emergency plans of the isolation flow channel. It analyzes the flow stability of hazardous materials in the isolation channel through a real-time monitoring system. If the stability is lower than the stability threshold, it adjusts the air pressure and separation distance of the isolation flow channel to determine the isolation status of the hazardous materials. The flow stability analysis module is used to extract similar scenario processing records from the historical safety accident database based on the isolation status of hazardous materials, use a decision tree algorithm to evaluate the level of potential threats, and determine the stability indicators of the overall system operation; The threat assessment module is used to obtain the stability index of the overall system operation. It compares the matching degree of the current operation data with the stability index through the real-time monitoring system. If the matching degree is lower than the matching degree threshold, it updates the parameter threshold in the dynamic control rule base and outputs new control instructions. The system control module is used to adjust the working status of all sorting equipment through a distributed computing framework according to new control instructions, use time series analysis to predict short-term changes in material flow trends, and obtain a stable operation plan for the system under high-load scenarios.

Images