Risk prediction method, device, equipment, and storage medium for fleet operation indicators

Abstract

The invention provides a risk prediction method and device for motorcade operation indexes, equipment and a storage medium, and relates to the technical field of data processing, in particular to the technical field of intelligent transportation, automatic driving, large models and data analysis. According to the specific implementation scheme, the method comprises the steps of obtaining time sequence data of at least one target operation index from a space-time knowledge graph representing a motorcade operation situation; according to a plurality of constituent parts of the time sequence data, identifying instantaneous operation abnormity of the target operation index, and predicting a future deterioration risk; and by using the analysis model, integrating the instantaneous operation abnormality and the future deterioration risk, and outputting an analysis result. Through deep analysis of a core operation index time sequence, dual and advanced early warning of instantaneous abnormity and trend risks is realized.

Description

(19) State Intellectual Property Office (12) Invention Patent Application (10) Application Publication Number (43) Application Publication Date (21) Application Number 202511351704.8 (22) Application Date 2025.09.19 (71) Applicant Apollo Intelligent Technology (Beijing) Co., Ltd. Address 105, 1st Floor, Building 1, No. 10, Shangdi 10th Street, Haidian District, Beijing 100085 (72) Inventors Yan Wenkai, Zhang Bo, Zhang Xiaojie (74) Patent Agency Beijing Yiguang Intellectual Property Agency Co., Ltd. 11596 Patent Attorney Wang Shanshan, Wang Ying (51) Int.Cl. G06Q 10 / 0635 (2023.01) G06Q 10 / 0639 (2023.01) G06Q 10 / 083 (2024.01) G08B 31 / 00 (2006.01) (54) Invention Title: Risk Prediction Method, Apparatus, Equipment, and Storage Medium for Fleet Operation Indicators (57) Abstract: This disclosure provides a risk prediction method, apparatus, equipment, and storage medium for fleet operation indicators, relating to the field of data processing technology, particularly to intelligent transportation, autonomous driving, large-scale models, and data analysis technology. The specific implementation scheme is as follows: Time series data of at least one target operation indicator is obtained from a spatiotemporal knowledge graph representing the fleet operation status; based on multiple components of the time series data, instantaneous operational anomalies of the target operation indicator are identified, and future deterioration risks are predicted; using an analytical model, the instantaneous operational anomalies and future deterioration risks are integrated, and the analytical results are output. Through in-depth analysis of the time series of core operation indicators, dual early warning of instantaneous anomalies and trend risks is achieved. Claims 3 pages, Description 13 pages, Drawings 2 pages, CN 121279775 A 2026.01.06 CN 1 21 27 97 75 A 1. A method for risk prediction of fleet operation indicators, comprising: obtaining time series data of at least one target operation indicator from a spatiotemporal knowledge graph representing the operational status of a fleet; identifying instantaneous operational anomalies of the target operation indicator and predicting future deterioration risks based on multiple components of the time series data; and using an analysis model to integrate the instantaneous operational anomalies and the future deterioration risks to output analysis results. 2. The method according to claim 1, wherein obtaining time series data of at least one target operation indicator from a spatiotemporal knowledge graph representing the operational status includes: real-time aggregation and calculation of the time series data of the target operation indicator from the spatiotemporal knowledge graph representing the operational status of a fleet; wherein the target operation indicator includes at least one of operational efficiency indicators, service quality indicators, safety and reliability indicators, and economic benefit indicators. 3. The method according to claim 1, wherein identifying instantaneous operational anomalies of the target operation indicator based on multiple components of the time series data and predicting future deterioration risks based on multiple components of the time series data.The method according to claim 3, wherein decomposing the time series data into multiple components includes: analyzing a first component of the multiple components to determine the instantaneous operational anomaly of the target operational indicator; and analyzing a second component of the multiple components to determine the future deterioration risk of the target operational indicator. 4. The method according to claim 4, wherein analyzing the first component of the multiple components to determine the instantaneous operational anomaly of the target operational indicator includes: inputting the residual sequence into an isolated forest or variational autoencoder model to calculate an anomaly score; and determining the instantaneous operational anomaly based on the anomaly score. 6. The method according to claim 4, wherein analyzing the second component of the plurality of components to determine the future deterioration risk of the target operating indicator includes: using a Prophet or autoregressive moving average model to extrapolate and predict the future trend of the trend item sequence; determining the slope of the future trend; and determining that there is a future deterioration risk if the degree of deterioration of the slope exceeds a preset threshold. 7. The method according to claim 4, wherein analyzing the second component of the plurality of components to determine the future deterioration risk of the target operating indicator includes: converting the trend item sequence and related contextual information into a situation description text in natural language form; inputting the situation description text into a large language model; and determining the future deterioration risk based on the output of the large language model. 8. The method according to claim 1, wherein the step of using an analysis model to synthesize the instantaneous operational anomaly and the stated future deterioration risk, and outputting analysis results, includes: converting the latest values ​​of the target operational indicators, the instantaneous operational anomaly, and the future deterioration risk into a comprehensive analysis prompt in natural language form; and inputting the comprehensive analysis prompt into a large language model to generate an analysis result containing the comprehensive health of the target operational indicators. 9. The method according to claim 8, wherein the step of converting the latest values ​​of the core indicators, the instantaneous operational anomaly, and the information on the future deterioration risk into a comprehensive analysis prompt in natural language form includes: in the presence of the instantaneous operational anomaly, extracting operational event information that is temporally adjacent to the instantaneous operational anomaly from the spatiotemporal knowledge graph;The latest values ​​of the target operational indicators, the instantaneous operational anomalies, the operational event information, and the future deterioration risks are transformed into comprehensive analysis prompts in natural language form. 10. The method according to claim 1 further includes: acquiring multi-source operational data reflecting the fleet's operational status from multiple data sources, including at least a scheduling system, vehicle-side equipment, and a work order system; and constructing a spatiotemporal knowledge graph based on the multi-source operational data. 11. The method according to claim 10, wherein acquiring multi-source operational data reflecting the unmanned vehicle's operational status from multiple data sources, including at least a scheduling system, vehicle-side equipment, and a work order system, includes: real-time access to heterogeneous data streams from multiple data sources, including at least a scheduling system, vehicle-side equipment, and a work order system; and identifying and matching the heterogeneous data streams based on preset operational rules to generate structured operational events. 12. The method according to claim 11, wherein the step of identifying and matching the heterogeneous data stream based on preset operation rules to generate structured operation events includes: matching event patterns that conform to the preset operation rules in the heterogeneous data stream; generating structured operation events containing event type, timestamp, and associated entity identifier when the event pattern is satisfied; wherein the preset operation rules include at least one of the following: time window statistical rules for global operation indicators; sequence pattern rules for the operating status of a single vehicle; and state consistency rules for the execution process of a single order. 13. A risk prediction device for fleet operation indicators, comprising: an acquisition module, configured to acquire time series data of at least one target operation indicator from a spatiotemporal knowledge graph characterizing the operational status of a fleet; an analysis module, configured to identify instantaneous operational anomalies of the target operation indicator and predict future deterioration risks based on multiple components of the time series data; and an output module, configured to use an analysis model to integrate the instantaneous operational anomalies and the future deterioration risks and output analysis results. 14. An electronic device comprising: at least one processor; and a memory communicatively connected to said at least one processor; wherein the memory stores instructions executable by said at least one processor, said instructions being executed by said at least one processor to enable said at least one processor to perform the method of any one of claims 1-12. 15. A non-transitory computer-readable storage medium storing computer instructions, wherein said computer instructions are used to cause the computer to perform the method of any one of claims 1-12. 16. A computer program product comprising a computer program that, when executed by a processor, implements the method of any one of claims 1-12.121279775 A Risk Prediction Method, Device, Equipment, and Storage Medium for Fleet Operation Indicators Technical Field

[0001] This disclosure relates to the field of data processing technology, and particularly to the fields of intelligent transportation, autonomous driving, large-scale models, and data analysis technology. Background Technology

[0002] With the increasing maturity of autonomous driving technology, driverless taxis have gradually moved from the technology verification stage to large-scale commercial pilot operation. In this transformation process, the industry's focus has expanded from simply improving the autonomous driving capabilities of individual vehicles to how to ensure the safe, efficient, and economical operation of the entire fleet. Large-scale fleet operation generates massive amounts of multi-dimensional operational data, providing a foundation for achieving refined management and intelligent decision-making. Summary of the Invention

[0003] This disclosure provides a risk prediction method, device, equipment, and storage medium for fleet operation indicators.

[0004] According to one aspect of the present disclosure, a risk prediction method for fleet operation indicators is provided, comprising:

[0005] acquiring time series data of at least one target operation indicator from a spatiotemporal knowledge graph characterizing the fleet operation status;

[0006] identifying instantaneous operational anomalies of the target operation indicator and predicting future deterioration risks based on multiple components of the time series data;

[0007] using an analysis model to integrate instantaneous operational anomalies and future deterioration risks, and outputting analysis results.

[0008] According to another aspect of the present disclosure, a risk prediction device for fleet operation indicators is provided, comprising:

[0009] an acquisition module for acquiring time series data of at least one target operation indicator from a spatiotemporal knowledge graph characterizing the fleet operation status;

[0010] an analysis module for identifying instantaneous operational anomalies of the target operation indicator and predicting future deterioration risks based on multiple components of the time series data;

[0011] an output module for using an analysis model to integrate instantaneous operational anomalies and future deterioration risks, and outputting analysis results.

[0012] According to another aspect of the present disclosure, an electronic device is provided, comprising:

[0013] at least one processor; and

[0014] a memory communicatively connected to the at least one processor; wherein,

[0015] the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform any method according to any embodiment of the present disclosure.

[0016] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to perform any method according to any embodiment of the present disclosure.

[0017] According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements any method according to any embodiment of the present disclosure.

[0018] According to the solution provided in this disclosure, through in-depth analysis of the time series of core operational indicators, a dual, early warning system for instantaneous anomalies and trend risks is achieved. Specification 1 / 13 pages 5 CN 121279775 A

[0019] It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily understood through the following description. Brief Description of the Drawings

[0020] The drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0021] Figure 1 is a flowchart illustrating a risk prediction method for fleet operational indicators provided according to an embodiment of this disclosure;

[0022] Figure 2 is a flowchart illustrating a risk prediction method for fleet operational indicators provided according to another embodiment of this disclosure;

[0023] Figure 3 is a structural schematic diagram of a risk prediction device for fleet operational indicators provided according to an embodiment of this disclosure;

[0024] Figure 4 is a block diagram of an electronic device used to implement an embodiment of this disclosure. Detailed Description of Embodiments

[0025] The exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, including various details of the embodiments of the present disclosure to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0026] This embodiment discloses a risk prediction method for fleet operation indicators, which aims to achieve dual early warning of instantaneous anomalies and trend risks through in-depth analysis of the time series of core operation indicators.

[0027] Figure 1 is a flowchart illustrating the risk prediction method for fleet operation indicators provided according to an embodiment of the present disclosure. As shown in Figure 1, the method includes at least the following steps:

[0028] S110: Obtain time series data of at least one target operation indicator from a spatiotemporal knowledge graph characterizing the fleet operation status.

[0029] In this embodiment of the present disclosure, it is first necessary to extract basic data for analysis from a pre-built data model. A spatiotemporal knowledge graph can be understood as a central knowledge base that integrates multi-source operational data and is stored in a graph structure, where entities and relationships all have temporal and spatial attributes. Target operational indicators can be key quantitative parameters used to measure the operational health of a fleet, such as vehicle utilization. Time series data refers to a data sequence formed by arranging the values ​​of a target operational indicator at multiple consecutive time points in chronological order.

[0030] S120. Based on the multiple components of the time series data, identify instantaneous operational anomalies of the target operational indicator and predict the risk of future deterioration.

[0031] In this step, a dual-track analysis will be performed on the time series data obtained in the previous step. Multiple components refer to subsequences with different characteristics (such as long-term trends, periodic patterns, random fluctuations, etc.) separated from the original data through time series decomposition technology. Then, by performing differential analysis on these different components, the two analytical objectives of instantaneous operational anomalies and future deterioration risks can be achieved simultaneously.

[0032] The identification of instantaneous operational anomalies can be understood as discovering sudden problems that are currently occurring and do not conform to the normal pattern. The prediction of future deterioration risks can be understood as discovering potential risks where indicators are currently normal, but their long-term trends have shown signs of deterioration.

[0033] S130. Using the analysis model, the instantaneous operational anomalies and future deterioration risks are integrated, and the analysis results are output.

[0034] In this step, the two analytical conclusions obtained in the previous step are integrated. The analysis model can be one or more modules used to process the analytical conclusions, such as a large language model. The analysis results can be a comprehensive operational health report (pages 2 / 13, CN 121279775 A), which not only includes the current abnormal state but also points out the future risk trend and can provide corresponding decision-making suggestions.

[0035] According to the scheme of the embodiment of this disclosure, by decomposing the time series of operational indicators, instantaneous anomaly identification and future risk prediction are performed on their different components, and an innovative dual-track monitoring and prediction mechanism is established. This method solves the problem that traditional threshold alarm methods are difficult to distinguish between normal fluctuations and real anomalies and can only respond after the fact, and realizes more accurate and forward-looking management of operational risks.

[0036] It should be noted that instantaneous operational anomalies and future deterioration risks can occur independently because of different detection methods, covering different types of problem scenarios, thus forming a complement.

[0037] Scenario 1: There is an instantaneous anomaly, but no future deterioration risk

[0038] Example: A payment gateway server suddenly crashes for half an hour, causing the order success rate indicator to plummet instantly. This is a typical instantaneous operational anomaly. However, after the fault was repaired, the indicator quickly returned to normal, and its long-term positive trend remained unchanged. At this time, a severe-level instantaneous anomaly alarm should be issued, but the future risk trend may still be stable.

[0039] Scenario 2: There is a risk of future deterioration, but no instantaneous anomaly

[0040] Example: A new, imperfect dispatch algorithm was launched, causing the vehicle empty-running rate to increase slightly by 0.1% each day compared to the previous day. In the initial stage, the daily empty-running rate is still within the normal range and will not trigger any instantaneous anomalies. However, by analyzing its trend, the system will predict that the indicator is in a continuous and slow deterioration risk. If no intervention is taken, it will exceed the acceptable range after one month. At this time, a "risk of trend deterioration" warning should be issued to remind the operations staff to pay attention to this "boiling frog" problem.

[0041] In more complex situations, the two can influence each other and even transform into one another. Transient anomalies may lead to future deterioration risks. This can be understood as a severe, poorly handled transient anomaly that may have a profound impact on the long-term trend of subsequent indicators.

[0042] Example: A large-scale dispatch system failure (transient anomaly) in a certain region caused a large number of passengers to wait for a long time, seriously damaging the user reputation of the region. After this event, the long-term trend of the region's "average daily new user registrations" indicator began to decline continuously, forming a risk of future deterioration.

[0043] On the other hand, the risk of future deterioration increases the probability of transient anomalies. This can be understood as a continuously deteriorating trend often means that the system is becoming more and more fragile, thus making it easier to generate severe transient anomalies under small disturbances.

[0044] Example: The trend of the fleet's "average maintenance response time" has been rising continuously over the past few months (risk of future deterioration), indicating that the overall health of the vehicles is slowly declining. This will inevitably lead to a significant increase in the probability of a serious vehicle failure (transient anomaly) at some point in the future.

[0045] It is evident that instantaneous operational anomalies and future deterioration risks together constitute a three-dimensional and comprehensive risk monitoring system. For the operating system, looking at only one aspect may lead to a one-sided judgment. Combining the two allows operators to both respond quickly to emergencies and proactively address slow-onset problems.

[0046] In one possible implementation, the process of obtaining time-series data in step S110 may further include:

[0047] S111: Real-time aggregation and calculation of time-series data of target operating indicators from the spatiotemporal knowledge graph representing the operational status.

[0048] The target operating indicators may include at least one of operating efficiency indicators, service quality indicators, safety and reliability indicators, and economic benefit indicators.

[0049] In this embodiment of the present disclosure, an indicator calculation engine may be configured, which will initiate aggregation queries to the spatiotemporal knowledge graph at a preset frequency (e.g., every 5 minutes). These queries will calculate macro-level values ​​that reflect the operational health of the entire fleet or a specific region based on the nodes, edges, and attributes in the graph, as described on page 3 / 13 of the manual (CN 121279775 A), and store these values ​​in chronological order to form a time series.

[0050] Specifically, in order to comprehensively assess the operational status, the target operational indicators can cover the following dimensions, which can be any number of indicators selected from the following dimensions and combined, etc.:

[0051] I. Operational efficiency indicators

[0052] This indicator mainly measures the utilization efficiency of fleet assets and may include:

[0053] 1. Empty mileage ratio

[0054] Required graph data: It is necessary to query all order nodes with a status of "completed" within a specified time period.

[0055] Calculation method: First, sum the actual paid mileage of all completed orders to obtain the total paid mileage. Then, sum the total mileage of all vehicles to obtain the total mileage. Empty mileage is "total mileage" minus "total paid mileage". Finally, the empty mileage ratio can be calculated by the formula (total mileage - total paid mileage) / total mileage.

[0056] 2. Average daily order completions per vehicle

[0057] Required map data: It is necessary to query the total number of order nodes with a status of "completed" within a specified date; at the same time, query the total number of vehicle nodes that have had "online" records within that date.

[0058] Calculation method: Calculated by the formula total number of completed orders within a specified date / total number of active vehicles within that date.

[0059] 3. Recharge Time Ratio

[0060] Required map data: It is necessary to query all work order nodes of type "charging" or "battery swapping" within a specified time period, and obtain their start time and end time attributes to calculate the duration; at the same time, query the total online time of all vehicle nodes within this period.

[0061] Calculation method: First, sum the duration of all recharge work orders to obtain the total recharge time. Then, sum the total online time of all vehicles. Finally, the recharge time ratio can be calculated by the formula total recharge time / total online time of vehicles.

[0062] II. Service Quality Indicators

[0063] This indicator mainly evaluates the operational performance from the perspective of passenger experience, and may include:

[0064] 1. Pick-up Time Accuracy

[0065] Required map data: It is necessary to query all completed order nodes within a specified time period, and obtain their estimated pick-up time and actual pick-up time attributes.

[0066] Calculation method: Traverse all queried order nodes and count the number of orders whose difference between "actual pick-up time" and "estimated pick-up time" is within a preset threshold (e.g., 60 seconds). Finally, the pick-up time accuracy can be calculated by the formula: number of orders with a difference within the threshold / total number of orders.

[0067] 2. Five-star order ratio

[0068] Required map data: It is necessary to query all completed order nodes with user ratings within a specified time period and obtain their user rating attributes.

[0069] Calculation method: Count the number of all orders with a user rating attribute of 5 stars, and then calculate it by the formula: number of 5-star orders / total number of valid rated orders.

[0070] 3. Average detour index instruction manual 4 / 13 pages 8 CN 121279775 A

[0071] Required map data: It is necessary to query all completed order nodes within a specified time period and obtain their planned route mileage and actual driving mileage attributes.

[0072] Calculation method: Calculate the detour index for each order, which is the actual mileage / planned route mileage, and then take the average of the index for all orders.

[0073] III. Safety and Reliability Indicators

[0074] This indicator focuses on the stability and safety of vehicle operation and may include:

[0075] 1. Takeover rate per 10,000 kilometers

[0076] Required map data: It is necessary to query the total number of all operational event nodes of type "remote takeover" within a specified time period; at the same time, query the total mileage of all vehicle nodes within the period.

[0077] Calculation method: Calculated by formula (total number of remote takeover events / total mileage) * 10000.

[0078] 2. Average fault interval mileage

[0079] Required map data: It is necessary to query the total number of all operational event nodes of type "system-level fault" within a specified time period; at the same time, query the total mileage of all vehicle nodes within the period.

[0080] Calculation method: Calculated by formula total mileage / total number of system-level fault events.

[0081] IV. Economic Benefit Indicators

[0082] These indicators are directly related to the financial performance of the operation and may include:

[0083] 1. Average Daily Revenue per Vehicle

[0084] Required Map Data: It is necessary to query all completed order nodes within a specified date and obtain their order amount attributes; at the same time, query the total number of active vehicle nodes within that date.

[0085] Calculation Method: First, sum the order amounts of all completed orders to obtain the total daily revenue. Then, calculate it using the formula Total Daily Revenue / Total Number of Active Vehicles per Day.

[0086] 2. Revenue per Kilometer

[0087] Required Map Data: It is necessary to query all completed order nodes within a specified time period and obtain their order amount and actual paid mileage attributes.

[0088] Calculation Method: Calculated using the formula Total Amount of All Orders / Total Actual Paid Mileage of All Orders.

[0089] 3. Peak Hour Recharge Cost Ratio

[0090] Required Graph Data: It is necessary to query all work order nodes of type "charging" within a specified time period, and obtain their recharge cost attribute and an electricity price period attribute that indicates whether it is a peak hour.

[0091] Calculation Method: First, filter out the charging work orders with electricity price periods of "peak" or "peak", and sum their recharge costs to obtain the peak hour recharge cost. Then, sum the recharge costs of all charging work orders to obtain the total recharge cost. Finally, calculate the total recharge cost using the formula Peak Hour Recharge Cost / Total Recharge Cost.

[0092] According to the scheme of this embodiment, by aggregating and calculating multi-dimensional core indicators in real time from a unified spatiotemporal knowledge graph, it is ensured that the input data for analysis is comprehensive, consistent and timely, providing a solid data foundation for the accuracy of subsequent risk prediction.

[0093] In one possible implementation, step S120, which determines the instantaneous abnormal state and future risk trend, may further include:

[0094] S121, decomposing the time series data into multiple components.

[0095] This step aims to decompose the original time series, which contains a variety of variation patterns, into subsequences that are easier to analyze, as described in the independent specification, page 5 / 13, CN 121279775 A.

[0096] S122, analyzing the first component among the multiple components to determine the instantaneous operational anomaly of the target operational indicator.

[0097] S123, analyzing the second component among the multiple components to determine the future deterioration risk of the target operational indicator.

[0098] In this embodiment of the present disclosure, different analysis strategies are used for the different components decomposed. For example, one component may be more suitable for identifying sudden anomalies, while another component may be more suitable for judging long-term trends.

[0099] According to the scheme of the present disclosure, by adopting the strategy of "decomposition first, then dual-track analysis", the same data can be subjected to targeted in-depth mining from different angles. This method solves the problem that traditional analysis methods attempt to process complex time series with a single model, resulting in insufficient information extraction or mutual interference.

[0100] In one possible implementation, the process of decomposing the time series in step S121 may further include:

[0101] S121a, decomposing the time series data into a trend term sequence, a seasonal term sequence, and a residual term sequence based on the seasonal trend decomposition algorithm.

[0102] In the embodiments of the present disclosure, the seasonal trend decomposition algorithm may specifically be the seasonal and trend decomposition algorithm using LOESS (STL) with local weighted scatter plot smoothing. The STL algorithm can decompose a time series into three core components:

[0103] Trend term sequence: reflects the main direction of change and long-term trend of the indicator over a longer time span. For example, the continuous upward trend of vehicle utilization over several weeks.

[0104] Seasonal Term Sequence: Reflects the regular fluctuations of the indicator that repeat at fixed periods (e.g., days, weeks). For example, the periodic peaks in order demand during the morning and evening rush hours each day.

[0105] Residual Term Sequence: The part remaining after removing trend and seasonal components from the original sequence. It represents unpredictable random fluctuations in the data that cannot be explained by regular patterns.

[0106] According to the scheme of the embodiments of this disclosure, by using mature decomposition algorithms such as STL, the predictable and regular fluctuations (trends and seasonality) and the unpredictable and random fluctuations (residuals) in the original indicator data can be reliably and effectively separated, providing a clean and high-quality input for subsequent differential analysis.

[0107] In one possible implementation, step S122, determining the instantaneous operational anomaly, may further include:

[0108] S122a: Inputting the residual term sequence into an isolated forest or variational autoencoder model to calculate anomaly scores.

[0109] In this embodiment, the first component may be the residual term sequence decomposed in the above embodiments. Since all normal fluctuations have been eliminated from the residual terms, the drastic changes that occur are very likely caused by real sudden anomalies. This sequence can be input into a specialized anomaly detection model, such as an isolated forest. This model can efficiently identify outliers in the data sequence.

[0110] S122b: Determining the instantaneous operational anomaly based on the anomaly score.

[0111] In this embodiment, models such as isolated forests calculate an anomaly score for each data point in the residual sequence. A threshold can be set, and when the anomaly score of a point exceeds the threshold, it is determined to be an instantaneous operational anomaly.

[0112] An example is: at 3 pm one afternoon, which is not a peak time, the residual term of "order cancellation rate" suddenly showed an extremely high positive spike. The Isolation Forest algorithm identified this as a point with a high anomaly score, and the system therefore determined that a momentary operational anomaly had occurred. Manual 6 / 13 pages 10 CN 121279775 A

[0113] In order to facilitate understanding by operators and system decision-making, numerical anomaly scores are usually used to compare with one or more preset thresholds, thereby converting them into a qualitative state level. The simplest form is "present" or "absent". For example: anomaly score < 0.7 => no anomaly, anomaly score >= 0.7 => anomaly.

[0114] In a more sophisticated system, there may be multiple thresholds, which divide it into richer states: anomaly score < 0.7 => normal, 0.7 <= anomaly score < 0.85 => warning, anomaly score >= 0.85 => severe anomaly.

[0115] According to the scheme of the embodiments of this disclosure, by specifically performing anomaly detection on the residual term sequence, it is possible to accurately identify real and sudden operational problems that are masked by traditional periodic fluctuations. This method greatly reduces the false alarm rate of traditional threshold alarm systems and improves the signal-to-noise ratio of anomaly detection.

[0116] In one possible implementation, the process of determining the future deterioration risk in step S123 may further include:

[0117] S123a, using a Prophet or autoregressive differential moving average model to extrapolate and predict the future trend of the trend term sequence.

[0118] In the embodiments of this disclosure, the second component may be the trend term sequence decomposed in the above embodiments. Since the trend term reflects the long-term trend of the indicator, predicting it can provide insight into future risks. This sequence can be input into...In a time series forecasting model, such as Prophet or Autoregressive Integrated Moving Average (ARIMA).

[0119] S123b, Determine the slope of the future trend.

[0120] S123c, If the degree of deterioration of the slope exceeds a preset threshold, determine that there is a risk of future deterioration.

[0121] The forecasting model extrapolates the future trend of the trend term and then determines the slope of the future trend line. The slope is a numerical value that represents the rate and direction of change of the indicator. A negative slope indicates a decline, and a positive slope indicates an increase. If the forecast slope continues to develop in a negative direction (for example, for "vehicle utilization rate", it is continuously negative), and its absolute value exceeds a preset risk threshold, it is determined that there is a risk of future deterioration. For example, a slope > 0.01 can be judged as an improving trend. -0.01 <= slope <= 0.01 can be judged as a stable trend, and a slope < -0.01 can be judged as a risk of deterioration.

[0122] For example: The current "vehicle utilization rate" value is 55%, which is still within the normal range, but the prediction model indicates that its trend will continue to accelerate downward in the next 3 hours. Based on the predicted slope, the risk of future deterioration is determined in advance.

[0123] According to the scheme of the embodiment of this disclosure, a forward-looking risk warning mechanism is realized by extrapolating and predicting the trend of the indicator and analyzing the slope. This method can identify the inherent and continuous deterioration trend of the core indicator in advance before the absolute value of the core indicator becomes abnormal, and significantly advance the window for risk intervention.

[0124] In one possible implementation, the process of determining the risk of future deterioration in step S123 can also be implemented in another way including the following steps:

[0125] S123d: Convert the trend sequence and related context information into a situation description text in natural language form.

[0126] S123e: Input the situation description text into the large language model.

[0127] S123f: Determine the risk of future deterioration based on the output of the large language model.

[0128] In this embodiment of the disclosure, the cognitive reasoning ability of a large language model is utilized. A sequence of trend items (e.g., "the trend value has been declining continuously over the past 5 hours") and related contextual information (e.g., "the weather forecast predicts rain," "there is a large event today") can be combined to form a natural language description, and the large language model can be asked: "Based on the above information, is there a risk of future deterioration of this indicator?" The large language model will provide an answer containing judgment and reasoning based on its knowledge and reasoning ability, thereby determining whether a risk exists. Specification 7 / 13 pages 11 CN 121279775 A

[0129] According to the scheme of this disclosure embodiment, by introducing a large language model to analyze trends, a new technical path is provided for risk prediction. This method can effectively integrate numerical trends with unstructured contextual information, enabling prediction to take into account more complex factors in the real world, thereby improving the intelligence level and accuracy of judgment.

[0130] In one possible implementation, the process of outputting analysis results using the analysis model in step S130 may further include:

[0131] S131, converting the latest values ​​of the target operation indicators, instantaneous operation anomalies, and future deterioration risks into comprehensive analysis prompts in natural language form.

[0132] S132, inputting the comprehensive analysis prompts into the large language model to generate an analysis result containing the comprehensive health of the target operation indicators.

[0133] In this disclosure embodiment, the analysis model may specifically use a large language model. In this step, all analysis conclusions from the preceding steps (i.e., the latest values ​​of the indicators, whether there are instantaneous anomalies, and whether there are future deterioration risks) will be integrated to form a comprehensive natural language text as a comprehensive analysis prompt for the large language model.

[0134] After receiving the prompt, the large language model will generate an analysis result that is easy for humans to understand and includes a comprehensive health assessment. An example could be: inputting into the large model: "The latest value of the indicator 'order cancellation rate' is 8%, the instantaneous abnormal state is 'severe', and the future risk trend is 'continuous deterioration'. Please conduct a comprehensive analysis."

[0135] According to the scheme of the embodiments of this disclosure, by using the large language model to perform the final synthesis and interpretation of multiple discrete analysis signals, the machine-generated, technical conclusions can be transformed into a more user-friendly and instructive comprehensive diagnostic report for operators, thus bridging the last mile from data analysis to human understanding.

[0136] In one possible implementation, step S131, which transforms the latest value of the target operational indicator, the instantaneous operational abnormality, and the future deterioration risk into a comprehensive analysis prompt in natural language form, may further include:

[0137] S131a, in the case of the existence of the instantaneous operational abnormality, extracting operational event information that is temporally adjacent to the instantaneous operational abnormality from the spatiotemporal knowledge graph.

[0138] S131b: The latest values ​​of the target operational indicators, instantaneous operational anomalies, operational event information, and future deterioration risks are transformed into a comprehensive analysis prompt in natural language form.

[0139] In this embodiment of the present disclosure, in order to enable the final analysis of the large model to trace back to its source, a key contextual event information is added when constructing the final prompt. When the system detects an instantaneous anomaly, it immediately queries the spatiotemporal knowledge graph for other operational events that occurred nearby (e.g., within 5 minutes before and after) centered on the time point of the anomaly. For example:The system detected an anomaly in the "order cancellation rate" indicator at 15:30 and immediately checked the knowledge graph, finding that an external event of "payment gateway service provider reporting service interruption" had occurred at 15:28. This event information will be added to the final analysis prompt.

[0140] According to the scheme of this embodiment, by dynamically extracting contextual events related to the anomaly from the knowledge graph in the final analysis stage, rich attribution clues are provided for the comprehensive analysis of the large model. This enables the large model not only to report "indicator anomaly", but also to further infer that "the indicator anomaly is likely caused by the payment gateway interruption", which greatly improves the depth and practical value of the analysis results.

[0141] In one possible implementation, the method may further include the following steps:

[0142] S210, Obtain multi-source operation data reflecting the fleet operation status from multiple data sources, including at least the scheduling system, vehicle-end equipment and work order system.

[0143] S220, Construct a spatiotemporal knowledge graph based on the multi-source operation data.

[0144] In this embodiment of the present disclosure, the spatiotemporal knowledge graph is obtained by deeply integrating and associating multi-source data from various corners of the unmanned vehicle operation ecosystem through a series of steps such as data access, entity recognition, and relationship construction.

[0145] According to the scheme of this embodiment of the present disclosure, by clarifying that the spatiotemporal knowledge graph is constructed based on multi-source operation data, it is ensured that all analysis and prediction of this method are based on comprehensive, real-time, and accurate data, thus guaranteeing the reliability and practical value of the final output results.

[0146] In one possible implementation, the process of obtaining multi-source operation data in step S210 may further include:

[0147] S211, real-time access to heterogeneous data streams from multiple data sources, including at least a scheduling system, vehicle-end equipment, and a work order system.

[0148] In this step, standardized data interfaces and protocols can be used to ensure that real-time data from various sources can be received continuously and stably.

[0149] Heterogeneous data streams refer to data with different sources, formats, and update frequencies. For example, vehicle-side devices report GPS data streams at a rate of seconds via the MQTT protocol, while the scheduling system may push order status change data at the minute level via API interfaces or database log subscriptions (such as Canal) in an event-driven manner. Real-time access means that the system has the ability to process these data streams with different characteristics and integrate them into subsequent processing flows.

[0150] S212. Based on preset operation rules, identify and match heterogeneous data streams to generate structured operation events.

[0151] In this step, the original, continuous data streams undergo preliminary intelligent processing to transform them into discrete events with clear business meanings.

[0152] Pre-defined operational rules are a series of logical conditions pre-defined by operators based on experience or data analysis, used to identify potential anomalies or critical states during operation.

[0153] Structured operational events refer to a standardized data record (e.g., a JSON object) automatically generated when data in a data stream meets a certain operational rule. This record clearly describes "when, where, and what happened". These operational events are one of the important sources of information for building a spatiotemporal knowledge graph.

[0154] According to the scheme of the embodiments of this disclosure, by preprocessing real-time, raw data streams into structured operational events, high-quality, high-information-density input is provided for the construction of the knowledge graph, which is a prerequisite for ensuring that the knowledge graph can accurately reflect the operational situation.

[0155] In one possible implementation, the process of identifying and matching in step S212 may further include:

[0156] S212a, matching event patterns that conform to preset operational rules in heterogeneous data streams.

[0157] In this step, the event pattern can be understood as a technical expression of the operational rules, which defines a specific data sequence or combination that needs to be found in one or more data streams. This process is typically executed by a Complex Event Processing (CEP) engine. For example, an event pattern of "vehicle stationary for a long time" can be defined as follows: within a 10-minute time window, for the same vehicle ID, the reported GPS location change distance is less than 50 meters. The CEP engine continuously monitors the GPS data stream of all vehicles to find instances that meet this pattern.

[0158] S212b: When the event pattern is met, a structured operational event containing the event type, timestamp, and associated entity identifier is generated.

[0159] The preset operational rules may include at least one of the following: time window statistical rules for global operational indicators, sequence pattern rules for the operating status of a single vehicle, and state consistency rules for the execution process of a single order. It may include one, two, three, or even any number of more rules.

[0160] In this step, once the CEP engine successfully matches an event pattern, it will immediately generate a standardized event object. This event object is structured, meaning that its fields are fixed and explicit. For example, for the above pattern "Vehicle manual 9 / 13 page 13 CN 121279775 A Long period of stillness", the generated event object might be as follows:

[0161] {"eventType":"VEHICLE_STATIC_LONG_TIME","timestamp":"2025-09-10T15: 30:00Z"

[0162] Among them, the associated entity identifier (such as vehicleId) is a crucial link, which enables this isolated event to be associated with specific vehicle nodes in the knowledge graph in subsequent steps.

[0163] According to the scheme of the embodiment of this disclosure, through a multi-dimensional rule system and event pattern matching mechanism, various key dynamics in the operation process can be captured comprehensively and deeply, providing the necessary atomic information for constructing an information-rich spatiotemporal knowledge graph that can reflect complex operation logic.

[0164] Figure 3 is a structural schematic diagram of the risk prediction device 300 for fleet operation indicators provided according to an embodiment of this disclosure. As shown in Figure 3, the device includes:

[0165] an acquisition module 301, used to acquire time series data of at least one target operation indicator from a spatiotemporal knowledge graph representing the fleet operation status;

[0166] an analysis module 302, used to identify instantaneous operation anomalies of the target operation indicator and predict future deterioration risks based on multiple components of the time series data;

[0167] an output module 303, used to use an analysis model to integrate instantaneous operation anomalies and future deterioration risks and output analysis results.

[0168] In one possible implementation, the acquisition module 301 is used to:

[0169] aggregate and calculate the time series data of the target operation indicator in real time from the spatiotemporal knowledge graph representing the fleet operation status;

[0170] wherein the target operation indicator includes at least one of operation efficiency indicators, service quality indicators, safety and reliability indicators, and economic benefit indicators.

[0171] In one possible implementation, the analysis module 302 is used to:

[0172] decompose the time series data into multiple components;

[0173] analyze the first component among the multiple components to determine the instantaneous operational anomaly of the target operational indicator;

[0174] analyze the second component among the multiple components to determine the future deterioration risk of the target operational indicator.

[0175] In one possible implementation, the analysis module 302 is used to:

[0176] decompose the time series data into a trend term sequence, a seasonal term sequence, and a residual term sequence based on a seasonal trend decomposition algorithm.

[0177] In one possible implementation, the analysis module 302 is used to:

[0178] input the residual term sequence into an isolated forest or variational autoencoder model to calculate anomaly scores;

[0179] determine instantaneous operational anomalies based on the anomaly scores.

[0180] In one possible implementation, the analysis module 302 is used to:

[0181] The future trend of the trend item sequence is extrapolated and predicted using a Prophet or autoregressive differential moving average model;

[0182] The slope of the future trend is determined;

[0183] If the deterioration of the slope exceeds a preset threshold, it is determined that there is a risk of future deterioration. Specification 10 / 13 pages 14 CN 121279775 A

[0184] In one possible implementation, the analysis module 302 is used to:

[0185] Convert the trend item sequence and related context information into a situation description text in natural language form;

[0186] Input the situation description text into a large language model;

[0187] And determine the risk of future deterioration based on the output of the large language model.

[0188] In one possible implementation, the output module 303 is used to:

[0189] Convert the latest value of the target operation indicator, instantaneous operation anomaly and the risk of future deterioration into a comprehensive analysis prompt in natural language form;

[0190] Input the comprehensive analysis prompt into the large language model to generate an analysis result containing the comprehensive health of the target operation indicator.

[0191] In one possible implementation, the output module 303 is used to:

[0192] extract operational event information that is temporally adjacent to the instantaneous operational anomaly from the spatiotemporal knowledge graph when an instantaneous operational anomaly exists;

[0193] convert the latest values ​​of the target operational indicators, the instantaneous operational anomaly, the operational event information, and the risk of future deterioration into a comprehensive analysis prompt in natural language form.

[0194] In one possible implementation, the device further includes a construction module, used to:

[0195] acquire multi-source operational data reflecting the fleet's operational status from multiple data sources, including at least a scheduling system, vehicle-side equipment, and a work order system;

[0196] construct a spatiotemporal knowledge graph based on the multi-source operational data.

[0197] In one possible implementation, the construction module is used to:

[0198] access heterogeneous data streams in real time from multiple data sources, including at least a scheduling system, vehicle-side equipment, and a work order system;

[0199] identify and match the heterogeneous data streams based on preset operational rules to generate structured operational events.

[0200] In one possible implementation, the construction module is used to:

[0201] match event patterns that conform to preset operation rules in a heterogeneous data stream;

[0202] generate a structured operation event containing event type, timestamp, and associated entity identifier when the event pattern is satisfied;

[0203] wherein the preset operation rules include at least one of the following:

[0204] time window statistical rules for global operation indicators;

[0205] sequence pattern rules for the operating status of a single vehicle;

[0206] state consistency rules for the execution process of a single order.

[0207] The specific functions and examples of each module and submodule of the apparatus in the embodiments of this disclosure can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0208] In the technical solutions of this disclosure, the acquisition, storage and application of user personal information involved all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0209] According to the embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0210] FIG4 shows a schematic block diagram of an example electronic device 400 that can be used to implement the embodiments of this disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers and other suitable computers. The electronic device may also represent various mobile devices in the form of CN 121279775 A, such as personal digital assistants, cellular phones, smartphones, wearable devices and other similar computing devices. The components, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0211] As shown in FIG4, device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 402 or a computer program loaded from storage unit 408 into random access memory (RAM) 403. Various programs and data required for the operation of device 400 may also be stored in RAM 403. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0212] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of displays, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0213] Computing unit 401 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing unit 401 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Computing unit 401 performs the various methods and processes described above.For example, a risk prediction method for fleet operation indicators. For example, in some embodiments, the risk prediction method for fleet operation indicators may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by computing unit 401, one or more steps of the risk prediction method for fleet operation indicators described above may be performed. Alternatively, in other embodiments, computing unit 401 may be configured to perform the risk prediction method for fleet operation indicators by any other suitable means (e.g., by means of firmware).

[0214] Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), payload programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a special-purpose or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0215] Program code for implementing the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus such that, when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, partially on a machine and partially on a remote machine as a standalone software package, or entirely on a remote machine or server.

[0216] In the context of this disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. Machine-readable media can be machine-readable signal media or machine-readable storage media. Machine-readable media can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any combination thereof.Suitable combinations. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0217] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0218] The systems and technologies described herein can be implemented in computing systems that include back-end components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include front-end components (e.g., user computers with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or computing systems that include any combination of such back-end components, middleware components, or front-end components. Components of the system can be interconnected via digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0219] Computer systems can include clients and servers. Clients and servers are generally geographically distant from each other and typically interact via communication networks. Client-server relationships are created by computer programs running on respective computers and having client-server relationships with each other. Servers can be cloud servers, servers for distributed systems, or servers incorporating blockchain.

[0220] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0221] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand thatIt should be noted that various modifications, combinations, sub-combinations, and substitutions can be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure shall be included within the scope of protection of this disclosure. Instruction sheet 13 / 13 Page 17 CN 121279775 A Figure 1 Figure 2 Figure 3 Instruction drawing 1 / 2 Page 18 CN 121279775 A Figure 4 Instruction sheet drawing 2 / 2 Page 19 CN 121279775 A Abstract This disclosure provides a risk prediction method, device, equipment, and storage medium for fleet operation indicators, relating to the field of data processing technology, particularly intelligent transportation, autonomous driving, large models, and data analysis technology. The specific implementation scheme is as follows: Obtain time series data of at least one target operation indicator from the spatiotemporal knowledge graph representing the operational situation of the fleet; identify instantaneous operational anomalies of the target operation indicator based on multiple components of the time series data, and predict future deterioration risks; utilize an analysis model to comprehensively consider instantaneous operational anomalies and future deterioration risks, and outputanalysis results. Through in-depth analysis of the time series of core operation indicators, dual and early warnings for instantaneous anomalies and trend-based risks are achieved.

Claims

1. A method for predicting risks of fleet operation indicators, comprising: obtaining time series data of at least one target operation indicator from a spatio-temporal knowledge graph representing a fleet operation situation; identifying instantaneous operation anomalies of the target operation indicator and predicting future deterioration risks according to multiple components of the time series data; outputting an analysis result by synthesizing the instantaneous operation anomalies and the future deterioration risks using an analysis model.

2. The method of claim 1, wherein, The obtaining time series data of at least one target operation indicator from a spatio-temporal knowledge graph representing a fleet operation situation comprises: real-time aggregating and calculating the time series data of the target operation indicator from the spatio-temporal knowledge graph representing a fleet operation situation; wherein the target operation indicator comprises at least one of operation efficiency indicators, service quality indicators, safety and reliability indicators, and economic benefit indicators.

3. The method of claim 1, wherein, The identifying instantaneous operation anomalies of the target operation indicator and predicting future deterioration risks according to multiple components of the time series data comprises: decomposing the time series data into multiple components; analyzing a first component of the multiple components to determine instantaneous operation anomalies of the target operation indicator; analyzing a second component of the multiple components to determine future deterioration risks of the target operation indicator.

4. The method of claim 3, wherein, The decomposing the time series data into multiple components comprises: decomposing the time series data into a trend item sequence, a seasonal item sequence, and a residual item sequence based on a seasonal trend decomposition algorithm.

5. The method of claim 4, wherein, The analyzing a first component of the multiple components to determine instantaneous operation anomalies of the target operation indicator comprises: inputting the residual item sequence into an isolation forest or a variational autoencoder model to calculate an anomaly score; determining instantaneous operation anomalies according to the anomaly score.

6. The method of claim 4, wherein, The analyzing a second component of the multiple components to determine future deterioration risks of the target operation indicator comprises: using a Prophet or an autoregressive integrated moving average model to extrapolate and predict a future trend of the trend item sequence; determining a slope of the future trend; determining that there is a future deterioration risk if a deterioration degree of the slope exceeds a preset threshold.

7. The method of claim 4, wherein, The analyzing a second component of the multiple components to determine future deterioration risks of the target operation indicator comprises: converting the trend item sequence and related context information into a situation description text in natural language form; inputting the situation description text into a large language model; and determining the future deterioration risks according to an output of the large language model.

8. The method of claim 1, wherein, The outputting an analysis result by synthesizing the instantaneous operation anomalies and the future deterioration risks using an analysis model comprises: converting a latest value of the target operation indicator, the instantaneous operation anomalies, and the future deterioration risks into a comprehensive analysis prompt in natural language form; inputting the comprehensive analysis prompt into a large language model to generate an analysis result including a comprehensive health degree of the target operation indicator.

9. The method of claim 8, wherein, The information of the latest value of the core indicator, the instantaneous operation anomaly and the future deterioration risk is converted into a comprehensive analysis prompt in natural language form, including: In the case of the existence of the instantaneous operation anomaly, the operation event information adjacent to the instantaneous operation anomaly in time is extracted from the spatio-temporal knowledge graph; The latest value of the target operation indicator, the instantaneous operation anomaly, the operation event information and the future deterioration risk are converted into a comprehensive analysis prompt in natural language form.

10. The method of claim 1, further comprising: acquiring multi-source operation data reflecting the operation state of the vehicle fleet from a plurality of data sources including at least a scheduling system, a vehicle terminal device and a work order system; constructing a spatio-temporal knowledge graph according to the multi-source operation data.

11. The method of claim 10, wherein, The multi-source operation data reflecting the operation state of the unmanned vehicle is acquired from a plurality of data sources including at least a scheduling system, a vehicle terminal device and a work order system, comprising: real-time access to heterogeneous data streams from a plurality of data sources including at least a scheduling system, a vehicle terminal device and a work order system; based on a preset operation rule, the heterogeneous data stream is identified and matched to generate a structured operation event.

12. The method of claim 11, wherein, The heterogeneous data stream is identified and matched based on the preset operation rule to generate a structured operation event, comprising: matching the event mode conforming to the preset operation rule in the heterogeneous data stream; in the case where the event mode is met, a structured operation event containing event type, timestamp and associated entity identifier is generated; wherein the preset operation rule comprises at least one of: time window statistical rule for global operation indicators; sequence mode rule for single vehicle operation state; state consistency rule for single order execution process.

13. A vehicle fleet operation indicator risk prediction device, comprising: an acquisition module for acquiring time series data of at least one target operation indicator from a spatio-temporal knowledge graph representing the operation situation of the vehicle fleet; an analysis module for identifying instantaneous operation anomalies of the target operation indicator and predicting future deterioration risks according to a plurality of components of the time series data; an output module for outputting analysis results by utilizing an analysis model to integrate the instantaneous operation anomalies and the future deterioration risks.

14. An electronic device, comprising: at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-12.

15. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to make the computer execute the method according to any one of claims 1-12.

16. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-12.

Images