CN113222287A

CN113222287A - Network appointment demand prediction method, device, equipment and storage medium

Info

Publication number: CN113222287A
Application number: CN202110610938.5A
Authority: CN
Inventors: 刘杰; 王健宗; 瞿晓阳
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2021-06-01
Filing date: 2021-06-01
Publication date: 2021-08-06

Abstract

The application relates to the technical field of artificial intelligence, and discloses a method, a device, equipment and a storage medium for predicting network taxi appointment demands, wherein the method comprises the following steps: inputting the target starting point region, the target prediction time and the target starting point region characteristic data into a target network vehicle reduction demand prediction model to carry out network vehicle reduction demand prediction, wherein the target network vehicle reduction demand prediction model is a model obtained on the basis of independent cyclic neural network, linear model, full connection layer and Sigmiod activation function training; and obtaining a target network car booking demand prediction result corresponding to the target prediction time of the target starting point region output by the target network car booking demand prediction model. Therefore, network car booking demand prediction is carried out according to various factors, accuracy of the network car booking demand prediction is improved, and the technical problem that accuracy of order demand prediction is not high by directly adopting network car booking historical order data is solved.

Description

Network appointment demand prediction method, device, equipment and storage medium

Technical Field

The application relates to the technical field of artificial intelligence, in particular to a method, a device, equipment and a storage medium for predicting network appointment demands.

Background

In order to reduce waiting time of passengers and improve order taking efficiency of drivers, the network booking platform needs to predict order demands of the passengers as accurately as possible. When the order demand of the network car booking is predicted, the accuracy of directly adopting the historical order data of the network car booking to predict the order demand is not high due to the fact that factors influencing the order demand are complex and various.

Disclosure of Invention

The application mainly aims to provide a method, a device, equipment and a storage medium for predicting the order demand of a network car booking, and aims to solve the technical problem that in the prior art, when the order demand of the network car booking is predicted, the accuracy of order demand prediction directly performed by adopting network car booking historical order data is low due to the fact that factors influencing the order demand are complex and various.

In order to achieve the above object, the present application provides a method for predicting a demand for a network taxi appointment, the method comprising:

acquiring a target starting point region, target prediction time and target starting point region characteristic data of the target starting point region corresponding to the target prediction time;

inputting the target starting point region, the target prediction time and the target starting point region feature data into a target network vehicle reduction demand prediction model to perform network vehicle reduction demand prediction, wherein the target network vehicle reduction demand prediction model is a model obtained based on independent cyclic neural network, linear model, full connection layer and Sigmiod activation function training;

and acquiring a target network car-booking demand prediction result, corresponding to the target prediction time, of the target starting point region output by the target network car-booking demand prediction model.

Further, before the step of inputting the target starting point region, the target prediction time and the target starting point region characteristic data into a target network vehicle-booking requirement prediction model to predict the network vehicle-booking requirement, the method further includes:

obtaining a plurality of target training sample sets, wherein each of the plurality of target training sample sets comprises: a plurality of training samples, each training sample of the plurality of training samples comprising: the order characteristic sample data, the regional characteristic sample data and the network appointment demand calibration data, wherein all the training samples in the target training sample group are subjected to positive sequence sorting according to the time period of the order in the order characteristic sample data, the target training sample group comprises the training samples in at least one starting point region, and all the target training sample groups in the plurality of target training sample groups are sorted according to the time sequence;

sequentially obtaining one target training sample group from the plurality of target training sample groups as a target training sample group to be trained;

inputting all the order characteristic sample data and all the regional characteristic sample data of the target training sample group to be trained into an initial model to carry out network appointment demand prediction to obtain a network appointment demand sample prediction value set;

calculating loss values according to the network appointment demand sample prediction value set and all the network appointment demand calibration data corresponding to the target training sample group to be trained to obtain target loss values, updating parameters of the initial model according to the target loss values, and using the updated initial model for calculating the network appointment demand sample prediction value set next time;

repeatedly executing the step of sequentially obtaining one target training sample group from the plurality of target training sample groups as a target training sample group to be trained until a training convergence condition is reached;

and taking the initial model reaching the training convergence condition as the target net car booking demand prediction model.

Further, the step of inputting all the order feature sample data and all the regional feature sample data of the target training sample group to be trained into an initial model to perform network appointment demand prediction to obtain a network appointment demand sample prediction value set includes:

sequencing all the order feature sample data of the target training sample set to be trained according to the time period of the order and forming time sequence data to obtain order feature sample time sequence data;

forming time series data by all the regional characteristic sample data of the target training sample group to be trained according to the arrangement sequence of the order characteristic sample time series data to obtain regional characteristic sample time series data;

inputting the order feature sample time series data into an independent time feature extraction module of the initial model to extract independent time features to obtain independent time feature prediction data, wherein the independent time feature extraction module is a module obtained based on the independent recurrent neural network;

performing multi-dimensional vector splicing on the regional characteristic sample time sequence data and the independent time characteristic prediction data through a characteristic splicing module of the initial model to obtain characteristic data to be subjected to importance analysis;

performing feature importance analysis on the feature data to be subjected to the feature importance analysis through a feature resolution module of the initial model to obtain a feature data set to be predicted, wherein the feature resolution module is a module obtained based on the linear model;

and performing network car booking requirement classification prediction on the feature data set to be predicted through a network car booking requirement classification prediction module of the initial model to obtain the network car booking requirement sample prediction value set, wherein the network car booking requirement classification prediction module is a module obtained based on the full connection layer and the Sigmiod activation function.

Further, the calculation formula f of the feature resolution module is as follows:

wherein X is the characteristic data to be analyzed for importance,

is the Hadamard product operator, W_bIs the weight matrix of the feature resolution module, and σ (W) is the Sigmiod activation function of W.

Further, the formula argmin (W) for calculating the target loss value_a,W_b)f_lossComprises the following steps:

wherein, W_bIs a weight matrix of the feature resolution module, W_aIs the splicing result of the weight matrix of the independent time feature extraction module and the weight matrix of the network appointment demand classification prediction module, alpha is a constant, beta is a constant, O is the network appointment demand sample prediction value set, T is all the network appointment demand calibration data corresponding to the target training sample set to be trained, | |₁Is a norm of L1 which is,

the square calculation and the root-cutting calculation are sequentially carried out.

Further, the step of obtaining a plurality of target training sample sets includes:

acquiring a historical network car booking data set;

dividing all historical network car booking data in the historical network car booking data set according to a starting point area to obtain a plurality of historical network car booking data sub-sets;

acquiring a historical network car booking data subset from the plurality of historical network car booking data subsets as a target historical network car booking data subset;

acquiring dividing data of preset time periods, and respectively carrying out order quantity statistics on each preset time period in the dividing data of the preset time periods according to the sub-set of the car booking data of the target historical network to obtain the quantity of target orders corresponding to each preset time period corresponding to the sub-set of the car booking data of the target historical network;

acquiring historical network car booking data from the target historical network car booking data subset as network car booking data to be analyzed;

order data acquisition and feature extraction are carried out on the network car booking data to be analyzed, and order feature data to be processed corresponding to the network car booking data to be analyzed are obtained;

obtaining non-order data and extracting characteristics of the network appointment data to be analyzed to obtain regional characteristic data to be processed corresponding to the network appointment data to be analyzed;

performing training sample generation according to the order characteristic data to be processed, the area characteristic data to be processed and the target order quantity corresponding to the time period of the order of the network appointment data to be analyzed to obtain a training sample to be classified corresponding to the network appointment data to be analyzed;

repeatedly executing the historical network car booking data acquired from the target historical network car booking data subset as data of the network car booking data to be analyzed until the acquisition of the historical network car booking data in the target historical network car booking data subset is completed, and taking all the training samples to be classified as training sample sets to be classified corresponding to the target historical network car booking data subset;

repeatedly executing the step of acquiring one historical network car booking data subset from the plurality of historical network car booking data subsets as a target historical network car booking data subset until the acquisition of the historical network car booking data subset from the plurality of historical network car booking data subsets is completed;

and generating a target training sample group according to all the training sample sets to be classified to obtain the plurality of target training sample groups.

Further, the step of generating a target training sample set according to all the training sample sets to be classified to obtain the plurality of target training sample sets includes:

respectively carrying out training sample subset division on each training sample set to be classified by adopting a preset training sample subset division rule and a method of sequentially obtaining the training sample subsets according to a time sequence to obtain a plurality of training sample subsets to be processed;

performing training sample group division on the plurality of training sample subsets to be processed by adopting a preset training sample group generation rule and a method of sequentially dividing the training sample groups according to a time sequence to obtain a plurality of training sample groups to be sequenced;

and respectively carrying out positive sequence sequencing on all the training samples in each training sample group to be sequenced in the plurality of training sample groups to be sequenced according to the time period of the order to obtain the plurality of target training sample groups.

The application also provides a net car appointment demand prediction device, the device includes:

the data acquisition module is used for acquiring a target starting point region, target prediction time and target starting point region characteristic data corresponding to the target starting point region in the target prediction time;

the network car-booking demand prediction module is used for inputting the target starting point region, the target prediction time and the target starting point region characteristic data into a target network car-booking demand prediction model to predict the network car-booking demand, wherein the target network car-booking demand prediction model is a model obtained by training based on an independent cyclic neural network, a linear model, a full connection layer and a Sigmiod activation function;

and the target network car-booking requirement prediction result determining module is used for acquiring a target network car-booking requirement prediction result, corresponding to the target starting point area in the target prediction time, output by the target network car-booking requirement prediction model.

The present application further proposes a computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the steps of any of the above methods when executing the computer program.

The present application also proposes a computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, carries out the steps of the method of any of the above.

According to the network car booking demand forecasting method, the device, the equipment and the storage medium, a target network car booking demand forecasting model is obtained through training based on an independent circulation neural network, a linear model, a full connection layer and a Sigmiod activation function, and then target starting point region, target forecasting time and target starting point region feature data are input into the target network car booking demand forecasting model to conduct network car booking demand forecasting to obtain a target network car booking demand forecasting result corresponding to the target starting point region in the target forecasting time, so that network car booking demand forecasting is achieved according to various factors, accuracy of network car booking demand forecasting is improved, and the technical problem that accuracy of directly adopting network car booking historical order data to conduct order demand forecasting is not high is solved.

Drawings

Fig. 1 is a schematic flow chart illustrating a method for predicting a vehicle booking requirement of a network according to an embodiment of the present application;

fig. 2 is a schematic block diagram illustrating a structure of a network appointment demand prediction apparatus according to an embodiment of the present application;

fig. 3 is a block diagram illustrating a structure of a computer device according to an embodiment of the present application.

The implementation, functional features and advantages of the objectives of the present application will be further explained with reference to the accompanying drawings.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

In order to solve the technical problem that in the prior art, when order demand prediction of a network car booking is carried out, due to the fact that factors influencing the order demand are complex and various, accuracy of order demand prediction directly carried out by adopting network car booking historical order data is low, the network car booking demand prediction method is provided, the method is applied to the technical field of artificial intelligence, and the method is further applied to the technical field of artificial intelligence neural networks. The method is carried out by the device capable of realizing the network taxi appointment demand prediction method, and the device comprises but not limited to a terminal and a server. The terminal comprises a desktop terminal and a mobile terminal, wherein the desktop terminal comprises but is not limited to a desktop computer, an industrial personal computer and a vehicle-mounted computer, and the mobile terminal comprises but is not limited to a mobile phone, a tablet computer, a notebook computer, an intelligent watch and other wearable equipment; the server includes a high performance computer and a cluster of high performance computers.

Referring to fig. 1, an embodiment of the present application provides a network taxi appointment demand prediction method, including:

s1: acquiring a target starting point region, target prediction time and target starting point region characteristic data of the target starting point region corresponding to the target prediction time;

s2: inputting the target starting point region, the target prediction time and the target starting point region feature data into a target network vehicle reduction demand prediction model to perform network vehicle reduction demand prediction, wherein the target network vehicle reduction demand prediction model is a model obtained based on independent cyclic neural network, linear model, full connection layer and Sigmiod activation function training;

s3: and acquiring a target network car-booking demand prediction result, corresponding to the target prediction time, of the target starting point region output by the target network car-booking demand prediction model.

According to the method, the target network car booking demand prediction model is obtained through training based on the independent circulation neural network, the linear model, the full connection layer and the Sigmiod activation function, and then the target starting point region, the target prediction time and the target starting point region feature data are input into the target network car booking demand prediction model to carry out the network car booking demand prediction to obtain the target network car booking demand prediction result corresponding to the target starting point region in the target prediction time, so that the network car booking demand prediction is carried out according to various factors, the accuracy of the network car booking demand prediction is improved, and the technical problem that the accuracy of directly adopting network car booking historical order data to carry out order demand prediction is not high is solved.

For S1, the target start point region, the target prediction time, and the target start point region feature data of the target start point region corresponding to the target prediction time may be acquired from a database, the target start point region, the target prediction time, and the target start point region feature data of the target start point region corresponding to the target prediction time, which are input by the user, may be acquired, and the target start point region, the target prediction time, and the target start point region feature data of the target start point region corresponding to the target prediction time may be acquired from a third-party application system.

The target starting area is a starting area of the network appointment which needs to be predicted. It is understood that the origin area is an area corresponding to the geographic location of origin of boarding of the passengers of the network appointment. The starting area may be an administrative level area or an area divided according to the density of the stream of people.

The target prediction time is a time point when the network taxi appointment demand needs to be predicted.

The target starting point region feature data is feature data obtained by performing feature extraction on non-order data of the target starting point region at the target prediction time.

Non-order data includes, but is not limited to: weather conditions, regional congestion conditions, number of cars in a region. The number of empty vehicles in the region, namely the number of empty vehicles of the taxi appointment in the region.

And S2, simultaneously inputting the target starting point area, the target prediction time and the target starting point area characteristic data into a target network vehicle-saving demand prediction model to perform network vehicle-saving demand prediction. It will be appreciated that the net appointment demand forecast is a forecast of the quantity of demand for net appointments.

Obtaining an initial model according to the independent cyclic neural network, the linear model, the full connection layer and the Sigmiod activation function; training the initial model by adopting a plurality of target training sample groups, and taking the initial model after training as the target network car booking demand prediction model; wherein each of the plurality of target training sample sets comprises: a plurality of training samples, each training sample of the plurality of training samples comprising: the order characteristic sample data, the regional characteristic sample data and the network appointment demand calibration data are obtained, all the training samples in the target training sample set are subjected to positive sequence sorting according to the time periods of orders in the order characteristic sample data, the target training sample set comprises the training samples in at least one starting point region, and all the target training sample sets in the plurality of target training sample sets are sorted according to the time sequence.

For step S3, a network vehicle booking demand prediction result output by the target network vehicle booking demand prediction model is obtained, and the network vehicle booking demand prediction result is used as a target network vehicle booking demand prediction result corresponding to the target prediction time in the target starting point area.

That is, the target network car-booking demand prediction result is a prediction result of the demand amount of the network car booking at the target prediction time in the target starting point area.

It is understood that the target network car booking requirement prediction result can be a numerical range or a specific numerical value.

In one embodiment, before the step of inputting the target starting point region, the target prediction time, and the target starting point region characteristic data into the target network vehicle reservation demand prediction model to perform the network vehicle reservation demand prediction, the method further includes:

s21: obtaining a plurality of target training sample sets, wherein each of the plurality of target training sample sets comprises: a plurality of training samples, each training sample of the plurality of training samples comprising: the order characteristic sample data, the regional characteristic sample data and the network appointment demand calibration data, wherein all the training samples in the target training sample group are subjected to positive sequence sorting according to the time period of the order in the order characteristic sample data, the target training sample group comprises the training samples in at least one starting point region, and all the target training sample groups in the plurality of target training sample groups are sorted according to the time sequence;

s22: sequentially obtaining one target training sample group from the plurality of target training sample groups as a target training sample group to be trained;

s23: inputting all the order characteristic sample data and all the regional characteristic sample data of the target training sample group to be trained into an initial model to carry out network appointment demand prediction to obtain a network appointment demand sample prediction value set;

s24: calculating loss values according to the network appointment demand sample prediction value set and all the network appointment demand calibration data corresponding to the target training sample group to be trained to obtain target loss values, updating parameters of the initial model according to the target loss values, and using the updated initial model for calculating the network appointment demand sample prediction value set next time;

s25: repeatedly executing the step of sequentially obtaining one target training sample group from the plurality of target training sample groups as a target training sample group to be trained until a training convergence condition is reached;

s26: and taking the initial model reaching the training convergence condition as the target net car booking demand prediction model.

In this embodiment, a plurality of target training sample groups are adopted to train the initial model, and the initial model after training is used as the target network car booking demand prediction model, and the training samples include: the order characteristic sample data, the regional characteristic sample data and the network car booking requirement calibration data are used for training an initial model according to multiple factors, so that the initial model reaching the training convergence condition can forecast the network car booking requirement according to the multiple factors, and the accuracy of the network car booking requirement forecast is improved.

For S21, a plurality of target training sample sets may be obtained from the database, a plurality of target training sample sets input by the user may be obtained, or a plurality of target training sample sets may be obtained from a third-party application system.

Each training sample includes: the system comprises an order characteristic sample data, an area characteristic sample data and a network appointment demand calibration data.

In the same training sample, the regional characteristic sample data is characteristic data obtained by extracting non-order data corresponding to the time period of the order corresponding to the order characteristic sample data in the starting region corresponding to the order characteristic sample data, and the network appointment demand calibration data is the order quantity of the network appointment in the time period of the order corresponding to the order characteristic sample data in the starting region corresponding to the order characteristic sample data.

All the target training sample sets in the plurality of target training sample sets are ordered in time sequence, that is, two adjacent target training sample sets in the plurality of target training sample sets sequentially include a first target training sample set and a second target training sample set, the time period of the order of the first target training sample set in the starting point area to be processed is earlier than the time period of the order of the second target training sample set in the starting point area to be processed, and the starting point area to be processed is any one of the starting point areas.

For step S22, one target training sample set is sequentially obtained from the plurality of target training sample sets, and the obtained target training sample set is used as a target training sample set to be trained.

For S23, extracting independent time characteristics of all the order characteristic sample data of the target training sample group to be trained by adopting an initial model to obtain independent time characteristic prediction data; and carrying out multi-dimensional vector splicing on the independent time feature prediction data and all the regional feature sample data of the target training sample group to be trained to obtain feature data to be subjected to importance analysis, and then carrying out feature importance analysis and net car booking demand classification prediction on the feature data to be subjected to importance analysis to obtain the net car booking demand sample prediction value set.

And S24, inputting the net appointment demand sample prediction value set and all the net appointment demand calibration data corresponding to the target training sample group to be trained into a loss function for loss value calculation, and taking the calculated data as a target loss value.

The specific steps of updating the parameters of the initial model according to the target loss value are not described herein.

For S25, steps S22 to S25 are repeatedly performed until the training convergence condition is reached.

The training convergence condition comprises: the target loss value reaches a first convergence condition or the number of iterations reaches a second convergence condition.

The first convergence condition means that the magnitudes of target loss values calculated two adjacent times satisfy a lipschitz condition (lipschitz continuity condition).

The iteration times refer to the times that the initial model is used for calculating the network appointment demand sample prediction value set, namely, the iteration times are increased by 1 when the network appointment demand sample prediction value set is calculated once.

The second convergence condition is a specific numerical value.

For S26, when the initial model reaches the training convergence condition, meaning that the training continues not to improve the quality of the initial model, the initial model reaching the training convergence condition may be used as the target net appointment demand prediction model.

In one embodiment, the step of inputting all the order feature sample data and all the region feature sample data of the target training sample group to be trained into the initial model to perform network appointment demand prediction to obtain a network appointment demand sample prediction value set includes:

s231: sequencing all the order feature sample data of the target training sample set to be trained according to the time period of the order and forming time sequence data to obtain order feature sample time sequence data;

s232: forming time series data by all the regional characteristic sample data of the target training sample group to be trained according to the arrangement sequence of the order characteristic sample time series data to obtain regional characteristic sample time series data;

s233: inputting the order feature sample time series data into an independent time feature extraction module of the initial model to extract independent time features to obtain independent time feature prediction data, wherein the independent time feature extraction module is a module obtained based on the independent recurrent neural network;

s234: performing multi-dimensional vector splicing on the regional characteristic sample time sequence data and the independent time characteristic prediction data through a characteristic splicing module of the initial model to obtain characteristic data to be subjected to importance analysis;

s235: performing feature importance analysis on the feature data to be subjected to the feature importance analysis through a feature resolution module of the initial model to obtain a feature data set to be predicted, wherein the feature resolution module is a module obtained based on the linear model;

s236: and performing network car booking requirement classification prediction on the feature data set to be predicted through a network car booking requirement classification prediction module of the initial model to obtain the network car booking requirement sample prediction value set, wherein the network car booking requirement classification prediction module is a module obtained based on the full connection layer and the Sigmiod activation function.

In this embodiment, all the order feature sample data and all the area feature sample data of the target training sample group to be trained are input into the initial model to perform the net appointment demand prediction, and each training sample in the target training sample group to be trained includes: the order characteristic sample data, the regional characteristic sample data and the network car booking requirement calibration data are used for training an initial model according to multiple factors, so that the initial model reaching the training convergence condition can forecast the network car booking requirement according to the multiple factors, and the accuracy of the network car booking requirement forecast is improved.

For S231, sorting all the order feature sample data of the target training sample set to be trained by adopting a positive order sorting method according to the time period of the order to obtain a sorted order feature sample data set; and forming time series data by all the order feature sample data in the ordered order feature sample data set to obtain the order feature sample time series data. That is, the order feature sample time series data is sorted in the same manner as the sorted order feature sample data set.

And (3) adopting a positive sequence sorting method according to the time period of the order, namely, the time period of the order corresponding to the order characteristic sample data at the first position in the sorted order characteristic sample data set is earlier than the time period of the order corresponding to the order characteristic sample data at the second position in the sorted order characteristic sample data set, and the position sequence number of the order characteristic sample data set at the first position after sorting is smaller than the position sequence number of the order characteristic sample data set at the second position after sorting.

For step S232, according to the order feature sample time series data, all the region feature sample data of the target training sample set to be trained are combined into time series data, that is, the sequencing mode of the region feature sample time series data is the same as the sequencing mode of the order feature sample time series data.

For step S233, the order feature sample time series data is input to the independent time feature extraction module of the initial model to extract an independent time feature, and the extracted independent time feature is used as independent time feature prediction data.

For step S234, performing multidimensional vector stitching on the region feature sample time series data and the independent time feature prediction data through the feature stitching module of the initial model, and taking a multidimensional vector obtained by the stitching as feature data to be subjected to importance analysis.

And the characteristic splicing modules are spliced by adopting a Concat method.

For S235, performing feature importance analysis on the feature data to be subjected to importance analysis through the feature resolution module of the initial model, so that important features obtain higher weight at an early stage, and non-important features obtain lower weight.

Wherein the feature resolution module employs the linear model.

For step S236, performing network car booking requirement classification prediction on each to-be-predicted feature data in the to-be-predicted feature data set through a network car booking requirement classification prediction module of the initial model, and taking each predicted network car booking requirement classification result as a network car booking requirement sample prediction value; and taking all network car booking requirement sample prediction values as the network car booking requirement sample prediction value set.

That is, the net appointment demand sample predicted value is a predicted value of the demand quantity of the net appointment of one training sample.

The method for obtaining the network appointment demand classification prediction module capable of performing classification prediction based on the full connection layer and the sigmood activation function is not described herein in detail.

In one embodiment, the calculation formula f of the feature resolution module is:

wherein X is the characteristic data to be analyzed for importance,

This embodiment uses a linear model

As a feature resolution module, support is provided for important features to get higher weight in the early stage and non-important features to get lower weight.

The parameters of the feature resolution module (i.e. the weight matrix of the feature resolution module) are gradually updated during the training.

In order to ensure fairness among the features, all values of the weight matrix of the feature resolution module are initialized according to a uniform distribution.

The Hadamard product operator makes each weight separately responsible for a feature, resulting in a unique weight for each feature. After the training is finished, the importance of any feature of any area can be clear at a glance, and the feature interpretation module is facilitated.

In one embodiment, the above target loss value is calculated by the formula argmin (W)_a,W_b)f_lossComprises the following steps:

The embodiment realizes that the loss value is calculated by adopting the mean square error loss function, and provides support for updating the parameters of the initial model.

Where α is a coefficient. β is a coefficient.

The training goal of the initial model is to make

The calculated value is minimal.

Updating the parameters of the initial model, i.e. updating W, in dependence on the target loss value_bAnd W_a。

In an embodiment, the step of obtaining a plurality of target training sample sets includes:

s211: acquiring a historical network car booking data set;

s212: dividing all historical network car booking data in the historical network car booking data set according to a starting point area to obtain a plurality of historical network car booking data sub-sets;

s213: acquiring a historical network car booking data subset from the plurality of historical network car booking data subsets as a target historical network car booking data subset;

s214: acquiring dividing data of preset time periods, and respectively carrying out order quantity statistics on each preset time period in the dividing data of the preset time periods according to the sub-set of the car booking data of the target historical network to obtain the quantity of target orders corresponding to each preset time period corresponding to the sub-set of the car booking data of the target historical network;

s215: acquiring historical network car booking data from the target historical network car booking data subset as network car booking data to be analyzed;

s216: order data acquisition and feature extraction are carried out on the network car booking data to be analyzed, and order feature data to be processed corresponding to the network car booking data to be analyzed are obtained;

s217: obtaining non-order data and extracting characteristics of the network appointment data to be analyzed to obtain regional characteristic data to be processed corresponding to the network appointment data to be analyzed;

s218: performing training sample generation according to the order characteristic data to be processed, the area characteristic data to be processed and the target order quantity corresponding to the time period of the order of the network appointment data to be analyzed to obtain a training sample to be classified corresponding to the network appointment data to be analyzed;

s219: repeatedly executing the historical network car booking data acquired from the target historical network car booking data subset as data of the network car booking data to be analyzed until the acquisition of the historical network car booking data in the target historical network car booking data subset is completed, and taking all the training samples to be classified as training sample sets to be classified corresponding to the target historical network car booking data subset;

s2110: repeatedly executing the step of acquiring one historical network car booking data subset from the plurality of historical network car booking data subsets as a target historical network car booking data subset until the acquisition of the historical network car booking data subset from the plurality of historical network car booking data subsets is completed;

s2111: and generating a target training sample group according to all the training sample sets to be classified to obtain the plurality of target training sample groups.

In this embodiment, a plurality of target training sample sets are extracted according to a historical network car booking data set, and because the historical network car booking data set includes order data and non-order data, and the plurality of target training sample sets includes order characteristic data extracted from the order data and area characteristic data extracted from the non-order data, a basis is provided for training an initial model according to a plurality of factors.

For step S211, a historical network car-booking data set may be obtained from the database, a historical network car-booking data set input by the user may be obtained, or a historical network car-booking data set may be obtained from a third-party application system.

The historical network car booking data set comprises a plurality of historical network car booking data.

The historical network appointment data comprises order data and non-order data, wherein the order data comprises but is not limited to: the non-order data is the weather condition, the area congestion condition and the number of the vehicles in the area of the time period of the order data of the starting point area of the order data.

For step S212, all historical network car booking data in the historical network car booking data set are divided according to the starting point region, that is, historical network car booking data of the same starting point region are divided into the same subset, and each subset obtained through division is used as a historical network car booking data subset.

The starting point areas of all historical network car booking data in the historical network car booking data subset are the same.

The starting regions of the car appointment data subsets of different historical networks are different.

And S213, sequentially acquiring a historical network car-booking data subset from the plurality of historical network car-booking data subsets, and taking the acquired historical network car-booking data subset as a target historical network car-booking data subset.

For S214, the preset time period division data may be obtained from the database, or the preset time period division data input by the user may be obtained, or the preset time period division data may be obtained from the third-party application system.

Alternatively, the preset time period division data may be such that each hour is taken as one preset time period. It is to be understood that the preset time period may be divided into other dividing manners, and is not limited herein.

And counting the quantity of the historical network car-booking data for each preset time period in the preset time period division data according to the target historical network car-booking data subset, and taking the quantity of the historical network car-booking data for each preset time period as the quantity of the target order.

And S215, sequentially acquiring one historical network car booking data from the target historical network car booking data subset, and taking the acquired historical network car booking data as the network car booking data to be analyzed.

And S216, order data acquisition is carried out on the network car booking data to be analyzed, feature extraction is carried out on the acquired order data, and all extracted feature data are used as to-be-processed order feature data corresponding to the network car booking data to be analyzed.

And S217, acquiring non-order data of the network car booking data to be analyzed, performing feature extraction on the acquired non-order data, and taking all extracted feature data as to-be-processed regional feature data corresponding to the network car booking data to be analyzed.

For step S218, the order characteristic data to be processed is used as order characteristic sample data of the training sample to be classified corresponding to the network car booking data to be analyzed, the area characteristic data to be processed is used as area characteristic sample data of the training sample to be classified corresponding to the network car booking data to be analyzed, and the target order number corresponding to the time period in which the order of the network car booking data to be analyzed is located is used as network car booking requirement calibration data of the training sample to be classified corresponding to the network car booking data to be analyzed.

And S219, repeating the step S215 to the step S219 until the historical car booking data in the target historical car booking data subset is obtained, taking all the training samples to be classified as a set, and taking the set as the training sample set to be classified corresponding to the target historical car booking data subset.

For step 2110, repeating step 213 to step 2110 until the obtaining of the historical network appointment data subset of the plurality of historical network appointment data subsets is completed.

For S2111, generating a target training sample set according to all the training sample sets to be classified, so that all the training samples in the target training sample set are subjected to positive sequence ordering according to the time period of the order in the order feature sample data, where the target training sample set includes the training samples in at least one starting point region, and all the target training sample sets in the plurality of target training sample sets are ordered in time sequence.

In an embodiment, the step of generating a target training sample set according to all the training sample sets to be classified to obtain the plurality of target training sample sets includes:

s21111: respectively carrying out training sample subset division on each training sample set to be classified by adopting a preset training sample subset division rule and a method of sequentially obtaining the training sample subsets according to a time sequence to obtain a plurality of training sample subsets to be processed;

s21112: performing training sample group division on the plurality of training sample subsets to be processed by adopting a preset training sample group generation rule and a method of sequentially dividing the training sample groups according to a time sequence to obtain a plurality of training sample groups to be sequenced;

s21113: and respectively carrying out positive sequence sequencing on all the training samples in each training sample group to be sequenced in the plurality of training sample groups to be sequenced according to the time period of the order to obtain the plurality of target training sample groups.

In this embodiment, a target training sample group is generated according to all the training sample sets to be classified, so that all the training samples in the target training sample group are subjected to positive sequence ordering according to the time period of the order in the order feature sample data, the target training sample group includes at least one training sample in a starting point region, and all the target training sample groups in the plurality of target training sample groups are ordered in a time sequence, thereby facilitating extraction of independent time features and providing a basis for training an initial model.

For S21111, the preset training sample subset partitioning rule refers to the number of training samples in the training sample subset when the training sample subset is partitioned for each training sample set to be classified. It is understood that the number of training samples in different subsets of training samples may be the same or different.

For example, the training sample set J1 to be classified includes, in time sequence: y1, Y2, Y3, Y4, Y5, and Y6, the preset training sample subset partition rule is that 2 training samples are extracted from J1 each time, and 3 training sample subsets are obtained by a method of sequentially obtaining the training samples according to the time sequence: (Y1, Y2), (Y3, Y4), (Y5, Y6), and examples thereof are not particularly limited.

The method comprises the steps of respectively carrying out training sample subset division on each training sample set to be classified by adopting a preset training sample subset division rule and a method of sequentially obtaining the training sample subsets according to time sequence, taking each training sample subset obtained by division as a training sample subset to be processed, and taking all training sample subsets to be processed as the plurality of training sample subsets to be processed.

For S21112, the preset training sample group generation rule refers to the number of extracted starting point regions when the to-be-processed training sample subset is extracted from the to-be-processed training sample subsets to form a training sample group.

For example, the subset of training samples to be processed, which is included in the plurality of subsets of training samples to be processed in time sequence, is as follows: YZ1, YZ2, YZ3, YZ4, YZ5, YZ6, YZ7, YZ8, YZ9 and YZ10, wherein YZ1, YZ2, YZ3 and YZ4 belong to the origin region QY1, YZ5, YZ6, YZ7, YZ8, YZ9 and YZ10 belong to the origin region QY2, and the preset training sample group generation rule is: extracting 2 each time from the starting point region QY1 and 3 each time from the starting point region QY2, the method of time-sequentially dividing will determine 3 training sample sets, the 3 training sample sets are: (YZ1, YZ2, YZ5, YZ6, YZ7), (YZ3, YZ4, YZ8, YZ9, YZ10), and the examples are not particularly limited.

The method comprises the steps of dividing a plurality of training sample subsets to be processed into training sample groups by adopting a preset training sample group generation rule and a method of sequentially dividing the training sample groups according to a time sequence, using each divided training sample group as a training sample group to be ordered, and using all training sample groups to be ordered as the training sample groups to be ordered.

For S21113, respectively performing positive sequence ordering on all the training samples in each training sample group to be ordered in the plurality of training sample groups to be ordered according to the time period in which the order is located, and using each training sample group to be ordered after positive sequence ordering as a target training sample group, so that all the training samples in the target training sample group are subjected to positive sequence ordering according to the time period in which the order in the order feature sample data is located.

Referring to fig. 2, the present application further provides a network appointment demand prediction apparatus, including:

a data obtaining module 100, configured to obtain a target starting point region, a target prediction time, and target starting point region feature data of the target starting point region corresponding to the target prediction time;

the network car-booking demand prediction module 200 is configured to input the target starting point region, the target prediction time and the target starting point region feature data into a target network car-booking demand prediction model to predict a network car-booking demand, where the target network car-booking demand prediction model is a model trained on an independent cyclic neural network, a linear model, a full connection layer and a Sigmiod activation function;

and a target network car-booking requirement prediction result determining module 300, configured to obtain a target network car-booking requirement prediction result, corresponding to the target prediction time, of the target starting point region output by the target network car-booking requirement prediction model.

Referring to fig. 3, a computer device, which may be a server and whose internal structure may be as shown in fig. 3, is also provided in the embodiment of the present application. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the computer designed processor is used to provide computational and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer equipment is used for storing data such as a network taxi appointment demand prediction method and the like. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a network appointment demand prediction method. The network taxi appointment demand prediction method comprises the following steps: acquiring a target starting point region, target prediction time and target starting point region characteristic data of the target starting point region corresponding to the target prediction time; inputting the target starting point region, the target prediction time and the target starting point region feature data into a target network vehicle reduction demand prediction model to perform network vehicle reduction demand prediction, wherein the target network vehicle reduction demand prediction model is a model obtained based on independent cyclic neural network, linear model, full connection layer and Sigmiod activation function training; and acquiring a target network car-booking demand prediction result, corresponding to the target prediction time, of the target starting point region output by the target network car-booking demand prediction model.

An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements a network taxi appointment demand prediction method, including the steps of: acquiring a target starting point region, target prediction time and target starting point region characteristic data of the target starting point region corresponding to the target prediction time; inputting the target starting point region, the target prediction time and the target starting point region feature data into a target network vehicle reduction demand prediction model to perform network vehicle reduction demand prediction, wherein the target network vehicle reduction demand prediction model is a model obtained based on independent cyclic neural network, linear model, full connection layer and Sigmiod activation function training; and acquiring a target network car-booking demand prediction result, corresponding to the target prediction time, of the target starting point region output by the target network car-booking demand prediction model.

According to the executed network car booking demand prediction method, the target network car booking demand prediction model is obtained through training based on the independent cyclic neural network, the linear model, the full connection layer and the Sigmiod activation function, then the target starting point region, the target prediction time and the target starting point region feature data are input into the target network car booking demand prediction model to carry out network car booking demand prediction to obtain a target network car booking demand prediction result corresponding to the target starting point region at the target prediction time, so that network car booking demand prediction is carried out according to various factors, the accuracy of network car booking demand prediction is improved, and the technical problem that the accuracy of directly adopting network car booking historical order data to carry out order demand prediction is not high is solved.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium provided herein and used in the examples may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), double-rate SDRAM (SSRSDRAM), Enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and bus dynamic RAM (RDRAM).

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, apparatus, article, or method that includes the element.

The above description is only a preferred embodiment of the present application, and not intended to limit the scope of the present application, and all modifications of equivalent structures and equivalent processes, which are made by the contents of the specification and the drawings of the present application, or which are directly or indirectly applied to other related technical fields, are also included in the scope of the present application.

Claims

1. A network appointment demand prediction method is characterized by comprising the following steps:

2. The method of claim 1, wherein the step of inputting the target starting point area, the target prediction time, and the target starting point area characteristic data into a target network vehicle reduction demand prediction model for vehicle reduction demand prediction further comprises:

3. The network appointment demand forecasting method according to claim 2, wherein the step of inputting all the order feature sample data and all the area feature sample data of the target training sample group to be trained into an initial model to forecast the network appointment demand to obtain a network appointment demand sample forecast value set includes:

4. The network taxi appointment demand prediction method according to claim 3, wherein the calculation formula f of the feature resolution module is as follows:

wherein X is the characteristic data to be analyzed for importance,

5. The method according to claim 3, wherein the target loss value is calculated by argmin (W)_a，W_b)f_lossComprises the following steps:

wherein, W_bIs a weight matrix of the feature resolution module, W_aIs the splicing result of the weight matrix of the independent time feature extraction module and the weight matrix of the network appointment demand classification prediction module, alpha is a constant, beta is a constant, O is the network appointment demand sample prediction value set, T is all the network appointment demand calibration data corresponding to the target training sample set to be trained, | | | |₁Is a norm of L1 which is,

6. The method of claim 2, wherein the step of obtaining a plurality of target training sample sets comprises:

acquiring a historical network car booking data set;

7. The method according to claim 6, wherein the step of generating a target training sample set according to all the training sample sets to be classified to obtain the plurality of target training sample sets comprises:

8. A network appointment demand prediction apparatus, comprising:

9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.