Catering distribution information intelligent management system based on big data

Abstract

The invention provides a catering distribution information intelligent management system based on big data, and the system comprises the steps: reading real-time distribution data collected by distribution equipment, and inputting the collected original information flow into a data processing module; executing a three-level serial cleaning operation on the collected original data to generate a structured data set, and inputting the structured data set into an anomaly detection engine; detecting the cleaned data, and outputting a detection result; performing classification processing according to detection results, and inputting arbitration results and manual intervention records into a self-optimization knowledge base; dynamically updating the anomaly detection rule, and feeding back an optimized decision strategy to a related module; and executing a path optimization algorithm and a resource scheduling strategy on the normal data, generating a distribution instruction and issuing the distribution instruction to an execution terminal. The method effectively solves the problems that a traditional system is low in data processing efficiency and lags behind abnormal response, and has the remarkable advantages of improving distribution timeliness and service quality.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a big data-based intelligent management system for catering delivery information. Background Art

[0002] The current catering delivery business faces a technical bottleneck in processing multi-source heterogeneous data. The order feature data, delivery dynamic data, and environmental status data generated during the delivery process have significant differences in the temporal and spatial dimensions. Traditional systems use a single threshold judgment and manual review method, which makes it difficult to effectively handle complex problems such as delivery trajectory drift, business logic conflicts, and system-level anomalies. Especially during peak hours, existing technologies are unable to effectively achieve intelligent filling of missing data, spatiotemporal alignment of multi-source data, and adaptive optimization of anomaly detection rules, resulting in high delivery delay rates. The system lacks a complete monitoring chain from data verification to business assurance. When encountering network fluctuations or terminal failures, it is difficult to switch to backup channels and implement self-healing strategies in a timely manner. These problems seriously restrict the improvement of delivery timeliness and service quality. To address the above problems, existing technologies urgently need to be improved. Summary of the Invention

[0003] In order to solve the problems existing in the prior art, the present invention provides a catering delivery information intelligent management system based on big data to improve delivery timeliness and service quality.

[0004] To achieve the above objectives, the present invention provides a catering delivery information intelligent management system based on big data, the system comprising a data acquisition module, a data processing module, an anomaly detection engine, a conflict arbitration module, a self-optimizing knowledge base, and a decision output module;

[0005] The data acquisition module converts the real-time delivery data collected from the delivery equipment into a multi-source heterogeneous data stream through a distributed message queue, and inputs the multi-source heterogeneous data stream into the data processing module. The multi-source heterogeneous data stream includes an order feature data stream, a delivery dynamic data stream, and an environmental status data stream;

[0006] The data processing module performs a three-stage serial cleaning operation on the multi-source heterogeneous data stream, obtains a structured data set from the cleaned multi-source heterogeneous data stream, and inputs the structured data set into the anomaly detection engine;

[0007] The anomaly detection engine performs anomaly detection on the structured data set, determines the data anomaly type as a single anomaly or a compound conflict based on the anomaly detection result, and outputs the detection result to the conflict arbitration module;

[0008] The conflict arbitration module classifies and handles data anomalies according to the detection results, issues handling instructions, and inputs handling records into the self-optimization knowledge base;

[0009] The self-optimizing database dynamically updates the anomaly detection rules of the anomaly detection engine according to the handling records, and feeds back the optimized decision strategy to the anomaly detection engine;

[0010] The decision output module executes a path optimization algorithm and a resource scheduling strategy on the handling instruction, generates a delivery instruction, and sends it to the execution terminal.

[0011] Optionally, the multi-source heterogeneous data stream includes an order feature data stream, a delivery dynamic data stream and an environmental status data stream. The order feature data stream is formed based on the merchant terminal collecting the order generation timestamp, delivery destination coordinates, estimated cooking time, and user special requirement identification. The delivery dynamic data stream is formed based on the delivery personnel's mobile device collecting the GPS positioning points, battery power status, network signal strength value, and moving speed vector per second. The environmental status data stream is formed by accessing the real-time traffic event coordinates, weather warning level, and road closure status sign information through a third-party API.

[0012] Optionally, the three-level serial cleaning operations are field integrity repair, spatiotemporal dimension alignment, and business logic verification in sequence;

[0013] When a required field is missing in the input structured dataset, the filling mechanism is automatically triggered. For numeric fields, the missing value is filled by using the average value of orders completed in the last hour within a 500-meter radius of the same business district. For coordinate fields, linear interpolation is performed using the previous valid positioning points. The processed structured dataset is then aligned in time and space.

[0014] For the input structured dataset, a time-based time axis is created based on the order, and the road network is modeled as a directed graph structure with time as the reference time axis. This ensures that multi-source data is collaboratively processed under a unified time-space reference. The processed structured dataset is then validated for business logic.

[0015] Predefined rules are applied to the input structured data set to filter illegal data. Geographic fence verification is used to automatically discard coordinate points that are more than 10 meters beyond the delivery area boundary. Speed ​​mutation verification is used to initiate trajectory smoothing for coordinate points whose speed change rate exceeds 50% for two consecutive seconds.

[0016] Optionally, the spatiotemporal dimension alignment specifically includes time axis reconstruction, space mapping, and cross-source data synchronization, wherein:

[0017] The timeline reconstruction uses the order creation time as the time origin, rearranges the deliveryman's trajectory points, and fills the missing time periods with cubic spline interpolation;

[0018] Spatial mapping abstracts roads into directed edges and intersections into nodes, establishes a road network topology, matches original coordinate points to the nearest road edge, marks points as outliers if the projection error exceeds 20 meters, and establishes a mapping relationship between GPS coordinates and road network topology.

[0019] Cross-source data synchronization starts a sliding window compensation mechanism when the time deviation between order data and delivery data exceeds 30 seconds, intercepting 60 seconds of data before and after the deviation period for trend extrapolation.

[0020] Optionally, the anomaly detection engine forms a complete monitoring chain from basic data verification to high-level business assurance by building a three-layer defense system at the data level, business level, and system level;

[0021] Data-level anomaly detection is completed through data reception and diversion, field integrity verification, numerical compliance verification, format specification review, and output control mechanisms to ensure the validity and compliance of the original data;

[0022] Through benchmark model construction, real-time progress tracking, dynamic threshold management, multi-dimensional root cause analysis, and early warning output specifications, business-level anomaly detection is completed to monitor operational deviations of core delivery business indicators.

[0023] System-level anomaly detection is completed through terminal health monitoring, network quality assessment, application service diagnosis and hierarchical self-healing strategies to ensure the reliable operation of terminal devices and service components.

[0024] Optionally, the data-level anomaly detection includes the following processing units:

[0025] Data reception and diversion: import the cleaned data stream into the detection channel and automatically divert it to a dedicated verifier according to the data type, including order data verifier, trajectory data verifier or environmental data verifier;

[0026] Field integrity verification scans for missing required fields and triggers an automatic filling mechanism. For missing numeric fields, the filling mechanism uses the historical average of the last 10 orders within a 300-meter radius of the same business district. For missing coordinate fields, linear interpolation of the previous and subsequent valid points is used to generate the missing fields. Furthermore, data sources with more than three consecutive missing fields are marked as high-risk.

[0027] Numerical compliance verification includes checking the value range boundaries, locking the moving speed range to 0-25 m / s, setting the lower limit of the order amount to the restaurant's minimum delivery price, and performing logical conflict detection. Compliance is verified when the angle between the moving direction and the planned path is greater than 70 degrees for 30 seconds or when the order status and payment status have enumeration value conflicts.

[0028] Format specification review: data format is verified through predefined pattern matchers. Timestamps must comply with the ISO 8601 standard. Coordinate point formats are forcibly converted to [longitude, latitude] arrays. When format errors are detected, the forced conversion engine is activated to automatically split string coordinates into value pairs. Local timestamps are converted to the UTC time zone.

[0029] Output control mechanism and establish health scoring model:

[0030] Score value = (number of fields that passed the check / total number of fields) × confidence coefficient

[0031] The confidence coefficient is a dynamically adjusted weight factor, quantified based on the historical performance of the data source and its real-time environmental impact. When the score is 0.9, it is injected into the business processing flow. When the score is 0.7-0.9, it enters the secondary compliance channel. When the score is 0.7, it is marked as an invalid data source.

[0032] Optionally, the business-level anomaly detection includes the following processing units:

[0033] Benchmark model construction, generating the optimal path set based on real-time road network topology, and calculating the theoretical delivery time T 理论 , where T 理论 = (road section length / standard speed) + fixed buffer time. Meanwhile, extract data from the same section and time period over the past 30 days, use the P85 quantile to establish an initial threshold, and generate a historical baseline.

[0034] Real-time progress tracking, progress comparison every 30 seconds, calculation of actual progress rate and planned progress rate, where actual progress rate = traveled distance / total route length, planned progress rate = (current time - order time) / T 理论 , Deviation rate = (actual progress rate - planned progress rate) / planned progress rate;

[0035] Dynamic threshold management, setting up a time-based adaptive mechanism to divide the peak hours into morning, afternoon, evening, and off-peak periods. The P80 percentile is used during peak hours, and the P90 percentile is used during off-peak hours. An environmental compensation strategy is also set up to adjust the threshold in extreme weather such as heavy rain or during traffic control.

[0036] Multi-dimensional root cause analysis: When an alert is triggered, correlation detection is initiated. At the restaurant level, the current order backlog rate is retrieved and compared with the restaurant's historical food delivery speed. At the rider level, the number of simultaneous delivery orders is detected to analyze recent operational behavior patterns. At the environmental level, real-time traffic congestion index is accessed to obtain meteorological disaster warning levels.

[0037] The warning output range adopts a graded response strategy based on the duration of the deviation rate > threshold. When the duration is 60 seconds, the third-level warning is triggered and the system records it. When the duration is 60-120 seconds, the second-level warning is triggered and pushed to the dispatcher. When the duration is 120 seconds, the first-level warning is triggered and automatic intervention is performed.

[0038] Optionally, the system-level anomaly detection includes the following processing units:

[0039] Terminal health monitoring, including heartbeat continuity review, receives terminal heartbeat signals every 2 seconds. A consecutive loss of three signals triggers a communication channel switch. Simultaneously, power attenuation mode analysis is performed. In normal mode, power decreases by 0.5%-1% per minute. If the decrease exceeds 5% per minute for 5 minutes, the system enters abnormal mode and activates the power-saving protocol.

[0040] Network quality assessment: Dynamic sampling of signal strength is performed. 4G network RSSI > -85dBm is considered healthy, and 5G network RSRP > -105dBm is considered healthy. Packet loss rate is monitored, and the network mode is switched when the packet loss rate is > 20% for 30 seconds.

[0041] Application service diagnosis includes process resource monitoring and service dependency checks. An overload is flagged when CPU usage exceeds 95% for 30 seconds. A memory leak check is performed when heap memory exceeds 85% after five consecutive GCs. A backup engine is activated when map service response latency exceeds 500ms. A circuit breaker is triggered if the payment gateway timeout exceeds three times per hour.

[0042] A hierarchical self-healing strategy uses an automated repair path. When an anomaly is detected, different solutions are adopted based on the fault type. When a communication interruption is detected, the system switches to the backup channel to restore data synchronization. When an application crash is detected, the service is remotely restarted and a status recovery check is performed. When resources are overloaded, containers are dynamically allocated for load rebalancing.

[0043] To achieve the above objectives, the present invention also provides an electronic device, including: a memory for storing computer software programs; a processor for reading and executing the computer software programs, thereby realizing any of the above-mentioned intelligent management systems for catering delivery information.

[0044] To achieve the above objectives, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned intelligent management systems for catering delivery information.

[0045] The present invention provides an intelligent management system for catering delivery information based on big data. It obtains multi-source heterogeneous data streams in real time through a data acquisition module, performs multi-level monitoring by an anomaly detection engine after three-level cleaning processing, and dynamically updates rules in combination with a self-optimizing knowledge base to ultimately output optimized delivery instructions. This effectively solves the problems of low data processing efficiency and delayed exception response in traditional systems, and has the significant advantage of improving delivery timeliness and service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a schematic diagram of the structure of the intelligent management system for catering delivery information provided by an embodiment of the present invention;

[0047] Figure 2 An embodiment diagram of an electronic device provided by an embodiment of the present invention;

[0048] Figure 3 An embodiment diagram of a computer-readable storage medium provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include one or more features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0051] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or illustration". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0052] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of the catering delivery information intelligent management system provided by the present invention, which includes:

[0053] The data acquisition module 10 forms a multi-source heterogeneous data stream from the real-time delivery data collected from the delivery equipment through a distributed message queue, and inputs the multi-source heterogeneous data stream into the data processing module. The multi-source heterogeneous data stream includes order feature data stream, delivery dynamic data stream and environmental status data stream.

[0054] Specifically, the data collection module refers to the component used to obtain various types of real-time data during the distribution process. It uses distributed message queue technology to achieve parallel access to multi-source heterogeneous data, and realizes real-time collection of high-throughput data streams through middleware such as Kafka or RabbitMQ. This module solves the problem of single data source and poor timeliness under the traditional manual review model.

[0055] Furthermore, the multi-source heterogeneous data streams include order feature data streams, delivery dynamic data streams and environmental status data streams. The merchant terminal collects the order generation timestamp, delivery destination coordinates, estimated cooking time, and user special requirement identification to form the order feature data stream. The delivery personnel's mobile device collects GPS positioning points, battery power status, network signal strength value, and moving speed vector every second to form the delivery dynamic data stream. The real-time traffic event coordinates, weather warning level, and road closure status sign information are obtained through third-party API access to form the environmental status data stream.

[0056] Furthermore, the order feature data stream generates structured order information through the merchant terminal. The delivery dynamics data stream captures physical state parameters in real time through mobile device sensors. The environmental state data stream obtains dynamic environmental variables through external system interfaces. The order generation timestamp is in UTC standard time format, the delivery destination coordinates are stored in geocoded form, the estimated cooking time is quantized in minutes, and user special requirements are categorized using preset enumeration values. GPS positioning points in the delivery dynamics data stream are recorded in the WGS-84 coordinate system, the battery charge status is transmitted with percentage accuracy, the network signal strength value is quantized in dBm, the movement speed vector includes rate and direction components, and the traffic event coordinates in the environmental state data stream are associated with road network topology nodes. The meteorological warning level is graded according to national emergency standards, and the road closure status flag is marked with a Boolean value.

[0057] Specifically, after the order feature data stream is generated by the merchant terminal, the timestamp is used for subsequent spatiotemporal alignment operations, the delivery destination coordinates are used as input parameters for the path planning algorithm, and the estimated cooking time is coordinated with the moving speed vector in the delivery dynamic data to calculate the estimated arrival time.

[0058] The GPS positioning points in the dynamic distribution data stream are collected every second through the built-in positioning module of the mobile device to form a continuous trajectory sequence. The battery power status and network signal strength value are used to evaluate the communication reliability of the terminal device. The mobile speed vector is generated by fusing the acceleration sensor and GPS differential data.

[0059] The environmental status data stream triggers dynamic route adjustment after the real-time traffic event coordinates are matched with the road network topology by calling a third-party API interface. The emergency delivery strategy is activated when the weather warning level exceeds the set threshold. The road closure status flag is updated through a polling mechanism and injected into the route optimization algorithm. The user special requirement identifier in the order feature data stream is converted into machine-readable instructions through predefined coding rules and combined with the moving speed vector in the delivery dynamic data to dynamically adjust the delivery priority. The weather warning level in the environmental status data stream is associated with the battery power status in the delivery dynamic data, triggering the equipment power compensation mechanism in low temperature environments.

[0060] In the specific implementation, the data collection module uses Apache Kafka as a distributed message queue and establishes three independent topics to receive order feature data, delivery dynamic data and environmental status data respectively. The order feature data is pushed in batches every second by the restaurant POS system, the delivery dynamic data is reported by the rider's mobile device with GPS coordinates every second, and the environmental status data is updated once a minute through a third-party API. The order feature data stream includes the order generation timestamp, delivery destination coordinates, estimated cooking time, and user special requirement identification. The delivery dynamic data stream includes the GPS positioning point per second, battery power status, network signal strength value, and moving speed vector. The environmental status data stream includes real-time traffic event coordinates, weather warning level, and road closure status sign.

[0061] In the order feature data stream, the order generation timestamp uses the ISO 8601 standard format, accurate to milliseconds. The delivery destination coordinates use the WGS84 coordinate system, accurate to 6 decimal places. The estimated cooking time is in minutes. User special requirements include Boolean values ​​such as whether contactless delivery is required and whether tableware is required. In the delivery dynamic data stream, the GPS positioning point collection frequency is once per second, using the WGS84 coordinate system. The battery power status is expressed in percentage, and the network signal strength value is in dBm units. The moving speed vector includes the speed magnitude and direction. The speed unit is meters per second, and the direction is 0 degrees with due north, rotating clockwise. In the environmental status data stream, the real-time traffic event coordinates also use the WGS84 coordinate system. The weather warning level is divided into four levels: blue, yellow, orange, and red. The road closure status flag is a Boolean value, true indicates closure, and false indicates normal traffic.

[0062] Through the above technical solution, comprehensive collection of multi-source heterogeneous data is achieved, providing a rich data foundation for subsequent anomaly detection and intelligent decision-making. By integrating real-time data streams in the three dimensions of orders, delivery, and environment, the system can fully grasp all key information in the delivery process, thereby improving the accuracy and timeliness of anomaly detection.

[0063] The data processing module 20 performs three-level serial cleaning operations on the multi-source heterogeneous data stream, obtains a structured data set after cleaning, and inputs the structured data set into the anomaly detection engine.

[0064] Specifically, the data processing module refers to the component that performs standardized processing on the original data. It uses a three-level serial cleaning operation to improve data quality. It filters invalid data layer by layer through the field repair algorithm, spatiotemporal alignment model and business rule engine, overcoming the high misjudgment rate and single data processing dimension caused by simple threshold judgment.

[0065] Furthermore, the three-level serial cleaning operations are field integrity repair, spatiotemporal dimension alignment, and business logic verification. In the field integrity repair phase, the filling mechanism is triggered when the required fields are detected to be missing. The missing numerical fields are filled by calling the average of the orders completed in the last hour within a radius of 500 meters in the same business district. The coordinate fields are filled by linear interpolation using the previous valid positioning points. In the time dimension alignment phase, the time axis is reconstructed and the cubic spline interpolation algorithm is used to fill the missing time period trajectory points to ensure trajectory continuity. Spatial mapping is used to project the original coordinates of the road network topology matching algorithm onto the actual road network. The projection error threshold is used to identify abnormal positioning points, and cross-source data synchronization is performed. The related data segments are intercepted based on the sliding window, and the linear regression model is applied to perform trend extrapolation to compensate for time deviation. In the business logic verification phase, predefined rules are applied to filter illegal data, and geographic fence verification is used to automatically discard coordinate points that are more than 10 meters beyond the delivery area boundary. Speed ​​mutation verification is used to initiate trajectory smoothing for coordinate points whose speed change rate exceeds 50% for two consecutive seconds.

[0066] Specifically, field integrity repair ensures data integrity through numerical filling and coordinate interpolation to avoid subsequent processing interruptions due to missing values. Time dimension alignment is achieved through timeline reconstruction, with the order creation time as the benchmark timeline. The delivery personnel's trajectory points are rearranged in chronological order, and continuous trajectory curves are generated by cubic spline interpolation for missing periods to eliminate trajectory breaks caused by device clock asynchrony. Spatial mapping is performed, and the road network is modeled as a directed graph structure. Coordinate points are matched to corresponding road edges through the nearest neighbor algorithm. When the projection distance exceeds 20 meters, an abnormal marking mechanism is triggered to avoid incorrect path planning. In the cross-source data synchronization stage, when it is detected that the time deviation between order and delivery data exceeds 30 seconds, a 60-second data window before and after the deviation period is extracted, and the least squares method is used to fit the data change trend to generate a compensated synchronized data stream to ensure that multi-source data are processed collaboratively under a unified time reference.

[0067] Business logic verification filters abnormal data through geographic fences and speed mutation rules. For example, it discards coordinate points that are 10 meters beyond the delivery area and smoothes the trajectories of coordinate points with speed mutations exceeding 50% to prevent abnormal data from entering subsequent modules. The three-level cleaning operation forms a serial processing flow, and the output of the previous step serves as the input of the next step. For example, the data after field repair enters the spatiotemporal alignment stage, and the structured data after spatiotemporal alignment enters the business verification stage, improving data quality layer by layer. Numerical parameters such as the 500-meter business district radius, the 10-meter geographic fence threshold, and the 50% speed change rate threshold are determined through historical data analysis to balance processing efficiency and accuracy.

[0068] In the specific implementation, the three-level serial cleaning operation sequentially performs field integrity repair, spatiotemporal dimension alignment, and business logic verification.

[0069] During the field integrity repair phase, the filling mechanism is automatically triggered when a required field is detected to be missing. For missing numeric fields, the average value of orders completed in the last hour within a 500-meter radius of the same business district is used to fill the missing field. For example, if the delivery fee field of an order is missing, the system automatically calculates the average delivery fee of orders completed in the past hour within a radius of 500 meters and fills the missing field with this average value.

[0070] For coordinate fields, linear interpolation is used to fill in the gaps using the previous valid positioning points. Specifically, if a GPS coordinate point is missing, the system will take the two valid coordinate points before and after the point and calculate the coordinate value of the missing point through linear interpolation. The spatiotemporal dimension alignment stage includes time axis reconstruction, spatial mapping, and cross-source data synchronization. The time axis reconstruction uses the order creation time as the time origin and rearranges the delivery person's trajectory points. The missing period is filled in using the cubic spline interpolation method. Specifically, valid data points before and after the missing period are selected to construct a cubic spline function to calculate the interpolation value of the missing period.

[0071] Spatial mapping abstracts roads as directed edges and intersections as nodes, establishes a road network topology, and matches the original GPS coordinate points to the nearest road edge. When the projection error exceeds 20 meters, the point is marked as an outlier. In this way, a mapping relationship between GPS coordinates and road network topology is established. Cross-source data synchronization is initiated when the time deviation between order data and delivery data exceeds 30 seconds. A 60-second sliding window is used to capture 60 seconds of data before and after the deviation period. Trend extrapolation is performed based on this 120-second data to achieve synchronization between different data sources.

[0072] During the business logic verification phase, predefined rules are applied to filter out illegal data. Geographic fence verification is used to automatically discard coordinate points that are more than 10 meters beyond the delivery area boundary. Speed ​​mutation verification is used to initiate trajectory smoothing when the speed change rate of a coordinate point exceeds 50% for two consecutive seconds. Specifically, the system first sets a virtual geographic fence to mark GPS points that are more than 10 meters beyond the delivery area as abnormal and eliminate them. Secondly, the system continuously monitors the delivery person's movement speed. If the speed change exceeds 50% within two consecutive seconds, it is considered that GPS drift may exist. At this time, trajectory smoothing algorithms such as Kalman filtering are activated to correct the abnormal trajectory.

[0073] Through the above technical solutions, the original distribution data is fully cleaned and standardized. Field integrity repair ensures the integrity of the data and reduces analysis deviations caused by missing data. The alignment of spatiotemporal dimensions improves the consistency of multi-source data and provides a unified spatiotemporal reference system for subsequent analysis. Business logic verification effectively filters out abnormal data and improves data quality.

[0074] The anomaly detection engine 30 performs anomaly detection on the structured data set, determines the data anomaly type as a single anomaly or a compound conflict based on the anomaly detection result, and outputs the detection result to the conflict arbitration module.

[0075] Specifically, the anomaly detection engine refers to an intelligent analysis component that identifies abnormal conditions in the distribution process. Specifically, a multi-layer detection architecture can be used to achieve comprehensive monitoring from basic data to business logic. It detects numerical anomalies through the isolation forest algorithm, identifies logical contradictions based on the business rule engine, and uses the system health model to diagnose operational failures. This module breaks through the limitations of traditional methods that only focus on a single type of anomaly.

[0076] Furthermore, the anomaly detection engine establishes a three-tiered defense system at the data, business, and system levels, forming a complete monitoring chain from basic data verification to high-level business assurance. Data-level anomaly detection includes five sequential execution units: data reception and diversion, field integrity verification, numerical compliance verification, format specification review, and output control mechanism. Business-level anomaly detection includes five progressive processing steps: baseline model construction, real-time progress tracking, dynamic threshold management, multi-dimensional root cause analysis, and early warning output specification. System-level anomaly detection includes four collaborative modules: terminal health monitoring, network quality assessment, application service diagnosis, and hierarchical self-healing strategy.

[0077] Furthermore, data-level anomaly detection imports the cleaned data stream into the detection channel through data reception and diversion, and automatically diverts it to a dedicated validator according to the data type, including the order data validator, trajectory data validator or environmental data validator. The field integrity verification scans the missing status of required fields and triggers the automatic filling mechanism. For missing numerical fields, the historical average of the last 10 orders within a radius of 300 meters in the same business district is called to fill the missing data. For coordinate fields, linear interpolation of the previous valid points and subsequent valid points is used to generate the missing data. At the same time, when more than three fields are missing continuously, it will be marked as a high-risk data source.

[0078] Numerical compliance verification checks the value range boundaries, locks the moving speed range to 0-25 meters per second, sets the lower limit of the order amount to the restaurant's minimum delivery price, and performs logical conflict detection. Compliance is verified when the angle between the moving direction and the planned path is greater than 70 degrees for 30 seconds or when the order status and payment status have enumeration values ​​that conflict. Format specification review verifies the data format through a predefined pattern matcher. Timestamps must comply with the ISO 8601 standard. The coordinate point format is forcibly converted to a [longitude, latitude] array. When a format error is detected, the forced conversion engine is activated, automatically splitting the string coordinates into numeric pairs and converting the local timestamp to the UTC time zone. The output control mechanism establishes a health scoring model. The confidence coefficient is a dynamically adjusted weight factor ranging from 0.5 to 1.2. It is quantified based on the historical performance of the data source and the real-time environmental impact. A score of 0.9 is injected into the business processing flow, a score of 0.7-0.9 enters the secondary compliance channel, and a score of 0.7 is marked as an invalid data source.

[0079] Among them, data reception and diversion adopt a multi-channel parallel processing architecture, and the data type is identified through the data header identifier. The order data validator is configured with a JSON parser, the trajectory data validator is integrated with a spatial index engine, and the environmental data validator has a built-in regular expression library. The field integrity verification sets a dual-path filling logic. The numerical filling adopts the sliding window average algorithm, and the coordinate filling applies the linear interpolation formula. Continuous missing triggers the data source credibility degradation mechanism. The numerical compliance verification constructs a multidimensional constraint matrix. The speed threshold is set based on traffic regulations. The lower limit of the order amount is associated with the merchant configuration parameters. Logical conflict detection uses a state machine model for enumeration value matching. The format specification review deploys a pattern matching rule library. The timestamp conversion calls the time zone mapping table. The coordinate splitting uses string segmentation and type conversion functions. The output control mechanism implements dynamic weight calculation. The confidence coefficient is adjusted in real time according to the data source error rate and the environmental interference level. The health scoring model adopts a weighted sum algorithm, and the weight factor is determined by the historical error frequency and real-time network delay.

[0080] Business-level anomaly detection achieves anomaly monitoring through benchmark model construction, real-time progress tracking, dynamic threshold management, multi-dimensional root cause analysis, and warning output range. The benchmark model construction generates an optimal path set based on the real-time road network topology, calculates the theoretical delivery time, and extracts data from the same period and road section in the past 30 days. The initial threshold is established using the P85 quantile. Real-time progress tracking performs progress comparison every 30 seconds, calculates the actual progress rate and the planned progress rate, and the deviation rate is calculated by the difference between the actual and planned progress rates.

[0081] Dynamic threshold management divides delivery times into morning, midday, evening, and off-peak periods. Peak periods use the 80th percentile, while off-peak periods use the 90th percentile. Thresholds are adjusted during heavy rain or traffic control. Multidimensional root cause analysis initiates correlation detection when an alert is triggered, drawing on restaurant order backlog rates, the number of delivery orders by riders, and the environmental traffic congestion index. The alert output range triggers a graded response based on the duration that the deviation rate exceeds the threshold. During the baseline model construction phase, theoretical delivery times are calculated using the road network topology. Initial thresholds are established based on the 85th percentile of historical data, providing a benchmark for dynamic adjustments. Real-time progress tracking utilizes a dual progress rate comparison mechanism, capturing delivery deviations through both temporal and spatial metrics. Dynamic threshold management enables flexible threshold adjustment through time division. During peak periods, threshold sensitivity is reduced to cover more potential anomalies, while threshold accuracy is improved during off-peak periods. An environmental compensation strategy introduces external variables for dynamic threshold correction. Multidimensional root cause analysis establishes a three-dimensional correlation detection matrix for restaurants, riders, and the environment, eliminating single-dimensional misjudgments through horizontal comparison. The graded warning response mechanism quantifies anomaly severity based on duration and matches it to different response strategies.

[0082] System-level anomaly detection ensures the reliable operation of terminal devices and service components through terminal health monitoring, network quality assessment, application service diagnosis and hierarchical self-healing strategies. Terminal health monitoring includes heartbeat continuity review and power decay pattern analysis. The heartbeat signal is received every 2 seconds. The loss of 3 consecutive signals triggers the communication channel switching. In normal mode, the power level drops by 0.5%-1% per minute. In abnormal mode, the power level drops by more than 5% per minute for 5 minutes to activate the power saving protocol. Network quality assessment is based on dynamic sampling of signal strength and packet loss rate monitoring. The RSSI health threshold for 4G networks is -85dBm, and the RSRP health threshold for 5G networks is -105dBm. When the packet loss rate exceeds 20% for 30 seconds, the network mode is switched. Application service diagnosis implements process resource monitoring and service dependency checks. CPU usage exceeds 95% for 30 seconds, which is marked as overloaded. Heap memory exceeds 85%, which triggers memory leak detection. If the map service response delay exceeds 500ms, the backup engine is enabled. The hierarchical self-healing strategy executes differentiated repair paths based on the fault type. When communication is interrupted, the backup channel is switched to synchronize data. When the application crashes, the service is remotely restarted and a status recovery check is performed. When resources are overloaded, containers are dynamically allocated to achieve load rebalancing.

[0083] In the specific implementation, data-level anomaly detection is achieved through the following steps: First, data reception and diversion imports the cleaned data stream into the detection channel, and automatically diverts it to a dedicated verifier according to the data type. Order data enters the order data verifier, trajectory data enters the trajectory data verifier, and environmental data enters the environmental data verifier.

[0084] Secondly, the field integrity verification scans the missing status of required fields. For example, it checks whether the order ID, timestamp, and coordinate points exist. When missing, the automatic filling mechanism is triggered. For numeric fields, the historical average of the last 10 orders within a radius of 300 meters in the same business district is called to fill the gap. For missing coordinate fields, linear interpolation of the previous valid points and subsequent valid points is used to generate the gap. When more than three fields are missing continuously, the system marks the data source as a high-risk data source.

[0085] Secondly, the numerical compliance verification performs a value range boundary check. The system locks the moving speed range to 0-25 meters per second, sets the lower limit of the order amount to the restaurant's minimum delivery price, and performs logical conflict detection. When the angle between the moving direction and the planned path is greater than 70 degrees for 30 seconds, or when the enumeration value of the order status and the payment status conflict, the system verifies compliance.

[0086] Then, the format specification review verifies the data format through predefined pattern matchers. The timestamp must comply with the ISO8601 standard. The coordinate point format is forced to be converted to a [longitude, latitude] array. When a format error is detected, the forced conversion engine is started, and the string coordinates are automatically split into value pairs. The local timestamp is converted to the UTC time zone.

[0087] Finally, the output control mechanism establishes a health score model. The confidence coefficient is a dynamically adjusted weight factor ranging from 0.5 to 1.2. It is quantified based on the historical performance of the data source and the real-time environmental impact. When the score is greater than or equal to 0.9, the data is injected into the business processing flow. When the score is between 0.7 and 0.9, the data enters the secondary compliance channel. When the score is less than or equal to 0.7, the system marks the data as an invalid data source. Through the above technical solutions, comprehensive automated anomaly detection of catering delivery information is achieved. The system can quickly identify and handle anomalies such as missing data, format errors, and logical conflicts, improving data quality and reliability. Through a multi-level verification mechanism, the system effectively reduces the workload of manual review and improves the efficiency and accuracy of anomaly detection. The dynamically adjusted health score model enables the system to adapt to the characteristics of different data sources and achieve more accurate anomaly identification and classification processing.

[0088] Business-level anomaly detection is achieved through benchmark model construction, real-time progress tracking, dynamic threshold management, multi-dimensional root cause analysis, and early warning output specifications. When constructing the benchmark model, the optimal path set is first generated based on the real-time road network topology, and the theoretical delivery time is calculated. The theoretical delivery time calculation formula is the road section length divided by the standard speed plus the fixed buffer time. At the same time, data from the same section at the same time in the past 30 days are extracted, and the initial threshold is established with the P85 quantile to generate a historical baseline. The real-time progress tracking link performs progress comparison every 30 seconds, and calculates the actual progress rate and planned progress rate. The actual progress rate is equal to the distance traveled divided by the total length of the path, and the planned progress rate is equal to the current time minus the order acceptance time divided by the theoretical delivery time. The deviation rate is calculated as the actual progress rate minus the planned progress rate divided by the planned progress rate.

[0089] Dynamic threshold management adopts a time-based adaptive mechanism, dividing a day into four periods: morning peak, noon peak, evening peak, and off-peak. The P80 percentile is used as the threshold during peak hours, and the P90 percentile is used during off-peak hours. At the same time, an environmental compensation strategy is set to adjust the threshold in extremely severe weather such as heavy rain or traffic control. Multi-dimensional root cause analysis starts correlation detection when the warning is triggered. In the restaurant dimension, the current order backlog rate is retrieved and compared with the restaurant's historical food delivery speed. In the rider dimension, the number of simultaneous delivery orders is detected to analyze recent operating behavior patterns. In the environmental dimension, the real-time traffic congestion index is connected to obtain the meteorological disaster warning level. The warning output specification adopts a graded response strategy based on the duration of the deviation rate exceeding the threshold. When the duration does not exceed 60 seconds, a third-level warning is triggered and recorded in the system. When the duration is between 60 and 120 seconds, a second-level warning is triggered and pushed to the dispatcher. When the duration exceeds 120 seconds, a first-level warning is triggered for automatic intervention.

[0090] The above technical solutions enable accurate identification and timely response to anomalies in delivery operations. The baseline model construction takes into account historical data and real-time traffic conditions, improving the accuracy of anomaly detection. Real-time progress tracking and dynamic threshold management can adapt to the characteristics of delivery under different time periods and environmental conditions, reducing false positives and missed reports. Multi-dimensional root cause analysis helps quickly locate the source of problems and improve processing efficiency. The graded warning output mechanism ensures the appropriate handling of anomalies of varying severity and optimizes system resource utilization. Overall, this solution significantly improves the controllability and stability of the delivery process, effectively reducing delivery delays and customer complaints.

[0091] During the terminal health monitoring process in system-level anomaly detection, heartbeat signals are received at intervals of every 2 seconds. When three consecutive signals are lost, the communication channel switching operation is performed, and the power consumption pattern is continuously analyzed. In normal mode, the power consumption drops by 0.5% to 1% per minute. When it is detected that the power consumption drops by more than 5% per minute and lasts for five minutes, the power saving protocol is activated to extend the device operation time. The network signal strength is monitored through dynamic sampling. For 4G networks, the received signal strength indicator value must be higher than -85dBm to maintain a healthy state, and the reference signal receiving power of 5G networks must be higher than -105dBm. When the data packet loss rate exceeds 20% for 30 seconds, the network mode switching operation is performed to ensure communication stability.

[0092] The running status of the application service is diagnosed through process resource monitoring. When the CPU usage exceeds 95% and lasts for 30 seconds, the system is marked as overloaded. In terms of memory management, if the heap memory usage still exceeds 85% after five consecutive garbage collections, the memory leak detection program is started. When the response delay of the map service exceeds 500 milliseconds, the backup engine is enabled to ensure service continuity. When the payment gateway has more than three timeouts per hour, the circuit breaker mechanism is triggered to prevent further service requests. For the detected anomaly type, an automated repair strategy is implemented. When communication is interrupted, the system automatically switches to the backup communication channel to restore data transmission. When an application service crash is identified, a remote restart operation is performed and the service status recovery verification is performed. In the resource overload scenario, container resources are dynamically reallocated to achieve load balancing optimization.

[0093] Through the above technical solutions, multi-dimensional anomaly detection and autonomous repair capabilities for terminal devices, network environments and application services are realized, effectively solving the response lag problem caused by traditional methods relying on manual processing. Through preset dynamic thresholds and automated repair mechanisms, the system can quickly locate and restore faults without human intervention, significantly improving the operational reliability and business continuity of service components. The hierarchical self-healing strategy implements precise handling of different types of system anomalies, avoiding secondary failures that may be caused by a single processing method, and ensuring the stable issuance of delivery instructions and the healthy operation of execution terminals.

[0094] The conflict arbitration module 40 classifies and handles data anomalies according to the detection results, issues handling instructions, and inputs handling records into the self-optimization knowledge base.

[0095] Specifically, the conflict arbitration module refers to the decision-making component for handling complex exception scenarios. It can generate disposal strategies by combining a rule engine with case-based reasoning. It uses a decision tree classifier to prioritize exception types and match them with preset disposal plans. This module solves the problem of difficulty in responding quickly to complex exceptions.

[0096] In the specific implementation, the conflict arbitration module handles the detected exceptions according to the preset priority rules. For a single exception, the corresponding processing flow is directly triggered. For a complex conflict where multiple exceptions occur simultaneously, the processing order and method are determined by comprehensively considering factors such as the exception level and the scope of impact.

[0097] The self-optimizing knowledge base 50 dynamically updates the anomaly detection rules of the anomaly detection engine according to the handling records, and feeds back the optimized decision strategy to the anomaly detection engine.

[0098] Specifically, the self-optimizing knowledge base refers to a knowledge management system that stores exception handling experience. Specifically, a graph database can be used to store the relationship between exception cases and handling plans, and the detection rule base and decision-making strategy base are dynamically updated through online learning algorithms. This module realizes the self-evolution capability of the system to continuously improve detection accuracy during operation.

[0099] In its specific implementation, the self-optimizing knowledge base uses machine learning algorithms to continuously update anomaly detection rules and decision-making strategies based on historical arbitration results and manual intervention records. For example, it dynamically adjusts the anomaly threshold of delivery time based on historical data, or automatically adjusts the path planning strategy based on weather conditions.

[0100] The decision output module 60 executes the path optimization algorithm and resource scheduling strategy for the handling instructions, generates the delivery instructions and sends them to the execution terminal.

[0101] Specifically, the decision output module refers to an intelligent scheduling component that generates optimized delivery instructions. It can use genetic algorithms to solve path optimization problems and dynamically adjust resource allocation strategies through reinforcement learning models. This module optimizes the response delay and strategy rigidity problems existing in traditional manual scheduling.

[0102] In the specific implementation, the decision output module uses an improved AI algorithm to optimize the route based on the latest road network conditions and order information, while taking into account the rider's current load and skill level to perform intelligent scheduling and allocation. The final delivery instructions are pushed to the rider through the mobile application for execution.

[0103] Reference Figure 2 , Figure 2 The electronic device provided by the embodiment of the present invention is shown in FIG. Figure 2 As shown, the present invention also provides an electronic device 200, including: a memory 210 for storing computer software programs; a processor 220 for reading and executing the computer software program 211, thereby realizing any of the above-mentioned intelligent management systems for catering delivery information.

[0104] Reference Figure 3 , Figure 3 An embodiment diagram of a computer-readable storage medium provided in an embodiment of the present invention, such as Figure 3 As shown, the present invention also provides a computer program product 300, including a computer program 211, which implements any of the above-mentioned intelligent management systems for catering delivery information when executed by a processor.

[0105] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0106] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A catering delivery information intelligent management system based on big data, characterized by: It includes data acquisition module, data processing module, anomaly detection engine, conflict arbitration module, self-optimization knowledge base and decision output module; The data acquisition module converts the real-time delivery data collected from the delivery equipment into a multi-source heterogeneous data stream through a distributed message queue, and inputs the multi-source heterogeneous data stream into the data processing module. The multi-source heterogeneous data stream includes an order feature data stream, a delivery dynamic data stream, and an environmental status data stream; The data processing module performs a three-stage serial cleaning operation on the multi-source heterogeneous data stream, obtains a structured data set from the cleaned multi-source heterogeneous data stream, and inputs the structured data set into the anomaly detection engine; The anomaly detection engine performs anomaly detection on the structured data set, determines the data anomaly type as a single anomaly or a compound conflict based on the anomaly detection result, and outputs the detection result to the conflict arbitration module; The conflict arbitration module classifies and handles data anomalies according to the detection results, issues handling instructions, and inputs handling records into the self-optimization knowledge base; The self-optimizing database dynamically updates the anomaly detection rules of the anomaly detection engine according to the handling records, and feeds back the optimized decision strategy to the anomaly detection engine; The decision output module executes a path optimization algorithm and a resource scheduling strategy on the handling instruction, generates a delivery instruction, and sends it to the execution terminal.

2. The system according to claim 1, characterized in that The multi-source heterogeneous data stream includes an order feature data stream, a delivery dynamic data stream and an environmental status data stream. The order feature data stream is formed based on the merchant terminal collecting the order generation timestamp, delivery destination coordinates, estimated cooking time, and user special requirement identification. The delivery dynamic data stream is formed based on the delivery personnel's mobile device collecting GPS positioning points, battery power status, network signal strength value, and moving speed vector per second. The environmental status data stream is formed by accessing the real-time traffic event coordinates, weather warning level, and road closure status sign information through a third-party API.

3. The system according to claim 1, characterized in that The three-level serial cleaning operations are field integrity repair, spatiotemporal dimension alignment, and business logic verification. When a required field is missing in the input structured dataset, the filling mechanism is automatically triggered. For numeric fields, the missing value is filled by using the average value of orders completed in the last hour within a 500-meter radius of the same business district. For coordinate fields, linear interpolation is performed using the previous valid positioning points. The processed structured dataset is then aligned in time and space. For the input structured dataset, a time-based time axis is created based on the order, and the road network is modeled as a directed graph structure with time as the reference time axis. This ensures that multi-source data is collaboratively processed under a unified time-space reference. The processed structured dataset is then validated for business logic. Predefined rules are applied to the input structured data set to filter illegal data. Geographic fence verification is used to automatically discard coordinate points that are more than 10 meters beyond the delivery area boundary. Speed ​​mutation verification is used to initiate trajectory smoothing for coordinate points whose speed change rate exceeds 50% for two consecutive seconds.

4. The system according to claim 3, characterized in that The spatiotemporal alignment specifically includes timeline reconstruction, spatial mapping, and cross-source data synchronization, wherein: The timeline reconstruction takes the order creation time as the time origin T0, rearranges the deliveryman's trajectory points according to T0, and fills the missing time periods with cubic spline interpolation; Spatial mapping abstracts roads into directed edges and intersections into nodes, establishes a road network topology, matches original coordinate points to the nearest road edge, marks points as outliers if the projection error exceeds 20 meters, and establishes a mapping relationship between GPS coordinates and road network topology. Cross-source data synchronization starts a sliding window compensation mechanism when the time deviation between order data and delivery data exceeds 30 seconds, intercepting 60 seconds of data before and after the deviation period for trend extrapolation.

5. The system according to claim 1, characterized in that: The anomaly detection engine builds a three-layer defense system at the data level, business level, and system level, forming a complete monitoring chain from basic data verification to high-level business assurance; Data-level anomaly detection is completed through data reception and diversion, field integrity verification, numerical compliance verification, format specification review, and output control mechanisms to ensure the validity and compliance of the original data; Through benchmark model construction, real-time progress tracking, dynamic threshold management, multi-dimensional root cause analysis, and early warning output specifications, business-level anomaly detection is completed to monitor operational deviations of core delivery business indicators. System-level anomaly detection is completed through terminal health monitoring, network quality assessment, application service diagnosis and hierarchical self-healing strategies to ensure the reliable operation of terminal devices and service components.

6. The system according to claim 5, characterized in that The data-level anomaly detection includes the following processing units: Data reception and diversion: import the cleaned data stream into the detection channel and automatically divert it to a dedicated verifier according to the data type, including order data verifier, trajectory data verifier or environmental data verifier; Field integrity verification scans for missing required fields and triggers an automatic filling mechanism. For missing numeric fields, the filling mechanism uses the historical average of the last 10 orders within a 300-meter radius of the same business district. For missing coordinate fields, linear interpolation of the previous and subsequent valid points is used to generate the missing fields. Furthermore, data sources with more than three consecutive missing fields are marked as high-risk. Numerical compliance verification includes checking the value range boundaries, locking the moving speed range to 0-25 m / s, setting the lower limit of the order amount to the restaurant's minimum delivery price, and performing logical conflict detection. Compliance is verified when the angle between the moving direction and the planned path is greater than 70 degrees for 30 seconds or when the order status and payment status have enumeration value conflicts. Format specification review: data format is verified through predefined pattern matchers. Timestamps must comply with the ISO 8601 standard. Coordinate point formats are forcibly converted to [longitude, latitude] arrays. When format errors are detected, the forced conversion engine is activated to automatically split string coordinates into value pairs. Local timestamps are converted to the UTC time zone. Output control mechanism and establish health scoring model: Score value = (number of fields that passed the check / total number of fields) × confidence coefficient The confidence coefficient is a dynamically adjusted weight factor, quantified based on the historical performance of the data source and its real-time environmental impact. When the score is >0.9, it is injected into the business processing flow. When the score is 0.7-0.9, it enters the secondary compliance channel. When the score is <0.7, it is marked as an invalid data source.

7. The system according to claim 5, characterized in that The business-level anomaly detection includes the following processing units: Benchmark model construction, generating the optimal path set based on real-time road network topology, and calculating the theoretical delivery time T 理论 , where T 理论 = (road section length / standard speed) + fixed buffer time. Meanwhile, extract data from the same section and time period over the past 30 days, use the P85 quantile to establish an initial threshold, and generate a historical baseline. Real-time progress tracking, progress comparison every 30 seconds, calculation of actual progress rate and planned progress rate, where actual progress rate = traveled distance / total route length, planned progress rate = (current time - order time) / T 理论 , Deviation rate = (actual progress rate - planned progress rate) / planned progress rate; Dynamic threshold management, setting up a time-based adaptive mechanism to divide the peak hours into morning, afternoon, evening, and off-peak periods. The P80 percentile is used during peak hours, and the P90 percentile is used during off-peak hours. An environmental compensation strategy is also set up to adjust the threshold in extreme weather such as heavy rain or during traffic control. Multi-dimensional root cause analysis: When an alert is triggered, correlation detection is initiated. At the restaurant level, the current order backlog rate is retrieved and compared with the restaurant's historical food delivery speed. At the rider level, the number of simultaneous delivery orders is detected to analyze recent operational behavior patterns. At the environmental level, real-time traffic congestion index is accessed to obtain meteorological disaster warning levels. The warning output range adopts a graded response strategy based on the duration of the deviation rate > threshold. When the duration is <60 seconds, a level 3 warning is triggered and the system records it. When the duration is 60-120 seconds, a level 2 warning is triggered and pushed to the dispatcher. When the duration is >120 seconds, a level 1 warning is triggered and automatic intervention is performed.

8. The system according to claim 5, characterized in that: The system-level anomaly detection includes the following processing units: Terminal health monitoring, including heartbeat continuity review, receives terminal heartbeat signals every 2 seconds. A consecutive loss of three signals triggers a communication channel switch. Simultaneously, power attenuation mode analysis is performed. In normal mode, power decreases by 0.5%-1% per minute. If the decrease exceeds 5% per minute for 5 minutes, the system enters abnormal mode and activates the power-saving protocol. Network quality assessment: Dynamic sampling of signal strength is performed. 4G network RSSI > -85dBm is considered healthy, and 5G network RSRP > -105dBm is considered healthy. Packet loss rate is monitored, and the network mode is switched when the packet loss rate is > 20% for 30 seconds. Application service diagnosis includes process resource monitoring and service dependency checks. An overload is flagged when CPU usage exceeds 95% for 30 seconds. A memory leak check is performed when heap memory exceeds 85% after five consecutive GCs. A backup engine is activated when map service response latency exceeds 500ms. A circuit breaker is triggered if the payment gateway timeout exceeds three times per hour. A hierarchical self-healing strategy uses an automated repair path. When an anomaly is detected, different solutions are adopted based on the fault type. When a communication interruption is detected, the system switches to the backup channel to restore data synchronization. When an application crash is detected, the service is remotely restarted and a status recovery check is performed. When resources are overloaded, containers are dynamically allocated for load rebalancing.

9. An electronic device comprising: Memory for storing computer software programs; A processor for reading and executing the computer software program, wherein when the processor executes the computer software program, it implements the intelligent management system for catering delivery information as claimed in any one of claims 1 to 8.

10. A non-transitory computer-readable storage medium storing a computer software program, wherein: When the computer software program is executed by a processor, the catering delivery information intelligent management system as claimed in any one of claims 1 to 8 is implemented.

Images