WO2021189950A1 - 短时公交站客流预测方法、装置、计算机设备及存储介质 - Google Patents

短时公交站客流预测方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2021189950A1
WO2021189950A1 PCT/CN2020/134955 CN2020134955W WO2021189950A1 WO 2021189950 A1 WO2021189950 A1 WO 2021189950A1 CN 2020134955 W CN2020134955 W CN 2020134955W WO 2021189950 A1 WO2021189950 A1 WO 2021189950A1
Authority
WO
WIPO (PCT)
Prior art keywords
joint
bus station
model
bus
target
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2020/134955
Other languages
English (en)
French (fr)
Inventor
王健宗
肖京
何安珣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2021189950A1 publication Critical patent/WO2021189950A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment, and storage medium for predicting passenger flow at a short-term bus station.
  • Public transportation plays an important role in providing transportation capacity and alleviating traffic congestion.
  • the current public transportation capacity in most cities in China is insufficient, especially during peak hours.
  • There are various methods for estimating the passenger flow on the bus It emerged as the times require, through the estimation of the passenger flow on the bus at the bus station, the traffic capacity can be adjusted in time, and the estimated result can even be sent to the client of the relevant user to facilitate people's travel arrangements.
  • the purpose of the embodiments of the present application is to propose a short-term bus stop passenger flow prediction method, device, computer equipment, and storage medium, so as to improve the accuracy of the regression prediction model for short-term bus passenger flow prediction.
  • the embodiments of the present application provide a short-term bus station passenger flow prediction method, which adopts the following technical solutions:
  • a method for forecasting passenger flow at short-term bus stations, applied to a central server includes the following steps:
  • Receive bus stop location information and a target bus stop construct a bus stop undirected network based on the bus stop location information, and determine a bus stop meeting preset conditions in the bus stop undirected network based on the target bus stop, as a joint A bus station, where the target bus station corresponds to a target server, and each joint bus station corresponds to a joint server;
  • the embodiments of the present application provide a short-term bus station passenger flow prediction method, which adopts the following technical solutions:
  • a method for forecasting passenger flow at short-term bus stations, applied to a joint server includes the following steps:
  • All the joint servers transmit the received random numbers to a central server, so that the central server uses the gradient information and the random number calculation model to update parameters;
  • the regression prediction model is iterated based on the model update parameters until the preset stopping condition is reached, and the trained regression prediction model is obtained.
  • the embodiment of the present application also provides a short-term bus station passenger flow prediction device, which adopts the following technical solutions:
  • the short-term bus stop passenger flow forecasting device includes a central server, and the central server includes:
  • a network construction module for receiving bus station location information and a target bus station, constructing a bus station undirected network based on the bus station location information, and determining that the bus station undirected network meets preset conditions based on the target bus station
  • the bus station in, as a joint bus station, where the target bus station corresponds to a target server, and each joint bus station corresponds to a joint server;
  • the parameter calculation module is used to receive the model gradient information transmitted by the target server corresponding to the target bus station, and the random number transmitted by the joint server corresponding to the joint bus station, and calculate model update parameters based on the model parameters and the random number, and The model update parameters are transmitted to the target server and each joint server, so that the target server and the joint server iteratively update the regression prediction model through the model update parameters.
  • the embodiment of the present application also provides a short-term bus station passenger flow prediction device, which adopts the following technical solutions:
  • the short-term bus stop passenger flow forecasting device includes a joint server, and the joint server includes:
  • the joint sample construction module is used to receive the passenger boarding information of the target time window and the corresponding joint bus station, and construct the training sample set of the corresponding joint bus station based on the passenger boarding information and the target time window;
  • the joint model construction module is used to construct the regression prediction model of the joint bus station based on the training sample set until the preset construction conditions are reached, and the model parameters of the respective constructed regression prediction models are obtained;
  • the distribution module is used to receive the side weight of the corresponding joint bus station transmitted by the central server, calculate the product of the side weight and the model parameters of the regression prediction model as a calculation result, and generate multiple random numbers based on the calculation result, Distributing the random number to each of the joint servers;
  • a joint data transmission module configured to transmit the received random number to a central server, so that the central server uses the gradient information and the random number calculation model to update parameters;
  • the joint iterative update module is used to iterate the regression prediction model based on the model update parameters when the model update parameter transmitted by the central server is received, until the preset stopping condition is reached to obtain the trained regression prediction model.
  • the embodiments of the present application also provide a computer device, which adopts the following technical solutions:
  • a computer device includes a memory and a processor, wherein computer readable instructions are stored in the memory, and when the processor executes the computer readable instructions, the steps of the short-term bus stop passenger flow prediction method as described below are implemented:
  • Receive bus stop location information and a target bus stop construct a bus stop undirected network based on the bus stop location information, and determine a bus stop meeting preset conditions in the bus stop undirected network based on the target bus stop, as a joint A bus station, where the target bus station corresponds to a target server, and each joint bus station corresponds to a joint server;
  • a computer device includes a memory and a processor, wherein computer readable instructions are stored in the memory, and when the processor executes the computer readable instructions, the steps of the short-term bus stop passenger flow prediction method as described below are implemented:
  • All the joint servers transmit the received random numbers to a central server, so that the central server uses the gradient information and the random number calculation model to update parameters;
  • the regression prediction model is iterated based on the model update parameter until the preset stopping condition is reached, and the trained regression prediction model is obtained.
  • the embodiments of the present application also provide a computer-readable storage medium, which adopts the following technical solutions:
  • a computer-readable storage medium having computer-readable instructions stored thereon, and when the computer-readable instructions are executed by a processor, the steps of the short-term bus stop passenger flow prediction method as described below are realized:
  • Receive bus stop location information and a target bus stop construct a bus stop undirected network based on the bus stop location information, and determine a bus stop meeting preset conditions in the bus stop undirected network based on the target bus stop, as a joint A bus station, where the target bus station corresponds to a target server, and each joint bus station corresponds to a joint server;
  • a computer-readable storage medium having computer-readable instructions stored thereon, and when the computer-readable instructions are executed by a processor, the steps of the short-term bus stop passenger flow prediction method as described below are realized:
  • All the joint servers transmit the received random numbers to a central server, so that the central server uses the gradient information and the random number calculation model to update parameters;
  • the regression prediction model is iterated based on the model update parameter until the preset stopping condition is reached, and the trained regression prediction model is obtained.
  • the central server establishes an undirected network of bus stations through the location information of the bus stations, so that it is convenient to calculate the qualified joint bus stations through the undirected network of bus stations.
  • the joint bus stop determined by the present application through the undirected network calculation of the bus stop has a stronger correlation with the target bus stop.
  • the joint server determines the passenger flow of the bus station based on the target time window through the target time window, and builds their own regression prediction models based on the passenger flow in the target time window to realize the short-term prediction model construction of the bus.
  • the joint server multiplies the parameters of the constructed regression prediction model with the corresponding side weights of the joint bus station, and generates random numbers for distribution, which protects the privacy of joint bus station data and prevents leakage during transmission.
  • the central server updates the parameters based on the model gradient information transmitted by the target server and the random number comprehensive calculation model transmitted by the joint server, which realizes the joint learning between the target bus station and the joint bus station with strong correlation with the target bus station, which is obtained through iteration
  • the regression prediction model after the final training improves the accuracy of the prediction of the regression prediction model.
  • Figure 1 is an exemplary system architecture diagram to which the present application can be applied;
  • Fig. 2 is a flowchart of an embodiment of a method for predicting passenger flow at a short-term bus station according to the present application
  • Fig. 3 is a schematic diagram of the structure of an undirected network of a bus station according to the method for predicting passenger flow at a short-term bus station according to the present application;
  • FIG. 4 is a schematic structural diagram of an embodiment of a passenger flow prediction device for a short-term bus station according to the present application
  • Fig. 5 is a schematic structural diagram of an embodiment of a computer device according to the present application.
  • the short-term bus passenger flow prediction method of the embodiment of the present application is applied to a short-term bus passenger flow prediction system.
  • the system architecture includes a central server, a target server, a joint server, a first client, and a second client.
  • the target server and the central server, between the federation server and the central server, between the first client and the target server, and between the second client and the federation server are all connected through the network.
  • the network may include various connection types, such as Wired, wireless communication link or fiber optic cable, etc.
  • the user can use the first client or the second client to interact with the target server or the joint server through the network to receive or send messages, etc.
  • Various communication client applications may be installed on the first client and the second client, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.
  • the first client and the second client can be various electronic devices that have a display screen and support web browsing, including but not limited to smart phones, tablets, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III , Motion Picture Experts compress standard audio layer 3), MP4 (Moving Picture Experts Group Audio Layer IV, Motion Picture Experts compress standard audio layer 4) Players, laptop portable computers and desktop computers, etc.
  • MP3 players Motion Picture Experts Group Audio Layer III , Motion Picture Experts compress standard audio layer 3
  • MP4 Motion Picture Experts compress standard audio layer 4
  • Players laptop portable computers and desktop computers, etc.
  • central server the target server, the federation server, the first client, and the second client in FIG. 1 are merely illustrative. According to implementation needs, there can be any number of central servers, target servers, federation servers, first clients, and second clients.
  • FIG. 2 there is shown a flowchart of an embodiment of a method for predicting passenger flow at a short-term bus station according to the present application.
  • the method for predicting passenger flow at short-term bus stations includes the following steps:
  • the central server receives the bus stop location information and the target bus stop, constructs a bus stop undirected network based on the bus stop location information, and determines the bus that meets preset conditions in the bus stop undirected network based on the target bus stop Station, as a joint bus station, where the target bus station corresponds to a target server, and each joint bus station corresponds to a joint server.
  • the joint bus station in the undirected network of bus stations is determined based on the target bus station, and the correlation between subsequent relevant data of the joint bus station and the data of the target bus station is improved.
  • the target bus station and each joint bus station correspond to a data center, and the subsequent model building process of each bus station based on training samples is completed in their own data center to avoid data leakage.
  • the electronic device (such as the central server shown in FIG. 1) on which the short-term bus stop passenger flow prediction method runs can receive the location information of the bus stop and the target bus stop through a wired connection or a wireless connection.
  • the above-mentioned wireless connection methods can include, but are not limited to, 3G/4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or developed in the future .
  • step S1 the undirected network of bus stops is constructed based on the location information of the bus stops, and the bus stop in the undirected network of bus stops that meets the preset conditions is determined based on the target bus stop, as
  • the steps of the United Bus Station include:
  • bus station as a node of the bus station undirected network based on the location information of the bus station;
  • a bus stop with a corresponding side weight less than a preset threshold is used as the joint bus stop.
  • the bus station is a node of the network, and the connection between any two bus stations is the edge of the network, and the joint bus station is determined by the edge weight, and the joint bus station is the bus station that has a close relationship with the target bus station i. It is convenient for subsequent joint learning with the target bus station, avoiding the situation that the training data dimension of a single bus station is low and the prediction result is inaccurate.
  • Fig. 3 is a schematic diagram of the structure of an undirected network of a bus station according to the short-term bus station passenger flow prediction method of the present application.
  • the edge weight S if is calculated by the following formula:
  • POP if is the difference between the total population of the target bus stop i and the bus stop f within a preset range.
  • the preset range is a circular coverage area with a bus stop as the center, N (km) as a radius, and N as a custom value.
  • Total population population density ⁇ area of the preset range, where population density means the number of people per unit area of land, which can be obtained by calling the browser to query. Using the above formula to calculate the side weight, the population density is fully considered, not only the passenger flow on the train, but also the dimension of the subsequent model update parameters.
  • the first client disturbs the locally stored passenger boarding information of the target bus station by using a preset local differential privacy algorithm, generates first disturbance data, and sends the first disturbance data to the target server.
  • the passenger boarding information in this application refers to the corresponding bus stop and boarding time when the passenger takes the bus.
  • the content that needs to be encrypted is whether the passenger boarded at the current bus stop.
  • Each record has its corresponding Time information.
  • the user's card swiping records are stored in their mobile phones and bank cards.
  • the first client uses a local differential privacy algorithm to disturb the passenger boarding information of the target bus station, which protects the user's privacy and avoids directly sending the user's card swiping records to the target server.
  • the target server estimates the total passenger flow on the bus based on the disturbance data. Specifically: Add disturbance to each data in the real set, that is, add a probability calculation to form a probability set, and send the probability set to the target server.
  • the first disturbance data is calculated by the following formula:
  • y ij ⁇ [0,1], 1 means that the user swipes the card to get on the bus at the target bus stop
  • 0 means that the user does not swipe the card to get on the bus at the target bus stop
  • ⁇ is a hyperparameter
  • the application adds disturbance to passenger ride information through the above formula, so that the data center cannot obtain actual data and protects the privacy of users, and can estimate the passenger flow on the car based on the disturbance data and perform subsequent calculations.
  • the target server corresponding to the target bus station receives a target time window and the first disturbance data sent by the first client, and constructs a training sample set of the target bus station based on the first disturbance data and the target time window .
  • the target time window in this application is self-defined.
  • the target server constructs a training sample set based on the target time window, so that the training sample set is correlated with the target time window, so that the subsequent trained regression prediction model can accurately predict the passenger flow on the train within the target time window.
  • step S3 the step of constructing a training sample set of a bus station based on the first disturbance data and the target time window includes:
  • the bus passenger flow of the target bus station in D time windows before the same period of n days and the target time window is used as the feature of the training sample, that is, the first input parameter X, and the target
  • the passenger flow of the regional bus station at the same time as the target time window in n days is used as the second input parameter Y of the training sample
  • the training sample ⁇ (X 1 ,Y 1 ),(X 2 ,Y 2 ),..., (X m ,Y m ),...(X n ,Y n ) ⁇ , each sample (X m ,Y m ) contains two input parameters, where X m has a characteristic dimension of D, and D is an integer .
  • the passenger flow from 1 pm to 2 pm on the first day and the passenger flow from 2 pm to 3 pm constitute X1 (two-dimensional vector including two passenger flows), and the passenger flow from 3 pm to 4 pm on the first day Form a pair of training samples for Y1, (X 1 , Y 1 ); the passenger flow from 1 pm to 2 pm and the passenger flow from 2 pm to 3 pm on the next day form X2 (2-dimensional vector, including two passenger flows Quantities), the passenger flow from 3pm to 4pm on the next day is Y2, and (X 2 , Y 2 ) forms a pair of training samples.
  • the passenger flow from 1pm to 2pm on the mth day and the passenger flow from 2pm to 3pm on the mth day constitute X m (two-dimensional vector including two passenger flows), and the passenger flow from 3pm to 4pm on the mth day
  • the quantity is Y m
  • (X m , Y m ) forms a pair of training samples.
  • the target server constructs the regression prediction model of the target bus station based on the training sample set, until the preset construction condition is reached, and obtains the model parameters of the constructed regression prediction model.
  • the regression prediction model of the target bus station is constructed, the preliminary training of the regression prediction model is completed, and the model parameters of the constructed regression prediction model are obtained, which is convenient for subsequent further processing.
  • step S4 the target server constructs a regression prediction model of the target bus station based on the training sample set until a preset construction condition is reached, and the step of obtaining model parameters of the constructed regression prediction model includes :
  • the target server randomly initializes the model parameters of the preset regression prediction model, where each day corresponds to a preset regression prediction model
  • the model parameters of the constructed regression prediction model are obtained, where the preset construction conditions are:
  • y m represents the passenger flow during the same period of the target time window on the mth day.
  • argmin refers to the value of the variable when the following formula reaches the minimum value. This application uses this construction condition to realize the preliminary training of the regression prediction model of each bus station based on the training sample corresponding to each bus station .
  • the target server calculates the model gradient information of the target bus station based on the model parameters, and transmits the model gradient information to the central server.
  • model gradient information of the target bus station is calculated according to the following formula:
  • the second client uses the preset local differential privacy algorithm to disturb the passenger boarding information of the joint bus station stored locally, respectively generates second disturbance data, and sends the second disturbance data to the respective joint server.
  • the passenger boarding information in this application refers to the corresponding bus stop and boarding time when the passenger takes the bus.
  • the content that needs to be encrypted is whether the passenger boarded at the current bus stop.
  • Each record has its corresponding Time information.
  • the user's card swiping records are stored in their mobile phones and bank cards.
  • the second client uses a local differential privacy algorithm to disturb the passenger boarding information of the target bus station, which protects the user's privacy and avoids directly sending the user's card swiping records to the joint server.
  • the joint server estimates the total passenger flow on the bus based on the disturbance data. Specifically: Add disturbance to each data in the real set, that is, add a probability calculation to form a probability set, and send the probability set to the joint server.
  • the second disturbance data is generated by the following formula:
  • y ij ⁇ [0,1], 1 means that the user swipes to board the bus at the current joint bus station
  • 0 means that the user does not swipe the card to get on the bus at the current joint bus station
  • ⁇ is a hyperparameter
  • the application adds disturbance to passenger ride information through the above formula, so that the data center cannot obtain actual data and protects the privacy of users, and can estimate the passenger flow on the car based on the disturbance data and perform subsequent calculations.
  • the joint server corresponding to the joint bus station receives the target time window and the respective corresponding second disturbance data, and constructs the training sample set of the respective joint bus station based on the second disturbance data and the target time window.
  • the target time window in this application is self-defined.
  • the joint server constructs respective training sample sets based on the target time window, so that the training sample set is correlated with the target time window, so that the subsequent trained regression prediction model can accurately predict the passenger flow in the target time window.
  • step S7 the step of constructing respective training sample sets of joint bus stations based on the second disturbance data and the target time window includes:
  • the bus passenger flow of the target bus station in D time windows before the same period of n days and the target time window is used as the feature of the training sample, that is, the first input parameter X, and the target
  • the passenger flow of the regional bus station at the same time as the target time window in n days is used as the second input parameter Y of the training sample
  • the training sample ⁇ (X 1 ,Y 1 ),(X 2 ,Y 2 ),..., (X m ,Y m ),...(X n ,Y n ) ⁇ , each sample (X m ,Y m ) contains two input parameters, where X m has a characteristic dimension of D, and D is an integer .
  • the passenger flow from 1 pm to 2 pm on the first day and the passenger flow from 2 pm to 3 pm constitute X1 (two-dimensional vector including two passenger flows), and the passenger flow from 3 pm to 4 pm on the first day Form a pair of training samples for Y1, (X 1 , Y 1 ); the passenger flow from 1 pm to 2 pm and the passenger flow from 2 pm to 3 pm on the next day form X2 (2-dimensional vector, including two passenger flows Quantities), the passenger flow from 3pm to 4pm on the next day is Y2, and (X 2 , Y 2 ) forms a pair of training samples.
  • the passenger flow from 1pm to 2pm on the mth day and the passenger flow from 2pm to 3pm on the mth day constitute X m (two-dimensional vector including two passenger flows), and the passenger flow from 3pm to 4pm on the mth day
  • the quantity is Y m
  • (X m , Y m ) forms a pair of training samples.
  • the joint server constructs the regression prediction model of the joint bus station based on the training sample set of the joint bus station, until the preset construction condition is reached, and obtains the model parameters of the respective constructed regression prediction model.
  • each constructs a regression prediction model of the joint bus station completes the preliminary training of the regression prediction model, and obtains the corresponding model parameters of the constructed regression prediction model, which is convenient for subsequent further processing.
  • step S8 the joint server constructs the regression prediction model of the joint bus station based on the training sample set, until a preset construction condition is reached, and the step of obtaining the model parameters of the constructed regression prediction model include:
  • the joint server each constructs the regression prediction model of the joint bus station based on the training sample set of the joint bus station until the preset construction conditions are reached, and the step of obtaining the corresponding model parameters of the regression prediction model of the joint bus station after construction include:
  • the joint server randomly initializes the model parameters of the preset regression prediction model, wherein each day corresponds to a preset regression prediction model
  • the joint servers each train a regression prediction model corresponding to each day based on the training samples in the training sample set
  • the formula of the regression prediction model is:
  • the model parameters of the respective constructed regression prediction models are obtained, wherein the preset construction conditions are:
  • y m represents the passenger flow during the same period of the target time window on the mth day.
  • the joint server receives the side weight of the corresponding joint bus station transmitted by the central server, calculates the product of the side weight and the model parameters of the regression prediction model as a calculation result, and generates multiple random numbers based on the calculation result, and The random number is distributed to each of the joint servers.
  • each joint server calculates the product of the side weight of the corresponding joint bus station and the model parameters of the corresponding joint bus station for t iterations. And the product of edge weights and model parameters K random numbers are randomly generated by the key sharing algorithm and distributed to each joint server. Wherein, the sum of the k numbers is equal to The sum of the random numbers received by all federation servers is The model parameters in the joint bus station are protected by random numbers, and the privacy of the data is improved.
  • All the joint servers transmit the received random numbers to a central server, so that the central server uses the model gradient information and the random number to calculate model update parameters.
  • all the joint servers transmit the received random numbers to the central server, and the sum of the random numbers received by all the joint servers is Improve data privacy by transmitting random numbers, and avoid data leakage and tampering during transmission.
  • the central server receives the model gradient information transmitted by the target server corresponding to the target bus station and the random number transmitted by the joint server corresponding to the joint bus station, and calculates model update parameters based on the model gradient information and the random number, and combines The model update parameters are transmitted to the target server and each joint server, so that the target server and the joint server iteratively update the regression prediction model through the model update parameters.
  • the model update parameters are calculated through the model gradient information of the target bus stop and the random number of the joint bus stop.
  • the joint bus station is determined by the side weight and has a strong correlation with the target bus station, which avoids the simple calculation of model update parameters only through a single bus station or regional bus station, which leads to inconsistent training results. Good, the situation where the forecast effect is poor.
  • the target bus station and the joint bus station federal training with the target bus station that meets the border rights screening conditions comprehensively calculate the model update parameters, and improve the prediction effect of the model.
  • the joint bus station in this application is different from the regional bus station.
  • the regional bus station only considers the geographic location relationship between the bus stations, and does not consider the population relationship between the bus stations in a certain area.
  • step S11 the step of calculating model update parameters based on the model gradient information and the random number, and transmitting the model update parameters to the target server and each joint server includes:
  • N(i) is the set of the joint bus stations, Is the sum of all random numbers received, ⁇ is the learning rate, Is model gradient information;
  • the model update parameters are transmitted to the target server and each joint server.
  • N(i) is a set of bus stations that perform federated learning with bus station i, that is, a set of joint bus stations.
  • ⁇ is the learning rate. In this article, ⁇ is set to 0.001. In practical applications, the learning rate can be changed according to actual needs, and it can be applied. Through this formula, the calculation method of the model update parameters can be clearly known.
  • the target server and each joint server both receive the model update parameters returned by the central server, update their regression prediction models according to the model update parameters, and then train according to the training samples, repeat the above training steps, and train
  • the result is that the model gradient information of the target bus station and the random number of the joint bus station are sent to the central server, waiting to receive the next model update parameter of the central server.
  • the respective regression prediction models are iterated in a loop until the preset number of iterations T is reached, and the trained regression prediction model is obtained.
  • the obtained training regression prediction model is used to predict the passenger flow of the target bus station at the same time period of the target time window.
  • the trained regression prediction model can be transmitted to the central server, or stored in the target server or the joint server for direct call.
  • the constructed prediction samples are received through the central server, the target server or the joint server, and the prediction samples are input into the trained regression prediction model to obtain the passenger flow of the target bus station in the target time window. For example: To predict the passenger flow Y from 3 pm to 4 pm on a certain day, when the dimension D is 2, the corresponding prediction sample X is the passenger flow from 1 pm to 2 pm and the passenger flow from 2 pm to 3 pm on that day. Passenger flow. At this time, enter X into the model to get it.
  • the central server establishes an undirected network of bus stations through the location information of the bus stations, so that it is convenient to calculate the qualified joint bus stations through the undirected network of bus stations.
  • the joint bus stop determined by the present application through the undirected network calculation of the bus stop has a stronger correlation with the target bus stop.
  • the joint server determines the passenger flow of the bus station based on the target time window through the target time window, and builds their own regression prediction models based on the passenger flow in the target time window to realize the short-term prediction model construction of the bus.
  • the joint server multiplies the parameters of the constructed regression prediction model with the corresponding side weights of the joint bus station, and generates random numbers for distribution, which protects the privacy of joint bus station data and prevents leakage during transmission.
  • the central server updates the parameters based on the model gradient information transmitted by the target server and the random number comprehensive calculation model transmitted by the joint server, which realizes the joint learning between the target bus station and the joint bus station with strong correlation with the target bus station, which is obtained through iteration
  • the regression prediction model after the final training improves the accuracy of the prediction of the regression prediction model.
  • the above-mentioned trained regression prediction model may also be stored in a node of a blockchain.
  • the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • This application can be applied in the field of smart transportation to promote the construction of smart cities.
  • the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through computer-readable instructions, which can be stored in a computer-readable storage medium.
  • the computer-readable instructions When executed, they may include the processes of the above-mentioned method embodiments.
  • the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.
  • this application provides an embodiment of a short-term bus station passenger flow prediction device, which corresponds to the method embodiment shown in FIG. 2.
  • the device can be specifically applied to various electronic devices.
  • the short-term bus station passenger flow prediction device in this embodiment includes: a central server, a target server and a joint server, a first client and a second client.
  • the central server includes a network construction module 101 and a parameter calculation module 102.
  • the network construction module 101 is configured to receive bus station location information and a target bus station, construct a bus station undirected network based on the bus station location information, and determine that the bus station undirected network conforms to a preset based on the target bus station
  • Conditional bus stations are used as joint bus stations, where the target bus station corresponds to a target server, and each joint bus station corresponds to a joint server;
  • the parameter calculation module 102 is configured to receive the model gradient information transmitted by the target server corresponding to the target bus station and the random number transmitted by the joint server corresponding to the joint bus station, and calculate model update parameters based on the model parameters and the random numbers, The model update parameters are transmitted to the target server and each joint server, so that the target server and the joint server iteratively update the regression prediction model through the model update parameters.
  • the first client terminal is used to perturb the passenger boarding information of the target bus station stored locally through a preset local differential privacy algorithm, generate first disturbance data, and send the first disturbance data to the target server.
  • the target server includes: a target sample construction module 201, a target model construction module 202, a target data transmission module 203, and a target iteration update module 204.
  • the target sample construction module 201 is configured to receive a target time window and the first disturbance data sent by the first client, and construct a training sample set of the target bus station based on the first disturbance data and the target time window;
  • the target model construction module 202 is configured to construct a regression prediction model of the target bus station based on the training sample set, until a preset construction condition is reached, and obtain model parameters of the constructed regression prediction model;
  • the target data transmission module 203 is configured to calculate the model gradient information of the target bus station based on the model parameters, and transmit the model gradient information to the central server;
  • the target iterative update module 204 is configured to, when the model update parameters transmitted by the central server are received, iterative regression prediction model based on the model update parameters until a preset stopping condition is reached to obtain a trained regression prediction model;
  • the second client is used to disturb the passenger boarding information of the joint bus station stored locally through the preset local differential privacy algorithm, respectively generate second disturbance data, and send the second disturbance data to the respective joint server.
  • the joint server includes: a joint sample construction module 301, a joint model construction module 302, a distribution module 303, a joint data transmission module 304, and a joint iteration update module 305.
  • the joint sample construction module 301 is configured to receive a target time window and respective corresponding second disturbance data, and construct a respective joint bus station training sample set based on the second disturbance data and the target time window;
  • the joint model construction module 302 is configured to construct the regression prediction model of the joint bus station based on the training sample set until the preset construction conditions are reached, and obtain the model parameters of the respective constructed regression prediction models;
  • the distribution module 303 is configured to receive the side weight of the corresponding joint bus station transmitted by the central server, calculate the product of the side weight and the model parameters of the regression prediction model as a calculation result, and generate multiple random numbers based on the calculation result , Distribute the random number to each of the joint servers;
  • the joint data transmission module 304 is configured to transmit the received random number to a central server, so that the central server uses the model gradient information and the random number to calculate model update parameters;
  • the joint iterative update module 305 is configured to, when the model update parameter transmitted by the central server is received, iterative regression prediction model based on the model update parameter until the preset stopping condition is reached to obtain the trained regression prediction model.
  • the central server establishes an undirected network of bus stations through the location information of the bus stations, so that it is convenient to calculate the qualified joint bus stations through the undirected network of bus stations.
  • the joint bus stop determined by the present application through the undirected network calculation of the bus stop has a stronger correlation with the target bus stop.
  • the joint server determines the passenger flow of the bus station based on the target time window through the target time window, and builds their own regression prediction models based on the passenger flow in the target time window to realize the short-term prediction model construction of the bus.
  • the joint server multiplies the parameters of the constructed regression prediction model with the corresponding side weights of the joint bus station, and generates random numbers for distribution, which protects the privacy of joint bus station data and prevents leakage during transmission.
  • the central server updates the parameters based on the model gradient information transmitted by the target server and the random number comprehensive calculation model transmitted by the joint server, which realizes the joint learning between the target bus station and the joint bus station with strong correlation with the target bus station, which is obtained through iteration
  • the regression prediction model after the final training improves the accuracy of the prediction of the regression prediction model.
  • the network construction module 101 includes a node sub-module, a connection sub-module, an edge weight sub-module, and a threshold sub-module.
  • the node sub-module is used to set the bus station as a node of the undirected network of the bus station based on the location information of the bus station; the connection sub-module is used to connect any two bus stations as the bus station.
  • the edge of the undirected network is used to obtain the undirected network of the bus station;
  • the edge weight sub-module is used to calculate the edge weight S if between the target bus station and each bus station;
  • the threshold sub-module is used to reduce the corresponding edge weight to less than A bus stop with a preset threshold is used as the joint bus stop.
  • , where, a and b are all hyperparameters, and satisfy a+b 1, DIS if is the distance between the target bus station i and any bus station f, and POP if is the target bus station i and bus station f in the preset range The difference in the total population within.
  • the aforementioned target sample construction module 201 is further configured to: based on the first disturbance data, calculate a preset number of days before the target time window at the same time.
  • the target model construction module 202 includes a first initialization sub-module, a first training sub-module, and a first construction sub-module.
  • the first initialization submodule is used to randomly initialize the model parameters of the preset regression prediction model, wherein each day corresponds to a preset regression prediction model; the first training submodule is used for training samples based on the training sample set Train the regression prediction model corresponding to each day
  • the formula of the regression prediction model is: in, Represents the passenger flow of boarding passengers in the same period of the target time window on the mth day predicted by the model; W 0 and W d are the random initialization parameters; X d represents the passenger flow of boarding in the d time window; the first construction
  • the sub-module is used to obtain the model parameters of the constructed regression prediction model until the regression prediction model reaches the preset construction condition, wherein the preset construction condition is: Among them, y m represents the passenger flow during the same period of the target time window on the mth day.
  • the target iterative update module 204 is further configured to calculate the model gradient information of the target bus station according to the following formula:
  • the aforementioned joint sample construction module 301 is further configured to calculate the number of days before the same period of the respective target time window based on the respective corresponding second disturbance data.
  • the passenger flow corresponding to the time window of the preset number D is used as the first input parameter X of the training sample, and the passenger flow of the same period of the target time window is used as the second input parameter Y of the training sample to generate respective The training sample set (X 1 , Y 1 ) (X 2 , Y 2 ) (X 3 , Y 3 )...(X m , Y m ) of the joint bus station, where X m has The feature dimension of is D, D is an integer, and m represents the mth day.
  • the joint model construction module 302 includes a second initialization sub-module, a second training sub-module, and a second construction sub-module.
  • the second initialization submodule is used to randomly initialize the model parameters of the preset regression prediction model, wherein each day corresponds to a preset regression prediction model; the second training submodule is used to based on the training samples in the training sample set Train the regression prediction model corresponding to each day
  • the formula of the regression prediction model is: in, Represents the passenger flow of boarding passengers in the same period of the target time window on the mth day predicted by the model; W 0 and W d are the random initialization parameters; X d represents the passenger flow of boarding in the d-th time window; the second construction
  • the sub-module is used to obtain the model parameters of the respective constructed regression prediction model until the regression prediction model reaches the preset construction condition, wherein the preset construction condition is: Among them, y m represents the passenger flow during the same period of the target time window on the
  • the aforementioned parameter calculation module 102 is further configured to: calculate the model update parameters by using the following formula:
  • N(i) is the set of the joint bus stations, Is the sum of all random numbers received, ⁇ is the learning rate, Is the model gradient information.
  • FIG. 5 is a block diagram of the basic structure of the computer device in this embodiment.
  • the computer device 400 includes a memory 401, a processor 402, and a network interface 403 that are connected to each other in communication via a system bus. It should be pointed out that only the computer device 400 with components 401-403 is shown in the figure, but it should be understood that it is not required to implement all of the illustrated components, and more or fewer components may be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculation and/or information processing in accordance with pre-set or stored instructions.
  • Its hardware includes, but is not limited to, a microprocessor, a dedicated Integrated Circuit (Application Specific Integrated Circuit, ASIC), Programmable Gate Array (Field-Programmable Gate Array, FPGA), Digital Processor (Digital Signal Processor, DSP), embedded equipment, etc.
  • ASIC Application Specific Integrated Circuit
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • DSP Digital Processor
  • the computer device may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server.
  • the computer device can interact with the user through a keyboard, a mouse, a remote control, a touch panel, or a voice control device.
  • the memory 401 includes at least one type of readable storage medium, the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static memory Random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disk, optical disk, etc.
  • the computer-readable storage medium may be non-volatile or volatile.
  • the memory 401 may be an internal storage unit of the computer device 400, such as a hard disk or a memory of the computer device 400.
  • the memory 401 may also be an external storage device of the computer device 400, such as a plug-in hard disk, a smart media card (SMC), and a secure digital device equipped on the computer device 400. (Secure Digital, SD) card, Flash Card, etc.
  • the memory 401 may also include both the internal storage unit of the computer device 400 and the external storage device thereof.
  • the memory 401 is generally used to store an operating system and various application software installed in the computer device 400, such as computer-readable instructions of a method for predicting passenger flow at a short-term bus station.
  • the memory 401 can also be used to temporarily store various types of data that have been output or will be output.
  • the processor 402 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips.
  • the processor 402 is generally used to control the overall operation of the computer device 400.
  • the processor 402 is configured to run computer-readable instructions or processed data stored in the memory 401, for example, computer-readable instructions for running the method for predicting passenger flow at a short-term bus station.
  • the network interface 403 may include a wireless network interface or a wired network interface, and the network interface 403 is generally used to establish a communication connection between the computer device 400 and other electronic devices.
  • the joint learning between the target bus station and the joint bus station that has a strong correlation with the target bus station is realized, and the accuracy of the prediction of the regression prediction model is improved.
  • the present application also provides another implementation manner, that is, a computer-readable storage medium is provided with computer-readable instructions stored thereon, and the computer-readable instructions can be executed by at least one processor to The at least one processor is made to execute the steps of the method for predicting passenger flow at a short-term bus station as described above.
  • the joint learning between the target bus station and the joint bus station that has a strong correlation with the target bus station is realized, and the accuracy of the prediction of the regression prediction model is improved.
  • the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to make a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods described in the various embodiments of the present application.
  • a terminal device which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Tourism & Hospitality (AREA)
  • Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Mathematical Physics (AREA)
  • Computational Mathematics (AREA)
  • Operations Research (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Quality & Reliability (AREA)
  • Probability & Statistics with Applications (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Algebra (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Educational Administration (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Traffic Control Systems (AREA)

Abstract

一种短时公交站客流预测方法、装置、计算机设备及存储介质，属于人工智能技术领域，应用于智慧交通领域中，包括：接收公交站位置信息和目标公交站，构建公交站无向网络，确定公交站无向网络中符合预设条件的公交站，作为联合公交站；接收目标公交站对应的目标服务器传输的模型梯度信息和联合公交站对应的联合服务器传输的随机数；基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。其中，训练后的回归预测模型可存储于区块链中。所述方法有效提高回归预测模型对短时公交车客流的预测效果。

Description

短时公交站客流预测方法、装置、计算机设备及存储介质
本申请要求于2020年10月29日提交中国专利局、申请号为202011182335.1，发明名称为“短时公交站客流预测方法、装置、计算机设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域，尤其涉及短时公交站客流预测方法、装置、计算机设备及存储介质。
背景技术
公共交通对提供交通运力，缓解交通拥堵起着重要作用，然而目前国内大部分城市的公共交通运力不足，特别是高峰时段的城市公交运力不足，于是预估公交站上车客流量的各种方法应运而生，通过对公交站上车客流量的预估，实现交通运力的及时调整，甚至可以将预估结果发送至相关用户的客户端，便于人们的出行安排等。
目前，有关于通过回归预测模型预估公交站上车客流量的方法包括基于单个公交站点或者区域公交站进行公交车客流预测。发明人意识到，对于单个公交站点的方案来说，训练样本较为单一，训练效果较差，未能考虑周边公交站的客流对目标公交站的影响。对于区域公交站的方案来说，仅基于区域来确定公交站，确定的公交站与目标公交站之间的相关性较弱，导致预测效果较差。
发明内容
本申请实施例的目的在于提出一种短时公交站客流预测方法、装置、计算机设备及存储介质，提升了回归预测模型对于短时公交车客流预测的准确性。
第一方面，本申请实施例提供一种短时公交站客流预测方法，采用了如下所述的技术方案：
一种短时公交站客流预测方法，应用于中心服务器，包括下述步骤：
接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器；
接收目标公交站对应的目标服务器传输的模型梯度信息和联合公交站对应的联合服务器传输的随机数，并基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
第二方面，本申请实施例提供一种短时公交站客流预测方法，采用了如下所述的技术方案：
一种短时公交站客流预测方法，应用于联合服务器，包括下述步骤：
接收目标时间窗和对应的联合公交站的乘客乘车信息，基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合；
基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数；
接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器；
所有所述联合服务器将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述梯度信息和所述随机数计算模型更新参数；
当目标服务器和各个联合服务器接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
第三方面，本申请实施例还提供一种短时公交站客流预测装置，采用了如下所述的技术方案：
一种短时公交站客流预测装置，所述短时公交站客流预测装置包括中心服务器，所述中心服务器包括：
网络构建模块，用于接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器；
参数计算模块，用于接收目标公交站对应的目标服务器传输的模型梯度信息，和联合公交站对应的联合服务器传输的随机数，并基于所述模型参数和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
第五方面，本申请实施例还提供一种短时公交站客流预测装置，采用了如下所述的技术方案：
一种短时公交站客流预测装置，所述短时公交站客流预测装置包括联合服务器，所述联合服务器包括：
联合样本构建模块，用于接收目标时间窗和对应的联合公交站的乘客乘车信息，基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合；
联合模型构建模块，用于基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得各自的构建后的回归预测模型的模型参数；
分发模块，用于接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器；
联合数据传输模块，用于将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述梯度信息和所述随机数计算模型更新参数；
联合迭代更新模块，用于当接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
第六方面，本申请实施例还提供一种计算机设备，采用了如下所述的技术方案：
一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现如下所述的短时公交站客流预测方法的步骤：
接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器；
接收目标公交站对应的目标服务器传输的模型梯度信息和联合公交站对应的联合服务器传输的随机数，并基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现如下所述的短时公交站客流预测方法的步骤：
接收目标时间窗和对应的联合公交站的乘客乘车信息，基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合；
基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数；
接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器；
所有所述联合服务器均将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述梯度信息和所述随机数计算模型更新参数；
当接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
第七方面，本申请实施例还提供一种计算机可读存储介质，采用了如下所述的技术方案：
一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如下所述的短时公交站客流预测方法的步骤：
接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器；
接收目标公交站对应的目标服务器传输的模型梯度信息和联合公交站对应的联合服务器传输的随机数，并基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如下所述的短时公交站客流预测方法的步骤：
接收目标时间窗和对应的联合公交站的乘客乘车信息，基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合；
基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数；
接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器；
所有所述联合服务器均将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述梯度信息和所述随机数计算模型更新参数；
当接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
与现有技术相比，本申请实施例主要有以下有益效果：
中心服务器通过公交站位置信息建立公交站无向网络，便于后续通过公交站无向网络来计算出符合条件的联合公交站。相对于仅是通过区域划分确定联合学习的公交站的方式来说，本申请通过公交站无向网络计算从而确定的联合公交站与目标公交站的相关联性较强。联合服务器均通过目标时间窗确定公交站基于目标时间窗的客流量，基于目标时间窗内的客流量构建各自的回归预测模型，实现公交车的短时预测的模型构建。联合服务器将构建后的回归预测模型的参数与联合公交站对应的边权相乘，并生成随机数进行分发，保护了联合公交站数据的隐私性，防止在传输过程中泄露。中心服务器基于目标服务器传输的模型梯度信息和联合服务器传输的随机数综合计算模型更新参数，实现了目标公交站和与目标公交站相关性较强的联合公交站之间的联合学习，通过迭代获得最终训练后的回归预测模型，提升了回归预测模型预测的准确性。
附图说明
为了更清楚地说明本申请中的方案，下面将对本申请实施例描述中所需要使用的附图作一个简单介绍，显而易见地，下面描述中的附图是本申请的一些实施例，对于本领域普 通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本申请可以应用于其中的示例性系统架构图；
图2是根据本申请的短时公交站客流预测方法的一个实施例的流程图；
图3是根据本申请的短时公交站客流预测方法的公交站无向网络的结构示意图；
图4是根据本申请的短时公交站客流预测装置的一个实施例的结构示意图；
图5是根据本申请的计算机设备的一个实施例的结构示意图。
附图标记：101、网络构建模块；102、参数计算模块；201、目标样本构建模块；202、目标模型构建模块；203、目标数据传输模块；204、目标迭代更新模块；301、联合样本构建模块；302、联合模型构建模块；303、分发模块；304、联合数据传输模块；305、联合迭代更新模块。
具体实施方式
除非另有定义，本文所使用的所有的技术和科学术语与属于本申请的技术领域的技术人员通常理解的含义相同；本文中在申请的说明书中所使用的术语只是为了描述具体的实施例的目的，不是旨在于限制本申请；本申请的说明书和权利要求书及上述附图说明中的术语“包括”和“具有”以及它们的任何变形，意图在于覆盖不排他的包含。本申请的说明书和权利要求书或上述附图中的术语“第一”、“第二”等是用于区别不同对象，而不是用于描述特定顺序。
在本文中提及“实施例”意味着，结合实施例描述的特定特征、结构或特性可以包含在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例，也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是，本文所描述的实施例可以与其它实施例相结合。
为了使本技术领域的人员更好地理解本申请方案，下面将结合附图，对本申请实施例中的技术方案进行清楚、完整地描述。
本申请实施例的短时公交客流预测方法应用于短时公交客流预测系统中。如图1所示，系统架构包括中心服务器、目标服务器、联合服务器、第一客户端和第二客户端。目标服务器与中心服务器之间、联合服务器与中心服务器之间、第一客户端与目标服务器之间、第二客户端与联合服务器之间均通过网络进行连接，网络可以包括各种连接类型，例如有线、无线通信链路或者光纤电缆等等。
用户可以使用第一客户端或第二客户端通过网络与目标服务器或联合服务器交互，以接收或发送消息等。第一客户端和第二客户端上可以安装有各种通讯客户端应用，例如网页浏览器应用、购物类应用、搜索类应用、即时通信工具、邮箱客户端、社交平台软件等。
第一客户端和第二客户端可以是具有显示屏并且支持网页浏览的各种电子设备，包括但不限于智能手机、平板电脑、电子书阅读器、MP3播放器(Moving Picture Experts Group Audio Layer III，动态影像专家压缩标准音频层面3)、MP4(Moving Picture Experts Group Audio Layer IV，动态影像专家压缩标准音频层面4)播放器、膝上型便携计算机和台式计算机等等。
应该理解，图1中的中心服务器、目标服务器、联合服务器、第一客户端和第二客户端的数目仅仅是示意性的。根据实现需要，可以具有任意数目的中心服务器、目标服务器、联合服务器、第一客户端和第二客户端。
继续参考图2，示出了根据本申请的短时公交站客流预测方法的一个实施例的流程图。所述的短时公交站客流预测方法，包括以下步骤：
S1：中心服务器接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器。
在本实施例中，可以根据实际需要进行调整。基于所述目标公交站确定所述公交站无向网络中的联合公交站，提升后续联合公交站的相关数据与目标公交站的数据之间的相关性。其中，目标公交站和每个联合公交站分别对应一个数据中心，后续各公交站基于训练样本的模型构建过程，在各自的数据中心完成，避免数据泄露。
在本实施例中，短时公交站客流预测方法运行于其上的电子设备(例如图1所示的中心服务器)可以通过有线连接方式或者无线连接方式接收公交站位置信息和目标公交站。需要指出的是，上述无线连接方式可以包括但不限于3G/4G连接、WiFi连接、蓝牙连接、WiMAX连接、Zigbee连接、UWB(ultra wideband)连接、以及其他现在已知或将来开发的无线连接方式。
具体的，在步骤S1中，即所述基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站的步骤包括：
基于所述公交站位置信息将所述公交站作为所述公交站无向网络的节点；
将任意两个公交站均进行连线，作为所述公交站无向网络的边，获得所述公交站无向网络；
计算所述目标公交站与各公交站之间的边权S if；
将对应的边权小于预设阈值的公交站作为所述联合公交站。
在本实施例中，公交站为网络的节点，任意两公交站的连线为网络的边，通过边权确定联合公交站，联合公交站即为与目标公交站i关系较为密切的公交站，便于后续与目标公交站联合学习，避免了单个公交站点训练数据维度低，预测结果不准确的情况。
如图3所示，图3是根据本申请的短时公交站客流预测方法的公交站无向网络的结构示意图。其中，通过如下公式计算所述边权S if：
S if＝a·DIS if+b·|POP if|
其中，a，b均为超参数，且满足a+b＝1，DIS if为所述目标公交站i与任意公交站f之间的距离，POP if为目标公交站i与公交站f在预设范围内的人口总数差值为。
在本实施例中，POP if为目标公交站i与公交站f在预设范围内的人口总数差值。其中，所述预设范围为以公交站为圆心，N(km)为半径的圆形覆盖区域，N为自定义数值。人口总数＝人口密度×预设范围的面积，其中，人口密度即表示单位土地面积上的人口数量，可通过调用浏览器查询获得。通过上述公式计算边权，充分考虑了人口密度的情况，而不仅仅只是考虑上车客流量，扩展了后续模型更新参数的维度。
S2：第一客户端通过预设的本地差分隐私算法，扰动本地存储的目标公交站的乘客乘车信息，生成第一扰动数据，并将第一扰动数据发送至目标服务器。
在本实施例中，本申请中乘客乘车信息指乘客搭公交车时的对应的公交站和上车时间,需要加密的内容为乘客是否在当前公交站上车,每个记录都有其对应的时间信息。用户刷卡记录存储在自己手机，银行卡中，第一客户端通过本地差分隐私算法扰动目标公交站的乘客乘车信息，保护了用户的隐私，避免了直接将用户的刷卡记录发送至目标服务器。目标服务器根据扰动数据估算上车总客流量。具体为：对于真实集合中的每一个数据都添加扰动，即添加一个概率计算，形成概率集合，将概率集合发送至目标服务器。目标服务器根据所述概率集合随机生成y′ ij＝0或者1。将随机生成的y ij’通过如下公式计算即可得到公交站的总上车人数
Figure PCTCN2020134955-appb-000001
其中，j指第j个用户，通过时间窗可以筛选扰动数据，进而通过上述方式获得时间窗对应的上车人数，即上车客流量。
具体的，通过如下公式计算所述第一扰动数据：
Figure PCTCN2020134955-appb-000002
P(y′ ij＝0)＝1-P(y′ ij＝1)
其中，y ij∈[0,1]，1表示用户在目标公交站刷卡上车，0表示用户未在目标公交站刷卡上车，ε为超参数。
在本实施例中，本申请通过上述公式对乘客乘车信息添加扰动，实现数据中心既无法获知实际数据，保护了用户的隐私，又可以根据扰动数据估算上车客流量，进行后续计算。
S3：所述目标公交站对应的目标服务器接收目标时间窗和第一客户端发送的第一扰动数据，基于所述第一扰动数据和所述目标时间窗构建所述目标公交站的训练样本集合。
在本实施例中，本申请中目标时间窗是自定义的。目标服务器基于目标时间窗构建训练样本集合，使得训练样本集合与目标时间窗具有相关性，从而能实现后续的通过训练后的回归预测模型准确预测目标时间窗内的上车客流量。
具体的，在步骤S3中，即所述基于所述第一扰动数据和所述目标时间窗构建公交站的训练样本集合的步骤包括：
基于所述第一扰动数据，计算在预设天数中，所述目标时间窗同时段之前的预设个数D的时间窗对应的上车客流量，作为训练样本的第一输入参数X，所述目标时间窗同时段的上车客流量，作为训练样本的第二输入参数Y，生成训练样本集合(X 1，Y 1)(X 2，Y 2)(X 3，Y 3)......(X m，Y m),其中，所述X m具有的特征维数为D，D为整数，m表示第m天。
在本实施例中，以所述目标公交站点在n天中与目标时间窗同时段之前的D个时间窗内的公交客流量作为训练样本的特征，即第一输入参数X，以所述目标区域公交站点在n天中与目标时间窗同时段的上车客流量作为训练样本的第二输入参数Y，构建训练样本{(X 1,Y 1),(X 2,Y 2),…,(X m,Y m),…(X n,Y n)},每一个样本(X m,Y m)都包含两个输入参数，其中，X m具有的特征维数为D，D为整数。
以下以D为2，即特征维度数为2进行举例说明：
第一天的下午1点至2点的客流量和下午2点至3点的客流量组成X1(2维向量，包括两个客流量)，第一天的下午3点至4点的客流量为Y1,(X 1,Y 1)组成一对训练样本；第二天的下午1点至2点的客流量和下午2点至3点的客流量组成X2(2维向量，包括两个客流量)，第二天的下午3点至4点的客流量为Y2,(X 2,Y 2)组成一对训练样本。第m天的下午1点至2点的客流量和下午2点至3点的客流量组成X m(2维向量，包括两个客流量)，第m天的下午3点至4点的客流量为Y m，(X m,Y m)组成一对训练样本。
S4：目标服务器基于所述训练样本集合构建所述目标公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数。
在本实施例中，构建目标公交站的回归预测模型，完成回归预测模型的初步训练，获得构建后的回归预测模型的模型参数，便于后续的进一步处理。
具体的，在步骤S4中，所述目标服务器基于所述训练样本集合构建所述目标公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数的步骤包括：
目标服务器随机初始化预设的回归预测模型的模型参数，其中，每一天分别对应一个预设的回归预测模型；
基于所述训练样本集合中的训练样本分别训练每天对应的回归预测模型
Figure PCTCN2020134955-appb-000003
其中，所述回归预测模型的公式为：
Figure PCTCN2020134955-appb-000004
其中，
Figure PCTCN2020134955-appb-000005
表示模型预测的第m天的目标时间窗同时段的上车客流量；W 0和W d为所述随机初始化参数；X d表示在第d个时间窗中的上车客流量；
直至所述回归预测模型达到预设的构建条件，获得构建后的回归预测模型的模型参数，其中，所述预设的构建条件为：
Figure PCTCN2020134955-appb-000006
其中，y m表示第m天的目标时间窗同时段的上车客流量。
在本实施例中，argmin指使其后面的式子达到最小值时的变量的取值，本申请通过该构建条件，实现基于各公交站对应的训练样本对各公交站的回归预测模型的初步训练。
S5：目标服务器基于所述模型参数计算目标公交站的模型梯度信息，并将所述模型梯度信息传输至中心服务器。
在本实施例中，根据如下公式计算所述目标公交站的模型梯度信息：
Figure PCTCN2020134955-appb-000007
Figure PCTCN2020134955-appb-000008
是目标公交站i在t次迭代时候的模型参数梯度，d表示维度，即第d个时间窗。
S6：第二客户端分别通过预设的本地差分隐私算法，扰动本地存储的联合公交站的乘客乘车信息，分别生成第二扰动数据，并将第二扰动数据发送至各自对应的联合服务器。
在本实施例中，本申请中乘客乘车信息指乘客搭公交车时的对应的公交站和上车时间,需要加密的内容为乘客是否在当前公交站上车,每个记录都有其对应的时间信息。用户刷卡记录存储在自己手机，银行卡中，第二客户端通过本地差分隐私算法扰动目标公交站的乘客乘车信息，保护了用户的隐私，避免了直接将用户的刷卡记录发送至联合服务器。联合服务器根据扰动数据估算上车总客流量。具体为：对于真实集合中的每一个数据都添加扰动，即添加一个概率计算，形成概率集合，将概率集合发送至联合服务器。联合服务器根据所述概率集合随机生成y′ ij＝0或者1。将随机生成的y ij’通过如下公式计算即可得到公交站的总上车人数
Figure PCTCN2020134955-appb-000009
其中，j指第j个用户，通过时间窗可以筛选扰动数据，进而通过上述方式获得时间窗对应的上车人数，即上车客流量。
具体的，通过如下公式生成第二扰动数据：
Figure PCTCN2020134955-appb-000010
P(y′ ij＝0)＝1-P(y′ ij＝1)
其中，y ij∈[0,1]，1表示用户在当前联合公交站刷卡上车，0表示用户未在当前联合公交站刷卡上车，ε为超参数。
在本实施例中，本申请通过上述公式对乘客乘车信息添加扰动，实现数据中心既无法获知实际数据，保护了用户的隐私，又可以根据扰动数据估算上车客流量，进行后续计算。
S7：所述联合公交站对应的联合服务器接收目标时间窗和各自对应的第二扰动数据，基于所述第二扰动数据和所述目标时间窗构建各自的联合公交站的训练样本集合。
在本实施例中，本申请中目标时间窗是自定义的。联合服务器分别基于目标时间窗构建各自的训练样本集合，使得训练样本集合与目标时间窗具有相关性，从而能实现后续的通过训练后的回归预测模型准确预测目标时间窗内的上车客流量。
具体的，在步骤S7中，即所述基于所述第二扰动数据和所述目标时间窗构建各自的联合公交站的训练样本集合的步骤包括：
基于各自对应的所述第二扰动数据，计算在预设天数中，各自所述目标时间窗同时段之前的预设个数D的时间窗对应的上车客流量，作为训练样本的第一输入参数X，所述目标时间窗同时段的上车客流量，作为训练样本的第二输入参数Y，生成各自的联合公交站的训练样本集合(X 1，Y 1)(X 2，Y 2)(X 3，Y 3)......(X m，Y m),其中，所述X m具有的特征维数为D，D为整数，m表示第m天。
在本实施例中，以所述目标公交站点在n天中与目标时间窗同时段之前的D个时间窗内的公交客流量作为训练样本的特征，即第一输入参数X，以所述目标区域公交站点在n 天中与目标时间窗同时段的上车客流量作为训练样本的第二输入参数Y，构建训练样本{(X 1,Y 1),(X 2,Y 2),…,(X m,Y m),…(X n,Y n)},每一个样本(X m,Y m)都包含两个输入参数，其中，X m具有的特征维数为D，D为整数。
以下以D为2，即特征维度数为2进行举例说明：
第一天的下午1点至2点的客流量和下午2点至3点的客流量组成X1(2维向量，包括两个客流量)，第一天的下午3点至4点的客流量为Y1,(X 1,Y 1)组成一对训练样本；第二天的下午1点至2点的客流量和下午2点至3点的客流量组成X2(2维向量，包括两个客流量)，第二天的下午3点至4点的客流量为Y2,(X 2,Y 2)组成一对训练样本。第m天的下午1点至2点的客流量和下午2点至3点的客流量组成X m(2维向量，包括两个客流量)，第m天的下午3点至4点的客流量为Y m，(X m,Y m)组成一对训练样本。
S8：联合服务器基于各自的联合公交站的训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得各自对应的构建后的回归预测模型的模型参数。
在本实施例中，各自构建联合公交站的回归预测模型，完成回归预测模型的初步训练，获得各自对应的构建后的回归预测模型的模型参数，便于后续的进一步处理。
具体的，在步骤S8中，即所述联合服务器基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数的步骤包括：
联合服务器各自基于所述联合公交站的训练样本集合构建各自联合公交站的回归预测模型，直至达到预设的构建条件，获得各自对应的构建后的联合公交站的回归预测模型的模型参数的步骤包括：
联合服务器各自随机初始化预设的回归预测模型的模型参数，其中，每一天分别对应一个预设的回归预测模型；
联合服务器各自基于所述训练样本集合中的训练样本分别训练每天对应的回归预测模型
Figure PCTCN2020134955-appb-000011
其中，所述回归预测模型的公式为：
Figure PCTCN2020134955-appb-000012
其中，
Figure PCTCN2020134955-appb-000013
表示模型预测的第m天的目标时间窗同时段的上车客流量；W 0和W d为所述随机初始化参数；X d表示在第d个时间窗中的上车客流量；
直至所述回归预测模型达到预设的构建条件，获得各自对应的构建后的回归预测模型的模型参数，其中，所述预设的构建条件为：
Figure PCTCN2020134955-appb-000014
其中，y m表示第m天的目标时间窗同时段的上车客流量。
S9：联合服务器接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器。
在本实施例中，各个联合服务器分别计算对应联合公交站的边权与对应的联合公交站在t次迭代时的模型参数之积
Figure PCTCN2020134955-appb-000015
并将边权与模型参数之积
Figure PCTCN2020134955-appb-000016
由密钥分享算法随机生成k个随机数分配给各个联合服务器。其中，所述k个数的和等于
Figure PCTCN2020134955-appb-000017
所有联合服务器接收到的随机数之和即为
Figure PCTCN2020134955-appb-000018
通过随机数的方式保护了联合公交站中的模型参数，提高了数据的隐私性。
S10：所有所述联合服务器将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述模型梯度信息和所述随机数计算模型更新参数。
在本实施例中，所有所述联合服务器将接收到的随机数传输至中心服务器，所有联合 服务器接收到的随机数之和即为
Figure PCTCN2020134955-appb-000019
通过传输随机数的方式提高数据隐私性，避免了传输过程中数据的泄露和篡改。
S11：中心服务器接收目标公交站对应的目标服务器传输的模型梯度信息和联合公交站对应的联合服务器传输的随机数，并基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
在本实施例中，通过目标公交站的模型梯度信息和联合公交站的随机数计算模型更新参数。其中，由上述可知，联合公交站是通过边权确定，与目标公交站有较强的关联性，避免了仅通过单个公交站点或区域公交站简单进行模型更新参数的计算，而导致训练结果不佳，预测效果差的情况。本申请通过目标公交站和与目标公交站之间具有符合边权筛选条件的联合公交站联邦训练，综合计算模型更新参数，提升了模型的预测效果。
需要说明的是：本申请的联合公交站区别于区域公交站，区域公交站仅考虑公交站之间的地理位置关系，并未考虑公交站之间在一定区域内的人口数量关系。
具体的，在步骤S11中，即所述基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器的步骤包括：
所述模型更新参数的计算公式为：
Figure PCTCN2020134955-appb-000020
其中，N(i)为所述联合公交站组成的集合，
Figure PCTCN2020134955-appb-000021
为接收到的所有随机数之和，α为学习率，
Figure PCTCN2020134955-appb-000022
为模型梯度信息；
将所述模型更新参数传输至目标服务器和各个联合服务器。
在本实施例中，N(i)是与公交站i进行联邦学习的公交站集合，即联合公交站组成的集合。其中，公式中的
Figure PCTCN2020134955-appb-000023
来自于上文提及的由联合服务器传输的随机数之和。α为学习率，本文中设置α为0.001，在实际应用中，可以根据实际需要改变学习率的大小，适用即可。通过此公式可以明确得知模型更新参数的计算方式。
S12：当目标服务器和各个联合服务器接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
在本实施例中，目标服务器和各个联合服务器均接收中心服务器传回的模型更新参数，根据模型更新参数更新各自的回归预测模型，再根据训练样本进行训练，重复执行上述训练步骤，并将训练结果即目标公交站的模型梯度信息和联合公交站的随机数发送至中心服务器，等待接收中心服务器下一次的模型更新参数。如此循环迭代各自的回归预测模型，直到达到预设的迭代次数T，获得训练后的回归预测模型。
需要说明的是，获得的训练后的回归预测模型用于预测目标公交站在目标时间窗同时段的客流量。训练后的回归预测模型可以传输给中心服务器，也可以存储在目标服务器或联合服务器中直接调用。完成模型训练后，通过中心服务器、目标服务器或联合服务器接收构建好的预测样本，将预测样本输入训练后的回归预测模型，即得到目标公交站在目标时间窗的客流量。例如：要预测某天下午3点至下午4点的客流量Y，在维度D为2时，对应的预测样本X为该天下午1点至2点的客流量和下午2点至3点的客流量。此时把X输进模型即可得到。
中心服务器通过公交站位置信息建立公交站无向网络，便于后续通过公交站无向网络来计算出符合条件的联合公交站。相对于仅是通过区域划分确定联合学习的公交站的方式来说，本申请通过公交站无向网络计算从而确定的联合公交站与目标公交站的相关联性较强。联合服务器均通过目标时间窗确定公交站基于目标时间窗的客流量，基于目标时间窗内的客流量构建各自的回归预测模型，实现公交车的短时预测的模型构建。联合服务器将 构建后的回归预测模型的参数与联合公交站对应的边权相乘，并生成随机数进行分发，保护了联合公交站数据的隐私性，防止在传输过程中泄露。中心服务器基于目标服务器传输的模型梯度信息和联合服务器传输的随机数综合计算模型更新参数，实现了目标公交站和与目标公交站相关性较强的联合公交站之间的联合学习，通过迭代获得最终训练后的回归预测模型，提升了回归预测模型预测的准确性。
需要强调的是，为进一步保证上述训练后的回归预测模型的私密和安全性，上述训练后的回归预测模型还可以存储于一区块链的节点中。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
本申请可应用于智慧交通领域中，从而推动智慧城市的建设。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，该计算机可读指令可存储于一计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等非易失性存储介质，或随机存储记忆体(Random Access Memory，RAM)等。
应该理解的是，虽然附图的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，其可以以其他的顺序执行。而且，附图的流程图中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，其执行顺序也不必然是依次进行，而是可以与其他步骤或者其他步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
进一步参考图3，作为对上述图2所示方法的实现，本申请提供了一种短时公交站客流预测装置的一个实施例，该装置实施例与图2所示的方法实施例相对应，该装置具体可以应用于各种电子设备中。
如图3所示，本实施例所述的短时公交站客流预测装置包括：中心服务器、目标服务器和联合服务器、第一客户端和第二客户端。
所述中心服务器包括网络构建模块101和参数计算模块102。
网络构建模块101，用于接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器；
参数计算模块102，用于接收目标公交站对应的目标服务器传输的模型梯度信息，和联合公交站对应的联合服务器传输的随机数，并基于所述模型参数和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
第一客户端用于通过预设的本地差分隐私算法，扰动本地存储的目标公交站的乘客乘车信息，生成第一扰动数据，并将第一扰动数据发送至目标服务器。
所述目标服务器包括：目标样本构建模块201、目标模型构建模块202、目标数据传输模块203和目标迭代更新模块204。
目标样本构建模块201，用于接收目标时间窗和第一客户端发送的第一扰动数据，基于所述第一扰动数据和所述目标时间窗构建目标公交站的训练样本集合；
目标模型构建模块202，用于基于所述训练样本集合构建所述目标公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数；
目标数据传输模块203，用于基于所述模型参数计算目标公交站的模型梯度信息，并将所述模型梯度信息传输至中心服务器；
目标迭代更新模块204，用于当接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型；
第二客户端用于分别通过预设的本地差分隐私算法，扰动本地存储的联合公交站的乘客乘车信息，分别生成第二扰动数据，并将第二扰动数据发送至各自对应的联合服务器。
所述联合服务器包括：联合样本构建模块301、联合模型构建模块302、分发模块303、联合数据传输模块304和联合迭代更新模块305。
联合样本构建模块301，用于接收目标时间窗和各自对应的各自对应的第二扰动数据，基于所述第二扰动数据和所述目标时间窗构建各自的联合公交站的训练样本集合；
联合模型构建模块302，用于基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得各自的构建后的回归预测模型的模型参数；
分发模块303，用于接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器；
联合数据传输模块304，用于将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述模型梯度信息和所述随机数计算模型更新参数；
联合迭代更新模块305，用于当接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
中心服务器通过公交站位置信息建立公交站无向网络，便于后续通过公交站无向网络来计算出符合条件的联合公交站。相对于仅是通过区域划分确定联合学习的公交站的方式来说，本申请通过公交站无向网络计算从而确定的联合公交站与目标公交站的相关联性较强。联合服务器均通过目标时间窗确定公交站基于目标时间窗的客流量，基于目标时间窗内的客流量构建各自的回归预测模型，实现公交车的短时预测的模型构建。联合服务器将构建后的回归预测模型的参数与联合公交站对应的边权相乘，并生成随机数进行分发，保护了联合公交站数据的隐私性，防止在传输过程中泄露。中心服务器基于目标服务器传输的模型梯度信息和联合服务器传输的随机数综合计算模型更新参数，实现了目标公交站和与目标公交站相关性较强的联合公交站之间的联合学习，通过迭代获得最终训练后的回归预测模型，提升了回归预测模型预测的准确性。
网络构建模块101包括节点子模块、连线子模块、边权子模块和阈值子模块。节点子模块用于基于所述公交站位置信息将所述公交站作为所述公交站无向网络的节点；连线子模块用于将任意两个公交站均进行连线，作为所述公交站无向网络的边，获得所述公交站无向网络；边权子模块用于计算所述目标公交站与各公交站之间的边权S if；阈值子模块用于将对应的边权小于预设阈值的公交站作为所述联合公交站。
在本实施例的一些可选的实现方式中，上述边权子模块进一步用于：通过如下公式计算所述边权S if：S if＝a·DIS if+b·|POP if|，其中，a，b均为超参数，且满足a+b＝1，DIS if为所述目标公交站i与任意公交站f之间的距离，POP if为目标公交站i与公交站f在预设范围内的人口总数差值。
在本实施例的一些可选的实现方式中，上述第一客户端进一步用于通过如下公式计算所述第一扰动数据：
Figure PCTCN2020134955-appb-000024
P(y′ ij＝0)＝1-P(y′ ij＝1)，其中，y ij∈[0,1]，1表示用户在目标公交站刷卡上车，0表示用户未在目标公交站刷卡上车，ε为超参数。
在本实施例的一些可选的实现方式中，上述目标样本构建模块201进一步用于：基于所述第一扰动数据，计算在预设天数中，所述目标时间窗同时段之前的预设个数D的时间 窗对应的上车客流量，作为训练样本的第一输入参数X，所述目标时间窗同时段的上车客流量，作为训练样本的第二输入参数Y，生成训练样本集合(X 1，Y 1)(X 2，Y 2)(X 3，Y 3)......(X m，Y m),其中，所述X m具有的特征维数为D，D为整数，m表示第m天。
目标模型构建模块202包括第一初始化子模块、第一训练子模块和第一构建子模块。第一初始化子模块用于随机初始化预设的回归预测模型的模型参数，其中，每一天分别对应一个预设的回归预测模型；第一训练子模块用于基于所述训练样本集合中的训练样本分别训练每天对应的回归预测模型
Figure PCTCN2020134955-appb-000025
其中，所述回归预测模型的公式为：
Figure PCTCN2020134955-appb-000026
Figure PCTCN2020134955-appb-000027
其中，
Figure PCTCN2020134955-appb-000028
表示模型预测的第m天的目标时间窗同时段的上车客流量；W 0和W d为所述随机初始化参数；X d表示在第d个时间窗中的上车客流量；第一构建子模块用于直至所述回归预测模型达到预设的构建条件，获得构建后的回归预测模型的模型参数，其中，所述预设的构建条件为：
Figure PCTCN2020134955-appb-000029
其中，y m表示第m天的目标时间窗同时段的上车客流量。
在本实施例的一些可选的实现方式中，上述目标迭代更新模块204进一步用于：根据如下公式计算所述目标公交站的模型梯度信息：
Figure PCTCN2020134955-appb-000030
其中，
Figure PCTCN2020134955-appb-000031
是目标公交站i在t次迭代时候的模型参数梯度，d表示维度，即第d个时间窗。
在本实施例的一些可选的实现方式中，上述第二客户端进一步用于：通过如下公式生成第二扰动数据：
Figure PCTCN2020134955-appb-000032
P(y′ ij＝0)＝1-P(y′ ij＝1)，其中，y ij∈[0,1]，1表示用户在当前联合公交站刷卡上车，0表示用户未在当前联合公交站刷卡上车，ε为超参数。
在本实施例的一些可选的实现方式中，上述联合样本构建模块301进一步用于基于各自对应的所述第二扰动数据，计算在预设天数中，各自所述目标时间窗同时段之前的预设个数D的时间窗对应的上车客流量，作为训练样本的第一输入参数X，所述目标时间窗同时段的上车客流量，作为训练样本的第二输入参数Y，生成各自的联合公交站的训练样本集合(X 1，Y 1)(X 2，Y 2)(X 3，Y 3)......(X m，Y m),其中，所述X m具有的特征维数为D，D为整数，m表示第m天。
联合模型构建模块302包括第二初始化子模块、第二训练子模块和第二构建子模块。第二初始化子模块用于随机初始化预设的回归预测模型的模型参数，其中，每一天分别对应一个预设的回归预测模型；第二训练子模块用于基于所述训练样本集合中的训练样本分别训练每天对应的回归预测模型
Figure PCTCN2020134955-appb-000033
其中，所述回归预测模型的公式为：
Figure PCTCN2020134955-appb-000034
Figure PCTCN2020134955-appb-000035
其中，
Figure PCTCN2020134955-appb-000036
表示模型预测的第m天的目标时间窗同时段的上车客流量；W 0和W d为所述随机初始化参数；X d表示在第d个时间窗中的上车客流量；第二构建子模块用于直至所述回归预测模型达到预设的构建条件，获得各自对应的构建后的回归预测模型的模型参数，其中，所述预设的构建条件为：
Figure PCTCN2020134955-appb-000037
其中，y m表示第m天的目标时间窗同时段的上车客流量。
在本实施例的一些可选的实现方式中，上述参数计算模块102进一步用于：通过如下公式计算所述模型更新参数：
Figure PCTCN2020134955-appb-000038
其中，N(i)为所述联合公交站组成的集合，
Figure PCTCN2020134955-appb-000039
为接收到的所有随机数之和，α为学习率，
Figure PCTCN2020134955-appb-000040
为模型梯度信息。
为解决上述技术问题，本申请实施例还提供计算机设备。具体请参阅图5，图5为本实施例计算机设备基本结构框图。
所述计算机设备400包括通过系统总线相互通信连接存储器401、处理器402、网络 接口403。需要指出的是，图中仅示出了具有组件401-403的计算机设备400，但是应理解的是，并不要求实施所有示出的组件，可以替代的实施更多或者更少的组件。其中，本技术领域技术人员可以理解，这里的计算机设备是一种能够按照事先设定或存储的指令，自动进行数值计算和/或信息处理的设备，其硬件包括但不限于微处理器、专用集成电路(Application Specific Integrated Circuit，ASIC)、可编程门阵列(Field－Programmable Gate Array，FPGA)、数字处理器(Digital Signal Processor，DSP)、嵌入式设备等。
所述计算机设备可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。所述计算机设备可以与用户通过键盘、鼠标、遥控器、触摸板或声控设备等方式进行人机交互。
所述存储器401至少包括一种类型的可读存储介质，所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。所述计算机可读存储介质可以是非易失性，也可以是易失性。在一些实施例中，所述存储器401可以是所述计算机设备400的内部存储单元，例如该计算机设备400的硬盘或内存。在另一些实施例中，所述存储器401也可以是所述计算机设备400的外部存储设备，例如该计算机设备400上配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。当然，所述存储器401还可以既包括所述计算机设备400的内部存储单元也包括其外部存储设备。本实施例中，所述存储器401通常用于存储安装于所述计算机设备400的操作系统和各类应用软件，例如短时公交站客流预测方法的计算机可读指令等。此外，所述存储器401还可以用于暂时地存储已经输出或者将要输出的各类数据。
所述处理器402在一些实施例中可以是中央处理器(Central Processing Unit，CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器402通常用于控制所述计算机设备400的总体操作。本实施例中，所述处理器402用于运行所述存储器401中存储的计算机可读指令或者处理数据，例如运行所述短时公交站客流预测方法的计算机可读指令。
所述网络接口403可包括无线网络接口或有线网络接口，该网络接口403通常用于在所述计算机设备400与其他电子设备之间建立通信连接。
在本实施例中，实现了目标公交站和与目标公交站相关性较强的联合公交站之间的联合学习，提升了回归预测模型预测的准确性。
本申请还提供了另一种实施方式，即提供一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令可被至少一个处理器执行，以使所述至少一个处理器执行如上述的短时公交站客流预测方法的步骤。
在本实施例中，实现了目标公交站和与目标公交站相关性较强的联合公交站之间的联合学习，提升了回归预测模型预测的准确性。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本申请各个实施例所述的方法。

Claims (20)

  1. 一种短时公交站客流预测方法，应用于中心服务器，包括下述步骤：
    接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器；
    接收目标公交站对应的目标服务器传输的模型梯度信息和联合公交站对应的联合服务器传输的随机数，并基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
  2. 根据权利要求1所述的短时公交站客流预测方法，其中，所述基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站的步骤包括：
    基于所述公交站位置信息将所述公交站作为所述公交站无向网络的节点；
    将任意两个公交站均进行连线，作为所述公交站无向网络的边，获得所述公交站无向网络；
    计算所述目标公交站与各公交站之间的边权S if；
    将对应的边权小于预设阈值的公交站作为所述联合公交站。
  3. 根据权利要求1所述的短时公交站客流预测方法，其中，所述基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器的步骤包括：
    所述模型更新参数的计算公式为：
    Figure PCTCN2020134955-appb-100001
    其中，N(i)为所述联合公交站组成的集合，
    Figure PCTCN2020134955-appb-100002
    为接收到的所有随机数之和，α为学习率，
    Figure PCTCN2020134955-appb-100003
    为模型梯度信息；
    将所述模型更新参数传输至目标服务器和各个联合服务器。
  4. 一种短时公交站客流预测方法，应用于联合服务器，包括下述步骤：
    接收目标时间窗和对应的联合公交站的乘客乘车信息，基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合；
    基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数；
    接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器；
    所有所述联合服务器均将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述梯度信息和所述随机数计算模型更新参数；
    当接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
  5. 根据权利要求4所述的短时公交站客流预测方法，其中，所述基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合的步骤包括：
    基于所述乘客乘车信息，计算各自的联合公交站在预设天数中，所述目标时间窗同时段之前的预设个数D的时间窗对应的上车客流量，作为训练样本的第一输入参数X，所述目标时间窗同时段的上车客流量，作为训练样本的第二输入参数Y，各自生成联合公交站的训练样本集合(X 1，Y 1)(X 2，Y 2)(X 3，Y 3)......(X m，Y m),其中，所述X m具有的特征维数为D，D为整数，m表示第m天。
  6. 根据权利要求5所述的短时公交站客流预测方法，其中，所述所述基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数的步骤包括：
    随机初始化预设的回归预测模型的模型参数，其中，每一天分别对应一个预设的回归预测模型；
    基于所述联合公交站的训练样本集合中的训练样本分别训练每天对应的回归预测模型
    Figure PCTCN2020134955-appb-100004
    其中，所述回归预测模型的公式为：
    Figure PCTCN2020134955-appb-100005
    其中，
    Figure PCTCN2020134955-appb-100006
    表示模型预测的第m天的目标时间窗同时段的上车客流量；W 0和W d为所述随机初始化参数；X d表示在第d个时间窗中的上车客流量；
    直至所述回归预测模型达到预设的构建条件，获得构建后的回归预测模型的模型参数，其中，所述预设的构建条件为：
    Figure PCTCN2020134955-appb-100007
    其中，y m表示联合公交站第m天的目标时间窗同时段的上车客流量。
  7. 一种短时公交站客流预测装置，所述短时公交站客流预测装置包括中心服务器，所述中心服务器包括：
    网络构建模块，用于接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器；
    参数计算模块，用于接收目标公交站对应的目标服务器传输的模型梯度信息，和联合公交站对应的联合服务器传输的随机数，并基于所述模型参数和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
  8. 一种短时公交站客流预测装置，所述短时公交站客流预测装置包括联合服务器，所述联合服务器包括：
    联合样本构建模块，用于接收目标时间窗和对应的联合公交站的乘客乘车信息，基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合；
    联合模型构建模块，用于基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得各自的构建后的回归预测模型的模型参数；
    分发模块，用于接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器；
    联合数据传输模块，用于将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述梯度信息和所述随机数计算模型更新参数；
    联合迭代更新模块，用于当接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
  9. 一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现如下所述的短时公交站客流预测方法的步骤：
    接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器；
    接收目标公交站对应的目标服务器传输的模型梯度信息和联合公交站对应的联合服 务器传输的随机数，并基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
  10. 根据权利要求9所述的计算机设备，其中，所述基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站的步骤包括：
    基于所述公交站位置信息将所述公交站作为所述公交站无向网络的节点；
    将任意两个公交站均进行连线，作为所述公交站无向网络的边，获得所述公交站无向网络；
    计算所述目标公交站与各公交站之间的边权S if；
    将对应的边权小于预设阈值的公交站作为所述联合公交站。
  11. 根据权利要求9所述的计算机设备，其中，所述基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器的步骤包括：
    所述模型更新参数的计算公式为：
    Figure PCTCN2020134955-appb-100008
    其中，N(i)为所述联合公交站组成的集合，
    Figure PCTCN2020134955-appb-100009
    为接收到的所有随机数之和，α为学习率，
    Figure PCTCN2020134955-appb-100010
    为模型梯度信息；
    将所述模型更新参数传输至目标服务器和各个联合服务器。
  12. 一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现如下所述的短时公交站客流预测方法的步骤：
    接收目标时间窗和对应的联合公交站的乘客乘车信息，基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合；
    基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数；
    接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器；
    所有所述联合服务器均将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述梯度信息和所述随机数计算模型更新参数；
    当接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
  13. 根据权利要求12所述的计算机设备，其中，所述基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合的步骤包括：
    基于所述乘客乘车信息，计算各自的联合公交站在预设天数中，所述目标时间窗同时段之前的预设个数D的时间窗对应的上车客流量，作为训练样本的第一输入参数X，所述目标时间窗同时段的上车客流量，作为训练样本的第二输入参数Y，各自生成联合公交站的训练样本集合(X 1，Y 1)(X 2，Y 2)(X 3，Y 3)......(X m，Y m),其中，所述X m具有的特征维数为D，D为整数，m表示第m天。
  14. 根据权利要求13所述的计算机设备，其中，所述所述基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数的步骤包括：
    随机初始化预设的回归预测模型的模型参数，其中，每一天分别对应一个预设的回归预测模型；
    基于所述联合公交站的训练样本集合中的训练样本分别训练每天对应的回归预测模型
    Figure PCTCN2020134955-appb-100011
    其中，所述回归预测模型的公式为：
    Figure PCTCN2020134955-appb-100012
    其中，
    Figure PCTCN2020134955-appb-100013
    表示模型预测的第m天的目标时间窗同时段的上车客流量；W 0和W d为所述随机初始化参数；X d表示在第d个时间窗中的上车客流量；
    直至所述回归预测模型达到预设的构建条件，获得构建后的回归预测模型的模型参数，其中，所述预设的构建条件为：
    Figure PCTCN2020134955-appb-100014
    其中，y m表示联合公交站第m天的目标时间窗同时段的上车客流量。
  15. 一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如下所述的短时公交站客流预测方法的步骤：
    接收公交站位置信息和目标公交站，基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站，其中，目标公交站对应目标服务器，每个所述联合公交站各自对应一个联合服务器；
    接收目标公交站对应的目标服务器传输的模型梯度信息和联合公交站对应的联合服务器传输的随机数，并基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器，以使所述目标服务器和联合服务器通过所述模型更新参数迭代更新回归预测模型。
  16. 根据权利要求15所述的计算机可读存储介质，其中，所述基于所述公交站位置信息构建公交站无向网络，并基于所述目标公交站确定所述公交站无向网络中符合预设条件的公交站，作为联合公交站的步骤包括：
    基于所述公交站位置信息将所述公交站作为所述公交站无向网络的节点；
    将任意两个公交站均进行连线，作为所述公交站无向网络的边，获得所述公交站无向网络；
    计算所述目标公交站与各公交站之间的边权S if；
    将对应的边权小于预设阈值的公交站作为所述联合公交站。
  17. 根据权利要求15所述的计算机可读存储介质，其中，所述基于所述模型梯度信息和所述随机数计算模型更新参数，并将所述模型更新参数传输至目标服务器和各个联合服务器的步骤包括：
    所述模型更新参数的计算公式为：
    Figure PCTCN2020134955-appb-100015
    其中，N(i)为所述联合公交站组成的集合，
    Figure PCTCN2020134955-appb-100016
    为接收到的所有随机数之和，α为学习率，
    Figure PCTCN2020134955-appb-100017
    为模型梯度信息；
    将所述模型更新参数传输至目标服务器和各个联合服务器。
  18. 一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如下所述的短时公交站客流预测方法的步骤：
    接收目标时间窗和对应的联合公交站的乘客乘车信息，基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合；
    基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数；
    接收中心服务器传输的对应的联合公交站的边权，计算所述边权与回归预测模型的模 型参数之积，作为计算结果，并基于所述计算结果生成多个随机数，将所述随机数分发至各个所述联合服务器；
    所有所述联合服务器均将接收到的随机数传输至中心服务器，以使所述中心服务器利用所述梯度信息和所述随机数计算模型更新参数；
    当接收到所述中心服务器传输的模型更新参数时，基于所述模型更新参数迭代回归预测模型，直至达到预设的停止条件，获得训练后的回归预测模型。
  19. 根据权利要求18所述的计算机可读存储介质，其中，所述基于所述乘客乘车信息和所述目标时间窗构建对应的联合公交站的训练样本集合的步骤包括：
    基于所述乘客乘车信息，计算各自的联合公交站在预设天数中，所述目标时间窗同时段之前的预设个数D的时间窗对应的上车客流量，作为训练样本的第一输入参数X，所述目标时间窗同时段的上车客流量，作为训练样本的第二输入参数Y，各自生成联合公交站的训练样本集合(X 1，Y 1)(X 2，Y 2)(X 3，Y 3)......(X m，Y m),其中，所述X m具有的特征维数为D，D为整数，m表示第m天。
  20. 根据权利要求19所述的计算机可读存储介质，其中，所述所述基于所述训练样本集合构建所述联合公交站的回归预测模型，直至达到预设的构建条件，获得构建后的回归预测模型的模型参数的步骤包括：
    随机初始化预设的回归预测模型的模型参数，其中，每一天分别对应一个预设的回归预测模型；
    基于所述联合公交站的训练样本集合中的训练样本分别训练每天对应的回归预测模型
    Figure PCTCN2020134955-appb-100018
    其中，所述回归预测模型的公式为：
    Figure PCTCN2020134955-appb-100019
    其中，
    Figure PCTCN2020134955-appb-100020
    表示模型预测的第m天的目标时间窗同时段的上车客流量；W0和Wd为所述随机初始化参数；Xd表示在第d个时间窗中的上车客流量；
    直至所述回归预测模型达到预设的构建条件，获得构建后的回归预测模型的模型参数，其中，所述预设的构建条件为：
    Figure PCTCN2020134955-appb-100021
    其中，y m表示联合公交站第m天的目标时间窗同时段的上车客流量。
PCT/CN2020/134955 2020-10-29 2020-12-09 短时公交站客流预测方法、装置、计算机设备及存储介质 Ceased WO2021189950A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011182335.1A CN112288162B (zh) 2020-10-29 2020-10-29 短时公交站客流预测方法、装置、计算机设备及存储介质
CN202011182335.1 2020-10-29

Publications (1)

Publication Number Publication Date
WO2021189950A1 true WO2021189950A1 (zh) 2021-09-30

Family

ID=74353244

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/134955 Ceased WO2021189950A1 (zh) 2020-10-29 2020-12-09 短时公交站客流预测方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN112288162B (zh)
WO (1) WO2021189950A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113919577A (zh) * 2021-10-15 2022-01-11 交控科技股份有限公司 一种轨道交通短时客流预测方法及装置
CN114626889A (zh) * 2022-03-18 2022-06-14 腾讯科技（深圳）有限公司 对象数量预估方法、装置、存储介质、设备及程序产品
CN115115082A (zh) * 2021-12-30 2022-09-27 南通大学 一种基于联邦学习的长距离高速公路交通流量预测方法
CN117033063A (zh) * 2023-10-08 2023-11-10 浪潮(山东)计算机科技有限公司 一种服务器漏液处理方法、系统、装置、电子设备及介质
CN117056785A (zh) * 2023-08-31 2023-11-14 西安电子科技大学 基于自适应模型扰动的联邦学习分类模型训练方法
CN117407774A (zh) * 2023-12-15 2024-01-16 深圳市地铁集团有限公司 一种基于人工智能的交通数据处理方法及系统

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114090407B (zh) * 2021-11-29 2024-11-08 平安科技（深圳）有限公司 基于线性回归模型的接口性能预警方法及其相关设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102169606A (zh) * 2010-02-26 2011-08-31 同济大学 一种城市轨道交通网络大客流影响的预测方法
CN102436603A (zh) * 2011-08-29 2012-05-02 北京航空航天大学 基于概率树d预测的轨道交通全路网客流预测方法
CN105095993A (zh) * 2015-07-22 2015-11-25 济南市市政工程设计研究院（集团）有限责任公司 一种轨道站点客流量预测系统及方法
US20150356458A1 (en) * 2014-06-10 2015-12-10 Jose Oriol Lopez Berengueres Method And System For Forecasting Future Events
CN107563540A (zh) * 2017-07-25 2018-01-09 中南大学 一种基于随机森林的短时公交上车客流量的预测方法
CN111754025A (zh) * 2020-05-25 2020-10-09 苏州大学文正学院 基于cnn+gru的公交短时客流预测方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108281027A (zh) * 2018-03-15 2018-07-13 浙江工业大学 一种交互式智能到站提醒方法及装置
CN110868326B (zh) * 2019-11-27 2022-07-19 武汉虹旭信息技术有限责任公司 网络服务质量的分析方法、边缘设备及中心服务器
CN111310932B (zh) * 2020-02-10 2024-09-27 深圳前海微众银行股份有限公司 横向联邦学习系统优化方法、装置、设备及可读存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102169606A (zh) * 2010-02-26 2011-08-31 同济大学 一种城市轨道交通网络大客流影响的预测方法
CN102436603A (zh) * 2011-08-29 2012-05-02 北京航空航天大学 基于概率树d预测的轨道交通全路网客流预测方法
US20150356458A1 (en) * 2014-06-10 2015-12-10 Jose Oriol Lopez Berengueres Method And System For Forecasting Future Events
CN105095993A (zh) * 2015-07-22 2015-11-25 济南市市政工程设计研究院（集团）有限责任公司 一种轨道站点客流量预测系统及方法
CN107563540A (zh) * 2017-07-25 2018-01-09 中南大学 一种基于随机森林的短时公交上车客流量的预测方法
CN111754025A (zh) * 2020-05-25 2020-10-09 苏州大学文正学院 基于cnn+gru的公交短时客流预测方法

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113919577A (zh) * 2021-10-15 2022-01-11 交控科技股份有限公司 一种轨道交通短时客流预测方法及装置
CN115115082A (zh) * 2021-12-30 2022-09-27 南通大学 一种基于联邦学习的长距离高速公路交通流量预测方法
CN114626889A (zh) * 2022-03-18 2022-06-14 腾讯科技（深圳）有限公司 对象数量预估方法、装置、存储介质、设备及程序产品
CN117056785A (zh) * 2023-08-31 2023-11-14 西安电子科技大学 基于自适应模型扰动的联邦学习分类模型训练方法
CN117033063A (zh) * 2023-10-08 2023-11-10 浪潮(山东)计算机科技有限公司 一种服务器漏液处理方法、系统、装置、电子设备及介质
CN117033063B (zh) * 2023-10-08 2024-02-09 浪潮(山东)计算机科技有限公司 一种服务器漏液处理方法、系统、装置、电子设备及介质
CN117407774A (zh) * 2023-12-15 2024-01-16 深圳市地铁集团有限公司 一种基于人工智能的交通数据处理方法及系统
CN117407774B (zh) * 2023-12-15 2024-03-26 深圳市地铁集团有限公司 一种基于人工智能的交通数据处理方法及系统

Also Published As

Publication number Publication date
CN112288162A (zh) 2021-01-29
CN112288162B (zh) 2024-05-10

Similar Documents

Publication Publication Date Title
CN112288162B (zh) 短时公交站客流预测方法、装置、计算机设备及存储介质
US20250005621A1 (en) Determining locations of interest based on user visits
CN112287244B (zh) 基于联邦学习的产品推荐方法、装置、计算机设备及介质
JP6693502B2 (ja) 情報処理装置、情報処理方法およびプログラム
WO2022126970A1 (zh) 金融欺诈风险识别方法、装置、计算机设备及存储介质
US20230054003A1 (en) Methods and internet of things systems for counting and regulating pedestrian volume in public places of smart cities
US11461858B2 (en) Information processing device, information processing method, and program
CN112508075B (zh) 基于横向联邦的dbscan聚类方法、及其相关设备
CN110110229A (zh) 一种信息推荐方法及装置
US10623888B2 (en) Computing system with crowd prediction mechanism and method of operation thereof
WO2021203728A1 (zh) 一种展业区域的选址方法、装置、计算机设备和介质
CN107908740A (zh) 信息输出方法和装置
CN113868532B (zh) 一种地点推荐方法、装置、电子设备及存储介质
WO2015084678A1 (en) Systems and methods for geo-location based message streams
CN105589939A (zh) 识别群体运动轨迹方法和装置
US20230316395A1 (en) Management methods and systems for affordable housing applications in smart city based on internet of things
CN105893537A (zh) 地理信息点的确定方法和装置
CN113780613A (zh) 一种路径推荐方法和装置
CN110276387B (zh) 一种模型的生成方法及装置
CN113011921B (zh) 信息推送方法、装置、电子设备及可读存储介质
CN114298421A (zh) 用户的兴趣点预测模型的建模、使用方法及相关设备
CN113781247B (zh) 协议数据推荐方法、装置、计算机设备及存储介质
CN111523802B (zh) 送达时间应答方法、装置、设备及介质
CN116631218A (zh) 停车场信息推荐方法、装置、计算机设备及存储介质
Nazareno et al. Changes in Mobile Broadband Infrastructure in Georgia During the COVID− 19 Pandemic

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20927333

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20927333

Country of ref document: EP

Kind code of ref document: A1