Gas station digital management method and system based on artificial intelligence

Abstract

The present application provides a gas station digital management method and system based on artificial intelligence, the method comprising: generating oil product flow transfer minimization data records based on the actual oil discharge process of the oil gun, constructing an oil product flow transfer topology instance containing tank nodes, oil gun nodes and order transaction nodes, and generating an edge constraint vector containing quantity deviation proportion, time deviation proportion and state marker. The system calculates the topology edge consistency value based on the edge constraint vector, generates a topology consistency feature package, and inputs it into an artificial intelligence anomaly positioning model to output an anomaly source positioning result. The present application organizes dispersed asynchronous data into link computing objects, identifies link anomaly features using an artificial intelligence model, solves the problem of difficult anomaly positioning caused by equipment collection cycle differences in the prior art, and realizes accurate positioning of inventory measurement, inventory synchronization, order generation and transaction state anomalies.

Description

Technical Field

[0001] This invention belongs to the field of data processing and artificial intelligence technology, and in particular relates to a digital management method and system for gas stations based on artificial intelligence. Background Technology

[0002] Digital management systems for gas stations typically integrate with devices or business systems such as level gauges, fuel dispenser controllers, order systems, inventory management systems, and transaction verification interfaces. They enable functions like inventory queries, sales statistics, order verification, document review, and anomaly alerts on a station-level or headquarters platform. A common approach in existing technologies is to collect data separately from each business system and then aggregate and compare it by station, fuel type, equipment number, or business time. For example, sales records are generated based on the fuel dispenser's output, tank inventory is adjusted based on level gauge data, and transaction completion is determined by the status returned by the order system or transaction verification interface. While this method can meet general operational reporting and basic inventory management needs, in real gas station operations, fuel is a continuous fluid, and tank inventory changes are affected by factors such as level fluctuations, temperature changes, pipeline residue, tank tilt, fuel dispenser metering errors, and inventory conversion methods. Furthermore, fuel nozzle start / stop, order generation, transaction status returns, inventory change collection, and ledger updates are not entirely synchronized, and the collection cycles, upload cycles, and caching mechanisms of different devices and systems also vary. Especially when multiple nozzles are simultaneously serving the same tank, multiple nozzle time windows are close together, order status returns are delayed, or level gauges upload data periodically, simply comparing based on absolute time or single thresholds can easily split the same oil transfer activity into multiple incomplete records, and can also easily misjudge normal equipment upload delays as anomalies. When discrepancies occur between inventory changes and nozzle output, between nozzle output and order output, missing order transaction status, or lagging inventory records, existing systems typically only output general prompts such as discrepancies between records and actual inventory, order anomalies, or inventory anomalies, making it difficult to further determine whether the anomaly originates from the tank level metering, inventory synchronization interface, fuel dispenser metering link, order generation interface, or transaction status return interface. These issues force managers to repeatedly check between level gauges, fuel dispensers, order systems, and inventory systems, resulting in long anomaly localization cycles, high risks of false alarms and missed alarms, and making it difficult for gas station digital management systems to evolve from simple data display and post-event statistics to a reliable verification and intelligent location system covering the entire oil transfer process. Therefore, based on existing gas station equipment interfaces and business systems, a key engineering problem to be solved in the digital management of gas stations is how to organize changes in tank inventory, fuel dispenser data, and order transaction status into a single oil product flow object, considering a real fuel dispensing process, and forming a linked feature that can be recognized by artificial intelligence models, while taking into account time delays, quantity deviations, and missing status information. The technical focus of this problem lies in aligning the asynchronously collected data from level gauges, fuel dispenser controllers, station control systems, inventory synchronization interfaces, and transaction status interfaces in terms of time, unit unification, link binding, and anomaly source localization, rather than simply providing operational management evaluations or manual review suggestions. Summary of the Invention

[0003] This invention discloses a digital management method and system for gas stations based on artificial intelligence, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the first aspect of the present invention provides a digital management method for gas stations based on artificial intelligence, comprising: The fuel dispenser controller obtains the fuel nozzle start / stop status to generate a fuel nozzle time window, and collects the fuel tank inventory change data, fuel nozzle dispensing data and order transaction status data corresponding to the fuel nozzle time window to synthesize a fuel product circulation minimized data record; Based on the minimized data records of oil product circulation, an oil product circulation topology instance is generated. The oil product circulation topology instance includes oil tank nodes, oil nozzle nodes, and order transaction nodes. The oil tank nodes are connected to the oil nozzle nodes through a first topology edge, and the oil nozzle nodes are connected to the order transaction nodes through a second topology edge. An edge constraint vector is generated for the first topology edge and the second topology edge. The edge constraint vector includes quantity deviation ratio, time deviation ratio, and status flag. The edge consistency values ​​of the first topological edge and the second topological edge are calculated based on the edge constraint vector, and a topological consistency feature package is generated. The topological consistency feature package includes the edge consistency value, the consistency difference between the first topological edge and the second topological edge, and the edge constraint vector. The topology consistency feature package is input into the artificial intelligence anomaly localization model, and the anomaly source localization result is output by the artificial intelligence anomaly localization model. The anomaly source localization result is used to indicate inventory measurement anomalies, inventory synchronization anomalies, order generation anomalies, or transaction status anomalies.

[0005] Furthermore, the process of collecting oil tank inventory change data, oil dispenser output data, and order transaction status data corresponding to the oil nozzle time window, and synthesizing a data record that minimizes oil product circulation, includes: Verify the device number, timestamp continuity, and monotonicity of the cumulative oil output value of the oil gun output data; When a timestamp reversal or an abnormal jump in the cumulative oil output value is detected, the oil gun time window is marked as an abnormal acquisition window; During the verification process, the oil gun time window identifier is used as the primary key to combine the oil tank inventory change data, the oil gun dispensing data, and the order transaction status data into a single minimal oil product circulation data record.

[0006] Furthermore, the process of collecting the order transaction status data corresponding to the oil gun time window includes: Search for order records within the specified oil gun time window and the preset allowable delay range; When multiple candidate order records are found within the allowed delay range, the order record with the same oil gun number and the smallest deviation between the order oil volume and the cumulative oil output value of the oil gun output data is selected as the matching order. If no order record is found within the allowed delay range, write an order mismatch status into the oil product circulation minimized data record.

[0007] Furthermore, the process of generating an oil flow topology instance based on the minimized oil flow data records includes: Read the tank number and window inventory change value from the oil product circulation minimization data record to generate an oil tank node; Read the oil gun number and the cumulative oil output value of the oil gun to generate an oil gun node; Read the order number, order fuel quantity, and transaction status to generate an order transaction node; When the order number is empty, a temporary order transaction node identifier is generated based on the oil gun time window identifier and the oil gun number, and the transaction status of the order transaction node is set to unmatched.

[0008] Furthermore, the process of generating edge constraint vectors for the first topological edge and the second topological edge includes: Calculate the actual quantity difference and actual time difference between the data bound to the nodes at both ends of the first topological edge; The quantity deviation ratio is obtained by dividing the actual quantity difference by the preset allowable quantity deviation. The time deviation ratio is obtained by dividing the actual time difference by the preset allowable time difference. When the data binding at the endpoint node of the first or second topological edge is missing, the status flag is set to the first flag value; when the binding data is complete, the status flag is set to the second flag value.

[0009] Furthermore, the process of calculating the edge consistency value of the first topological edge and the second topological edge based on the edge constraint vector includes: Calculate the sum of squares based on the quantity deviation ratio and the time deviation ratio; Multiplying the quantity deviation ratio by the time deviation ratio yields the coupling product term; The attenuation exponent term is obtained by adding the sum of squares term, the coupling product term, and the synchronization disturbance coefficient; The exponential function value is calculated based on the decay exponential term and the state flag, and the exponential function value is used as the edge consistency value. The state flag is used to limit the edge consistency value to zero when the bound data is missing.

[0010] Furthermore, the process of generating the topology-consistent feature package includes: The edge consistency value corresponding to the first topological edge is determined as the inventory link consistency value; The edge consistency value corresponding to the second topological edge is determined as the order transaction link consistency value; The consistency difference is obtained by calculating the difference between the inventory link consistency value and the order transaction link consistency value. The inventory link consistency value, the order transaction link consistency value, the consistency difference, and the edge constraint vectors of the first topological edge and the second topological edge are serialized into the topological consistency feature package in a preset order.

[0011] Furthermore, the artificial intelligence anomaly localization model includes a link feature encoding layer, a link correlation analysis layer, and an anomaly source output layer; The link feature coding layer receives the topology consistency feature packet and extracts hidden features; The link association analysis layer receives the hidden features and calculates the association weight between the inventory link consistency value and the order transaction link consistency value; The anomaly source output layer receives the associated weights and outputs the original response values ​​for inventory measurement anomalies, inventory synchronization anomalies, order generation anomalies, and transaction status anomalies.

[0012] Furthermore, the process of outputting the anomaly source localization result through the artificial intelligence anomaly localization model includes: Obtain the product of the consistency difference and the preset link offset coefficient; The product is then added to the original response values ​​of the inventory measurement anomaly, the inventory synchronization anomaly, the order generation anomaly, and the transaction status anomaly to obtain the adjusted response value. The adjusted response values ​​are normalized to obtain response values ​​for four types of anomaly sources; The anomaly category corresponding to the response value with the largest value among the four types of anomaly source response values ​​is determined as the anomaly source localization result.

[0013] A second aspect of the invention provides an artificial intelligence-based digital management system for gas stations, comprising: The data recording generation module is used to obtain the start / stop status of the fuel nozzle from the fuel dispenser controller to generate a fuel nozzle time window, collect the fuel tank inventory change data, fuel nozzle dispensing data and order transaction status data corresponding to the fuel nozzle time window, and synthesize a minimal data record of oil product circulation. The topology instance generation module is used to generate an oil product circulation topology instance based on the oil product circulation minimized data record. The oil product circulation topology instance includes oil tank nodes, oil nozzle nodes, and order transaction nodes. The oil tank nodes are connected to the oil nozzle nodes through a first topology edge, and the oil nozzle nodes are connected to the order transaction nodes through a second topology edge. An edge constraint vector is generated for the first topology edge and the second topology edge. The edge constraint vector includes quantity deviation ratio, time deviation ratio, and status flag. The feature package generation module is used to calculate the edge consistency value of the first topological edge and the second topological edge based on the edge constraint vector, and generate a topological consistency feature package, wherein the topological consistency feature package includes the edge consistency value, the consistency difference between the first topological edge and the second topological edge, and the edge constraint vector; An anomaly localization module is used to input the topology consistency feature package into the artificial intelligence anomaly localization model, and output the anomaly source localization result through the artificial intelligence anomaly localization model. The anomaly source localization result is used to indicate inventory measurement anomalies, inventory synchronization anomalies, order generation anomalies, or transaction status anomalies.

[0014] The beneficial technical effects of the present invention are at least as follows: To address the aforementioned issues, this invention provides an AI-based digital management method and system for gas stations. It aligns and binds data from dispersed and asynchronous level gauges, fuel dispenser controllers, order systems, and transaction interfaces using a real fuel dispenser process as a benchmark, constructing a topological instance containing fuel tanks, fuel dispensers, and order nodes. By generating edge constraint vectors and topological consistency feature packages, quantity deviations, time delays, and missing states are transformed into structured features input into the AI ​​model, enabling intelligent classification and location of anomaly sources. This solution effectively overcomes the problem of traditional single-point threshold comparisons easily fragmenting or misjudging delays as anomalies, refining general discrepancies between accounts and actual inventory into precise indicators of inventory measurement, inventory synchronization, order generation, or transaction status anomalies. It significantly shortens the manual cross-system verification cycle, reduces the risk of false alarms and missed alarms, and improves the efficiency of inventory verification, order review, and equipment inspection. This upgrades the gas station management system from passive data display to a reliable verification and intelligent location technology system covering the entire fuel circulation process. Attached Figure Description

[0015] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0016] Figure 1 This is a flowchart of the AI-based digital management method for gas stations according to the present invention.

[0017] Figure 2 This is a framework diagram of the AI-based digital management system for gas stations according to the present invention. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0019] In one or more embodiments, such as Figure 1 As shown, a digital management method for gas stations based on artificial intelligence is disclosed, the method including the following: S1: Obtain the start / stop status of the fuel dispenser controller to generate a fuel dispenser time window, collect the fuel tank inventory change data, fuel dispenser oil dispensing data and order transaction status data corresponding to the fuel dispenser time window, and synthesize a data record of minimized oil product circulation.

[0020] Specifically, step one is used to generate a minimal data record of oil product circulation. The system uses the actual dispensing process of a single fuel nozzle as the record organization benchmark, rather than the order generation time or liquid level upload time. In a gas station setting, fuel nozzle dispensing is the direct trigger event for the actual circulation of oil products. Order status and inventory changes usually have upload delays. If the three types of data are directly matched at the same time point, it is easy to separate data that should belong to the same refueling behavior, leading to incorrect connections in subsequent topology instances. The system receives three types of necessary data: the first type is tank inventory change data, collected by the liquid level gauge or inventory management system, recording the tank number, oil product number, collection time, and inventory change value; the second type is fuel nozzle dispensing data, collected by the fuel dispenser controller, recording the fuel nozzle number, oil product number, start time, stop time, and cumulative dispensing value; the third type is order transaction status data, provided by the order system or desensitized transaction verification interface, recording the order number, fuel nozzle number, order time, order oil quantity, and transaction status. The order number is used only as an identifier for de-identified business transactions to determine whether a single oil product transfer transaction constitutes an order record. The system first reads the nozzle status uploaded by the fuel dispenser controller. When a nozzle changes from the attached state to the activated state, the activation time is recorded; when the nozzle returns to the attached state, the deactivation time is recorded. The interval between the activation and deactivation times is defined as a nozzle time window, and a nozzle time window identifier is generated for this window. For example, if a nozzle activates at 08:15:12 and deactivates at 08:16:03, this time interval is identified as a nozzle time window, which is used to associate the corresponding order record and inventory change record. Before generating the nozzle time window, the system verifies the continuity of the device number, nozzle number, oil product number, timestamp, and the monotonicity of the cumulative oil dispensing value in the frame uploaded by the fuel dispenser controller. If there are duplicate uploads, timestamp reversals, missing oil product numbers, or abnormal jumps in the cumulative oil dispensing value, the nozzle process is marked as an abnormal acquisition window and is not used as a normal training sample or normal anomaly location input.

[0021] The system searches for order transaction status data within the fuel nozzle's time window and the allowed delay range thereafter, based on the nozzle number and fuel type number. The reason for setting the allowed delay range is that in actual station-level systems, order generation and transaction status return after a fuel nozzle stops typically require passing through the station control system, the order system, and the transaction verification interface, resulting in a brief delay. For example, if a fuel nozzle stops at 08:16:03, and the order system generates a transaction completion status at 08:16:09, then the order is bound to that fuel nozzle's time window. If no corresponding order is found within the allowed delay range, the system retains the order's unmatched status within that fuel nozzle's time window; this status is subsequently entered into the topology instance and used as a basis for anomaly localization. If multiple candidate orders exist within the allowed delay range, the system prioritizes the order with the same fuel nozzle and fuel type number, the smallest deviation between the order's fuel quantity and the fuel nozzle's cumulative output value, and the order time closest to the stop time, excluding already bound orders from other window candidate sets. If multiple indistinguishable candidate orders still exist, the window is marked as an order mapping uncertainty state. For oil tank inventory change data, the system determines the oil tank number corresponding to the current oil gun based on the configuration relationship between the oil gun and the oil tank, and searches for inventory change records uploaded by the level gauge or inventory management system within the oil gun time window and its adjacent inventory collection period. Since the level gauge usually uploads periodically, and the liquid level in the oil tank may be affected by fluctuations, the system uses a window accumulation method to form the window inventory change value. If the same oil tank serves multiple oil guns simultaneously within an overlapping time period, the system first allocates the same inventory change record to a window based on the proportion of the cumulative oil output value of each oil gun to the total oil output value of the same oil tank during the same period; if the concurrent window cannot be reliably allocated based on the oil output value of the oil gun, the corresponding window is marked as a concurrent indivisible state, and the inventory change value is not used as the basis for normal inventory link consistency calculation. This formula is derived from finite summation in mathematics and is used to combine multiple inventory change records within the same time window into a window-level inventory change result; this application limits it to the oil gun time window to handle the time misalignment problem between the periodic collection of the level gauge and the real-time oil output of the oil gun.

[0022] ; in, Indicates the first The window inventory change value formed within a time window of each oil gun is accumulated from the inventory change records of the level gauge or inventory management system within that window. This indicates the first [number] in this window. The inventory change values ​​in each inventory change record come from the level gauge or inventory management system, and are relevant when multiple oil nozzles are dispensing oil from the same tank and the amount can be allocated. This represents the inventory change value after allocating it according to the oil dispensing ratio of the oil gun; Indicates the first The number of inventory change records matched within a given fuel nozzle time window is determined by the system based on tank number, oil product number, and collection time. Both the left and right sides of the formula represent the same type of inventory change value. If the original level gauge reading differs in scale from the order and fuel nozzle data, the system first converts them to the same scale according to the oil product volume unit, temperature compensation method, and inventory conversion accuracy recorded in the station-level parameter area before accumulation. In a given fuel nozzle time window, the system matched three inventory change records with change values ​​of -12.30, -8.40, and -4.25, respectively. Therefore, the total inventory change value for that window is -24.95. If the cumulative oil dispenser output value within the same window is 24.90, the order oil quantity is 24.90, and the transaction status is completed, then a complete minimal oil flow data record is generated for that oil dispenser time window. If the order does not return to the transaction completion status, the record is generated and written with a "not matched" or "not completed" status. If no inventory change record is obtained within the window, the window inventory change value is marked as missing, and the source of the missing data is recorded as: level gauge not uploaded, inventory system not synchronized, or concurrent non-decomposable. Finally, the system uses the oil dispenser time window identifier as the primary key to synthesize the tank inventory change data, oil dispenser output data, and order transaction status data within the same window into a minimal oil flow data record. This record must contain at least the oil dispenser time window identifier, tank number, oil dispenser number, order number, oil product number, cumulative oil dispenser output value, order oil quantity, transaction status, window inventory change value, data validity flag, and data source flag. The output of this step is the minimal oil flow data record, which serves as the input for the next step to generate an oil flow topology instance.

[0023] S2: Generate an oil product circulation topology instance based on the minimized data record of the oil product circulation. The oil product circulation topology instance includes oil tank nodes, oil nozzle nodes and order transaction nodes. The oil tank nodes are connected to the oil nozzle nodes through a first topology edge, and the oil nozzle nodes are connected to the order transaction nodes through a second topology edge. Generate edge constraint vectors for the first topology edge and the second topology edge. The edge constraint vectors include quantity deviation ratio, time deviation ratio and status flag.

[0024] Specifically, step two is used to generate oil product circulation topology instances with bound data records. The system takes the minimized oil product circulation data records output in the previous step as input and creates an oil product circulation topology instance corresponding to each nozzle time window. Since the minimized oil product circulation data records output in step one already include the nozzle time window identifier, tank number, nozzle number, order number, oil product number, cumulative nozzle output value, order oil volume, transaction status, and window inventory change value, this step writes all these fields into the topology instance, ensuring that each topology instance corresponds to a specific oil product circulation behavior. The system reads the nozzle time window identifier and uses it as the instance identifier for the current topology instance; it reads the tank number and oil product number, generates a tank node, and binds the window inventory change value to that tank node; it reads the nozzle number and oil product number, generates a nozzle node, and binds the cumulative nozzle output value to that nozzle node; it reads the order number, nozzle number, and oil product number, generates an order transaction node, and binds the order oil volume and transaction status to that order transaction node. If the order number is empty, the system generates a temporary order transaction node identifier using the fuel nozzle time window identifier and the fuel nozzle number, and writes the node's status as "unmatched" to ensure the topology structure remains complete with the three types of nodes: fuel tank nodes, fuel nozzle nodes, and order transaction nodes. In actual gas stations, the same fuel tank can serve multiple fuel nozzles simultaneously, and multiple fuel nozzles may be dispensing fuel at the same time. Therefore, when generating topology instances, the system uses the fuel nozzle time window identifier, fuel nozzle number, and fuel product number to jointly define a single fuel product flow behavior, ensuring that business records between different fuel nozzles, different fuel products, and different time windows are entered into their respective corresponding topology instances. The system also reads the version number of the fuel nozzle-fuel tank configuration relationship. If the current fuel nozzle-fuel tank mapping relationship is missing or the configuration version is invalid, the topology instance is marked as an invalid configuration instance and is not used as normal model input.

[0025] After node generation, the system establishes topological edges according to the actual flow order of oil products within the station. The first topological edge points from the tank node to the nozzle node, representing the correspondence between changes in tank inventory and actual nozzle dispensing. The second topological edge points from the nozzle node to the order transaction node, representing the correspondence between actual nozzle dispensing and order transaction status. Each topological edge includes the starting node, ending node, oil product number, nozzle time window identifier, allowable time difference, and allowable quantity deviation. The allowable time difference is configured by the station-level system based on the level gauge acquisition cycle, fuel dispenser controller upload cycle, order generation cycle, and transaction status return cycle. The allowable quantity deviation is configured by the station-level system based on the fuel dispenser metering error range, tank level fluctuation range, and inventory conversion accuracy. For topological edges from tank nodes to nozzle nodes, the system uses the window inventory change value and the cumulative nozzle dispensing value as the binding data at both ends of the edge. For topological edges from nozzle nodes to order transaction nodes, the system uses the cumulative nozzle dispensing value, order oil quantity, and transaction status as the binding data at both ends of the edge. If the order number is empty or the transaction status is unmatched, the system still generates an order transaction node and writes the unmatched status into the binding data of the node and the corresponding topology edge, so that the next step can directly read the link status from the topology instance. If the allowed quantity deviation or allowed time difference is not configured, is zero, or is lower than the system's minimum effective threshold, the system does not perform normal deviation normalization calculation for that edge, but instead marks the edge as having invalid constraint configuration, and subsequent edge consistency values ​​are treated as invalid links.

[0026] To ensure that topological edges can represent the on-site constraints within the current fuel nozzle time window, the system generates an edge constraint vector for each topological edge. This calculation originates from the normalized deviation metric in mathematics, which involves ratioizing the actual deviation to the allowable deviation, giving each type of deviation a clear threshold meaning. This application applies this idea to the fuel product circulation chain at gas stations, storing quantity deviation, time deviation, and status missing data in different components of the same edge constraint vector. The resulting edge constraint vector is not a single weighted score but retains three types of engineering meaning: the quantity component indicates whether the fuel product quantity difference exceeds the on-site allowable range; the time component indicates whether the data upload or business generation exceeds the on-site allowable range; and the status component indicates whether the order or inventory binding status is complete. The system generates the edge constraint vector in the following manner: ; in, Indicates the first Within the first oil gun time window The edge constraint vector of a topological edge is generated from the binding data of the nodes at both ends of the topological edge; This indicates the difference in quantity between the nodes at both ends of the topological edge. For the topological edge from the oil tank node to the oil gun node, it is obtained by subtracting the cumulative oil output value from the oil gun after unifying the window inventory change value according to the direction of oil outflow. For the topological edge from the oil gun node to the order transaction node, it is obtained by subtracting the cumulative oil output value from the order oil quantity. Indicates the first The allowable deviation in the number of topological edges is configured by the station-level system based on the range of on-site metering error, the range of liquid level fluctuation, and the accuracy of inventory conversion. This indicates the time difference between the records bound at both ends of the topological edge. For the topological edge from the oil tank node to the oil gun node, it is calculated from the midpoint of the corresponding time period of the inventory collection time or inventory change record and the center time of the oil gun time window. For the topological edge from the oil gun node to the order transaction node, it is calculated from the order time and the oil gun stop time. Indicates the first The allowed time difference between topological edges is configured by the station-level system based on the upload cycle of relevant devices and business systems; This represents the state flag of the topology edge. It is 0 when the data required for binding to the endpoint node of the edge is complete and the state is valid; it is 1 when the endpoint node has unmatched orders, missing transaction states, missing inventory records, indivisible concurrency, or invalid constraint configurations. The first and second components in the formula are both ratios of the actual deviation to the allowable deviation. A component less than or equal to 1 indicates that the corresponding deviation is within the allowable range, and a component greater than 1 indicates that the corresponding deviation exceeds the allowable range. The third component records whether the link state is complete. These three components are saved separately and added to the topology consistency feature package in the next step, enabling the subsequent AI anomaly localization model to distinguish between three different sources: "inconsistent quantity," "abnormal time delay," and "missing records." If the binding data is missing, it will cause... or If a value cannot be generated, the system does not forcibly calculate the quantity or time component, but instead writes it to a status flag. Based on the corresponding reason for the missing link, subsequent steps will directly treat this edge as a missing link.

[0027] For example, the time window identifier corresponding to a minimum data record of a certain oil product flow is W20240501081512, the oil tank number is T01, the oil gun number is N03, the order number is O8891, the oil product number is F92, the window inventory change value is -24.95, the cumulative oil dispensed by the oil gun is 24.90, the order oil quantity is 24.90, and the transaction status is completed. The system first generates oil tank nodes T01-F92, oil gun nodes N03-F92, and order transaction nodes O8891-N03-F92, and then establishes two topological edges: one from the oil tank node to the oil gun node, and the other from the oil gun node to the order transaction node. For the topological edge from the oil tank node to the oil nozzle node, the system converts the window inventory change value to 24.95 according to the oil outflow direction and compares it with the cumulative oil dispensed value of 24.90 from the oil nozzle, obtaining a quantity difference of 0.05. If the allowable quantity deviation for this edge is configured to 0.30, the difference between the inventory collection time and the corresponding time of the oil nozzle time window is 8, the allowable time difference is configured to 30, and the data bound to this edge is complete, then the constraint vector for this edge is... That is, approximately For the topological edge from the fuel dispenser node to the order transaction node, the cumulative fuel dispenser output is 24.90, the order fuel quantity is 24.90, and the quantity difference is 0. If the difference between the order time and the corresponding time in the fuel dispenser time window is 6, the allowed time difference configuration is 20, and the transaction status is complete, then the constraint vector for this edge is... That is If the order transaction status is unmatched in the same scenario, the status flag of the corresponding edge is set to 1, and the third component of the edge constraint vector directly records the missing link status, so that the next step can retain this type of abnormal clue when generating the topology consistency feature package.

[0028] The system uses the following structure to represent the first... Each time window of the oil pump corresponds to an instance of the oil product flow topology. This structure is obtained by adding a binding data set and an edge constraint vector set to the node set and edge set representation in graph theory. The node set and edge set are used to represent the oil product flow link, the binding data set is used to receive the business records output from step one, and the edge constraint vector set is used to store the quantity constraint, time constraint, and state constraint results of the current link under this time window. ; in, Indicates the first The oil flow topology instance corresponding to the oil gun time window is given by the first... Minimize the generation of data records for oil product circulation; This represents the set of nodes in the topology instance, including oil tank nodes, oil nozzle nodes, and order transaction nodes, which are generated by the oil tank number, oil nozzle number, and order number or temporary order transaction node identifier in the record, respectively. This represents the set of edges in the topology instance, including the topology edges from the oil tank node to the oil nozzle node and the topology edges from the oil nozzle node to the order transaction node; This represents the set of bound data in the topology instance, including the binding relationship between the window inventory change value and the oil tank node, the binding relationship between the cumulative oil dispenser value and the oil gun node, the binding relationship between the order oil quantity and transaction status and the order transaction node, and the binding relationship between the oil product number and the oil gun time window identifier and each node and each edge. This represents the set of edge constraint vectors in this topology instance, consisting of the topological edges within the current topology instance. The system is composed of two formulas that form a relationship from edge-level constraint expression to instance-level topology encapsulation. Specifically, the system first generates edge constraint vectors with clear engineering meaning for each topological edge, and then encapsulates all edge constraint vectors along with nodes, edges, and binding data into an oil product flow topology instance. The system also... The system stores the original data collection record index, configuration version number, unit conversion version number, and missing reason field, enabling the topology instance to trace back to specific level gauge records, fuel dispenser controller records, and order interface records.

[0029] After completing the above processing, the system outputs an oil product circulation topology instance. This output includes a topology instance identifier, a set of nodes, a set of edges, a set of bound data, and a set of edge constraint vectors. In the next step, when generating the topology consistency feature package, the system directly reads the tank nodes, nozzle nodes, order transaction nodes, two topology edges, node bound data, and edge constraint vectors from this oil product circulation topology instance, and forms consistency features between adjacent nodes along the directions from the tank node to the nozzle node and from the nozzle node to the order transaction node. Through this instantiated topology generation step, the minimized oil product circulation data records from step one are organized into computational objects with oil product circulation direction, business binding relationships, and on-site constraint expressions, enabling the subsequent AI anomaly localization model to make judgments based on specific links rather than isolated fields.

[0030] S3: Calculate the edge consistency value of the first topological edge and the second topological edge based on the edge constraint vector, and generate a topological consistency feature package. The topological consistency feature package includes the edge consistency value, the consistency difference between the first topological edge and the second topological edge, and the edge constraint vector.

[0031] Specifically, step three is used to generate a topology consistency feature package. The system uses the oil product flow topology example output in the previous step. As input, the system reads the node set, edge set, bound data set, and edge constraint vector set, and performs consistency calculations on the inventory change link and order transaction link within a single oil pump time window according to the actual flow direction of oil products within the station. Step two has already organized the oil product flow minimization data records from step one into topology instances with edge constraints. Therefore, this step does not re-retrieve data from the original business system, but directly reads the node bound data and edge constraint vectors along the topology edges, so that the subsequently generated topology consistency feature package simultaneously possesses three types of information: link direction, business association, and on-site constraints. The system first bases... Determine the oil tank node, oil nozzle node, and order transaction node in the current topology instance based on... Determine the first topological edge from the oil tank node to the oil nozzle node and the second topological edge from the oil nozzle node to the order transaction node, based on... Read the window inventory change value, the cumulative oil dispenser output value, the order oil quantity, and the transaction status, based on... Read the edge constraint vectors corresponding to the two edges. The resulting computational object is no longer a simple data record, but a link structure that includes the direction of oil flow, node binding relationships, and edge constraint expressions. If the topology instance output in step two is marked as an invalid configuration instance, or if both edges are missing links, the system will not treat the instance as a normal anomaly location sample, but will output the data link undeterminable state and record the corresponding device or interface configuration reason.

[0032] The system generates an edge consistency value for each topological edge. This calculation is derived from the Gaussian similarity decay model in statistics, the basic idea of ​​which is that as the actual deviation gradually increases, the similarity decreases exponentially. This application introduces this model into the gas station oil circulation scenario and adds a link synchronization disturbance term to the traditional Gaussian decay term to express the on-site characteristic of a rapid increase in link anomaly risk when "quantity deviation and time deviation increase simultaneously". In actual gas station operation, individual quantity deviations usually originate from liquid level fluctuations, temperature changes, or fuel dispenser metering errors, while individual time deviations usually originate from order return delays or inventory upload cycles. When the same topological edge exhibits both large quantity deviations and large time deviations, it is more likely to correspond to inventory synchronization anomalies, order mapping anomalies, or equipment upload anomalies. Therefore, the system adds a synchronization disturbance term to the edge consistency value to create a coupling relationship between quantity deviations and time deviations. The system uses the following formula to generate the edge consistency value: ; in, Indicates the first Within the first oil gun time window The edge consistency value of a topological edge; Indicates the first Within the first oil gun time window The deviation ratio of the number of topological edges originates from the edge constraint vector in step two. The first component is obtained by the ratio of the actual quantity difference to the allowable quantity deviation; Indicates the first Within the first oil gun time window The time deviation ratio of each topological edge originates from the edge constraint vector in step two. The second component is obtained by the ratio of the actual time difference to the allowable time difference; Indicates the first Within the first oil gun time window The state markers of the topological edges are derived from the edge constraint vectors in step two. The third component is set to 0 when the order status, inventory record, or transaction status corresponding to the edge is complete, and to 1 when there are unmatched orders, missing transaction statuses, missing inventory records, non-decomposable concurrency, or invalid constraint configurations. Indicates the first The synchronization disturbance coefficient of each topological edge is configured by the station-level system according to the edge type, and its value ranges from 0 to 1. It is used to adjust the impact of both quantity deviation and time deviation on the edge consistency value. (The formula is incomplete and requires further context.) , and All are normalized ratio values, therefore the exponential term as a whole remains dimensionless; This is the state flag value, whose function is to directly decay the edge consistency value to 0 when the link state is missing. (Left side of the formula) This represents the edge consistency ratio value, with the right side showing the exponential decay result. Both sides are dimensionless ratio values, thus ensuring consistent calculation results. If If not configured, the system uses the default synchronization disturbance coefficient in the station-level parameter area; if or If the input exceeds the upper limit of the model, the system will truncate it according to the upper limit before recalculating to avoid distortion of the model input caused by extreme acquisition errors, while retaining the original over-limit value and truncation mark.

[0033] For example, in a certain oil product circulation topology example, the edge constraint vector corresponding to the oil tank node to the oil nozzle node is: This indicates that the current edge's quantity deviation ratio is 0.17, the time deviation ratio is 0.27, and the state is complete; the synchronization disturbance coefficient of this edge is configured to 0.5, then the system calculates sequentially. , , The sum of the three terms is 0.12475. Then calculate... Therefore, the consistency value of this edge is approximately 0.88, indicating that there are some time and quantity deviations in the current inventory chain, but overall it still maintains a high degree of consistency. For the oil gun node to the order transaction node, if the edge constraint vector is... If the synchronization disturbance coefficient is also configured to 0.5, then the edge consistency value is... This indicates that the order chain mainly suffers from time delay, while the quantity consistency is relatively good. If the order transaction status is unmatched, the status flag value becomes 1, and the edge consistency value is directly 0. The system can distinguish between "time delay in the chain" and "missing chain status". The calculation results in this example correspond one-to-one with the aforementioned edge constraint vectors. The quantity deviation, time deviation, and status flag can all be traced from the topology instance to the original level gauge record, fuel dispenser controller record, or order interface record.

[0034] After obtaining the edge consistency values ​​of the two topological edges, the system further generates a topological consistency feature package. This process originates from the concept of serialized feature encapsulation, which combines multiple related features in the same business link into a unified input object according to the business direction. This application limits it to the oil product circulation link, enabling the subsequent artificial intelligence anomaly localization model to simultaneously read the consistency of the inventory link, the consistency of the order link, and the consistency differences between links. The system uses the following structure to generate the topological consistency feature package: ; in, Indicates the first Topological consistency feature package corresponding to each oil gun time window; This represents the edge consistency value of the topological edge from the oil tank node to the oil nozzle node, expressed as the edge constraint vector corresponding to the oil tank node to the oil nozzle node. Substituting into the aforementioned edge consistency formula, we obtain: This represents the edge consistency value of the topological edge from the fuel dispenser node to the order transaction node, expressed as the edge constraint vector corresponding to the fuel dispenser node to the order transaction node. Substituting into the aforementioned edge consistency formula, we obtain: This indicates the consistency difference between two topological edges, used to express the relative position of an anomaly in the oil product circulation chain; This represents the edge constraint vector from the oil tank node to the oil nozzle node; This represents the edge constraint vector from the oil gun node to the order transaction node. There is a clear derivation relationship between the two formulas: the system first generates edge consistency values ​​based on the edge constraint vectors, and then encapsulates the edge consistency values ​​and their differences, along with the original edge constraint vectors, into a topology consistency feature package. Because... and All of these originate from the exponential decay of the normalization bias, therefore Always within a finite range, and maintaining the physical meaning of consistent differences. If an edge causes [something] due to missing state or invalid configuration... If the value is 0, the system will still correspond to... and status flag writing This enables AI anomaly localization models to distinguish between true quantity deviations, true time deviations, and inconsistencies caused by missing data.

[0035] For example, when the consistency value from the tank node to the nozzle node is 0.88 and the consistency value from the nozzle node to the order transaction node is 0.91, the consistency difference between the two links is -0.03, indicating that the inventory link and the order link are generally in a similar state. If the consistency value of the inventory link is 0.42 and the consistency value of the order link is 0.89, the consistency difference is -0.47, indicating that anomalies are mainly concentrated in the inventory-side link. If the consistency value of the inventory link is 0.93 and the consistency value of the order link is 0.21, the consistency difference is 0.72, indicating that anomalies are mainly concentrated in the order transaction link. This consistency difference is used as a key link location feature in the subsequent AI anomaly localization model.

[0036] After completing the above processing, the system outputs a topology consistency feature packet. The output includes the inventory link consistency value, order link consistency value, inter-link consistency difference, and edge constraint vectors corresponding to the two topological edges for the current oil pump time window. The AI ​​anomaly localization model in the next step directly reads... The system uses link consistency features and edge constraint vectors to locate anomalies in inventory measurement, order generation, inventory synchronization, and transaction status. Through the link consistency feature generation in this step, the oil product circulation topology instance in step two is further transformed into a structured input object with the ability to express link direction, deviation source, and anomaly location, enabling the subsequent AI anomaly localization model to identify the source of anomalies around specific oil product circulation links. The system also saves... The corresponding topological instance identifier and model input version number ensure that the same feature meaning, component order, and normalized boundary are used in the model training and online inference phases.

[0037] S4: Input the topology consistency feature package into the artificial intelligence anomaly localization model, and output the anomaly source localization result through the artificial intelligence anomaly localization model. The anomaly source localization result is used to indicate inventory measurement anomalies, inventory synchronization anomalies, order generation anomalies, or transaction status anomalies.

[0038] Specifically, step four generates anomaly source localization results based on an AI-powered anomaly localization model. The system uses the topology consistency feature package output in the previous step... As input, the inventory link consistency value is used. Consistency value of order transaction chain Link consistency differences Inventory link edge constraint vector and order transaction link edge constraint vector All are written into the same input vector. Specifically, if The three components are, in order, the quantity deviation ratio of the inventory chain, the time deviation ratio, and the status flag. The three components in the input vector are, in order, the quantity deviation ratio, the time deviation ratio, and the status flag of the order transaction chain. Therefore, the system sets the input vector to a nine-dimensional structure and writes it sequentially. , , , The three components and The nine-dimensional input vector has three components. It can simultaneously express whether the inventory chain has deviated, whether the order transaction chain has deviated, whether the deviation occurred upstream or downstream, and whether the deviation was due to quantity, time, or missing status. After the input vector enters the artificial intelligence anomaly localization model, the model outputs response values ​​for four types of anomaly sources: inventory measurement anomaly, inventory synchronization anomaly, order generation anomaly, and transaction status anomaly. The model employs a three-layer feedforward neural network. The first layer is the link feature encoding layer, with 9 input nodes and 12 output neurons. Each output neuron receives all components of the nine-dimensional input vector, multiplies each component by the trained connection weights, adds them to the bias, and then performs linear rectified activation. The linear rectified activation outputs 0 when the weighted sum is less than 0, and retains the weighted sum when it is greater than or equal to 0. The second layer is the link correlation analysis layer, with 8 output neurons. Each output neuron receives the 12 output values ​​from the first layer and continues calculation using the trained connection weights and biases to learn the correspondence between inventory link features and order transaction link features. For example, low inventory link consistency and high order transaction link consistency indicate a bias towards inventory-side anomalies, while high inventory link consistency and low order transaction link consistency indicate a bias towards order-side anomalies. The third layer is the anomaly source output layer, with 4 linear output neurons. Each linear output neuron receives the 8 output values ​​from the second layer and generates raw response values ​​for four types of anomalies using the trained connection weights and biases. , , and The connection weights and biases of the three-layer network are obtained through training using historical oil gun time window samples. These historical samples consist of verified records of inventory measurement anomalies, inventory synchronization anomalies, order generation anomalies, and transaction status anomalies. Each sample's input uses the topology consistency feature package generated in step three. Sample labels are determined by anomaly source categories based on manual verification, equipment calibration, level gauge verification, and inventory or transaction interface logs. During training, these four anomaly source labels are used as supervisory signals to update the connection weights and biases of each layer. After training, the model weights, biases, feature component order, normalization boundaries, and category definitions are written to the station-level platform or headquarters platform along with the model version number. During online inference, the system only calls the published model version and does not use unverified real-time output results to update model parameters, thus avoiding the contamination of subsequent models by erroneous localization results.

[0039] The system introduces a link offset term based on the original response values ​​at the third layer, generating four types of anomaly source response values. This calculation originates from the Softmax normalization model in multi-class logistic regression, whose basic function is to convert the original response values ​​of multiple categories into comparable category response ratios. This application adds a link consistency difference term to this model, allowing the relative deviation direction between the inventory link and the order transaction link to directly participate in anomaly source localization. When the value is less than 0, it indicates that the consistency of the inventory chain is lower than that of the order transaction chain, and the model tends to improve the response to inventory measurement anomalies and inventory synchronization anomalies; when A value greater than 0 indicates that the consistency of the order transaction chain is lower than that of the inventory chain, and the model tends to improve its response to order generation anomalies and transaction status anomalies. The system uses the following formula to generate the first... Exception source response values: ; in, Indicates the first The first oil gun time window corresponds to the first Response values ​​of exception sources; This represents the output of the third layer of the artificial intelligence anomaly localization model. The original response value of the anomaly source is processed by the link feature encoding layer and the link correlation analysis layer. Each component is calculated layer by layer; Indicates the first The link offset coefficient corresponding to the anomaly source is obtained by training with historical anomaly samples and is used to adjust the impact of link consistency differences on the response value of the anomaly source. This indicates the link consistency differences generated in step three; This represents the anomaly category index, with values ​​corresponding to four anomaly sources. (The formula contains...) , , and All values ​​are model response values ​​or scale values; the exponential term and normalization result are both expressed as scales. There are four categories. The sum of these values ​​equals 1, which can be used as the relative response levels of each anomaly source within the same oil gun time window. To avoid numerical overflow during exponential calculation, the system can subtract the same maximum response offset from all four original response values ​​before performing the exponential calculation. This process does not change the relative proportions of the four response values ​​or the maximum response category.

[0040] For example, the inventory chain consistency value corresponding to a certain oil gun time window. The consistency value of the order transaction chain is 0.42. If the value is 0.89, then the link consistency difference is -0.47. The original response values ​​of the four anomaly sources output by the third layer of the AI ​​anomaly localization model are: inventory measurement anomaly 1.9, inventory synchronization anomaly 1.5, order generation anomaly 0.4, and transaction status anomaly 0.2; the corresponding link offset coefficients, after training with historical anomaly samples, are -0.8, -0.6, 0.5, and 0.4, respectively. The system first calculates the exponential term corresponding to the inventory measurement anomaly: The index item corresponding to inventory synchronization anomaly: The index item corresponding to the order generation anomaly: The index item corresponding to the abnormal transaction status: The sum of the four values ​​is approximately 17.87. Therefore, the response values ​​for inventory measurement anomalies are approximately 0.545, inventory synchronization anomalies are approximately 0.332, order generation anomalies are approximately 0.066, and transaction status anomalies are approximately 0.057. This result indicates that the current anomalies are mainly concentrated in the inventory-side link, with inventory measurement anomalies having the highest response, followed by inventory synchronization anomalies. If another oil gun time window... 0.93 If the coefficient is 0.21, then the link consistency difference is 0.72. The link offset coefficient corresponding to order-side anomalies will improve the response to order generation anomalies and transaction status anomalies, thus making the model output more biased towards order transaction link anomalies. The input values ​​in the above examples all come from the output of step three. The output values, along with the published model parameters, are written to the abnormal location record along with the oil gun time window identifier.

[0041] The system generates anomaly source location results based on the response values ​​of four anomaly sources. The result uses the largest response category as the anomaly source category and retains all four response categories for subsequent review. The system outputs the anomaly source location results using the following structure: ; in, Indicates the first Anomaly source location results corresponding to each oil gun time window; This indicates the anomaly source category corresponding to the current oil gun time window, which is determined by the category with the largest value among the four anomaly source response values. This represents the set of abnormal source response values ​​corresponding to the current oil gun time window, consisting of four categories. Composition. This formula originates from the maximum response principle in classification decision-making, which takes the category with the highest normalized response as the output category of the current sample. This application binds the decision result to the oil gun time window, so that each oil product flow behavior can form a corresponding abnormal source category and response value set. Taking the above calculation results as an example, the inventory measurement abnormal response value is the highest among the four response values, therefore It was determined to be an abnormal inventory measurement. The system records 0.545 for inventory measurement anomalies, 0.332 for inventory synchronization anomalies, 0.066 for order generation anomalies, and 0.057 for transaction status anomalies. If the maximum response value is lower than the trusted threshold set by the station-level system, or if the difference between the maximum response value and the second-highest response value is lower than the distinguishing threshold, the system will still record the anomalies. and However, the results are marked as low-confidence locations and transferred to manual review or equipment log review processes to avoid directly using boundary samples as the source of anomalies.

[0042] The system outputs the result of the anomaly source location. The system is integrated into the station-level digital management system for inventory verification, order verification, and equipment inspection. For fuel nozzle time windows with high inventory metering anomaly response values, the system writes the corresponding tank number, fuel nozzle number, and fuel nozzle time window identifier into the inventory verification task, and associates it with the level gauge sampling record, tank volume table, or temperature compensation record. For fuel nozzle time windows with high inventory synchronization anomaly response values, the system writes the corresponding inventory management system link into the synchronization check task, and associates it with the inventory interface upload time, cache status, and synchronization log. For fuel nozzle time windows with high order generation anomaly or transaction status anomaly response values, the system writes the corresponding order number, fuel nozzle number, and transaction status into the order verification task, and associates it with the order system log and transaction status return interface log. Through this processing, the topology consistency feature package generated in step three is converted into anomaly source location results, enabling the gas station digital management system to identify the source of anomalies around specific oil product flow links. This result is not only used to display the anomaly category, but also to indicate the sensors, fuel dispenser controllers, inventory synchronization interfaces, or transaction status interfaces that need to be verified, thus forming a closed loop from equipment data acquisition, link feature generation, artificial intelligence inference to technical anomaly location.

[0043] In one or more embodiments, such as Figure 2 As shown, an artificial intelligence-based digital management system for gas stations is disclosed, the system comprising: The data recording generation module is used to obtain the start / stop status of the fuel nozzle from the fuel dispenser controller to generate a fuel nozzle time window, collect the fuel tank inventory change data, fuel nozzle dispensing data and order transaction status data corresponding to the fuel nozzle time window, and synthesize a minimal data record of oil product circulation. The topology instance generation module is used to generate an oil product circulation topology instance based on the oil product circulation minimized data record. The oil product circulation topology instance includes oil tank nodes, oil nozzle nodes, and order transaction nodes. The oil tank nodes are connected to the oil nozzle nodes through a first topology edge, and the oil nozzle nodes are connected to the order transaction nodes through a second topology edge. An edge constraint vector is generated for the first topology edge and the second topology edge. The edge constraint vector includes quantity deviation ratio, time deviation ratio, and status flag. The feature package generation module is used to calculate the edge consistency value of the first topological edge and the second topological edge based on the edge constraint vector, and generate a topological consistency feature package, wherein the topological consistency feature package includes the edge consistency value, the consistency difference between the first topological edge and the second topological edge, and the edge constraint vector; An anomaly localization module is used to input the topology consistency feature package into the artificial intelligence anomaly localization model, and output the anomaly source localization result through the artificial intelligence anomaly localization model. The anomaly source localization result is used to indicate inventory measurement anomalies, inventory synchronization anomalies, order generation anomalies, or transaction status anomalies.

[0044] It is worth noting that the specific workflow of the AI-based digital management system for gas stations provided in this embodiment is the same as that of the AI-based digital management method for gas stations described in the above embodiments, and will not be repeated here.

[0045] This invention also provides an artificial intelligence-based digital management device for gas stations, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps described in the above embodiments of the artificial intelligence-based digital management method for gas stations, for example... Figure 1 The steps S1 to S4 described above; or, when the processor executes the computer program, it implements the functions of each module in the above system embodiments.

[0046] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the AI-based digital management device for gas stations.

[0047] The AI-based digital management device for gas stations can be a desktop computer, laptop, handheld computer, or cloud server, among other computing devices. This device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the AI-based digital management device may also include input / output devices, network access devices, and buses.

[0048] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the AI-based digital management equipment for gas stations, connecting all parts of the equipment via various interfaces and lines.

[0049] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the AI-based digital management equipment for gas stations by running or executing the computer programs and / or modules stored in the memory and by calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created according to the operation of the controller, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0050] The modules integrated into the AI-based digital management equipment for gas stations, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0051] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0052] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A digital management method for gas stations based on artificial intelligence, characterized in that: include: The fuel dispenser controller obtains the fuel nozzle start / stop status to generate a fuel nozzle time window, and collects the fuel tank inventory change data, fuel nozzle dispensing data and order transaction status data corresponding to the fuel nozzle time window to synthesize a fuel product circulation minimized data record; Based on the minimized data records of oil product circulation, an oil product circulation topology instance is generated. The oil product circulation topology instance includes oil tank nodes, oil nozzle nodes, and order transaction nodes. The oil tank nodes are connected to the oil nozzle nodes through a first topology edge, and the oil nozzle nodes are connected to the order transaction nodes through a second topology edge. An edge constraint vector is generated for the first topology edge and the second topology edge. The edge constraint vector includes quantity deviation ratio, time deviation ratio, and status flag. The edge consistency values ​​of the first topological edge and the second topological edge are calculated based on the edge constraint vector, and a topological consistency feature package is generated. The topological consistency feature package includes the edge consistency value, the consistency difference between the first topological edge and the second topological edge, and the edge constraint vector. The topology consistency feature package is input into the artificial intelligence anomaly localization model, and the anomaly source localization result is output by the artificial intelligence anomaly localization model. The anomaly source localization result is used to indicate inventory measurement anomalies, inventory synchronization anomalies, order generation anomalies, or transaction status anomalies.

2. The artificial intelligence-based digital management method for gas stations according to claim 1, characterized in that, The process of collecting data on changes in oil tank inventory, oil dispensing data, and order transaction status corresponding to the oil nozzle time window, and synthesizing a minimal data record of oil product circulation, includes: Verify the device number, timestamp continuity, and monotonicity of the cumulative oil output value of the oil gun output data; When a timestamp reversal or an abnormal jump in the cumulative oil output value is detected, the oil gun time window is marked as an abnormal acquisition window; During the verification process, the oil gun time window identifier is used as the primary key to combine the oil tank inventory change data, the oil gun dispensing data, and the order transaction status data into a single minimal oil product circulation data record.

3. The artificial intelligence-based digital management method for gas stations according to claim 1, characterized in that, The process of collecting the order transaction status data corresponding to the oil gun time window includes: Search for order records within the specified oil gun time window and the preset allowable delay range; When multiple candidate order records are found within the allowed delay range, the order record with the same oil gun number and the smallest deviation between the order oil volume and the cumulative oil output value of the oil gun output data is selected as the matching order. If no order record is found within the allowed delay range, write an order mismatch status into the oil product circulation minimized data record.

4. The artificial intelligence-based digital management method for gas stations according to claim 1, characterized in that, The process of generating an oil flow topology instance based on the oil flow minimization data record includes: Read the tank number and window inventory change value from the oil product circulation minimization data record to generate an oil tank node; Read the oil gun number and the cumulative oil output value of the oil gun to generate an oil gun node; Read the order number, order fuel quantity, and transaction status to generate an order transaction node; When the order number is empty, a temporary order transaction node identifier is generated based on the oil gun time window identifier and the oil gun number, and the transaction status of the order transaction node is set to unmatched.

5. The artificial intelligence-based digital management method for gas stations according to claim 1, characterized in that, The process of generating edge constraint vectors for the first topological edge and the second topological edge includes: Calculate the actual quantity difference and actual time difference between the data bound to the nodes at both ends of the first topological edge; The quantity deviation ratio is obtained by dividing the actual quantity difference by the preset allowable quantity deviation. The time deviation ratio is obtained by dividing the actual time difference by the preset allowable time difference. When the data binding at the endpoint node of the first or second topological edge is missing, the status flag is set to the first flag value; when the binding data is complete, the status flag is set to the second flag value.

6. The artificial intelligence-based digital management method for gas stations according to claim 1, characterized in that, The process of calculating the edge consistency value of the first topological edge and the second topological edge based on the edge constraint vector includes: Calculate the sum of squares based on the quantity deviation ratio and the time deviation ratio; Multiplying the quantity deviation ratio by the time deviation ratio yields the coupling product term; The attenuation exponent term is obtained by adding the sum of squares term, the coupling product term, and the synchronization disturbance coefficient; The exponential function value is calculated based on the decay exponential term and the state flag, and the exponential function value is used as the edge consistency value. The state flag is used to limit the edge consistency value to zero when the bound data is missing.

7. The artificial intelligence-based digital management method for gas stations according to claim 1, characterized in that, The process of generating the topology-consistent feature package includes: The edge consistency value corresponding to the first topological edge is determined as the inventory link consistency value; The edge consistency value corresponding to the second topological edge is determined as the order transaction link consistency value; The consistency difference is obtained by calculating the difference between the inventory link consistency value and the order transaction link consistency value. The inventory link consistency value, the order transaction link consistency value, the consistency difference, and the edge constraint vectors of the first topological edge and the second topological edge are serialized into the topological consistency feature package in a preset order.

8. The artificial intelligence-based digital management method for gas stations according to claim 1, characterized in that, The artificial intelligence anomaly localization model includes a link feature encoding layer, a link correlation analysis layer, and an anomaly source output layer. The link feature coding layer receives the topology consistency feature packet and extracts hidden features; The link association analysis layer receives the hidden features and calculates the association weight between the inventory link consistency value and the order transaction link consistency value; The anomaly source output layer receives the associated weights and outputs the original response values ​​for inventory measurement anomalies, inventory synchronization anomalies, order generation anomalies, and transaction status anomalies.

9. The artificial intelligence-based digital management method for gas stations according to claim 1, characterized in that, The process of outputting the anomaly source localization result through the artificial intelligence anomaly localization model includes: Obtain the product of the consistency difference and the preset link offset coefficient; The product is then added to the original response values ​​of the inventory measurement anomaly, the inventory synchronization anomaly, the order generation anomaly, and the transaction status anomaly to obtain the adjusted response value. The adjusted response values ​​are normalized to obtain response values ​​for four types of anomaly sources; The anomaly category corresponding to the response value with the largest value among the four types of anomaly source response values ​​is determined as the anomaly source localization result.

10. A digital management system for gas stations based on artificial intelligence, characterized in that: include: The data recording generation module is used to obtain the start / stop status of the fuel nozzle from the fuel dispenser controller to generate a fuel nozzle time window, collect the fuel tank inventory change data, fuel nozzle dispensing data and order transaction status data corresponding to the fuel nozzle time window, and synthesize a minimal data record of oil product circulation. The topology instance generation module is used to generate an oil product circulation topology instance based on the oil product circulation minimized data record. The oil product circulation topology instance includes oil tank nodes, oil nozzle nodes, and order transaction nodes. The oil tank nodes are connected to the oil nozzle nodes through a first topology edge, and the oil nozzle nodes are connected to the order transaction nodes through a second topology edge. An edge constraint vector is generated for the first topology edge and the second topology edge. The edge constraint vector includes quantity deviation ratio, time deviation ratio, and status flag. The feature package generation module is used to calculate the edge consistency value of the first topological edge and the second topological edge based on the edge constraint vector, and generate a topological consistency feature package. The topological consistency feature package includes the edge consistency value, the consistency difference between the first topological edge and the second topological edge, and the edge constraint vector. An anomaly localization module is used to input the topology consistency feature package into the artificial intelligence anomaly localization model, and output the anomaly source localization result through the artificial intelligence anomaly localization model. The anomaly source localization result is used to indicate inventory measurement anomalies, inventory synchronization anomalies, order generation anomalies, or transaction status anomalies.

Images