Intelligent gas station operation optimization system and method based on big data and artificial intelligence

Abstract

The invention relates to the technical field of gas station operation optimization, in particular to an intelligent gas station operation optimization system and method based on big data and artificial intelligence. The data acquisition module acquires and uploads multi-source data in real time, the data processing module cleans and standardizes the multi-source data to generate auditable tagged data, the AI analysis module performs demand prediction based on a demand prediction model, and performs grouping through a K-means algorithm to generate user portraits. The intelligent scheduling module carries out intelligent dynamic scheduling according to a demand prediction result, optimizes service efficiency and reduces client waiting time, the operation optimization module carries out precise marketing according to a client portrait and formulates a personalized preferential strategy, and the feedback suggestion collection module monitors operation data and client feedback in real time. And the result is returned to the AI analysis module to optimize the prediction model and the marketing strategy. According to the invention, the scheduling and prediction precision is improved, an operation decision basis is provided for managers, and the gas station operation efficiency and customer satisfaction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of gas station operation optimization, and in particular to a smart gas station operation optimization system and method based on big data and artificial intelligence. Background Art

[0002] As society's demand for convenient and efficient services increases, the operation and management model of gas stations has gradually exposed some problems. First, it is difficult for gas stations to effectively allocate resources during peak hours, resulting in longer waiting times and reduced customer experience. Second, traditional management methods lack an in-depth understanding of customers' personalized needs and are unable to provide targeted services, resulting in low customer loyalty. In addition, inventory management also faces difficulties. The seasonality and volatility of oil demand often lead to inventory backlogs or shortages, increasing operating costs. Overall, gas stations have significant deficiencies in responding to changing market demands and improving operational efficiency, and are in urgent need of an innovative means of operational optimization.

[0003] A Chinese invention patent with publication number CN114139759A discloses an operation cost optimization method and system. The method constructs underlying data based on operation cost related data extracted from various data source systems; determines the optimization mode to be adopted, obtains the operation strategy rule configuration when the first optimization mode is adopted, and determines the parameter space according to the operation strategy rule configuration, wherein an operation strategy rule configuration includes one or more strategy configuration parameters; based on the underlying data, uses an optimization algorithm to find the optimal operation strategy rule in the parameter space, and the optimal operation strategy rule includes various strategy configuration parameter values ​​for obtaining optimized operation costs, which solves the problems of long effect observation period, poor timeliness, and inability of manual data analysis to find the global optimal point, improves calculation efficiency and accuracy, and achieves optimization and reduction of operation costs.

[0004] However, the current operation of smart gas stations faces many challenges, especially in terms of customer flow fluctuations, inventory management, equipment maintenance and personalized services. In addition, the management model is difficult to accurately predict customer needs, resulting in unreasonable resource allocation during peak periods and waste of resources during off-peak periods. Insufficient or excess inventory leads to increased operating costs, and delayed equipment maintenance may cause service interruptions. In addition, since data involves multiple links and systems in the process of collection, transmission, storage and processing, data deviation, loss, duplication and unauthorized access are very likely to occur. Especially in the process of data use and sharing, there is a lack of effective verification methods to ensure the authenticity and reliability of data, which makes it difficult to discover and resolve potential risks in a timely manner. Summary of the invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and to propose a smart gas station operation optimization system and method based on big data and artificial intelligence.

[0006] The technical solution of the present invention is a method for optimizing the operation of a smart gas station based on big data and artificial intelligence, which includes the following specific implementation steps: S1. Collect and upload multi-source data in gas stations; S2. Clean the multi-source data, remove abnormal data, and use ETL technology to standardize data from different sources. Then, label the pre-processed multi-source data and generate an audit mark for the processed labeled multi-source data to obtain {audit mark, labeled multi-source data}. S3. Audit the labeled multi-source data, then build a demand forecasting model to generate demand forecasting results, and identify customer demand characteristics to generate user portraits; S4. Generate oil replenishment suggestions based on demand forecast results and historical replenishment records. According to demand forecast results, make estimates of customer flow for each day and time period, adjust the number of fuel dispensers and service personnel configuration to ensure service efficiency during peak hours, and dynamically adjust the opening and closing of fuel service counters based on the real-time flow of gas stations; S5. Develop targeted marketing strategies based on user portraits and use multiple channels for delivery; S6. Monitor various feedback data in real time, use the feedback data to tune the demand forecasting model parameters, collect customer feedback regularly, calculate customer retention rate, and develop new customer care measures by analyzing the customer retention rate.

[0007] Preferably, the process of generating the identification to be audited is as follows: S21, define a common parameter group {p, F p ,a,b,Group,G,q,H}; Among them, p is a large prime number, F p is a finite field, a and b are in the finite field F p The elliptic curve equation y defined above 2 =x 3 +ax+b mod p coefficients, the group Group satisfies the elliptic curve equation y 2 =x 3 +ax+b mod p is an Abelian group, q is the order of the group, G is the base point of the group, H is a hash function, and its output string length is 256 bits; S22, randomly select an element α∈Z q * ={1, 2, ..., q-1}, calculate the identification audit parameter PIa=α×G, and publish the public parameter group {p, F p, a, b, Group, G, q, H} and identify audit parameters PIa; S23, calculate the identification generation parameter PIg = H( l 2||ID||a||b||x G ||y G ||x A ||y A ); Among them, ID is the length of l 1-bit module number, l 2 for l 1 is converted into 2 bytes, x G ,y G The base point G is the elliptic curve equation y 2 =x 3 +ax+b mod p on the horizontal and vertical coordinates, x A ,y A To identify the audit parameter PIa in the elliptic curve equation y 2 =x 3 +ax+b mod p on the horizontal and vertical coordinates, || is the string concatenation symbol; S24, select a random number β∈[1,q-1], and calculate the elliptic curve point point=β×G=(x1,y1); S25. Calculate the first-level audit flag La1=[H(PIg||data)+x1] mod q; Wherein, data is a binary string form of labeled multi-source data, and if La1=0 or La1+β=q, S24 is returned; S26, calculate the secondary audit mark La2 = [(1 + α) -1 ×(β-La1×α)] mod q; If La2=0, then return to S24; S27. Generate an audit pending mark La={first-level audit pending mark La1, second-level audit pending mark La2}.

[0008] Preferably, the implementation process of auditing labeled multi-source data is as follows: S31. Extract the first-level to-be-audited identifier La1 and the second-level to-be-audited identifier La2 from the to-be-audited identifier La={first-level to-be-audited identifier La1, second-level to-be-audited identifier La2}. If La1∈[1, q-1] is not true or La2∈[1, q-1] is not true, the labeled multi-source data audit fails and a security warning is issued. Where q is the order of the group Group, and the group Group satisfies the definition in the finite field F p Elliptic curve equation y on2 =x 3 +ax+b mod p constitutes an Abelian group of points; S32. Calculate the first-level audit parameter Pa1=(La1+La2) mod q. If Pa1=0, the labeled multi-source data audit fails and a security warning is issued. S33, calculate the elliptic curve point point'=(x1', y1')=La2×G+Pa1×PIa; Among them, G is the base point of the group Group, and PIa is the identification audit parameter; S34, secondary audit parameter Pa2=[H(PIg||data)+x1'] mod q, if Pa2=La1, the labeled multi-source data audit passes, and the labeled multi-source data can be analyzed.

[0009] Preferably, the generation process of the demand forecast result is as follows: S41, selecting data related to demand forecasting from labeled multi-source data; S42, construct an input feature dataset X={x1, x2, ..., x i , ..., x n}; Among them, x i Represents any data features related to demand forecasting; S43, building a demand forecasting model; S44. The daily input data X in actual operation is input into the demand forecasting model, and the demand forecasting result is obtained through the gating mechanism of the LSTM unit.

[0010] Preferably, the construction process of the demand forecasting model is as follows: S51,The demand forecasting model is built based on the LSTM model, which contains three gate control units: forget gate, input gate and output gate; S52, forget gate: controls whether the current time step t retains the memory h of the previous moment t-1 : ; Where, X t represents the input feature vector of the current time step t, that is, the input data at time t; W f represents the weight of the forget gate; b f represents the bias term; σ represents the Sigmoid activation function, the output range is between 0 and 1, which is used to adjust the information transmission ratio; h t-1 Represents the hidden state of the previous time step t-1, which is the output of the LSTM unit passed from the previous moment, and it contains the memory information of the previous time step; Represents the output of the forget gate, ranging from 0 to 1, which determines the cell state C in the previous time step t-1 Whether to keep it depends on the current input X t and the hidden state h t-1 Calculated by the forget gate; S53, input gate: controls whether the current input X t Add to memory unit: ; ; Where W i , W C represents the weight matrix; b i 、b C represents the bias term; σ represents the Sigmoid activation function, the output range is between 0 and 1, which is used to adjust the information transmission ratio; i t Represents the output of the input gate, with a value range of 0 to 1, which controls whether the current input updates the memory unit; Represents the candidate cell state at the current time step t, which is obtained by h t-1 and X t Generated candidate memory information; S54, memory update: ; In the formula, C t-1 Represents the cell state at the previous time step t−1, the core memory part of LSTM, which stores long-term dependent memory information; S55, output gate: determine the final output state: ; ; In the formula, o t Represents the output of the output gate, which controls the ratio of the cell state transmitted to the output, and its value range is between 0 and 1; h t It represents the hidden state of the current time step t, which is also the output of the LSTM unit, representing the prediction information at that moment, and is passed to the next layer or output layer; σ represents the Sigmoid activation function, and the output range is between 0 and 1, which is used to adjust the information transmission ratio; S56. Use mean square error as the loss function to evaluate the error between the model prediction value and the true value: ; In the formula, y i represents the actual refueling demand of the i-th data; represents the predicted refueling demand of the i-th data; N represents the number of samples, which is used to calculate the average error in the loss function; MSE represents the mean square error.

[0011] Preferably, the process of generating a user portrait is as follows: S61. Selecting customer behavior data at a gas station from labeled multi-source data, including: Refueling frequency F: the number of times a customer refuels per month; Fueling quantity Q: the average fueling quantity of each refueling by the customer; Consumption amount A: the total amount of consumption by the customer at the gas station; Arrival time T: the time period when customers usually arrive at the station; Payment method P: the payment method used by customers at gas stations; S62: construct the above behavior data into feature variables that can be used for analysis, and construct any customer vector X i ={F i , Q i , A i , T i , P i}; S63. Use K-means clustering algorithm to group customers according to their behavioral characteristics: randomly select K initial cluster centers, and determine the group to which each customer belongs by calculating the Euclidean distance between each customer and the cluster center: ; In the formula, X i,j represents the jth eigenvalue of the i-th customer; C k,j represents the vector of the kth cluster center, which represents the characteristic average of the cluster center; d(X i ,C k ) represents the Euclidean distance between the i-th customer vector and the k-th cluster center; S64, repeatedly executing S63, in each round of iteration, assigning customers to the nearest cluster center, and then updating the cluster center position until the cluster center position converges, after which the customers are divided into a high-frequency customer group, a general customer group, and a potential customer group, and the customers in each group have similar consumption characteristics; S65. Generate a description label for each cluster and generate a customer profile {label name, average number of refueling times, average consumption amount, main time period}.

[0012] Preferably, the specific implementation steps of S4 are as follows: S71, obtaining oil quantity data of the oil depot, and updating the oil quantity inventory data of the oil depot; S72. Analyze the upcoming customer traffic and oil demand based on the demand forecast results, predict the consumption of each oil product in a specific period in the future, integrate historical replenishment records, seasonal changes, and holiday factors, formulate replenishment standards, and combine demand forecasts with inventory data to set a minimum inventory threshold for each oil product. Once the inventory falls below this threshold, a replenishment reminder will be triggered; S73. Generate replenishment suggestions based on real-time oil inventory data and demand forecast results, and analyze customer traffic in different time periods according to demand forecast results to determine peak and off-peak hours. Automatically increase the number of fuel dispensers enabled during peak hours, dynamically adjust the arrangement of service personnel according to traffic changes, and automatically close some fueling service counters during off-peak hours. S74. Integrate the real-time flow of gas stations and customer feedback information into feedback information.

[0013] The technical solution of the present invention is a smart gas station operation optimization system based on big data and artificial intelligence, which is used to execute the above-mentioned smart gas station operation optimization method based on big data and artificial intelligence, including: The data collection module is used to configure several IoT devices to collect multi-source data in the smart gas station; The data processing module is used to clean, standardize and label the collected multi-source data, and generate an audit mark for the processed labeled multi-source data; AI analysis module, which is used to build a demand forecasting model, predict the daily refueling volume, output the demand forecast results, and use the K-means clustering algorithm to classify customers and build user portraits; Intelligent scheduling module, which is used to generate oil replenishment suggestions based on demand forecast results and historical replenishment records, and to make estimates of customer flow per day and time period, adjust the number of fuel dispensers enabled and service personnel configuration, and dynamically adjust the opening and closing of fuel service counters based on the real-time flow of gas stations; Operation optimization module, used to formulate targeted marketing strategies based on user portraits and use multiple channels for delivery; The feedback and suggestion collection module is used to monitor the system's feedback data in real time, feed the data back to the AI ​​analysis module, and regularly collect customer feedback data and transmit the customer feedback data to the operation optimization module.

[0014] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects: The present invention designs a smart gas station operation optimization system and method based on big data and artificial intelligence. The system first collects multi-source data such as oil volume in the oil depot, customer arrival time, and refueling volume in real time through Internet of Things devices, and stores the data after cleaning and standardization to realize real-time sharing and transmission of data. Then, the received labeled multi-source data is audited, and by building a demand prediction model and user portrait, using the LSTM time series algorithm and clustering algorithm, the refueling demand and customer flow are estimated to achieve accurate customer portrait and group management, thereby providing personalized marketing and services. Afterwards, based on the predicted demand and inventory status, replenishment suggestions are intelligently generated to ensure stable oil supply. By real-time monitoring of equipment status and customer flow, the configuration of gas pumps and service personnel is dynamically adjusted to improve service efficiency and reduce waiting time during peak hours. Finally, the system continuously analyzes the operation data, improves the accuracy of scheduling and prediction, and provides managers with scientific operation support and decision-making basis, thereby improving the overall operation efficiency and customer satisfaction of gas stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a system architecture diagram of a smart gas station operation optimization system based on big data and artificial intelligence proposed in the present invention. DETAILED DESCRIPTION

[0016] Embodiment 1, as Figure 1 As shown, the present invention proposes a smart gas station operation optimization system based on big data and artificial intelligence, including a data acquisition module, a data processing module, an AI analysis module, an intelligent scheduling module, an operation optimization module and a feedback suggestion collection module.

[0017] The data collection module is used to configure several IoT devices to collect multi-source data in the smart gas station, including but not limited to historical fueling volume, customer preferences, and traffic data; The data processing module is used to clean, standardize and label the collected multi-source data, and generate an audit mark for the processed labeled multi-source data; The AI ​​analysis module is used to build a demand forecasting model, predict the daily refueling volume, output the demand forecast results, and use the K-means clustering algorithm to classify customers and build user portraits; The intelligent scheduling module is used to generate oil replenishment suggestions based on demand forecast results and historical replenishment records, and to make estimates of customer flow for each day and time period, adjust the number of fuel dispensers enabled and the configuration of service personnel, and dynamically adjust the opening and closing of the fuel service counter based on the real-time flow of gas stations; The operation optimization module is used to formulate targeted marketing strategies based on user portraits and use multiple channels for delivery; The feedback and suggestion collection module is used to monitor the system's feedback data in real time, feed the data back to the AI ​​analysis module, and regularly collect customer feedback and transmit it to the operation optimization module.

[0018] Embodiment 2: A smart gas station operation optimization method based on big data and artificial intelligence proposed in the present invention is applied to a smart gas station operation optimization system based on big data and artificial intelligence proposed in Embodiment 1. The specific implementation steps are as follows: S1. The data acquisition module is equipped with several IoT (Internet of Things) devices to collect and upload multi-source data in the gas station in real time, including but not limited to the oil level in the oil depot, the status of the gas station, the amount of fuel, the customer's arrival time and consumption behavior data, and then transmit the collected multi-source data in the gas station to the data processing module.

[0019] S2. The data processing module cleans the collected multi-source data, removes abnormal data, and uses ETL (extract-transform-load) technology to standardize data from different sources. It then labels the pre-processed multi-source data and generates an audit mark for the processed labeled multi-source data. It then transmits {audit mark, labeled multi-source data} to the AI ​​analysis module. The above-mentioned audit pending mark is generated as follows: S21, define a common parameter group {p, F p ,a,b,Group,G,q,H}; Among them, p is a large prime number, F p is a finite field, a and b are in the finite field F p The elliptic curve equation y defined above 2 =x 3 +ax+b mod p coefficients, the group Group satisfies the elliptic curve equation y 2 =x 3 +ax+b mod p is an Abelian group, q is the order of the group, G is the base point of the group, H is a hash function, and its output is a string of 256 bits; S22, randomly select an element α∈Z q * ={1, 2, ..., q-1}, calculate the identification audit parameter PIa=α×G, and disclose the public parameter group {p, F p , a, b, Group, G, q, H} and identify audit parameters PIa; S23, calculate the identification generation parameter PIg = H( l 2||ID||a||b||x G ||yG ||x A ||y A ); Among them, ID is the length assigned to the data processing module during the system construction process. l 1-bit module number, l 2 for l 1 is converted into 2 bytes, x G ,y G The base point G is the elliptic curve equation y 2 =x 3 +ax+b mod p on the horizontal and vertical coordinates, x A ,y A To identify the audit parameter PIa in the elliptic curve equation y 2 =x 3 +ax+b mod p on the horizontal and vertical coordinates, || is the string concatenation symbol; S24, select a random number β∈[1,q-1], and calculate the elliptic curve point point=β×G=(x1,y1); S25. Calculate the first-level audit flag La1=[H(PIg||data)+x1] mod q; Wherein, data is a binary string form of labeled multi-source data, and if La1=0 or La1+β=q, S24 is returned; S26, calculate the secondary audit mark La2 = [(1 + α) -1 ×(β-La1×α)] mod q; If La2=0, then return to S24; S27. Generate an audit pending mark La={first-level audit pending mark La1, second-level audit pending mark La2}.

[0020] S3, AI analysis module, on the one hand, builds a demand forecasting model to predict the customer flow of gas stations every day and time period, and supports the early adjustment of resources during peak demand periods. On the other hand, it identifies customer demand characteristics and builds user portraits to provide a basis for personalized services and precision marketing. The specific operation process of the AI ​​analysis module is as follows: S31. The AI ​​analysis module receives {the identification to be audited, the labeled multi-source data}, and performs a comprehensive audit check on the received labeled multi-source data. The audit process is as follows: S3101. Extract the first-level to-be-audited identifier La1 and the second-level to-be-audited identifier La2 from the to-be-audited identifier La={first-level to-be-audited identifier La1, second-level to-be-audited identifier La2}. If La1∈[1, q-1] is not true or La2∈[1, q-1] is not true, the audit of the labeled multi-source data received by the AI ​​analysis module fails, and the AI ​​analysis module issues a security warning. Where q is the order of the group Group, and the group Group satisfies the definition in the finite field F p Elliptic curve equation y on 2 =x 3 +ax+b mod p constitutes an Abelian group of points; S3102. Calculate the first-level audit parameter Pa1=(La1+La2) mod q. If Pa1=0, the audit of the labeled multi-source data received by the AI ​​analysis module fails, and the AI ​​analysis module issues a security warning. S3103, calculate the elliptic curve point point'=(x1', y1')=La2×G+Pa1×PIa; Among them, G is the base point of the group Group, and PIa is the identification audit parameter; S3104, secondary audit parameter Pa2=[H(PIg||data)+x1'] mod q, if Pa2=La1, the audit of the labeled multi-source data received by the AI ​​analysis module passes, and the labeled multi-source data can be analyzed; S32. Build a demand forecasting model, use the LSTM (Long Short-Term Memory) model to predict future refueling demand, and transmit the demand forecasting results to the intelligent scheduling module. The specific implementation steps are as follows: S3201. Select data related to demand forecasting from labeled multi-source data, including but not limited to: Historical refueling volume: Record daily and hourly refueling volume data over a period of time to identify typical demand fluctuations; Customer traffic: including but not limited to traffic flow data and customer arrival time; Weather data: including but not limited to temperature, precipitation, and wind speed meteorological data; Holiday data: Holidays or major events will affect demand, so add categorized data by date type (weekdays, weekends, holidays); S3202: Construct an input feature dataset X={x1, x2, ..., x i , ..., x n}; Among them, x iRepresents any data features related to demand forecasting; S3203. Build an LSTM model, which includes three gate control units: forget gate, input gate and output gate: Forget gate: controls whether the current time step t retains the memory h of the previous moment t-1 : ; Where, X t represents the input feature vector at the current time step t, that is, the input data at time t (including but not limited to the historical refueling volume, traffic flow, and weather data of the day); W f represents the weight of the forget gate; b f represents the bias term; σ represents the Sigmoid activation function, the output range is between 0 and 1, which is used to adjust the information transmission ratio; h t-1 Represents the hidden state of the previous time step t-1, which is the output of the LSTM unit passed from the previous moment, and it contains the memory information of the previous time step; Represents the output of the forget gate, ranging from 0 to 1, which determines the cell state C in the previous time step t-1 Whether to keep or not, determined by the current input X t and the hidden state h t-1 Calculated by the forget gate; Input gate: controls whether the current input X t Add to memory unit: ; ; Where W i , W C represents the weight matrix; b i 、b C represents the bias term; σ represents the Sigmoid activation function, the output range is between 0 and 1, which is used to adjust the information transmission ratio; i t Represents the output of the input gate, with a value range of 0 to 1, which controls whether the current input updates the memory unit; Represents the candidate cell state at the current time step t, which is obtained by h t-1 and X t Generated candidate memory information; Memory Update: ; In the formula, C t-1 Represents the cell state (or memory unit state) at the previous time step t−1, the core memory part of LSTM, which stores long-term dependent memory information; Output gate: determines the final output state: ; ; In the formula, o t Represents the output of the output gate, which controls the ratio of the cell state transmitted to the output, and its value range is between 0 and 1; h t It represents the hidden state of the current time step t, which is also the output of the LSTM unit, representing the prediction information at that moment, and is passed to the next layer or output layer; σ represents the Sigmoid activation function, and the output range is between 0 and 1, which is used to adjust the information transmission ratio; S3204. Use mean square error (MSE) as the loss function to evaluate the error between the model prediction value and the true value: ; In the formula, y i represents the actual refueling demand of the i-th data; represents the predicted refueling demand of the i-th data; N represents the number of samples, which is used to calculate the average error in the loss function; S3205, the daily input data in actual operation X t Input the demand forecasting model and process it through the gating mechanism of the LSTM unit to obtain the predicted refueling demand : ; In the formula, represents the prediction result output by the LSTM model, i.e., the predicted refueling demand at time step t; S33. Select and collect customer consumption habits, member preferences and refueling frequency data from labeled multi-source data, use K-means clustering analysis algorithm to divide customer groups into high-frequency customers, ordinary customers and potential customer groups, build user portraits, and transmit user portraits to the operation optimization module. The specific implementation process is as follows: S3301. Selecting customer behavior data at a gas station from labeled multi-source data, including but not limited to: Refueling frequency F: the number of times a customer refuels per month; Fueling quantity Q: the average fueling quantity of each refueling by the customer; Consumption Amount A: The total amount of consumption by customers at gas stations, including but not limited to fuel consumption and consumption in convenience stores; Arrival time T: the time period when customers usually arrive at the station, such as the peak hours on weekdays (7-9 a.m.) and daytime on weekends; Payment method P: The payment method used by customers at the gas station, including but not limited to: Alipay, WeChat, bank card, and cash; S3302: construct the above behavior data into feature variables that can be used for analysis, and construct any customer vector X i={F i , Q i , A i , T i , P i}; S3303, using K-means clustering algorithm to group customers according to their behavioral characteristics: randomly select K initial cluster centers (i.e., customer portrait categories, the customer category set in the present invention is 3), and determine the group to which each customer belongs by calculating the Euclidean distance between each customer and the cluster center: ; In the formula, X i,j represents the jth eigenvalue of the i-th customer; C k,j represents the vector of the kth cluster center, which represents the characteristic average of the cluster center; d(X i ,C k ) represents the Euclidean distance between the i-th customer vector and the k-th cluster center; S3304, repeatedly executing S3303, in each iteration, assigning customers to the nearest cluster center, and then updating the cluster center position until the cluster center position converges, after which the customers are divided into different groups (high-frequency customers, ordinary customers, and potential customer groups), and customers in each group have similar consumption characteristics; S3305. Generate a description label for each cluster and generate a customer profile {label name, average number of refueling times, average consumption amount, main time period}: For example, the description label of high-frequency customers k ={Tag name: "High-frequency customers", average number of refueling times: , average consumption amount: , main periods: }.

[0021] S4, the intelligent scheduling module generates oil replenishment suggestions based on demand forecast results and historical replenishment records to avoid insufficient inventory. According to the demand forecast results, it estimates the customer flow of each day and time period, adjusts the number of fuel dispensers and service personnel configuration to ensure service efficiency during peak hours, and dynamically adjusts the opening and closing of the fuel service counter based on the real-time flow of gas stations to reduce customer waiting time. The specific implementation steps are as follows: S41, the intelligent scheduling module receives the oil volume data of the oil depot collected by the data collection module, and the intelligent scheduling module updates the oil volume inventory data of the oil depot; S42. Analyze the upcoming customer flow and oil demand based on the demand forecast results, predict the consumption of each oil product in a specific period in the future, integrate historical replenishment records, seasonal changes, and holiday factors, formulate replenishment standards, and combine demand forecasts with inventory data to set the minimum inventory threshold for each oil product. Once the inventory is lower than this threshold, the intelligent scheduling module will trigger a replenishment reminder; S43. Based on the real-time oil inventory data and demand forecast results, the intelligent scheduling module generates replenishment suggestions, including the required oil type, replenishment quantity and priority, and analyzes customer traffic in different time periods according to the demand forecast results to determine peak and off-peak hours. During peak hours, the intelligent scheduling module automatically increases the number of fuel dispensers enabled, dynamically adjusts the arrangement of service personnel according to traffic changes, and automatically closes some fueling service counters during off-peak hours to improve the utilization efficiency of manpower and resources; S44: Integrate the real-time flow of the gas station and customer feedback information into feedback information, and transmit it to the feedback suggestion collection module.

[0022] S5. The operation optimization module formulates targeted marketing strategies based on user portraits, including but not limited to recommending member discounts, points programs, and pushing additional customized discounts or points reward plans to member customers, and uses multiple channels for delivery, including but not limited to pushing precise advertisements or promotions through in-site advertising screens or mobile terminals.

[0023] S6. The feedback and suggestion collection module monitors various operational data of the system in real time, including but not limited to the real-time flow of gas stations and customer feedback, and feeds the data back to the AI ​​analysis module. The AI ​​analysis module uses the feedback data to tune the parameters of the demand forecasting model to ensure the accuracy of forecasting and scheduling. The feedback and suggestion collection module regularly collects customer feedback, and understands customer satisfaction with marketing activities and services through surveys including but not limited to questionnaires and online evaluations, and calculates customer retention rates. The operation optimization module analyzes customer retention rates and develops new customer care measures to improve customer satisfaction and loyalty; The formula for calculating customer retention rate is as follows: ; In the formula, CRP is the customer retention rate, N retained represents the number of customers who are still active in a specific time period, N start Represents the total number of initial customers.

[0024] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A smart gas station operation optimization method based on big data and artificial intelligence, characterized in that: The specific implementation steps include the following: S1. Collect and upload multi-source data in gas stations; S2. Clean the multi-source data, remove abnormal data, and use ETL technology to standardize data from different sources. Then, label the pre-processed multi-source data and generate an audit mark for the processed labeled multi-source data to obtain {audit mark, labeled multi-source data}. S3. Audit the labeled multi-source data, then build a demand forecasting model to generate demand forecasting results, and identify customer demand characteristics to generate user portraits; S4. Generate oil replenishment suggestions based on demand forecast results and historical replenishment records. According to demand forecast results, make estimates of customer flow for each day and time period, adjust the number of fuel dispensers and service personnel configuration to ensure service efficiency during peak hours, and dynamically adjust the opening and closing of fuel service counters based on the real-time flow of gas stations; S5. Develop targeted marketing strategies based on user portraits and use multiple channels for delivery; S6. Monitor various feedback data in real time, use the feedback data to tune the demand forecasting model parameters, collect customer feedback regularly, calculate customer retention rate, and develop new customer care measures by analyzing the customer retention rate.

2. According to claim 1, a smart gas station operation optimization method based on big data and artificial intelligence is characterized in that: The process of generating the audit mark is as follows: S21, define a common parameter group {p, F p ,a,b,Group,G,q,H}; Among them, p is a large prime number, F p is a finite field, a and b are in the finite field F p The elliptic curve equation y defined above 2 =x 3 +ax+b modp coefficient, group Group satisfies the elliptic curve equation y 2 =x 3 +ax+b mod p is an Abelian group, q is the order of the group, G is the base point of the group, H is a hash function, and its output string length is 256 bits; S22, randomly select an element α∈Z q * ={1, 2, ..., q-1}, calculate the identification audit parameter PIa=α×G, and publish the public parameter group {p, F p , a, b, Group, G, q, H} and identify audit parameters PIa; S23, calculate the identification generation parameter PIg = H( l 2||ID||a||b||x G ||y G ||x A ||y A ); Among them, ID is the length of l 1-bit module number, l 2 for l 1 is converted into 2 bytes, x G ,y G The base point G is the elliptic curve equation y 2 =x 3 +ax+b mod p on the horizontal and vertical coordinates, x A ,y A To identify the audit parameter PIa in the elliptic curve equation y 2 =x 3 + ax+b mod p on the horizontal and vertical coordinates, || is the string concatenation symbol; S24, select a random number β∈[1,q-1], and calculate the elliptic curve point point=β×G=(x1,y1); S25. Calculate the first-level audit flag La1=[H(PIg||data)+x1] mod q; Wherein, data is a binary string form of labeled multi-source data, and if La1=0 or La1+β=q, S24 is returned; S26, calculate the secondary audit mark La2 = [(1 + α) -1 ×(β-La1×α)] mod q; If La2=0, then return to S24; S27. Generate an audit pending mark La={first-level audit pending mark La1, second-level audit pending mark La2}.

3. The method for optimizing the operation of a smart gas station based on big data and artificial intelligence according to claim 1 is characterized in that: The implementation process of auditing labeled multi-source data is as follows: S31. Extract the first-level to-be-audited identifier La1 and the second-level to-be-audited identifier La2 from the to-be-audited identifier La={first-level to-be-audited identifier La1, second-level to-be-audited identifier La2}. If La1∈[1, q-1] is not true or La2∈[1, q-1] is not true, the labeled multi-source data audit fails and a security warning is issued. Where q is the order of the group Group, and the group Group satisfies the definition in the finite field F p Elliptic curve equation y 2 =x 3 +ax+bmod p constitutes an Abelian group of points; S32. Calculate the first-level audit parameter Pa1=(La1+La2) mod q. If Pa1=0, the labeled multi-source data audit fails and a security warning is issued. S33, calculate the elliptic curve point point'=(x1', y1')=La2×G+Pa1×PIa; Among them, G is the base point of the group Group, and PIa is the identification audit parameter; S34, secondary audit parameter Pa2=[H(PIg||data)+x1'] mod q, if Pa2=La1, the labeled multi-source data audit passes, and the labeled multi-source data can be analyzed.

4. The method for optimizing the operation of a smart gas station based on big data and artificial intelligence according to claim 1 is characterized in that: The process of generating demand forecast results is as follows: S41, selecting data related to demand forecasting from labeled multi-source data; S42, construct an input feature dataset X={x1, x2, ..., x i , ..., x n }; Among them, x i Represents any data features related to demand forecasting; S43, building a demand forecasting model; S44. The daily input data X in actual operation is input into the demand forecasting model, and the demand forecasting result is obtained through the gating mechanism of the LSTM unit.

5. The method for optimizing the operation of a smart gas station based on big data and artificial intelligence according to claim 4 is characterized in that: The construction process of the demand forecasting model is as follows: S51,The demand forecasting model is built based on the LSTM model, which contains three gate control units: forget gate, input gate and output gate; S52, forget gate: controls whether the current time step t retains the memory h of the previous moment t-1 : ； Where, X t represents the input feature vector of the current time step t, that is, the input data at time t; W f represents the weight of the forget gate; b f represents the bias term; σ represents the Sigmoid activation function, the output range is between 0 and 1, and is used to adjust the information transmission ratio; h t-1 Represents the hidden state of the previous time step t-1, which is the output of the LSTM unit passed from the previous moment, and it contains the memory information of the previous time step; Represents the output of the forget gate, ranging from 0 to 1, which determines the cell state C in the previous time step t-1 Whether to keep it depends on the current input X t and the hidden state h t-1 Calculated by the forget gate; S53, input gate: controls whether the current input X t Add to memory unit: ； ； Where W i , W C represents the weight matrix; b i 、b C represents the bias term; σ represents the Sigmoid activation function, the output range is between 0 and 1, and it is used to adjust the information transmission ratio; i t Represents the output of the input gate, with a value range of 0 to 1, which controls whether the current input updates the memory unit; Represents the candidate cell state at the current time step t, which is obtained by h t-1 and X t Generated candidate memory information; S54, memory update: ； In the formula, C t-1 Represents the cell state at the previous time step t−1, the core memory part of LSTM, which stores long-term dependent memory information; S55, output gate: determine the final output state: ； ； In the formula, o t Represents the output of the output gate, which controls the ratio of the cell state transmitted to the output, and its value range is between 0 and 1; h t It represents the hidden state of the current time step t, which is also the output of the LSTM unit, representing the prediction information at that moment, and is passed to the next layer or output layer; σ represents the Sigmoid activation function, and the output range is between 0 and 1, which is used to adjust the information transmission ratio; S56. Use mean square error as the loss function to evaluate the error between the model prediction value and the true value: ； In the formula, y i represents the actual refueling demand of the i-th data; represents the predicted refueling demand of the i-th data; N represents the number of samples, which is used to calculate the average error in the loss function; MSE represents the mean square error.

6. The method for optimizing the operation of a smart gas station based on big data and artificial intelligence according to claim 1 is characterized in that: The process of generating a user portrait is as follows: S61. Selecting customer behavior data at a gas station from labeled multi-source data, including: Refueling frequency F: the number of times a customer refuels per month; Fueling quantity Q: the average fueling quantity of each refueling by the customer; Consumption amount A: the total amount of consumption by the customer at the gas station; Arrival time T: the time period when customers usually arrive at the station; Payment method P: the payment method used by customers at gas stations; S62: construct the above behavior data into feature variables that can be used for analysis, and construct any customer vector X i ={F i , Q i , A i , T i , P i }; S63. Use K-means clustering algorithm to group customers according to their behavioral characteristics: randomly select K initial cluster centers, and determine the group to which each customer belongs by calculating the Euclidean distance between each customer and the cluster center: ； Where, X i,j represents the jth eigenvalue of the i-th customer; C k,j represents the vector of the kth cluster center, which represents the characteristic average of the cluster center; d(X i ,C k ) represents the Euclidean distance between the i-th customer vector and the k-th cluster center; S64, repeatedly executing S63, in each round of iteration, assigning customers to the nearest cluster center, and then updating the cluster center position until the cluster center position converges, after which the customers are divided into a high-frequency customer group, a general customer group, and a potential customer group, and the customers in each group have similar consumption characteristics; S65. Generate a description label for each cluster and generate a customer profile {label name, average number of refueling times, average consumption amount, main time period}.

7. The method for optimizing the operation of a smart gas station based on big data and artificial intelligence according to claim 1 is characterized in that: The specific implementation steps of S4 are as follows: S71, obtaining oil quantity data of the oil depot, and updating the oil quantity inventory data of the oil depot; S72. Analyze the upcoming customer traffic and oil demand based on the demand forecast results, predict the consumption of each oil product in a specific period in the future, integrate historical replenishment records, seasonal changes, and holiday factors, formulate replenishment standards, and combine demand forecasts with inventory data to set a minimum inventory threshold for each oil product. Once the inventory falls below this threshold, a replenishment reminder will be triggered; S73. Generate replenishment suggestions based on real-time oil inventory data and demand forecast results, and analyze customer traffic in different time periods according to demand forecast results to determine peak and off-peak hours. Automatically increase the number of fuel dispensers enabled during peak hours, dynamically adjust the arrangement of service personnel according to traffic changes, and automatically close some fueling service counters during off-peak hours. S74. Integrate the real-time flow of gas stations and customer feedback information into feedback information.

8. A smart gas station operation optimization system based on big data and artificial intelligence, which is used to execute a smart gas station operation optimization method based on big data and artificial intelligence as described in any one of claims 1 to 7, characterized in that: include: The data collection module is used to configure several IoT devices to collect multi-source data in the smart gas station; The data processing module is used to clean, standardize and label the collected multi-source data, and generate an audit mark for the processed labeled multi-source data; AI analysis module, which is used to build a demand forecasting model, predict the daily refueling volume, output the demand forecast results, and use the K-means clustering algorithm to classify customers and build user portraits; Intelligent scheduling module, which is used to generate oil replenishment suggestions based on demand forecast results and historical replenishment records, and to make estimates of customer flow per day and time period, adjust the number of fuel dispensers enabled and service personnel configuration, and dynamically adjust the opening and closing of fuel service counters based on the real-time flow of gas stations; Operation optimization module, used to formulate targeted marketing strategies based on user portraits and use multiple channels for delivery; The feedback and suggestion collection module is used to monitor the system's feedback data in real time, feed the data back to the AI ​​analysis module, and regularly collect customer feedback data and transmit the customer feedback data to the operation optimization module.

Images