CN111191839A

CN111191839A - Electricity swapping prediction method and system and storage medium

Info

Publication number: CN111191839A
Application number: CN201911386061.5A
Authority: CN
Inventors: 杨磊; 黄茗
Original assignee: Shanghai Junzheng Network Technology Co Ltd
Current assignee: Shanghai Junzheng Network Technology Co Ltd
Priority date: 2019-12-26
Filing date: 2019-12-26
Publication date: 2020-05-22
Anticipated expiration: 2039-12-26
Also published as: CN111191839B

Abstract

The invention provides a power conversion prediction method, which comprises the following steps: acquiring one or more characteristics related to battery replacement; processing the features according to a preset algorithm to obtain a feature extraction result; respectively obtaining the predicted riding probability and the predicted riding duration based on the feature extraction result; and judging whether the battery replacement is needed or not based on the estimated riding probability, the estimated riding time and the current battery power. The method creatively combines the multi-task idea and the DeepFM algorithm, is applied to the problem of battery replacement of the shared battery car, selects a reasonable deep learning model and relatively accurate characteristic data, links the two steps, predicts whether the battery of the direct rental car needs to be replaced end to end, excavates important characteristics hidden behind the data, gives full play to the influence of various factors, and provides a reasonable scheme for whether the battery of the shared battery car needs to be replaced.

Description

Electricity swapping prediction method and system and storage medium

Technical Field

The invention relates to the field of electric vehicle battery replacement, in particular to a real-time prediction method and a real-time prediction system for whether the electric quantity of a shared electric vehicle needs to be replaced.

Background

With the development of internet technology, shared battery cars are raised in all big cities in China, and the shared battery cars are easy to ride in a labor-saving way, are preferential and are convenient to accept by the public. However, as the number of shared battery cars increases and the service of shared battery cars matures, more and more human resources are needed for the battery replacement service of the shared battery cars. The unreasonable power changing scheme can also cause a great deal of waste of human resources and electric quantity.

The battery of the shared battery car should be replaced when the remaining amount is too small, and the remaining amount should be calculated. Therefore, when the shared battery car needs to be changed, a complex scene is realized, and intensive research is needed.

At present, some schemes use all rules of one-time switching, namely, as long as the electric quantity of the battery car is lower than a threshold value, operation and maintenance personnel can replace the battery regardless of other states of the battery car. Only the rules are used and no algorithm is used to predict whether the battery needs to be replaced.

This solution lacks much thought and has a number of drawbacks, and the final goal to be achieved is 'supply-demand' which is the surplus energy of the battery providing the electricity, and 'demand' which is the amount of electricity the user needs to use. And the current cutting scheme only considers supply and does not consider demand. Factors influencing the user requirements are many, such as the geographical position of the battery car and the activity degree of surrounding users; the current time, the time from the peak period; whether the user can select the battery car to go out or not under the clear weather condition; the life cycle of the battery is that the distance that the new battery and the old battery can ride is different when the residual electric quantity is the same; the riding habits and riding distances of users in different cities are greatly different due to geographical conditions, traffic conditions and the like; in addition, the weather temperature has a great influence on the battery power consumption and the descending speed. It can be seen that the manual address selection using a priori knowledge leads to two problems:

firstly, the method basically depends on the self business capability of related workers, and the personal knowledge and cognitive subjective factors of the workers can directly determine the income;

secondly, the deep association between features and benefits is difficult to capture by manual selection. The prediction of whether the battery is replaced or not is effective to obtain a result through feature interaction modeling, and some feature interactions can be easily understood, so that feature engineering can be designed manually. However, most features are hidden behind data, difficult to form a priori knowledge, and can only be automatically generated through machine learning.

Therefore, those skilled in the art are devoted to developing a method and system capable of accurately predicting whether a shared battery car should be replaced at the moment.

Disclosure of Invention

In view of the above-mentioned defects of the prior art, an object of the present invention is to provide a method for timely and accurately predicting the power of a battery of a shared battery car to be replaced and assisting an operation and maintenance service worker to replace the battery of the battery car. The technical scheme of the invention utilizes a neural network algorithm so as to technically support and realize accurate prediction.

In order to achieve the purpose, the invention provides a battery replacement prediction method and a battery replacement prediction system, which are based on a multi-task thought, utilize a deep FM algorithm to predict whether the electric quantity of a shared battery car needs to be replaced in real time, and aim at predicting whether the battery needs to be replaced by the algorithm, so that the electric quantity of the shared battery car can play the maximum commercial value on the basis of reducing the cost. Where multi-task refers to a given m learning tasks, where all or a portion of the tasks are related but not identical, the goal of multi-task learning is to help improve the performance of each task by using the knowledge contained in the m tasks.

Wide & Deep Learning is a neural network Model combining a depth module and a breadth module proposed by Google, and is a structural design combining a Wide linear Model (Wide Model) and a Deep neural network Model (Deep Model) in parallel. The Wide end uses a common LR model, and combines common discrete features and low-dimensional features as input, so that the memory capability of the model is realized. The Deep end converts the discrete features into dense feature vectors through an embedding method, so that fuzzy query of tag vectors is actually realized, and the generalization capability of the model is expanded. Wide for Memourisation, the Wide side remembers those common, high frequency patterns in the history. Valuable and obvious features and feature combinations are input into the wide side according to manual experience and business background. The Deep side of Deep for Generation vectorizes tag through embedding, changes the precise matching of tag and is the fuzzy query of tag vector, so that the model has good expansion capability.

The Deep learning model is an end-to-end Deep learning model which is obtained by replacing a Logistic Regression (LR) part at the side of the Wide of an original paper with FM on the basis of a classic paper Wide & Deep learning of Google. The Deep FM replaces LR at the Wide side in the Wide and Deep model with FM, overcomes the defect that the original model still needs to perform characteristic engineering on low-dimensional features, and realizes an end to end model without any artificial characteristic engineering. DeepFM shares the embedding feature vector on the wide side and deep side. The deep FM achieves good effects on both enterprise data sets and public data sets, and the method is used for reference and is innovatively applied to the battery changing scene of the shared battery car industry. The FM model is a feature combination method, which extracts feature combinations by using an implicit vector inner product of features of each dimension, but generally only uses second-order feature combinations due to computational complexity.

The multi-task thought and the deep FM algorithm are innovatively combined and applied to the problem of battery replacement of the shared battery car, a reasonable deep learning model and accurate characteristic data are selected, and a reasonable scheme is provided for judging whether the shared battery car needs to replace the battery.

In a preferred embodiment of the present invention, the present invention provides a power swapping prediction method, including the following steps: acquiring one or more characteristics related to battery replacement; processing the features according to a preset algorithm to obtain a feature extraction result; and respectively obtaining the estimated riding probability and the estimated riding duration based on the feature extraction result.

In some embodiments, optionally, the method further includes: and judging whether the battery is required to be replaced or not based on the estimated riding probability, the estimated riding time and the current battery power.

In some embodiments, optionally, the step of determining whether battery replacement is needed includes: and processing the estimated riding probability, the estimated riding time and the current battery power by using DNN, and activating by using Sigmoid to obtain a classification result score, wherein the classification result score is greater than a battery replacement threshold value to indicate that battery replacement is not needed, and the classification result score is less than or equal to the battery replacement threshold value to indicate that battery replacement is needed.

In some embodiments, optionally, the swapping threshold is 0.5.

In some embodiments, optionally, the expected riding probability, the expected riding duration and the current battery level are processed by using a DNN including two hidden layers.

In some embodiments, optionally, the step of processing the features according to a preset algorithm comprises: classifying the features into category features and numerical features; performing onehot coding on the category characteristics; and performing embedding operation on data obtained after onehot coding so as to enable the category characteristics to be used as numerical characteristics.

In some embodiments, optionally, the step of processing the features according to a preset algorithm comprises: obtaining a first processing result by utilizing FM processing characteristics; obtaining a second processing result by utilizing DNN processing characteristics; and combining the first processing result and the second processing result to obtain a feature extraction result.

In some embodiments, optionally, the first processing result and the second processing result are merged using a concat method.

In some embodiments, optionally, the step of obtaining the expected riding probability comprises: the feature extraction results are processed with DNN and activated using Sigmoid.

In some embodiments, optionally, the step of obtaining the expected length of time to be ridden comprises: the feature extraction results are processed with DNN and linear activation is used.

In some embodiments, optionally, the feature extraction results are processed using DNN comprising two hidden layers.

In some embodiments, optionally, the step of processing the feature extraction result by using DNN includes: training was performed using grid search and cross validation.

In some embodiments, optional features include: features relating to the hardware characteristics of the vehicle itself, including one or more of the following: the method comprises the following steps of (1) total vehicle putting time, battery version, vehicle software and hardware version, unlocking success rate in the past day and unlocking failure error code in the past day; features relating to the environment in which the vehicle is located, including one or more of the following: the method comprises the steps of obtaining information of a geohash grid where a vehicle is located, user activity levels around the current moment, weather clear conditions, weather temperatures, current riding states and current user characteristics; and features associated with the order, including one or more of the following: the average daily order number of the past week, the riding time of each order, the specific time of the order start and the specific time of the order end.

In another preferred embodiment of the present invention, the present invention further provides a swapping prediction system, which includes: a feature acquisition module configured to acquire one or more features related to battery swapping; the characteristic extraction module is configured to process the characteristics according to a preset algorithm to obtain a characteristic extraction result; and the riding prediction module is configured to obtain the predicted riding probability and the predicted riding duration respectively based on the feature extraction result.

In another preferred embodiment of the present invention, the present invention further provides a computer-readable storage medium, on which a computer program is stored, wherein the computer program is capable of implementing the steps of the above-mentioned swapping prediction method when being executed by a processor.

The battery replacement prediction method and the battery replacement prediction system fully consider the principle of supply balance, reduce the replacement cost of the moped and improve the replacement efficiency of the moped on the principle of not influencing the single riding quantity of the moped. The deep FM algorithm in the recommendation system is introduced into the recommendation system, the multi-task thought and the deep learning algorithm deep FM algorithm are utilized, two steps are connected, whether the direct renting vehicle battery needs to be replaced or not is predicted end to end, the important characteristic hidden behind data is mined, the influence of various factors is fully exerted, and important technical support is provided for improving the direct renting vehicle battery replacement quality and operation and maintenance service. And the obtained intermediate result, such as the probability and the expected duration of the vehicle ridden by the user, can also be used for making other strategies, such as making direct renting vehicle scheduling and releasing strategies, so that the expandability of the model is greatly enhanced.

The conception, the specific structure and the technical effects of the present invention will be further described with reference to the accompanying drawings to fully understand the objects, the features and the effects of the present invention.

Drawings

FIG. 1 is a flow chart illustrating a preferred embodiment of a swapping prediction scheme according to the present invention;

FIG. 2 is a diagram illustrating the overall architecture of a preferred embodiment of the swapping prediction scheme of the present invention;

FIG. 3 is a schematic diagram of a computer device, equipment or terminal according to a preferred embodiment of the present invention.

Detailed Description

The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.

It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention, and the drawings only show the components related to the present invention rather than the number, shape and size of the components in actual implementation, and the type, quantity and proportion of the components in actual implementation may be changed freely, and the layout of the components may be more complicated.

Some exemplary embodiments of the invention have been described for illustrative purposes, and it is to be understood that the invention may be practiced otherwise than as specifically described.

Fig. 1 is a flowchart illustrating an embodiment of a swapping prediction scheme according to the present invention. As shown in fig. 1, the power swapping prediction method includes the following steps:

step S110, one or more characteristics related to battery replacement are collected. These battery change related features may include features related to hardware characteristics of the vehicle itself, features related to the environment in which the vehicle is located, features related to orders, and the like.

Features related to the hardware characteristics of the vehicle itself may include: the method comprises the following steps of vehicle total putting time, battery version, vehicle software and hardware version, unlocking success rate in the past day, unlocking failure error code in the past day and the like.

The features relating to the environment in which the vehicle is located may include: the method includes the steps of obtaining the information of the geohash grid where the vehicle is located, the user activity level around the current time, the weather sunny condition, the weather temperature, the current riding state, the characteristics of the current user (the user who has made/just finished the order at the moment), and the like. The geohash is a code for a latitude and longitude address, and uniquely identifies the physical location of the address on the map. The geoHash is essentially a way of spatial indexing, and its basic principle is to understand the earth as a two-dimensional plane, and recursively decompose the plane into smaller sub-blocks, each having the same code in a certain latitude and longitude range. The space index is established in a geoHash mode, and the efficiency of performing longitude and latitude retrieval on the space data can be improved.

The features associated with the order, both affected by the vehicle itself and the environment, may include: the order data of the past week, the daily average order number of the past week, the riding time of each order, the specific time of the start of the order, the specific time of the end of the order, and the like.

And step S120, processing the features according to a preset algorithm to obtain a feature extraction result.

First, the collected features are classified and transformed to a numerical form that is easy to handle. As an embodiment, the method specifically includes the following steps: classifying the features into category features and numerical features, wherein numerical features such as riding time of each bicycle are used as numerical features, and category features such as clear weather conditions are used as category features; performing onehot coding on the category characteristics; and performing embedding operation on data obtained after onehot coding so as to enable the category characteristics to be used as numerical characteristics.

onehot encoding, also known as one-bit-efficient encoding, mainly uses a bit state register to encode each state, with each state being represented by its own independent register bit and only one bit being active at any time. In the actual application task of machine learning, a feature is sometimes not always a continuous value, but may be some classification value, and for such a feature, it is usually required to digitize the feature, while the onehot coding is used to digitize according to the classification dimension, and in the case of a large classification dimension, the data becomes very sparse. The embedding can convert a large sparse vector into a low-dimensional space with a reserved semantic relation, so that the features which are too sparse and occupy resources excessively are converted into a dense vector with a fixed size.

Then, the converted features are used as input and are processed by FM (Factorization) and DNN (Deep Neural Networks) respectively, so as to perform feature intersection of low and high dimensions simultaneously. As an embodiment, using FM processing features, a first processing result is obtained; obtaining a second processing result by utilizing DNN processing characteristics; and merging the first processing result and the second processing result by using a concat method to obtain a feature extraction result. The concat method is used for connecting two or more arrays so as to combine the arrays, and can splice two or more features according to a certain dimension, wherein the other dimensions are consistent. The concat method does not change the existing array, but returns a copy of the connected array.

FM aims to solve the problem of feature combination under sparse data, is more suitable for highly sparse data scenes, has linear computational complexity, and is used for modeling the interactive relationship among features. The DNN is to obtain high-order features by exploiting the propagation of deep learning data between networks. The FM model usually can only express the relationship between two combinations of features, and cannot model the deep relationship between two features or the interaction relationship between multiple features, so the relationship between higher-order features can be modeled by DNN. The parallel structure of FM and DNN can simultaneously consider the feature intersection of low dimension and high dimension.

And step S130, respectively obtaining the predicted riding probability and the predicted riding duration based on the feature extraction result. As an embodiment, a DNN including two hidden layers is used for processing a feature extraction result, and Sigmoid activation is used for obtaining a predicted riding probability; and processing the feature extraction result by using DNN containing two hidden layers, and using linear activation to obtain the expected riding time.

In the process of utilizing DNN to process the feature extraction result, grid search and cross validation are further used for training. Grid Search (Grid Search) is an optimization method, and exhaustive Search is performed in a parameter list, and each condition is trained to find the optimal parameter. Cross Validation (Cross Validation) is the grouping of raw data (dataset) in a sense, one part as a training set (train set) and the other part as a Validation set or test set. Firstly, training a classifier by using a training set, and then testing a model (model) obtained by training by using a verification set to be used as a method for evaluating the performance index of the classifier so as to obtain the optimal parameter combination within a specified parameter range.

In some embodiments, whether the battery replacement is needed or not can be judged based on the expected riding probability, the expected riding time and the current battery power. The step of judging whether the battery replacement is needed comprises the following steps: and processing the estimated riding probability, the estimated riding time and the current battery power by using DNN containing two hidden layers, and activating by using Sigmoid to obtain a classification result score, wherein the classification result score is greater than a battery replacement threshold value to indicate that battery replacement is not needed, and the classification result score is less than or equal to the battery replacement threshold value to indicate that battery replacement is needed. As an example, the swapping threshold is 0.5. The sigmoid function is used for hidden layer neuron output, the value range is (0, 1), a real number can be mapped to an interval of (0, 1), and the interval can be used for two classifications. The advantage of Sigmoid as an activation function is that it is smooth and easy to derive. The Sigmoid function is defined by the following equation: s (x) 1/(1+e^-x)。

FIG. 2 is a block diagram of the overall architecture of a preferred embodiment of the swapping prediction scheme of the present invention. The whole algorithm flow of the invention relates to two steps, firstly, a multi-task algorithm is utilized, low-dimensional and high-dimensional feature intersection is considered in the modeling process according to deep FM, and the riding probability and the expected riding time of the moped are predicted; and then adding the real-time electric quantity factor to carry out the second classification of whether to change the electric quantity.

The bottom layer is the input sparse original feature (onehot result), and then enters into the Embedding layer, limiting the feature to a limited vector space. The Embedding layer plays a role in compressing and reducing dimension and relieves parameter explosion. And then, entering an FM layer and a DNN layer, wherein the FM layer and the DNN layer share the embedding data of the bottom layer, so that the efficiency is high, and the whole data embedding only needs to be calculated once. In the FM layer, the main work is to obtain low-order features by crossing features, and the second-order features are the main. The characteristic crossing of the FM algorithm is not directly calculated by mutually crossing the original characteristics, but is a result obtained after the characteristic factor decomposition is crossed, so that the deep information of the characteristics can be further mined. And the DNN layer obtains high-order characteristics by utilizing the propagation of deep learning data among networks.

Compared with the traditional deep FM algorithm, the method provided by the invention is improved according to the actual service requirement, the activation function and the output layer for direct output are removed, the features extracted by DNN and FM are concat and merged, and then the two DNN are respectively input for next processing. The purpose of this is that the traditional deep FM algorithm can only output one result, but the method provided by the invention can output two results of the predicted riding probability and the predicted riding time, so the results are not directly output after the characteristics are extracted, and the training is respectively carried out.

As a specific embodiment, the flow of the swapping prediction scheme of the present invention is described as follows:

(1) under the known current situation, a battery is replaced at six o' clock every morning, the battery replacement time is 30 minutes, and 24-hour characteristic data of 3 thousands of shared battery cars are taken as samples.

(2) The method for labeling the supervised learning training model comprises the following steps: and taking the order riding time length data after five and a half points as training riding time length label data, wherein the riding probability is that an order is 1 within half an hour, and no order is 0, and the order is taken as a training riding probability label.

(3) 5000 samples in 3 thousands of samples are used as a verification data set, 2 thousands of samples in 5 thousands of samples are used as a training set, the training set is used for training a model, and the verification set is used for verifying the accuracy of the training model to prevent the model from being over-fitted.

(4) Inputting a sample into an algorithm module for feature extraction, wherein the module comprises three sub-modules, a feature processing module can carry out onehot and embedding on features, the feature extraction module uses a deepfm algorithm and finally a multi-task, two tasks of predicting the probability and predicting the riding time length are separately processed, two dnn containing two hidden layers are respectively used for extracting a feature model, and two times of training are respectively carried out to obtain the predicted riding time length and the predicted riding probability.

(5) And inputting the probability of the predicted riding and the predicted riding time length of the two intermediate output results, the two characteristics and the battery power acquired from the server at the moment into an dnn algorithm module containing two hidden layers.

(6) And obtaining the final classification result score after the sigmoid activation function, wherein the value of 0.5 represents that the power conversion is not needed, and the value of less than or equal to 0.5 represents that the power conversion is needed.

The second step is to train a model of electric quantity and riding state according to label data of whether the shared battery car which is manually played can be comfortably ridden or not in the process of testing the energy consumption of the battery of the low-electric-quantity vehicle by a tester in advance, and deploy the model in a server. The two-class riding type bicycle can be directly called, the probability of being expected to be ridden and the duration and the electric quantity of being expected to be ridden are input, and whether the two classes can be ridden comfortably or not, namely the expected result, and whether the battery is replaced or not can be obtained.

In some embodiments, the feature processing module of the deepfm algorithm may implement the following:

the algorithm data characteristics related to the invention mainly comprise three types, wherein the first type is the hardware characteristics of a direct renting vehicle, and comprises the total time for putting the direct renting vehicle, the battery putting time, the battery version, the vehicle software and hardware version, the unlocking success rate in the past day and the unlocking failure error code in the past day; the second type is an environment where the direct renting vehicle is located, and comprises the geohash grid information of the battery vehicle, the user activity level around the current time, the clear weather condition, the weather temperature, the riding state at the moment, and the user characteristics (of the order at the moment/the order just finished); the third type is the order data of the last week of the vehicle, which is influenced by the vehicle and the environment, and comprises the order number of the vehicle in one day, the riding time of each order, the specific time of the order beginning and the specific time of the order ending. Wherein the geohash is a code for a latitude and longitude address that uniquely identifies the physical location of the address on the map.

In some embodiments, the specific operational flow of the feature processing module is as follows:

(1) and classifying the features, wherein the numerical feature such as the riding time of each single is used as the numerical feature, and the category feature such as the clear weather condition is used as the category feature.

(2) And onehot coding is carried out on the classification characteristics.

(3) And performing characteristic embedding on the data obtained after the classification characteristic onehot coding.

(4) After the processing of the steps (2) and (3), the category features can be used as numerical features, and the numerical features in the step (1) and the features obtained in the step (3) are transmitted into the model together for subsequent operation.

In some embodiments, the training module of the deepfm algorithm may be implemented as follows:

the deepfm algorithm is improved in the training module according to actual service requirements, an activation function and an output layer are removed, the dnn and fm extracted features are concat, and then the two dnn are respectively connected for further processing. The purpose of this is that the original deepfm algorithm can only output one result, and two results of the riding probability and the predicted riding mileage need to be output, so the results are not directly output after the characteristics are extracted by using the deepfm algorithm, and are respectively trained. In addition, in the training process, the optimal parameter combination in the specified parameter range can be obtained by using a grid searching and cross validation method.

In some embodiments, the present invention further provides a power swapping prediction system, which can perform the above method and steps. The battery replacement prediction system comprises: a feature acquisition module capable of acquiring one or more features related to battery swapping; the characteristic extraction module can process the characteristics according to a preset algorithm to obtain a characteristic extraction result; and the riding prediction module can respectively obtain the predicted riding probability and the predicted riding duration based on the feature extraction result. In some embodiments, the battery replacement prediction system may further include a battery replacement determination module, which may determine whether battery replacement is required based on the predicted riding probability, the predicted riding duration, and the current battery power.

In some embodiments, the present invention also provides a computer apparatus, device or terminal, the internal structure of one embodiment of which may be as shown in fig. 3. The computer apparatus, device or terminal includes a processor, a memory, a network interface, a display screen and an input device connected by a system bus. The processor is used for providing calculation and control capability, and the memory comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program to run in the non-volatile storage medium. The network interface is used for communicating with an external terminal through network connection. The computer program is executed by a processor to implement the various methods, procedures, steps disclosed in the present invention, or the processor executes the computer program to implement the functions of the respective modules or units in the embodiments disclosed in the present invention. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell, an external keyboard, a touch pad or a mouse and the like.

Illustratively, a computer program may be divided into one or more modules or units, which are stored in a memory and executable by a processor to implement the inventive arrangements. These modules or units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution of a computer program in an apparatus, device or terminal.

The device, the equipment or the terminal can be computing equipment such as a desktop computer, a notebook computer, a mobile electronic device, a palm computer, a cloud server and the like. It will be appreciated by those skilled in the art that the arrangements shown in the drawings are merely block diagrams of some of the arrangements relevant to the inventive arrangements and do not constitute limitations on the apparatus, devices or terminals to which the arrangements are applied, and that a particular apparatus, device or terminal may include more or less components than shown in the drawings, or may combine certain components, or have a different arrangement of components.

The Processor may be a Central Processing Unit (CPU), other general or special purpose Processor, a microprocessor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. The processor is the control center of the above-mentioned apparatus, device or terminal, and connects the respective parts of the apparatus, device or terminal by using various interfaces and lines.

The memory may be used to store computer programs, modules and data, and the processor may implement various functions of the apparatus, device or terminal by executing or executing the computer programs and/or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program (such as a sound playing function, an image playing function, etc.) required by at least one function, and the like; the data storage area may store various types of data (such as multimedia data, documents, operation histories, etc.) created according to the application, and the like. In addition, the memory may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), a magnetic disk storage device, a Flash memory device, or other volatile solid state storage device.

The invention also provides a computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the above-mentioned method. 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, databases, or other media used in embodiments provided herein 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 Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The above-described apparatus or terminal device integrated modules and units, if implemented in the form of software functional units and sold or used as separate products, may be stored in a computer-readable storage medium. Based on such understanding, the present invention can realize all or part of the procedures of the disclosed methods, and can also be realized by relevant hardware instructed by a computer program, which can be stored in a computer readable storage medium, and when the computer program is executed by a processor, the steps of the methods can be realized. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying computer program code, recording medium, U.S. disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution media, and the like. It should be noted that the computer readable medium may contain content that is appropriately increased or decreased as required by legislation and patent practice in the jurisdiction.

In some embodiments, the various methods, procedures, modules, devices, apparatuses, or systems disclosed herein may be implemented or performed in one or more processing devices (e.g., digital processors, analog processors, digital circuits designed to process information, analog circuits designed to process information, state machines, computing devices, computers, and/or other mechanisms for electronically processing information). The one or more processing devices may include one or more devices that perform some or all of the operations of a method in response to instructions stored electronically on an electronic storage medium. The one or more processing devices may include one or more devices configured through hardware, firmware, and/or software to be specifically designed for performing one or more operations of a method. The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.

Embodiments of the invention may be implemented in hardware, firmware, software, or various combinations thereof, and may also be implemented as instructions stored on a machine-readable medium, which may be read and executed using one or more processing devices. In some implementations, a machine-readable medium may include various mechanisms for storing and/or transmitting information in a form readable by a machine (e.g., a computing device). For example, a machine-readable storage medium may include read-only memory, random-access memory, magnetic disk storage media, optical storage media, flash-memory devices, and other media for storing information, and a machine-readable transmission medium may include various forms of propagated signals (including carrier waves, infrared signals, digital signals), and other media for transmitting information. While firmware, software, routines, or instructions may be described in the above disclosure in terms of performing certain exemplary aspects and embodiments of certain actions, it will be apparent that such descriptions are merely for convenience and that such actions in fact result from a machine device, computing device, processing device, processor, controller, or other device or machine executing the firmware, software, routines, or instructions.

The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.

Claims

1. A power swapping prediction method is characterized by comprising the following steps:

acquiring one or more characteristics related to battery replacement;

processing the features according to a preset algorithm to obtain a feature extraction result; and

and respectively obtaining the predicted riding probability and the predicted riding duration based on the feature extraction result.

2. The battery swap prediction method of claim 1, further comprising:

and judging whether the battery replacement is needed or not based on the estimated riding probability, the estimated riding time and the current battery electric quantity.

3. The battery swap prediction method of claim 2, wherein the step of determining whether battery swap is required comprises: and processing the estimated riding probability, the estimated riding time and the current battery power by using DNN, and activating by using Sigmoid to obtain a classification result score, wherein the classification result score is greater than a battery replacement threshold value and indicates that battery replacement is not needed, and the classification result score is less than or equal to the battery replacement threshold value and indicates that battery replacement is needed.

4. The swapping prediction method of claim 3, wherein the swapping threshold is 0.5.

5. The battery swapping prediction method of claim 3, wherein the predicted cycling probability, the predicted cycling duration and the current battery level are processed by using a DNN comprising two hidden layers.

6. The battery swapping prediction method of claim 1, wherein the step of processing the features according to a predetermined algorithm comprises:

classifying the features into category features and numerical features;

performing onehot coding on the category characteristics; and

and performing embedding operation on data obtained after onehot coding so as to enable the category characteristics to be used as numerical characteristics.

7. The battery swapping prediction method of claim 1, wherein the step of processing the features according to a predetermined algorithm comprises:

processing the characteristics by using FM to obtain a first processing result;

processing the features by using DNN to obtain a second processing result; and

and combining the first processing result and the second processing result to obtain the feature extraction result.

8. The swapping prediction method of claim 7, wherein the first processing result and the second processing result are merged using a concat method.

9. The battery swapping prediction method of claim 1, wherein the step of obtaining the predicted probability of being ridden comprises: the feature extraction results are processed with DNN and activated using Sigmoid.

10. The battery swapping prediction method of claim 1, wherein the step of obtaining the predicted riding time comprises: the feature extraction results are processed with DNN and linear activation is used.

11. The swapping prediction method of claim 9 or 10, wherein the feature extraction result is processed using a DNN comprising two hidden layers.

12. The swapping prediction method of claim 11, wherein the step of processing the feature extraction result by DNN comprises: training was performed using grid search and cross validation.

13. The swapping prediction method of any of the preceding claims, wherein the features comprise:

features relating to the hardware characteristics of the vehicle itself, including one or more of the following: the method comprises the following steps of (1) total vehicle putting time, battery version, vehicle software and hardware version, unlocking success rate in the past day and unlocking failure error code in the past day;

features relating to the environment in which the vehicle is located, including one or more of the following: the method comprises the steps of obtaining information of a geohash grid where a vehicle is located, user activity levels around the current moment, weather clear conditions, weather temperatures, current riding states and current user characteristics; and

features associated with the order, including one or more of the following: the average daily order number of the past week, the riding time of each order, the specific time of the order start and the specific time of the order end.

14. A system for predicting battery swapping, comprising:

a feature acquisition module configured to acquire one or more features related to battery swapping;

the characteristic extraction module is configured to process the characteristics according to a preset algorithm to obtain a characteristic extraction result; and

and the riding prediction module is configured to obtain the predicted riding probability and the predicted riding duration respectively based on the feature extraction result.

15. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, is capable of carrying out the steps of the swapping prediction method according to any one of claims 1-13.