A railway freight service dynamic matching method and system based on customer characteristics

Abstract

The present application relates to a kind of railway freight service dynamic matching method and system based on customer characteristics, comprising the following steps: obtaining the customer data of each customer in each customer, customer index and customer traffic data;For each customer, according to customer data, determine customer classification, customer classification includes at least one of customer value, customer loyalty, customer potential value, customer transportation characteristics and customer transfer cost;For each customer, according to customer index, determine customer rating;For each customer, according to customer traffic data, determine customer loss trend;For each customer, according to the customer classification, customer rating and customer loss trend corresponding to customer, match the target freight service corresponding to customer from each freight service, each freight service includes transport capacity guarantee, freight product and value-added service, transport price strategy and transport timeliness.Provide differentiated transport capacity tilt and product service for the needs of different customers and the contribution to railway.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of railway technology, and in particular to a railway freight service dynamic matching method and system based on customer characteristics. BACKGROUND

[0002] The core of foreign railway customer management is to provide services to large customers. For example, in Germany, the United States, Canada and other countries, these countries generally follow the 80 / 20 freight marketing strategy, focusing on large customers in freight marketing work, and building modern freight owner service centers to provide differentiated and customized services according to customer information service needs. After the overall upgrade of China Railway 95306, railway business activities are basically handled online, which involves major changes in railway marketing system and customer service methods, and strengthens the organic connection between marketing services and transport resources. However, with the deepening of transport structure adjustment, customer demand is becoming more diversified, and railway freight services still have the following problems: the customer demand is not clear, the customer classification management method is not perfect, it is difficult to provide targeted customer exclusive services, it is difficult to provide balanced transport resources during the period of transport capacity shortage, there is a risk of customer loss, and the like. It is urgent to provide fine customer rating management and differentiated transport inclination and product services according to the demand characteristics of different customers and the contribution to the railway. SUMMARY

[0003] In order to overcome the problem of providing fine customer rating management and differentiated transport inclination and product services according to the demand of different customers and the contribution to the railway, the present application provides a railway freight service dynamic matching method and system based on customer characteristics.

[0004] In a first aspect, in order to solve the above technical problems, the present application provides a railway freight service dynamic matching method based on customer characteristics, comprising the following steps:

[0005] Obtain customer data, customer indicators and customer volume data of each customer in each customer;

[0006] For each customer, determine the customer classification according to the customer data, the customer classification including at least one of customer value, customer loyalty, customer potential value, customer transportation characteristics and customer transfer cost;

[0007] For each customer, determine the customer rating according to the customer indicators, the customer rating representing the contribution, loyalty and potential value of the customer to the railway;

[0008] For each customer, determine the customer loss trend according to the customer volume data;

[0009] For each customer, according to the customer classification corresponding to the customer, the customer rating and the customer loss trend, the target freight service corresponding to the customer is matched from various freight services, and the various freight services include capacity guarantee, freight product and value-added service, freight price strategy and transportation time limit.

[0010] The beneficial effect of the railway freight service dynamic matching method based on customer characteristics provided by the application is that for each customer, the customer classification corresponding to the customer, the customer rating and the customer loss trend are obtained according to the obtained customer data, customer indicators and customer traffic data, and then the target freight service is matched according to the customer classification, the customer rating and the customer loss trend of different customers, thereby solving the problem of providing fine customer rating management and differentiated capacity inclination and product service according to the needs (customer classification) of different customers and the contribution degree (customer rating and customer loss trend) to the railway.

[0011] On the basis of the above technical solution, the railway freight service dynamic matching method based on customer characteristics can be further improved as follows.

[0012] Further, the customer data includes at least one of customer sending volume, customer income, customer off-season traffic volume, railway transportation share, order realization rate, customer capacity, customer revenue, railway freight rate, railway freight rate fluctuation, railway equipment cost, potential investment cost, opportunity cost caused by transfer and railway service life;

[0013] If the customer data includes customer sending volume and customer income, for each customer, the customer classification is determined according to the customer data, including:

[0014] For each customer, the customer value is determined according to the customer sending volume and the customer income;

[0015] If the customer data includes customer off-season traffic volume, railway transportation share and order realization rate, for each customer, the customer classification is determined according to the customer data, including:

[0016] For each customer, the customer loyalty is determined according to the customer off-season traffic volume, the railway transportation share and the order realization rate;

[0017] If the customer data includes customer capacity, customer revenue and customer sending volume, for each customer, the customer classification is determined according to the customer data, including:

[0018] For each customer, the customer potential value is determined according to the customer capacity, the customer revenue and the customer sending volume;

[0019] If the customer data includes railway freight rate fluctuation, for each customer, the customer classification is determined according to the customer data, including:

[0020] determining, for each customer, customer shipping characteristics based on fluctuations in railroad rates;

[0021] If the customer data includes railroad equipment costs, potential investment costs, opportunity costs due to diversion, and railroad useful life, then for each customer, determining, based on the customer data, a customer classification including:

[0022] For each customer, determining a customer diversion cost based on railroad equipment costs, potential investment costs, opportunity costs due to diversion, and railroad useful life.

[0023] The beneficial effect of the above further scheme is that according to different customer data, the corresponding customer classification of the customer can be determined, and the customer is classified from the customer value, the customer loyalty, the customer potential value, the customer shipping characteristics and the customer diversion cost.

[0024] Further, the above for each customer, determining a customer rating based on customer indicators, includes:

[0025] For each customer, determining a weight vector based on each customer indicator, each element in the weight vector corresponding to a customer indicator, the weight representing the importance of the customer indicator;

[0026] For each customer, obtaining a plurality of evaluation levels of the customer from a plurality of evaluators to obtain an evaluation level set;

[0027] For each customer, determining a fuzzy relationship matrix based on each customer indicator and the evaluation level set, each element in the fuzzy relationship matrix representing the membership degree of each customer indicator and each evaluation level, the membership degree representing the membership relationship between the evaluation level and the customer indicator;

[0028] For each customer, determining a customer rating based on the weight vector and the fuzzy relationship matrix.

[0029] The beneficial effect of the above further scheme is that according to the customer indicators, the weight vector is obtained, so that the weight corresponding to each customer indicator is obtained, and then according to the customer indicators and the evaluation level set, the fuzzy relationship matrix between each customer indicator and each evaluation level is obtained, so that the membership relationship between each customer indicator and the evaluation levels of each evaluator is obtained, and finally the final customer rating is obtained according to the weight vector and the fuzzy relationship matrix.

[0030] Further, the above customer shipping volume data includes customer projected shipment volume in a predetermined period, shipment frequency in a predetermined period, last shipment time in a predetermined period, shipment volume of other transportation modes except for freight transportation in a predetermined period, and railroad monitored shipment volume in a predetermined period;

[0031] For each customer, according to the customer traffic data, the customer churn trend is determined, including:

[0032] For each customer, according to the customer expected sending volume in the preset period, the sending frequency in the preset period, the last sending time in the preset period, the sending volume of other transportation modes except freight transportation in the preset period, and the railway monitoring volume in the preset period, the customer churn trend is determined.

[0033] The beneficial effect of the above further scheme is that by comparing the customer expected sending volume in the preset period, the sending frequency in the preset period, the last sending time in the preset period, the sending volume of other transportation modes except freight transportation in the preset period, and the railway monitoring volume in the preset period, it can be determined whether the customer is lost, thereby determining the customer churn trend.

[0034] Further, for each customer, according to the customer classification, the customer rating, and the customer churn trend, the target freight transportation service corresponding to the customer is matched from the various freight transportation services, including:

[0035] For each customer, according to the customer classification, the customer rating, and the customer churn trend, and the various freight transportation services, a first score corresponding to each freight transportation service is determined by a first matching model, the first score representing the recommendation degree of each freight transportation service, the first matching model including a first feature extraction module, a second feature extraction module, and a scoring module, the first feature extraction module including a convolutional neural network and a recurrent neural network, and the second feature extraction module including a convolutional neural network and a recurrent neural network.

[0036] For each customer, the customer classification, the customer rating, and the customer churn trend are input into the first feature extraction module to extract the features of the customer classification, the customer rating, and the customer churn trend respectively, and a customer dynamic feature vector is determined.

[0037] For each customer, the various freight transportation services are input into the second feature extraction module to extract the features of the various freight transportation services, and a freight transportation service feature vector is obtained.

[0038] For each customer, the customer dynamic feature vector and the freight transportation service feature vector are input into the scoring module to obtain a predicted score corresponding to each freight transportation service.

[0039] For each customer, according to the customer classification, the customer rating, the customer churn trend, and the various freight transportation services, a second score corresponding to each freight transportation service is determined by a second matching model, the second score representing the recommendation degree of each freight transportation service, and the second matching model being a deep factorization machine.

[0040] For each customer, according to the various first scores and the various second scores, a target freight transportation service corresponding to the customer is determined.

[0041] The beneficial effect of the above further scheme is that by constructing the first matching model, the customer classification, customer rating and customer churn trend can be matched with each freight service to obtain a first score corresponding to each freight service, by constructing the second matching model, the relationship between the customer classification, customer rating and customer churn trend and each freight service can be obtained to obtain a second score corresponding to each freight service, and finally the target freight service is determined according to the first score and the second score, so as to realize differentiated transport inclination and product service according to the needs of different customers and the contribution to the railway.

[0042] Further, for each customer, the customer classification, customer rating and customer churn trend are input into the first feature extraction module to extract the features of the customer classification, customer rating and customer churn trend respectively to determine the customer dynamic feature vector, including:

[0043] For each customer, the customer classification, customer rating and customer churn trend are input into the embedding layer for preprocessing to obtain a first preprocessing result, and then the preprocessing result is input into the recurrent neural network to extract features by circulation to obtain a first recurrent feature vector, and then the first recurrent feature vector is input into the convolution layer and the pooling layer in sequence for convolution operation and pooling operation respectively to obtain a first feature extraction vector, and then the first feature extraction vector is input into the full connection layer to obtain the customer dynamic feature vector.

[0044] For each customer, each freight service is input into the second feature extraction module to extract the features of each freight service to obtain a freight service feature vector, including:

[0045] For each customer, each freight service is input into the embedding layer for preprocessing to obtain a second preprocessing result, and then the second preprocessing result is input into the recurrent neural network to extract features by circulation to obtain a second recurrent feature vector, and then the second recurrent feature vector is input into the convolution layer and the pooling layer in sequence for convolution operation and pooling operation respectively to obtain a second feature extraction vector, and then the second feature extraction vector is input into the full connection layer to obtain the freight service feature vector.

[0046] The beneficial effect of the above further scheme is that by combining the recurrent neural network and the deep neural network, the customer dynamic feature vector can be obtained through the customer classification, customer rating and customer churn trend, and the freight service feature vector can be obtained through each freight service.

[0047] Further, for each customer, the customer dynamic feature vector and the freight service feature vector are input into the scoring module to obtain a predicted score corresponding to each freight service, including:

[0048] For each customer, the customer dynamic feature vector and the freight service feature vector are input into a corresponding first formula of the scoring module to obtain a predicted score corresponding to each freight service, wherein the first formula is:

[0049] Px=SoftmaxλCustomerZc+λFreightZfx+ξ;

[0050] wherein Px represents the predicted score corresponding to the xth freight service, ξ represents an offset, λCustomer represents a preset weight corresponding to the first feature extraction module, Zc is the customer dynamic feature vector, λFreight represents a preset weight corresponding to the second feature extraction module, Zf is the freight service feature vector, and Softmax represents an activation function.

[0051] The beneficial effects of the above further scheme are that the elements in the customer dynamic feature vector are matched with the elements in the freight service feature vector by the first formula to obtain the predicted score corresponding to each freight service.

[0052] In a second aspect, the present application provides a railway freight service dynamic matching system based on customer features, comprising:

[0053] A first data acquisition module is configured to acquire customer data, customer indicators and customer traffic volume data of each customer in each customer;

[0054] A customer classification module is configured to determine a customer classification for each customer according to the customer data, wherein the customer classification includes at least one of customer value, customer loyalty, customer potential value, customer transportation characteristics and customer transfer cost;

[0055] A customer rating module is configured to determine a customer rating for each customer according to the customer indicators, wherein the customer rating represents the contribution, loyalty and potential value of the customer to the railway;

[0056] A customer loss trend module is configured to determine a customer loss trend for each customer according to the customer traffic volume data;

[0057] A matching module is configured to match a target freight service corresponding to each customer from each freight service according to the customer classification, customer rating and customer loss trend corresponding to the customer, wherein each freight service includes capacity guarantee, freight product and value-added service, freight pricing strategy and transportation timeliness.

[0058] In a third aspect, the present application further provides an electronic device comprising a memory, a processor and a program stored in the memory and running on the processor, wherein the processor implements the steps of the railway freight service dynamic matching method based on customer features as described above when executing the program.

[0059] Fourthly, the present invention also provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform steps of a dynamic matching method for railway freight services based on customer characteristics. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0061] Figure 1 This is a flowchart illustrating a dynamic matching method for railway freight services based on customer characteristics, according to an embodiment of the present invention.

[0062] Figure 2 A customer indicator system diagram constructed for embodiments of the present invention;

[0063] Figure 3 This is a schematic diagram of the structure of a dynamic matching system for railway freight services based on customer characteristics, according to an embodiment of the present invention. Detailed Implementation

[0064] The following embodiments are further explanations and supplements to the present invention and do not constitute any limitation on the present invention.

[0065] The following describes, with reference to the accompanying drawings, an embodiment of the present invention of a dynamic matching method and system for railway freight services based on customer characteristics.

[0066] This invention discloses a dynamic matching method for railway freight services based on customer characteristics. The method is applied to a terminal device. In this application, the terminal device is used as the execution subject to describe the solution. The terminal device can be a computer, server, etc., and is used to execute the steps of a dynamic matching method for railway freight services based on customer characteristics. The terminal device is also connected to a database, which is used to store customer data, customer indicators, customer freight volume data, and freight services.

[0067] like Figure 1 As shown, the present invention provides a dynamic matching method for railway freight services based on customer characteristics, comprising the following steps:

[0068] S1. Obtain customer data, customer metrics, and customer traffic data for each customer in each customer group;

[0069] S2. For each customer, based on customer data, determine the customer category, which includes at least one of customer value, customer loyalty, customer potential value, customer transportation characteristics, and customer switching costs.

[0070] S3, for each customer, determining a customer rating according to the customer index, the customer rating representing the contribution, loyalty and potential value of the customer to the railway;

[0071] S4, for each customer, determining a customer churn trend according to the customer volume data;

[0072] S5, for each customer, matching a target freight service corresponding to the customer from each freight service according to the customer classification, the customer rating and the customer churn trend corresponding to the customer, each freight service including capacity guarantee, freight product and value-added service, freight pricing strategy and transportation timeliness.

[0073] Optionally, the customer data includes at least one of customer sending volume, customer income, customer off-season volume, railway transportation share, order realization rate, customer capacity, customer revenue, railway freight rate, railway freight rate fluctuation, railway equipment cost and railway service life;

[0074] If the customer data includes customer sending volume and customer income, for each customer, determining the customer classification according to the customer data, including:

[0075] For each customer, determining the customer value according to the customer sending volume and the customer income;

[0076] If the customer data includes customer off-season volume, railway transportation share and order realization rate, for each customer, determining the customer classification according to the customer data, including:

[0077] For each customer, determining the customer loyalty according to the customer off-season volume, the railway transportation share and the order realization rate;

[0078] If the customer data includes customer capacity, customer revenue and customer sending volume, for each customer, determining the customer classification according to the customer data, including:

[0079] For each customer, determining the customer potential value according to the customer capacity, the customer revenue and the customer sending volume;

[0080] If the customer data includes railway freight rate fluctuation, for each customer, determining the customer classification according to the customer data, including:

[0081] For each customer, determining the customer transportation characteristics according to the railway freight rate fluctuation;

[0082] If the customer data includes railway equipment cost, potential investment cost, opportunity cost caused by transfer and railway service life, for each customer, determining the customer classification according to the customer data, including:

[0083] For each customer, the customer transfer cost is determined according to the railway equipment cost, potential investment cost, opportunity cost caused by transfer and railway service life.

[0084] Optionally, the meanings of the respective data in the customer data are as follows:

[0085] The customer sending volume refers to the total amount of products sent by the customer through the railway freight service.

[0086] The customer income refers to the profit of the customer.

[0087] The customer off-season volume refers to the total amount of products sent by the customer through the railway freight service in the off-season of freight transportation.

[0088] The railway transportation share refers to the proportion of the railway transportation mode in all transportation modes used by the customer.

[0089] The order fulfillment rate refers to the proportion of the delivered orders in the total orders of the railway freight orders of the customer.

[0090] The customer capacity refers to the production capacity of the products of the customer.

[0091] The customer operating revenue refers to the operating revenue of the customer.

[0092] The railway freight price fluctuation refers to the upward or downward fluctuation proportion of the transportation price of the railway freight service transacted by the customer in any preset period compared with that in the previous period.

[0093] The railway equipment cost refers to the equipment investment cost in the process of railway operation.

[0094] The railway service life refers to the life cycle of the railway.

[0095] In this embodiment, the customers ranked in the top x% in the customer value are high-potential customers, and the customers ranked after the top x% in the customer value are general-potential customers, and the value of x can be adjusted according to actual needs.

[0096] The customers whose railway freight price upward adjustment rate is within y1% and whose railway freight volume downward adjustment rate is more than z1% in the customer transportation characteristics are economic customers, and the customers whose railway freight price upward adjustment rate is more than y2% and whose railway freight volume downward adjustment rate is within z2% in the customer transportation characteristics are investment customers, and the values of y1, y2, z1 and z2 can be adjusted according to actual needs.

[0097] The customers with a high ratio of railway equipment cost to railway service life are high-transfer-cost customers, and the customers with a low ratio of railway equipment cost to railway service life or a ratio of 0 are low-transfer-cost customers.

[0098] Optionally, the above determining the customer rating according to the customer indicators for each customer comprises:

[0099] For each customer, a weight vector is determined based on various customer metrics. Each element in the weight vector corresponds to a customer metric, and the weight represents the importance of the customer metric.

[0100] For each customer, obtain the ratings from multiple evaluators to get a set of ratings;

[0101] For each customer, a fuzzy relation matrix is ​​determined based on the set of customer indicators and evaluation levels. Each element in the fuzzy relation matrix represents the membership degree between each customer indicator and each evaluation level. The membership degree represents the membership relationship between the evaluation level and the customer indicator.

[0102] For each customer, a customer rating is determined based on the weight vector and the fuzzy relation matrix.

[0103] Optional, such as Figure 2 As shown, customer metrics are as follows Figure 2 As shown, customer metrics are constructed from three main aspects: contribution, loyalty, and potential value. Contribution is further divided into metric contribution and revenue contribution. Metric contribution includes freight volume, number of wagons loaded, and turnover. Revenue contribution includes freight revenue and other revenue. Loyalty is divided into fidelity, integrity, and satisfaction. Fidelity includes the most recent shipment date, shipment frequency, railway transport share, and off-season support rate. Integrity includes plan fulfillment rate and number of cancellations. Satisfaction includes complaint and suggestion rate. Potential value is divided into expected revenue, potential revenue, and customer size. Expected revenue includes predicted value-added services for the next stage. Potential revenue includes potential value-added services. Customer size includes adjustments to industry planning, changes in production capacity layout, and customer brand image.

[0104] Optionally, once customer metrics are built, Figure 2 The most basic items, such as freight volume and number of truckloads, are used as customer metrics. Customer metrics are represented by a set U = (u1, u2, ..., u...). m Let ) represent, where u i Let m represent the i-th customer metric, and m represent the total number of customer metrics.

[0105] In addition, we collect ratings from multiple evaluators for the customer to obtain a set of ratings, denoted as P = (p1, p2, ..., p...). n ) indicates that, where p i Let represent the rating of the customer by the i-th evaluator, and n represent the total number of evaluators.

[0106] In addition, based on the various customer metrics and rating levels, a fuzzy relation matrix is ​​determined, which is represented as follows:

[0107]

[0108] wherein R represents a fuzzy relation matrix, each element in each row of the fuzzy relation matrix represents a membership degree of each customer index to each evaluation grade, for example, r nm represents a membership degree of the r m th customer index to the n th evaluation grade, n represents a total number of evaluation grades, and m represents a total number of evaluation indexes.

[0109] In addition, for the weight of each customer index, an expert scoring method is used to calculate, each expert scores each customer index, and a weighted average method is used to calculate the weight of each customer index, so as to obtain a weight vector, which is represented by ω , wherein ω i represents the weight of the i th customer index, and m represents a total number of customer indexes.

[0110] In addition, according to the weight vector and the fuzzy relation matrix, a comprehensive evaluation set of each customer is obtained through fuzzy composition operation, wherein the fuzzy composition operation is:

[0111]

[0112] wherein h represents the comprehensive evaluation set of the customer, and h i represents the i th comprehensive evaluation.

[0113] In addition, according to the comprehensive evaluation set of the customer, the maximum value in the comprehensive evaluation set is selected as the customer rating of the customer.

[0114] Optionally, the customer traffic data includes customer expected sending amount in a preset period, sending frequency in the preset period, last sending time in the preset period, sending amount of other transportation modes except for freight transportation in the preset period, and railway monitoring traffic amount in the preset period.

[0115] For each customer, a customer churn trend is determined according to the customer traffic data, including:

[0116] For each customer, a customer churn trend is determined according to customer expected sending amount in a preset period, sending frequency in the preset period, last sending time in the preset period, sending amount of other transportation modes except for freight transportation in the preset period, and railway monitoring traffic amount in the preset period.

[0117] Optionally, the preset period can be a day, a decade, a month, a season, or a year, wherein the customer expected sending amount refers to the total amount of freight transportation expected in the current preset period according to the actual traffic amount of the customer in the last preset period or a specified period, and the railway monitoring traffic amount refers to the actual traffic amount of the customer monitored in the current preset period.

[0118] Optionally, as shown in the following Table 1 (all are data in the current preset period):

[0119]

[0120] In Table 1, X represents the expected sending amount of the customer, Y represents the railway monitoring volume, the absolute value is |Y-X|, and the percentage is (Y-X) / Y. According to the triggering condition, when Y-X≥x1 or (Y-X) / Y≥x2%, it is a primary early warning, indicating that the customer has a loss trend, but it is not obvious. When Y-X≥y1 or (Y-X) / Y≥y2% or Y decreases continuously for two years, it is a moderate early warning, indicating that the customer loss trend is strengthened. When Y-X≥z1 or (Y-X) / Y≥z2% or Y decreases continuously for more than three years, it is a serious early warning, indicating that the customer loss trend is serious. The values of x1, x2, y1, y2, z1 and z2 can be adjusted according to actual needs.

[0121] Optionally, for each customer, according to the customer classification, the customer rating and the customer loss trend, a target freight service corresponding to the customer is matched from each freight service, including:

[0122] For each customer, according to the customer classification, the customer rating and the customer loss trend and each freight service, a first score corresponding to each freight service is determined through a first matching model, the first score representing a recommendation degree of each freight service. The first matching model includes a first feature extraction module, a second feature extraction module and a scoring module. The first feature extraction module includes a convolutional neural network and a recurrent neural network. The second feature extraction module includes a convolutional neural network and a recurrent neural network.

[0123] For each customer, the customer classification, the customer rating and the customer loss trend are input into the first feature extraction module to extract features of the customer classification, the customer rating and the customer loss trend respectively, and a customer dynamic feature vector is determined.

[0124] For each customer, each freight service is input into the second feature extraction module to extract features of each freight service, and a freight service feature vector is obtained.

[0125] For each customer, the customer dynamic feature vector and the freight service feature vector are input into the scoring module to obtain a predicted score corresponding to each freight service.

[0126] For each customer, according to the customer classification, the customer rating, the customer loss trend and each freight service, a second score corresponding to each freight service is determined through a second matching model, the second score representing a recommendation degree of each freight service. The second matching model is a deep factorization machine.

[0127] For each customer, a target freight service corresponding to the customer is determined according to the respective first score and the respective second score.

[0128] Optionally, the deep factorization machine obtains a feature relationship among the customers, the customer labels (customer classification, customer rating and customer churn trend) and the freight services through collaborative filtering, and then matches the customer labels with the freight services and predicts the respective second scores of the freight services through matching.

[0129] Optionally, for each customer, the customer classification, the customer rating and the customer churn trend are input into a first feature extraction module to extract features of the customer classification, the customer rating and the customer churn trend respectively, and to determine a customer dynamic feature vector, including:

[0130] For each customer, the customer classification, the customer rating and the customer churn trend are input into an embedding layer for preprocessing to obtain a first preprocessing result, and then the preprocessing result is input into a recurrent neural network to extract features through a loop to obtain a first loop feature vector, the first loop feature vector is sequentially input into a convolution layer and a pooling layer for convolution operation and pooling operation respectively to obtain a first feature extraction vector, and the first feature extraction vector is input into a full connection layer to obtain the customer dynamic feature vector.

[0131] For each customer, each freight service is input into a second feature extraction module to extract features of each freight service to obtain a freight service feature vector, including:

[0132] For each customer, each freight service is input into an embedding layer for preprocessing to obtain a second preprocessing result, and then the second preprocessing result is input into a recurrent neural network to extract features through a loop to obtain a second loop feature vector, the second loop feature vector is sequentially input into a convolution layer and a pooling layer for convolution operation and pooling operation respectively to obtain a second feature extraction vector, and the second feature extraction vector is input into a full connection layer to obtain the freight service feature vector.

[0133] Optionally, the first feature extraction module and the second feature extraction module both include a convolutional neural network and a recurrent neural network, wherein the convolutional neural network includes an embedding layer, a convolution layer, a pooling layer and a full connection layer connected in sequence, wherein the first layer is the embedding layer for receiving data corresponding to the customer classification, the customer rating, the customer churn trend and the freight service in the form of an embedding matrix composed of word vectors, the second layer is the convolution layer using a feature extractor to perform convolution operation on the embedding matrix to obtain respective feature values corresponding to the embedding matrix, the third layer is the pooling layer for extracting the feature values output by each feature extractor in the convolution layer, and the fourth layer is the full connection layer for outputting the respective feature values as customer dynamic features or freight service features.

[0134] Optionally, after the embedding layer, the data in the received embedding matrix is processed for features using a recurrent neural network, the neuron connection mode of the neural network is full connection, and the data in the embedding layer is input into the full connection layer of the recurrent neural network to start network training. For the features of the freight service, the obtained training features are returned to the first full connection layer for recurrent training to obtain the freight service features. For the features of customer classification, customer rating and customer loss trend, the features are trained through two full connection layers, and the output values of the full connection layers are returned to the first full connection layer for recurrent training to extract the customer dynamic features.

[0135] Optionally, for each customer, the customer dynamic feature vector and the freight service feature vector are input into a scoring module to obtain a predicted score corresponding to each freight service, including:

[0136] For each customer, the customer dynamic feature vector and the freight service feature vector are input into a first formula corresponding to the scoring module to obtain a predicted score corresponding to each freight service, wherein the first formula is:

[0137] Px=SoftmaxλCustomerZc+λFreightZfx+ξ;

[0138] wherein Px represents the predicted score corresponding to the xth freight service, ξ represents an offset, λCustomer represents a preset weight corresponding to the first feature extraction module, Zc is the customer dynamic feature vector, λFreight represents a preset weight corresponding to the second feature extraction module, Zf is the freight service feature vector, and Softmax represents an activation function.

[0139] Optionally, for each customer, when the first score and the second score of each customer are determined, the target freight service corresponding to the customer is determined by using a logistic regression method, i.e., the first score and the second score of the same freight service are added, and then the logistic regression function is used to convert the added scores into a probability value, so as to obtain the final predicted score value of each freight service. The freight service corresponding to the maximum value in the predicted score value is selected as the target freight service.

[0140] Optionally, the target freight service includes capacity guarantee, freight product and value-added service, freight pricing strategy and transportation timeliness, wherein:

[0141] The capacity guarantee refers to the capacity guarantee that the railway freight service can provide for the customer, and the capacity guarantee from low to high is A-level guarantee, B-level guarantee, C-level guarantee and D-level guarantee.

[0142] The freight product and value-added service refers to the freight product category that the railway freight service can provide for the customer, including express train, fast train, ordinary fast train and ordinary train, and the value-added service at each level provided.

[0143] The freight pricing strategy refers to the freight price range that the railway freight service can provide for the customer, including no freight reduction, freight reduction of 0-x1%, freight reduction of x1-x2%, and freight reduction of x2% or more above, and the values of x1 and x2 can be adjusted according to actual conditions.

[0144] The transport time efficiency refers to the priority transport that the railway freight service can provide for the customer, from low to high, including A-level time efficiency, B-level time efficiency, C-level time efficiency and D-level time efficiency.

[0145] As shown in Figure 3 The embodiment of the present application also provides a railway freight service dynamic matching system based on customer characteristics, which comprises:

[0146] A first data acquisition module 201 is configured to acquire customer data, customer indicators and customer traffic volume data of each customer in each customer.

[0147] A customer classification module 202 is configured to determine customer classification according to the customer data for each customer, and the customer classification comprises at least one of customer value, customer loyalty, customer potential value, customer transport characteristics and customer transfer cost.

[0148] A customer rating module 203 is configured to determine customer rating according to the customer indicators for each customer, and the customer rating represents the contribution, loyalty and potential value of the customer to the railway.

[0149] A customer loss trend module 204 is configured to determine customer loss trend according to the customer traffic volume data for each customer.

[0150] A matching module 205 is configured to match the target freight service corresponding to each customer from each freight service according to the customer classification, customer rating and customer loss trend corresponding to the customer, and each freight service comprises transport capacity guarantee, freight product and value-added service, freight pricing strategy and transport time efficiency.

[0151] Optionally, the system further comprises:

[0152] The second data acquisition module is configured to acquire at least one of customer sending volume, customer income, customer off-season traffic volume, railway transport share, order realization rate, customer capacity, customer revenue, railway freight, railway freight fluctuation, railway equipment cost, potential investment cost, opportunity cost caused by transfer and railway service life.

[0153] The customer classification module 202 is specifically configured to:

[0154] For each customer, determine customer value according to customer shipment volume and customer revenue;

[0155] If the customer data includes customer off-season volume, railway transportation share and order fulfillment rate, for each customer, determine customer classification according to the customer data, including:

[0156] For each customer, determine customer loyalty according to customer off-season volume, railway transportation share and order fulfillment rate;

[0157] For each customer, determine customer potential value according to customer capacity, customer revenue and customer shipment volume;

[0158] For each customer, determine customer transportation characteristics according to railway price fluctuation;

[0159] For each customer, determine customer transfer cost according to railway equipment cost, potential investment cost, opportunity cost caused by transfer and railway service life.

[0160] Optionally, the customer rating module 203 comprises:

[0161] A weight vector module, configured to determine, for each customer, a weight vector according to the customer indicators, each element in the weight vector corresponding to a customer indicator, the weight representing the importance of the customer indicator;

[0162] An evaluation grade set module, configured to obtain, for each customer, evaluation grades of the customer from multiple evaluators, to obtain an evaluation grade set;

[0163] A fuzzy relation matrix module, configured to determine, for each customer, a fuzzy relation matrix according to the customer indicators and the evaluation grade set, each element in each row in the fuzzy relation matrix representing a membership degree of each customer indicator and each evaluation grade, the membership degree representing a membership relationship between the evaluation grade and the customer indicator;

[0164] A determination module, configured to determine, for each customer, a customer rating according to the weight vector and the fuzzy relation matrix.

[0165] Optionally, the system further comprises:

[0166] A third data acquisition module, configured to customer shipment data includes customer estimated shipment volume in a preset period, shipment frequency in the preset period, last shipment time in the preset period, shipment volume of other transportation modes except for freight transportation in the preset period and railway monitored shipment volume in the preset period;

[0167] The customer loss trend module 204 is specifically configured to:

[0168] For each customer, according to the customer's expected sending volume in a preset period, the sending frequency in a preset period, the last sending time in a preset period, the sending volume of other transportation modes except freight transportation in a preset period and the railway monitoring volume in a preset period, the customer loss trend is determined.

[0169] Optionally, the matching module 205 is specifically used for:

[0170] For each customer, according to the customer classification, the customer rating and the customer loss trend and each freight service, a first score corresponding to each freight service is determined through a first matching model, the first score representing the recommendation degree of each freight service, the first matching module including a first feature extraction module, a second feature extraction module and a scoring module, the first feature extraction module including a convolutional neural network and a recurrent neural network, and the second feature extraction module including a convolutional neural network and a recurrent neural network.

[0171] For each customer, the customer classification, the customer rating and the customer loss trend are input into the first feature extraction module to extract the features of the customer classification, the customer rating and the customer loss trend respectively, and a customer dynamic feature vector is determined.

[0172] For each customer, each freight service is input into the second feature extraction module to extract the features of each freight service, and a freight service feature vector is obtained.

[0173] For each customer, the customer dynamic feature vector and the freight service feature vector are input into the scoring module to obtain a predicted score corresponding to each freight service.

[0174] For each customer, according to the customer classification, the customer rating, the customer loss trend and each freight service, a second score corresponding to each freight service is determined through a second matching model, the second score representing the recommendation degree of each freight service, and the second matching module being a deep factorization machine.

[0175] For each customer, according to each first score and each second score, a target freight service corresponding to the customer is determined.

[0176] Optionally, the matching module 205 is specifically used for:

[0177] For each customer, the customer classification, the customer rating and the customer loss trend are input into an embedding layer to preprocess the customer classification, the customer rating and the customer loss trend, and a first preprocessing result is obtained, and then the preprocessing result is input into a recurrent neural network to extract features through a loop to obtain a first recurrent feature vector, the first recurrent feature vector being input into a convolution layer and a pooling layer in sequence to perform convolution operation and pooling operation respectively, and a first feature extraction vector is obtained, and the first feature extraction vector is input into a fully connected layer to obtain a customer dynamic feature vector.

[0178] For each customer, the respective freight service input is embedded into the embedding layer, the respective freight service is preprocessed to obtain a second preprocessing result, the second preprocessing result is input into the recurrent neural network, the features are extracted through the recurrence to obtain a second recurrent feature vector, the second recurrent feature vector is sequentially input into the convolution layer and the pooling layer, the convolution operation and the pooling operation are performed respectively to obtain a second feature extraction vector, and the second feature extraction vector is input into the full connection layer to obtain a freight service feature vector.

[0179] Optionally, the prediction score module is specifically used for:

[0180] For each customer, the customer dynamic feature vector and the freight service feature vector are input into the first formula corresponding to the score module to obtain a prediction score corresponding to each freight service, wherein the first formula is:

[0181] Px = Softmax (lambda Customer Zc + lambda Freight Zf + xi) ;

[0182] wherein, Px represents the prediction score corresponding to the xth freight service, xi represents an offset, lambda Customer represents a preset weight corresponding to the first feature extraction module, Zc is the customer dynamic feature vector, lambda Freight represents a preset weight corresponding to the second feature extraction module, Zf is the freight service feature vector, and Softmax represents an activation function.

[0183] An electronic device according to an embodiment of the present application comprises a memory, a processor, and a program stored in the memory and running on the processor, and the processor implements part or all steps of the above-mentioned railway freight service dynamic matching method based on customer characteristics when executing the program.

[0184] Correspondingly, the program is computer software, and the parameters and steps in the above-mentioned electronic device can refer to the parameters and steps in the above-mentioned railway freight service dynamic matching method based on customer characteristics, which will not be repeated here.

[0185] Those skilled in the art know that the present application can be embodied as a system, method or computer program product. Therefore, the present disclosure can be embodied in the form of a complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, which is generally referred to as "circuitry", "module" or "system". In addition, in some embodiments, the present application can also be embodied in the form of a computer program product in one or more computer readable media, which contains computer readable program codes. The computer readable storage medium may, for example, be but not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above.

[0186] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0187] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-described embodiments within the scope of the present application.

Claims

1. A method for dynamic matching of railway freight services based on customer characteristics, characterized in that, The method comprises the following steps: obtaining customer data, customer indicators and customer volume data of each of the customers; for each of the customers, determining a customer classification according to the customer data, the customer classification comprising at least one of customer value, customer loyalty, customer potential value, customer transportation characteristics and customer transfer cost; for each of the customers, determining a customer rating according to the customer indicators, the customer rating representing the contribution, loyalty and potential value of the customer to the railway; for each of the customers, determining a customer churn trend according to the customer volume data; for each of the customers, matching a target freight service corresponding to the customer from various freight services according to the customer classification, customer rating and customer churn trend of the customer, each of the freight services comprising capacity guarantee, freight product and value-added service, freight pricing strategy and transportation timeliness; for each of the customers, the matching of the target freight service corresponding to the customer from the various freight services according to the customer classification, customer rating and customer churn trend of the customer comprises: for each of the customers, determining a first score corresponding to each of the freight services according to the customer classification, customer rating and customer churn trend and each of the freight services by a first matching model, the first score representing the recommendation degree of each of the freight services, the first matching model comprising a first feature extraction module, a second feature extraction module and a scoring module, the first feature extraction module comprising a convolutional neural network and a recurrent neural network, and the second feature extraction module comprising a convolutional neural network and a recurrent neural network; for each of the customers, inputting the customer classification, customer rating and customer churn trend into the first feature extraction module to extract features of the customer classification, customer rating and customer churn trend respectively and determine a customer dynamic feature vector; for each of the customers, inputting each of the freight services into the second feature extraction module to extract features of each of the freight services and obtain a freight service feature vector; for each of the customers, inputting the customer dynamic feature vector and the freight service feature vector into the scoring module to obtain a predicted score corresponding to each of the freight services; for each of the customers, determining a second score corresponding to each of the freight services according to the customer classification, customer rating, customer churn trend and each of the freight services by a second matching model, the second score representing the recommendation degree of each of the freight services, the second matching model being a deep factorization machine; for each of the customers, determining the target freight service corresponding to the customer according to each of the first scores and each of the second scores.

2. The method of claim 1, wherein, The customer data comprises at least one of customer sending volume, customer income, customer off-season volume, railway transportation share, order realization rate, customer capacity, customer revenue, railway freight rate, railway freight rate fluctuation, railway equipment cost, potential investment cost, opportunity cost caused by transfer and railway service life. If the customer data comprises the customer sending quantity and customer income, for each of the customers, determining a customer classification according to the customer data, comprising: For each of the customers, determining the customer value according to the customer sending quantity and the customer income; If the customer data comprises the customer slack season traffic volume, the railway transportation share and the order fulfillment rate, for each of the customers, determining a customer classification according to the customer data, comprising: For each of the customers, determining the customer loyalty according to the customer slack season traffic volume, the railway transportation share and the order fulfillment rate; If the customer data comprises the customer capacity, the customer revenue and the customer sending quantity, for each of the customers, determining a customer classification according to the customer data, comprising: For each of the customers, determining the customer potential value according to the customer capacity, the customer revenue and the customer sending quantity; If the customer data comprises the railway price fluctuation, for each of the customers, determining a customer classification according to the customer data, comprising: For each of the customers, determining the customer transportation feature according to the railway price fluctuation; If the customer data comprises the railway equipment cost, the potential investment cost, the opportunity cost caused by transfer and the railway service life, for each of the customers, determining a customer classification according to the customer data, comprising: For each of the customers, determining the customer transfer cost according to the railway equipment cost, the potential investment cost, the opportunity cost caused by transfer and the railway service life.

3. The method of claim 1, wherein, For each of the customers, determining a customer rating according to the customer indicators, comprising: For each of the customers, determining a weight vector according to each of the customer indicators, each element in the weight vector corresponding to one of the customer indicators, the weight representing the importance of the customer indicator; For each of the customers, obtaining evaluation grades of the customers from multiple evaluators to obtain an evaluation grade set; For each of the customers, determining a fuzzy relation matrix according to each of the customer indicators and the evaluation grade set, each element in the fuzzy relation matrix representing the membership degree of each of the customer indicators and each of the evaluation grades, the membership degree representing the membership relationship between the evaluation grade and the customer indicator; For each of the customers, determining the customer rating according to the weight vector and the fuzzy relation matrix.

4. The method of claim 1, wherein, The customer traffic data comprises the customer expected sending quantity in a preset period, the sending frequency in a preset period, the last sending time in a preset period, the sending quantity of other transportation modes except for freight transportation in a preset period and the railway monitored traffic volume in a preset period; For each of the customers, determining a customer loss trend according to the customer traffic data, comprising: For each of the customers, determining a customer loss trend according to the customer expected sending quantity in a preset period, the sending frequency in a preset period, the last sending time in a preset period, the sending quantity of other transportation modes except for freight transportation in a preset period and the railway monitored traffic volume in a preset period.

5. The method of claim 1, wherein, For each of the customers, the customer classification, the customer rating and the customer churn trend are input into the first feature extraction module, features of the customer classification, the customer rating and the customer churn trend are extracted respectively, and a customer dynamic feature vector is determined, including: For each of the customers, the customer classification, the customer rating and the customer churn trend are input into the embedding layer, the customer classification, the customer rating and the customer churn trend are preprocessed to obtain a first preprocessing result, and then the preprocessing result is input into the recurrent neural network to extract features through cycles to obtain a first recurrent feature vector, the first recurrent feature vector is sequentially input into the convolution layer and the pooling layer to perform convolution operation and pooling operation respectively to obtain a first feature extraction vector, and the first feature extraction vector is input into the full connection layer to obtain the customer dynamic feature vector; For each of the customers, the freight services are input into the second feature extraction module, features of the freight services are extracted to obtain a freight service feature vector, including: For each of the customers, the freight services are input into the embedding layer, the freight services are preprocessed to obtain a second preprocessing result, and then the second preprocessing result is input into the recurrent neural network to extract features through cycles to obtain a second recurrent feature vector, the second recurrent feature vector is sequentially input into the convolution layer and the pooling layer to perform convolution operation and pooling operation respectively to obtain a second feature extraction vector, and the second feature extraction vector is input into the full connection layer to obtain the freight service feature vector.

6. The method of claim 5, wherein, For each of the customers, the customer dynamic feature vector and the freight service feature vector are input into the scoring module to obtain a predicted score corresponding to each of the freight services, including: For each of the customers, the customer dynamic feature vector and the freight service feature vector are input into the scoring module corresponding to a first formula to obtain a predicted score corresponding to each of the freight services, wherein the first formula is: ； wherein, denotes the prediction score corresponding to the xth freight service, denotes the offset amount, denotes the preset weight corresponding to the first feature extraction module, is the customer dynamic feature vector, denotes the preset weight corresponding to the second feature extraction module, is the freight service feature vector, and Softmax denotes an activation function.

7. A dynamic matching system for rail freight services based on customer characteristics, characterized in that, including: The first data acquisition module is configured to acquire customer data, customer indicators and customer volume data of each of the customers; The customer classification module is configured to determine, for each of the customers, a customer classification according to the customer data, the customer classification including at least one of customer value, customer loyalty, customer potential value, customer transportation feature and customer transfer cost; The customer rating module is configured to determine, for each of the customers, a customer rating according to the customer indicators, the customer rating representing customer contribution, loyalty and potential value to the railway; The customer churn trend module is configured to determine, for each of the customers, a customer churn trend according to the customer volume data; The matching module is configured to match, according to the customer classification, the customer rating and the customer churn trend corresponding to the customer, a target freight service corresponding to the customer from the freight services, and each of the freight services includes capacity guarantee, freight product and value-added service, freight pricing strategy and transportation timeliness. The matching module is specifically configured to: For each customer, according to the customer classification, the customer rating, the customer churn trend and each freight service, a first score corresponding to each freight service is determined through a first matching model, the first score representing a recommendation degree of each freight service, the first matching model comprising a first feature extraction module, a second feature extraction module and a scoring module, the first feature extraction module comprising a convolutional neural network and a recurrent neural network, the second feature extraction module comprising a convolutional neural network and a recurrent neural network; For each customer, the customer classification, the customer rating and the customer churn trend are input into the first feature extraction module to extract features of the customer classification, the customer rating and the customer churn trend respectively, and a customer dynamic feature vector is determined; For each customer, each freight service is input into the second feature extraction module to extract features of each freight service, and a freight service feature vector is obtained; For each customer, the customer dynamic feature vector and the freight service feature vector are input into the scoring module to obtain a predicted score corresponding to each freight service; For each customer, according to the customer classification, the customer rating, the customer churn trend and each freight service, a second score corresponding to each freight service is determined through a second matching model, the second score representing a recommendation degree of each freight service, the second matching model being a deep factorization machine; For each customer, according to each first score and each second score, a target freight service corresponding to the customer is determined.

8. An electronic device comprising a memory, a processor, and a program stored on the memory and running on the processor, characterized in that, The processor implements the steps of the railway freight service dynamic matching method based on customer features according to any one of claims 1 to 6 when executing the program.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores instructions, when the instructions run on the terminal device, make the terminal device execute the steps of the railway freight service dynamic matching method based on customer features according to any one of claims 1 to 6.

Images