Energy charging customer conversion tendency prediction method and related product

Abstract

The invention discloses an energy charging customer conversion tendency prediction method and a related product. According to the method, energy charging consumption behaviors of non-conversion customers are predicted, then potential conversion customers are selected, three kinds of clustering operations are carried out, and finally the predicted purchase probability of the potential conversion customers for each kind of non-energy charging consumption is obtained. According to the method, the whole process from source data to predictive analysis to formalized output can be achieved, the output result directly meets multi-angle business requirements, and the effect that business personnel directly use zero codes is achieved.

Description

Technical Field

[0001] The present invention relates to a method for predicting charging customer conversion tendency and related products. Background Art

[0002] After refueling at a gas station, refueling customers may purchase other products at the gas station. After refueling at a gas station, gas customers may also purchase other non-gas products at the gas station. It is necessary to accurately predict the non-gas consumption tendency of gas customers to provide reliable auxiliary information for marketers. Summary of the invention

[0003] The present invention provides a charging customer conversion tendency prediction method and related products to obtain rich consumption tendency information of charging customers.

[0004] The technical solution of the present invention is as follows: A method for predicting the conversion tendency of charging customers, comprising:

[0005] Input the charging consumption feature time series of the unconverted customers within the first set time period up to the current period into a pre-trained prediction model, and obtain the predicted charging consumption feature time series of the unconverted customers for the second set time period after the current period, wherein the prediction model is trained with the charging consumption feature time series of the second set time period before the first non-charging consumption of the converted customers as output and the charging consumption feature time series of the first set time period before as input;

[0006] The charging consumption feature time series of the unconverted customers within the first set time period up to the current period is spliced ​​with the predicted charging consumption feature time series of the unconverted customers within the second set time period after the current period to obtain the spliced ​​charging consumption feature time series, and the similarity between the spliced ​​charging consumption feature time series and the charging consumption feature time series of the sum of the first set time period and the second set time period before the first non-charging consumption of the converted customers is calculated, and the unconverted customers whose similarity is greater than the set threshold are selected as potential conversion customers, and the corresponding similarity is used as the shopping probability of the potential conversion customers;

[0007] Cluster the charging consumption feature time series of potential conversion customers within the first set time period up to the current period and the charging consumption feature time series of converted customers within the second set time period before the first non-charging consumption and the first set time period before that, to obtain the first cluster center and a set number of sample points closest to the first cluster center;

[0008] Clustering the predicted charging consumption feature time series of the potential conversion customers for the second set time after the current period and the charging consumption feature time series of the converted customers for the second set time before the first non-charging consumption, to obtain the second cluster center and the set number of sample points closest to the second cluster center;

[0009] Clustering the concatenated charging consumption feature time series of potential conversion customers and the charging consumption feature time series of the sum of the first set time and the second set time before the first non-charging consumption of converted customers, to obtain the third cluster center and the set number of sample points closest to the third cluster;

[0010] The predicted purchase probability of various types of non-charging consumption of potential converted customers is predicted based on the frequency of various types of non-charging consumption of converted customers and the distances between potential converted customers and the first cluster center, the second cluster center and the third cluster center.

[0011] Optionally, the method further includes: normalizing the predicted purchase probabilities of various types of non-charging consumption of potential conversion customers, and outputting the maximum normalized value of the potential conversion customer and the corresponding non-charging consumption type.

[0012] Optionally, the normalization formula is as follows: i =p i (Sigmoid((Max(r ij )-r kj ) / (Max(r ij )-Min(r ij )))),l i is the maximum normalized value of potential conversion customer i, p i is the purchase probability of potential conversion customer i, sigmoid is the activation function, Max is the maximum value function, Min is the minimum value function, r ij is the predicted purchase probability of potential conversion customer i in category j of non-charging consumption.

[0013] Optionally, it also includes: outputting the purchase probability of the potential converted customer.

[0014] Optionally, the predicted purchase probability of each type of non-charging consumption of the potential converted customer is equal to the corresponding type of non-charging consumption frequency of the converted customer multiplied by the sum of the distances between the potential converted customer and the first type cluster center, the second type cluster center and the third type cluster center.

[0015] Optionally, the prediction model is a bidirectional long short-term memory neural network.

[0016] Optionally, the charging customer is a refueling customer or a gas filling customer.

[0017] The technical solution of the present invention is as follows: A device for predicting conversion tendency of charging customers, comprising:

[0018] A prediction module, used to input the charging consumption feature time series of the unconverted customers within a first set time period up to the current period into a pre-trained prediction model, and obtain the predicted charging consumption feature time series of the unconverted customers for a second set time period after the current period, wherein the prediction model is trained with the charging consumption feature time series of the second set time period before the first non-charging consumption of the converted customers as output and the charging consumption feature time series of the first set time period before as input;

[0019] A screening module is used to splice the charging consumption feature time series of unconverted customers within a first set time period up to the current period with the predicted charging consumption feature time series of unconverted customers within a second set time period after the current period to obtain a spliced ​​charging consumption feature time series, calculate the similarity between the spliced ​​charging consumption feature time series and the charging consumption feature time series of the sum of the first set time period and the second set time period before the first non-charging consumption of the converted customers, select the unconverted customers whose similarity is greater than a set threshold as potential conversion customers, and use the corresponding similarity as the shopping probability of the potential conversion customers;

[0020] The first clustering module is used to cluster the charging consumption feature time series of the potential conversion customers within the first set time period up to the current period and the charging consumption feature time series of the converted customers within the second set time period before the first non-charging consumption and the first set time period before, to obtain the first cluster center and the set number of sample points closest to the first cluster center;

[0021] The second clustering module is used to cluster the predicted charging consumption feature time series of the potential conversion customers for a second set time after the current period and the charging consumption feature time series of the converted customers for a second set time before the first non-charging consumption, to obtain the second cluster center and the set number of sample points closest to the second cluster center;

[0022] The third clustering module is used to cluster the concatenated charging consumption feature time series of potential conversion customers and the charging consumption feature time series of the sum of the first set time and the second set time before the first non-charging consumption of the converted customers, to obtain the third cluster center and the set number of sample points closest to the third cluster;

[0023] The prediction module is used to predict the predicted purchase probability of various types of non-charging consumption of potential converted customers based on the frequencies of various types of non-charging consumption of converted customers and the distances between potential converted customers and the first cluster center, the second cluster center and the third cluster center.

[0024] The technical solution of the present invention is as follows: A device for predicting the conversion tendency of charging customers, comprising: a memory and a processor, the memory stores a program, and the processor runs the program to execute the above-mentioned method for predicting the conversion tendency of charging customers.

[0025] The technical solution of the present invention is as follows: a program product, which executes the above-mentioned charging customer conversion tendency prediction method when running on a processor.

[0026] By only providing the time series of charging consumption characteristics of converted and non-converted customers and the time series of non-charging consumption characteristics of converted customers, we can predict non-converted customers and determine the predicted purchase probability of different types of non-charging products for potential converted customers. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of the charging customer conversion tendency prediction method of the present invention.

[0028] Figure 2 It is a structural block diagram of the charging customer conversion tendency prediction device of the present invention.

[0029] Figure 3 It is another structural block diagram of the charging customer conversion tendency prediction device of the present invention. DETAILED DESCRIPTION

[0030] The present invention is further described below in conjunction with specific embodiments, but the protection scope of the present invention is not limited thereto.

[0031] refer to Figure 1 , an embodiment of the present invention provides a method for predicting charging customer conversion tendency, comprising the following steps.

[0032] Step 101, input the charging consumption feature time series of the unconverted customers within the first set time period up to the current period into a pre-trained prediction model, and obtain the predicted charging consumption feature time series of the unconverted customers for the second set time period after the current period, wherein the prediction model is trained with the charging consumption feature time series of the second set time period before the first non-charging consumption of the converted customers as output and the charging consumption feature time series of the first set time period before as input.

[0033] The following is an example of energy consumption at a gas station. The following table is an exemplary display of a data point in a time series of energy consumption characteristics of a specific customer.

[0034] Field Name Business Implications Data Sample uid User unified identity 9sadjfb723r timestamp Timestamp 20210601 age age 31 sex gender male type Oil type 95 litter Refueling liters 5 amount Refueling amount 300 station Gas station name A gas station

[0035] The following table is an exemplary display of a data point in the time series of non-charging consumption characteristics of a specific converted customer.

[0036]

[0037]

[0038] The time slice corresponding to each of the above data points is, for example, 1 day. The first set time period is, for example, 100 days, and the second set time period is, for example, 10 days. The refueling consumption data of the unconverted customers in the next 10 days is predicted based on the refueling consumption data of the unconverted customers in the last 100 days.

[0039] The scale of the time slice is, for example, one week or one month.

[0040] The prediction model is trained using the refueling data of the converted customers, especially to extract the correlation between the refueling behavior characteristics of the converted customers before their first non-refueling consumption and their previous refueling behavior characteristics. The entire operation of the method is driven by known historical data.

[0041] Step 102: Concatenate the charging consumption feature time series of the unconverted customers within the first set time period up to the current period with the predicted charging consumption feature time series of the unconverted customers within the second set time period after the current period to obtain a concatenated charging consumption feature time series, calculate the similarity between the concatenated charging consumption feature time series and the charging consumption feature time series of the converted customers that is the sum of the first set time period and the second set time period before the first non-charging consumption, select the unconverted customers whose similarity is greater than the set threshold as potential converted customers, and use the corresponding similarity as the shopping probability of the potential converted customers.

[0042] Using the previous example, the refueling data of the unconverted customers in the last 100 days and the predicted next 10 days are compared with the refueling data of the converted customers in the 110 days before their first non-refueling consumption. The similarity can be measured by the Euclidean distance. The threshold is set to any threshold such as 0.6 or 0.7. A higher similarity calculation result indicates that the behavioral characteristics of the unconverted customers and the converted customers in refueling consumption are more similar.

[0043] Step 103: Cluster the charging consumption feature time series of the potential converted customers within the first set time period up to the current period and the charging consumption feature time series of the converted customers within the second set time period before the first non-charging consumption and the first set time period before that, to obtain the first cluster center and a set number of sample points closest to the first cluster center.

[0044] Any known clustering algorithm may be used, such as the K-means clustering algorithm or the K-means++ clustering algorithm.

[0045] Continuing with the previous example, cluster the refueling data of potential conversion customers in the last 100 days and the refueling data of converted customers from 110 days before their first non-refueling consumption to 10 days before their first non-refueling consumption.

[0046] Step 104: cluster the predicted charging consumption feature time series of the potential converted customers for the second set time after the current period and the charging consumption feature time series of the converted customers for the second set time before the first non-charging consumption to obtain the second cluster center and the set number of sample points closest to the second cluster center.

[0047] Continuing with the previous example, cluster the refueling data of potential conversion customers predicted for the next 10 days with the refueling data of converted customers within 10 days before their first non-refueling consumption.

[0048] Step 105: Cluster the concatenated charging consumption feature time series of the potential converted customers and the charging consumption feature time series of the sum of the first set time and the second set time before the first non-charging consumption of the converted customers to obtain the third cluster center and the set number of sample points closest to the third cluster.

[0049] Continuing with the previous example, cluster the refueling data of potential conversion customers in the past 100 days and the next 10 days and the refueling data of converted customers in the first 110 days before their first non-refueling consumption.

[0050] The above analysis of the refueling behaviors of potential conversion customers and converted customers is conducted from three different dimensions to analyze the similarities between the two.

[0051] Step 106: predict the predicted purchase probability of various types of non-charging consumption of potential conversion customers based on the frequencies of various types of non-charging consumption of converted customers and the distances between the potential conversion customers and the first cluster center, the second cluster center, and the third cluster center.

[0052] The converted customers' food consumption frequency, drug consumption frequency, toilet paper consumption frequency, etc. are divided into categories, and statistics are made based on the consumption records of the converted customers. For example, the converted customer's food consumption frequency is once a week, and the drug consumption frequency is 0.1 times a week.

[0053] For example, the prediction result is that the predicted purchase probability of food consumption of a certain potential conversion customer in the next 10 days is 0.7, and the predicted purchase probability of medicine consumption in the next 10 days is 0.1.

[0054] Optionally, the predicted purchase probability of each type of non-charging consumption of the potential converted customer is equal to the corresponding type of non-charging consumption frequency of the converted customer multiplied by the sum of the distances between the potential converted customer and the first type cluster center, the second type cluster center and the third type cluster center.

[0055] When a potential conversion customer is not located near any of the first-category cluster centers, the distance between its data point and the cluster center can be set to a larger set value.

[0056] The distance between the data point of a potential conversion customer and the cluster center can be measured using the Euclidean distance.

[0057] Optionally, the method further includes: normalizing the predicted purchase probabilities of various types of non-charging consumption of potential conversion customers, and outputting the maximum normalized value of the potential conversion customer and the corresponding non-charging consumption type. The normalization process is to make the displayed data have a unified standard.

[0058] Optionally, the normalization formula is as follows: li=pi(Sigmoid((Max(rij)-rkj) / (Max(rij)-Min(rij)))), where li is the maximum normalized value of potential conversion customer i, pi is the shopping probability of potential conversion customer i, sigmoid is the activation function, Max is the maximum value function, Min is the minimum value function, and rij is the predicted purchase probability of potential conversion customer i in type j non-charging consumption.

[0059] Optionally, it also includes: outputting the purchase probability of the potential converted customer.

[0060] In one embodiment, an example of the method output is: the probability of potential conversion customer a making non-gasoline purchases in the next 10 days is 0.5 (i.e., the pi value), and his predicted purchase probability of food consumption (i.e., number j) is the highest, and the predicted purchase probability of food consumption is 0.5.

[0061] Optionally, the prediction model is a bidirectional long short-term memory neural network. The prediction model needs to mine the causal relationship between data sequences, and any known appropriate model can be used.

[0062] Optionally, the charging customer is a refueling customer or a gas filling customer.

[0063] refer to Figure 2 Based on the same inventive concept, an embodiment of the present invention further provides a charging customer conversion tendency prediction device, comprising:

[0064] Prediction module 1, used for inputting the charging consumption characteristic time series of the unconverted customers within the first set time period up to the current period into a pre-trained prediction model, and obtaining the predicted charging consumption characteristic time series of the unconverted customers for the second set time period after the current period, wherein the prediction model is trained with the charging consumption characteristic time series of the second set time period before the first non-charging consumption of the converted customers as output and the charging consumption characteristic time series of the first set time period before as input;

[0065] Screening module 2 is used to splice the charging consumption feature time series of the unconverted customers within the first set time period up to the current period with the predicted charging consumption feature time series of the unconverted customers within the second set time period after the current period to obtain a spliced ​​charging consumption feature time series, calculate the similarity between the spliced ​​charging consumption feature time series and the charging consumption feature time series of the sum of the first set time period and the second set time period before the first non-charging consumption of the converted customers, select the unconverted customers whose similarity is greater than the set threshold as potential conversion customers, and use the corresponding similarity as the shopping probability of the potential conversion customers;

[0066] The first clustering module 3 is used to cluster the charging consumption characteristic time series of the potential conversion customers within the first set time period up to the current period and the charging consumption characteristic time series of the converted customers within the second set time period before the first non-charging consumption and the first set time period before, to obtain the first cluster center and the set number of sample points closest to the first cluster center;

[0067] The second clustering module 4 is used to cluster the predicted charging consumption feature time series of the potential conversion customers for the second set time after the current period and the charging consumption feature time series of the converted customers for the second set time before the first non-charging consumption, to obtain the second cluster center and the set number of sample points closest to the second cluster center;

[0068] The third clustering module 5 is used to cluster the concatenated charging consumption feature time series of potential conversion customers and the charging consumption feature time series of the sum of the first set time and the second set time before the first non-charging consumption of the converted customers, to obtain the third cluster center and the set number of sample points closest to the third cluster;

[0069] Prediction module 6 is used to predict the predicted purchase probability of various non-charging consumption of potential conversion customers based on the frequency of various non-charging consumption of converted customers and the distances between potential conversion customers and the first cluster center, the second cluster center and the third cluster center.

[0070] Optionally, the prediction module 6 is further used to normalize the predicted purchase probabilities of various types of non-charging consumption of potential conversion customers, and output the maximum normalized value of the potential conversion customer and the corresponding non-charging consumption type.

[0071] Optionally, the prediction module 6 is further configured to output a purchase probability of a potential conversion customer.

[0072] refer to Figure 3 Based on the same inventive concept, an embodiment of the present invention also provides a charging customer conversion tendency prediction device, including: a memory and a processor, the memory stores a program, and the processor runs the program to execute the above-mentioned charging customer conversion tendency prediction method.

[0073] The memory is, for example, any known memory such as a hard disk, a flash memory, an optical disk, etc. The processor is, for example, a known processor such as a central processing unit (CPU), a graphics processing unit (GPU), etc.

[0074] Based on the same inventive concept, an embodiment of the present invention further provides a program product, which executes the above-mentioned charging customer conversion tendency prediction method when running on a processor.

[0075] The various embodiments of the present invention are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments.

[0076] The protection scope of the present invention is not limited to the above-mentioned embodiments. Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the scope and spirit of the present invention. If these changes and modifications fall within the scope of the claims of the present invention and their equivalents, the present invention is intended to include these changes and modifications.

Claims

1. A method for predicting charging customer conversion tendency, characterized in that: include: Input the charging consumption feature time series of the unconverted customers within the first set time period up to the current period into a pre-trained prediction model, and obtain the predicted charging consumption feature time series of the unconverted customers for the second set time period after the current period, wherein the prediction model is trained with the charging consumption feature time series of the second set time period before the first non-charging consumption of the converted customers as output and the charging consumption feature time series of the first set time period before as input; The charging consumption feature time series of the unconverted customers within the first set time period up to the current period is spliced ​​with the predicted charging consumption feature time series of the unconverted customers within the second set time period after the current period to obtain the spliced ​​charging consumption feature time series, and the similarity between the spliced ​​charging consumption feature time series and the charging consumption feature time series of the sum of the first set time period and the second set time period before the first non-charging consumption of the converted customers is calculated, and the unconverted customers whose similarity is greater than the set threshold are selected as potential conversion customers, and the corresponding similarity is used as the shopping probability of the potential conversion customers; Cluster the charging consumption feature time series of potential conversion customers within the first set time period up to the current period and the charging consumption feature time series of converted customers within the second set time period before the first non-charging consumption and the first set time period before that, to obtain the first cluster center and a set number of sample points closest to the first cluster center; Clustering the predicted charging consumption feature time series of the potential conversion customers for the second set time after the current period and the charging consumption feature time series of the converted customers for the second set time before the first non-charging consumption, to obtain the second cluster center and the set number of sample points closest to the second cluster center; Clustering the concatenated charging consumption feature time series of potential conversion customers and the charging consumption feature time series of the sum of the first set time and the second set time before the first non-charging consumption of converted customers, to obtain the third cluster center and the set number of sample points closest to the third cluster; The predicted purchase probability of various types of non-charging consumption of potential converted customers is predicted based on the frequency of various types of non-charging consumption of converted customers and the distances between potential converted customers and the first cluster center, the second cluster center and the third cluster center.

2. The method according to claim 1, characterized in that Also includes: The predicted purchase probabilities of various types of non-charging consumption of potential conversion customers are normalized, and the maximum normalized value of the potential conversion customer and the corresponding non-charging consumption type are output.

3. The method according to claim 2, characterized in that The normalization formula is as follows: i =p i (Sigmoid((Max(r ij )-r kj ) / (Max(r ij )-Min(r ij )))),l i is the maximum normalized value of potential conversion customer i, p i is the purchase probability of potential conversion customer i, sigmoid is the activation function, Max is the maximum value function, Min is the minimum value function, r ij is the predicted purchase probability of potential conversion customer i in category j of non-charging consumption.

4. The method according to claim 1, characterized in that: Also includes: Output the purchase probability of potential conversion customers.

5. The method according to claim 1, characterized in that: The predicted purchase probability of each type of non-charging consumption of a potential conversion customer is equal to the frequency of the corresponding type of non-charging consumption of the converted customer multiplied by the sum of the distances between the potential conversion customer and the center of the first cluster, the center of the second cluster, and the center of the third cluster.

6. The method according to claim 1, characterized in that The prediction model is a bidirectional long short-term memory neural network.

7. The method according to claim 1, characterized in that The charging customer is a refueling customer or a gas filling customer.

8. A device for predicting charging customer conversion tendency, characterized in that: include: A prediction module, used to input the charging consumption feature time series of the unconverted customers within a first set time period up to the current period into a pre-trained prediction model, and obtain the predicted charging consumption feature time series of the unconverted customers for a second set time period after the current period, wherein the prediction model is trained with the charging consumption feature time series of the second set time period before the first non-charging consumption of the converted customers as output and the charging consumption feature time series of the first set time period before as input; A screening module is used to splice the charging consumption feature time series of unconverted customers within a first set time period up to the current period with the predicted charging consumption feature time series of unconverted customers within a second set time period after the current period to obtain a spliced ​​charging consumption feature time series, calculate the similarity between the spliced ​​charging consumption feature time series and the charging consumption feature time series of the sum of the first set time period and the second set time period before the first non-charging consumption of the converted customers, select the unconverted customers whose similarity is greater than a set threshold as potential conversion customers, and use the corresponding similarity as the shopping probability of the potential conversion customers; The first clustering module is used to cluster the charging consumption feature time series of the potential conversion customers within the first set time period up to the current period and the charging consumption feature time series of the converted customers within the second set time period before the first non-charging consumption and the first set time period before, to obtain the first cluster center and the set number of sample points closest to the first cluster center; The second clustering module is used to cluster the predicted charging consumption feature time series of the potential conversion customers for a second set time after the current period and the charging consumption feature time series of the converted customers for a second set time before the first non-charging consumption, to obtain the second cluster center and the set number of sample points closest to the second cluster center; The third clustering module is used to cluster the concatenated charging consumption feature time series of potential conversion customers and the charging consumption feature time series of the sum of the first set time and the second set time before the first non-charging consumption of the converted customers, to obtain the third cluster center and the set number of sample points closest to the third cluster; The prediction module is used to predict the predicted purchase probability of various types of non-charging consumption of potential converted customers based on the frequencies of various types of non-charging consumption of converted customers and the distances between potential converted customers and the first cluster center, the second cluster center and the third cluster center.

9. A device for predicting charging customer conversion tendency, characterized in that: include: A memory and a processor, wherein the memory stores a program, and the processor runs the program to execute the charging customer conversion tendency prediction method according to any one of claims 1 to 7.

10. A program product, characterized in that When running on a processor, the method for predicting charging customer conversion tendency according to any one of claims 1 to 7 is executed.

Images