CN112328914A - Task allocation method based on space-time crowdsourcing worker behavior prediction - Google Patents

Task allocation method based on space-time crowdsourcing worker behavior prediction Download PDF

Info

Publication number
CN112328914A
CN112328914A CN202011232008.2A CN202011232008A CN112328914A CN 112328914 A CN112328914 A CN 112328914A CN 202011232008 A CN202011232008 A CN 202011232008A CN 112328914 A CN112328914 A CN 112328914A
Authority
CN
China
Prior art keywords
task
crowdsourcing
space
time
worker
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.)
Pending
Application number
CN202011232008.2A
Other languages
Chinese (zh)
Inventor
孟祥福
谢晶
张霄雁
孙德伟
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.)
Liaoning Technical University
Original Assignee
Liaoning Technical University
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 Liaoning Technical University filed Critical Liaoning Technical University
Priority to CN202011232008.2A priority Critical patent/CN112328914A/en
Publication of CN112328914A publication Critical patent/CN112328914A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06311—Scheduling, planning or task assignment for a person or group

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Strategic Management (AREA)
  • Educational Administration (AREA)
  • Game Theory and Decision Science (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Development Economics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a task allocation method based on crowdsourcing worker space-time behavior prediction, which comprises the following steps of: s1, dividing the search space into a series of continuous space-time environments according to space-time characteristics; s2, inputting the spatiotemporal data of crowdsourcing workers and external influence factors into an AutoST spatiotemporal prediction model, and predicting the possible arrival place of the crowdsourcing workers at the next time stamp; s3, acquiring all to-be-completed space-time crowdsourcing task sets in the current environment state, and numbering in sequence; s4, acquiring a space-time crowdsourcing worker set predicted by the AutoST model, and numbering in sequence; and S5, recommending the optimal task for each crowdsourcing worker by adopting an improved greedy algorithm, and implementing a global optimal task allocation strategy. According to the task allocation method based on the crowdsourcing worker time-space behavior prediction, the dynamic characteristics of crowdsourcing workers and tasks are considered, the total travel cost of the crowdsourcing workers is reduced as far as possible under the condition that the time constraint is met, efficient multi-task allocation is achieved, and the quality and efficiency of task allocation are improved.

Description

Task allocation method based on space-time crowdsourcing worker behavior prediction
Technical Field
The invention belongs to the technical field of neural networks and recommendation algorithms, and particularly relates to a task allocation method based on space-time crowdsourcing worker behavior prediction.
Background
In recent years, with the development of mobile internet, rapid popularization of intelligent devices is promoted, space-time crowdsourcing draws people's attention, and gradually becomes an innovative and revolutionary platform, and people can perform crowdsourcing tasks by actually moving to a specified position, such as acquiring real-time road condition information of a road, collecting location information and the like. Emerging crowdsourcing application platforms include taxi service such as dribble and Uber, ordering and delivery service such as American group and Eleme, and dynamic information collection service such as GigWalk. How to assign the best space-time task to the workers still serves as a core hot problem for the research of the space-time crowdsourcing field.
The conventional crowdsourcing task allocation mode is generally assumed to be based on a static scene, the time-space crowdsourcing tasks and the time-space information of workers are completely known before task allocation, and the time-space crowdsourcing tasks and the workers are irregularly and dynamically appeared in a real scene, so that the prediction of the time-space information has great potential for improving the task allocation of a time-space crowdsourcing platform. In addition, the problem of space-time crowdsourcing task allocation is solved, constraint conditions such as task effective time and total travel cost are not comprehensively considered in most task allocation modes based on static scenes, optimal global optimal task allocation strategies cannot be carried out on crowdsourcing workers, idle rates of space-time crowdsourcing tasks and workers are too high, and accordingly the task allocation efficiency is not high.
Disclosure of Invention
In view of the above, the present invention provides a task allocation method based on temporal-spatial crowdsourcing worker behavior prediction, which analyzes the worker behavior trajectory prediction problem by comprehensively considering the temporal dependency and the spatial dependency of temporal-spatial crowdsourcing data, analyzes the global optimal task allocation problem by comprehensively considering constraints such as task effective time and total travel cost, and recommends appropriate tasks for workers in order to improve the quality and efficiency of task allocation.
In order to achieve the purpose, the invention is realized by the following technical scheme:
the invention relates to a task allocation method based on crowdsourcing worker space-time behavior prediction, which is characterized by comprising the following steps of:
step S1, dividing the search space into a series of continuous space-time environment according to space-time characteristics, dividing the area where the task is to be distributed into M multiplied by N grids along longitude and latitude, and controlling the resolution of grid division by M and N according to application requirements;
step S2, inputting the spatio-temporal data of crowdsourcing workers and external influence factors into a high-efficiency neural network learning model (AutoST model) facing spatio-temporal prediction, and predicting the possible arrival place of the crowdsourcing workers at the next time stamp;
step S3, obtaining all to-be-completed space-time crowdsourcing task sets in the current environment state, and numbering in sequence, thereby forming a space-time crowdsourcing task set T ═ { T ═ T {1,t2,...,tn}；tnRepresenting an nth spatiotemporal crowdsourcing task; furthermore, the nth spatial crowdsourcing task contains 4 attributes, namely: t is tn＝<ln,rn,dn,bn>Wherein l isnThe position of the nth crowdsourced task is represented and is taken as a two-dimensional geographic coordinate rnAnd dnRespectively representing the release time of the nth crowdsourcing task and the deadline of the task, taking the values as time points, bnRepresenting the benefit given to the platform when the nth crowdsourcing task is completed;
step S4, obtaining the space-time crowdsourcing worker set predicted by the AutoST model, and numbering the space-time crowdsourcing worker set in sequence to form a space-time crowdsourcing worker set W ═ { W }1,w2,...,wnIn which wnRepresenting the nth spatiotemporal crowdsourcing worker; furthermore, the nth spatiotemporal crowdsourcing worker contains 2 attributes, namely: w is an＝<pn,on>Wherein p isnThe position of the nth space-time crowdsourcing worker is represented, and the value is a two-dimensional geographic coordinate onRepresenting connections of the nth spatiotemporal crowdsourcing workerA single state, 0 represents no order, 1 represents an order that has been accepted;
and step S5, recommending the optimal task for each crowdsourcing worker by adopting an improved greedy algorithm, and implementing a global optimal task allocation strategy.
Further, step S2 includes the following substeps:
s21, designing an efficient space-time search space and dynamically capturing spatial correlation;
s22, inputting the intimacy (Closeness) of crowdsourcing workers and tasks, the movement Period (Period) of crowdsourcing workers and the Trend direction (Trend) into a ternary (CPT) paradigm, and inputting the ternary (CPT) paradigm and external influence factors (such as weather, traffic and the like) into an AutoST model for space-time prediction;
s23, further optimizing the model through a search algorithm;
and S24, calculating accuracy, and selecting the average absolute error and the average absolute percentage error to measure the accuracy of prediction.
Further, step S5 includes the following substeps:
s51, constructing the space-time crowdsourcing task set acquired in the step S3 into a task queue, and finding a new task for a worker from the task queue by an algorithm when the worker arrives or finishes the last task of the worker;
s52, recommending a new task for crowdsourcing workers by using the formulated greedy function, comprehensively considering the effective time of the crowdsourcing task and the distance ratio between the workers and the task, and calculating as follows:
Figure BDA0002765520810000031
where Ben (w, t) represents the recommended quality of the task, dn-rnIndicating the effective time of the task, dis (p)n-ln) The distance between crowdsourcing workers and the task is represented, and under the condition that the upper limit time of the task is similar, the larger the Ben (w, t) value is, the more the task is worth recommending.
Further, step S21 includes the following substeps:
s211, processing a candidate unit; in the space-time prediction, adjacent areas may affect each other, so the convolution operation is very important for establishing a local geographical correlation model; in addition, the spatial correlation of convolution models with different kernel sizes is different, so that different convolution kernels are considered, low convolution is used for capturing local correlation, and high convolution coding is used for obtaining global correlation to conduct crowd-sourced worker behavior prediction;
spatio-temporal information prediction is different from image prediction, and does not require a pooling operation, since the pooling operation may cause information loss, thus deleting the pooling operation, and dividing search candidate units into two categories:
a 3 × 3 standard convolution, a 5 × 5 standard convolution, a 3 × 3 separation convolution, and a 5 × 5 separation convolution, which are composed of convolution operations;
jump connection operations, including connectionless operations and identification operations;
to ensure that the shape of the output is the same as the shape of the input, we use a convolution kernel with the same step size equal to 1 and the filter size of the input and output are the same.
S212, processing by an operation module: in order to solve the problem of low memory efficiency and perform efficient search, the ST-NASONT (system-independent network) key module of the AutoST mainly performs two types of operations, namely a hybrid convolution operation and a hybrid connection operation; the mixed convolution operation searches different convolution kernels in each layer and calculates the weighted sum of all convolution outputs; whether the mixed connection operation is performed or not is learned among different layers, and the connection probability is multiplied by the output of each layer so as to perform fusion of different levels of features;
s213, constructing a neural structure search network: the fixed structure of an external network may influence the prediction performance, and in order to make the network more efficient, the search space is constrained from the following two aspects according to the characteristics of space-time prediction; firstly, only one convolution exists in adjacent layers, and the large-range spatial dependence can be captured; secondly, external feature conversion is not needed when multi-level features are fused, and only the output of the previous layer is added to the current layer; the output of the l-th layer can be expressed as:
Figure BDA0002765520810000041
wherein o isiIs the output of the i-th layer,
Figure BDA0002765520810000042
is a high-level feature of the l-th layer,
Figure BDA0002765520810000043
is the result of fusing the characteristics of the ith layer and the ith layer.
Further, step S22 includes the following substeps:
inputting the intimacy of crowdsourcing workers and tasks, the movement period and trend of crowdsourcing workers and external influence factors into an AutoST model by adopting a CPT paradigm, wherein the intimacy of crowdsourcing workers and tasks is calculated by a Pearson correlation coefficient method; for an AutoST model with L layers, the structure parameters a and model parameters M may be defined as:
Figure BDA0002765520810000051
Figure BDA0002765520810000052
wherein n isc4 is the number of convolution kernels, n s2 represents the number of convolution kernels and connections respectively,
Figure BDA0002765520810000053
is the convolution parameter of the i-th layer,
Figure BDA0002765520810000054
is the convolution parameter for the s-th layer,
Figure BDA0002765520810000055
is the weight of the c unit in the l-th layer, θc1，θc2，θfcParameters of the convolutional layer 1, convolutional layer 2 and fusion characteristic layer respectively;and finally, the output of the space-time prediction search network is as follows:
o＝Relu(f(oL；θc2)+f(xe；θfc))
wherein o isLFor output of AutoST, xeRepresenting an external influencing factor.
Further, step S23 includes the following substeps:
training in AutoST includes two phases, a search phase and a verification phase. In the search phase, the data is first decomposed into a training set and a validation set, and then the training loss L is usedtrainOptimizing a common neural network training parameter theta; and using the verification loss LvalidTo optimize the structural parameter a. The updating process of θ and a is as follows:
θ'＝θ-βΔθθLtrain,a'＝a-γΔaLvalid
where β and γ are the learning rates, the optimal convolution and join operations for each layer are calculated as follows:
Figure BDA0002765520810000056
wherein, clSum oflRepresents the optimal convolution and join operations, s, respectively, of the ith layercRepresenting the set of all candidate sets, ssRepresents the set of all the categories of the connection,
Figure BDA0002765520810000057
is the weight of the s unit in the l layer; in the training stage, the AutoST model obtains different training structures in the training of the data set, and finally, the optimal structure is selected for training.
Compared with the prior art, the task allocation method based on the space-time crowdsourcing worker behavior prediction has the beneficial effects that:
the invention is based on the actual situation, adds the space-time prediction into the task allocation strategy, and has great help for improving the utility of the space-time crowdsourcing platform. Constraint conditions of task effective time and total travel cost are considered from a more comprehensive angle, and the optimization problem of time-space crowdsourcing task allocation from a global angle is met, so that the quality and efficiency of task allocation are improved, and a time-space crowdsourcing platform is fully utilized.
Drawings
FIG. 1 is a flow chart of a task allocation method based on spatiotemporal crowdsourcing worker behavior prediction according to the present invention.
Fig. 2 is a flowchart of the specific step of step S2.
Fig. 3 is a flowchart illustrating the detailed steps of step S21.
FIG. 4 is a structural operation diagram of the model AutoST for establishing an efficient search space.
FIG. 5 is an illustration of spatiotemporal prediction of spatiotemporal crowdsourcing workers by the model AutoST.
FIG. 6 is a graph comparing the performance of the AutoST model and the ENAS algorithm and DARTS algorithm.
FIG. 7 is a graph comparing the AutoST model with the computing time of the ENAS algorithm and DARTS algorithm.
Fig. 8 is a schematic diagram of tasks to be allocated.
FIG. 9 is a schematic diagram of a global optimization task allocation process based on an improved greedy algorithm.
Detailed Description
Other aspects, features and advantages of the present invention will become apparent from the following detailed description, taken in conjunction with the accompanying drawings, illustrating by way of example the principles of the invention, when considered in its entirety and in part by the present specification.
First embodiment
Referring to fig. 1, a flowchart of a task allocation method based on spatiotemporal crowdsourcing worker behavior prediction according to the present invention is provided. The specific process of fig. 1 will be described in detail below.
Step S1: dividing a search space into a series of continuous space-time environments according to space-time characteristics, dividing an area where a task to be distributed is located into M multiplied by N grids along longitude and latitude, and controlling the resolution of grid division by M and N according to application requirements. The grid-based approach uses a multi-resolution network data structure that quantifies the object space into a finite number of cells that form the network structure, which has the advantage of fast processing speed.
Step S2: spatiotemporal data of crowdsourcing workers and external influencing factors are input into an AutoST model, and the positions where the crowdsourcing workers are likely to arrive at the next time stamp are predicted. Further, as shown in fig. 2, the method specifically includes:
step S21: and designing an efficient space-time search space and dynamically capturing spatial correlation. Further, as shown in fig. 3, the specific step of step S21 includes:
step S211: and processing the candidate units. In spatio-temporal prediction, neighboring regions may influence each other. Convolution operations are therefore very important for building local geo-correlation models. In addition, the spatial correlation of convolution models of different kernel sizes are different, so it is contemplated to use different convolution kernels, capture local correlation using low convolution and obtain global correlation with high convolution coding for crowd-sourced worker behavior prediction.
Spatio-temporal information prediction is different from image prediction, does not require a pooling operation, eliminates the pooling operation considering that the pooling operation may cause information loss, and classifies search candidate units into two categories:
(1) the convolution operations consist of a 3 × 3 standard convolution, a 5 × 5 standard convolution, a 3 × 3 discrete convolution, and a 5 × 5 discrete convolution.
(2) And the jump connection operation comprises a connectionless operation and an identification operation.
To ensure that the shape of the output is the same as the shape of the input, we use a convolution kernel with the same step size equal to 1 and the filter size of the input and output are the same.
Step S212: and processing by an operation module. It is known that in order to solve the memory inefficiency and thus perform efficient search, the ST-NASNet, a key module of AutoST, mainly performs two types of operations, a hybrid convolution operation and a hybrid join operation. As shown in FIG. 4, the blue arrows represent the mixed convolution blocks
Figure BDA0002765520810000085
The mixed convolution operation searches different convolution kernels in each layer and calculates the weighted sum of all convolution outputs; green arrow indicates hybrid junction block
Figure BDA0002765520810000086
The hybrid join operation learns whether to join operations between different layers, and multiplies the output of each layer by the join probability in order to perform fusion of different levels of features.
Convolution unit { a }0,a1,a2,a3And a connection unit { a }4,a5The trainable parameter of (c) is defined as the structural parameter a, which controls the likelihood of a candidate cell being selected, where the value in a is initialized with 0 instead of a random number. Further, the parameter of each convolution unit is defined as a model parameter M. Assuming that Sc and Ss are search units of the convolution operation and the skip join operation, respectively, the calculation of the hybrid convolution block is defined as:
Figure BDA0002765520810000081
the calculation of the hybrid junction block is defined as:
Figure BDA0002765520810000082
where f is the convolution operation, θ is a parameter of f, and σ represents the sigmoid activation function.
Figure BDA0002765520810000083
Represents the weight of candidate unit c at layer i, and S is the jump connection function. For the model used in this patent, there are L mixed volume blocks,
Figure BDA0002765520810000084
a hybrid junction block, which greatly reduces the search space.
Step S213: and constructing a neural structure search network. The fixed structure of the external network may affect the prediction performance, and in order to make the network more efficient, the search space is constrained from the following two aspects according to the characteristics of space-time prediction. First, there is only one convolution in adjacent layers, which can capture a wide range of spatial dependencies. And secondly, when multi-level features are fused, external feature conversion is not needed, and only the output of the previous layer is added to the current layer. The output of the l-th layer can be expressed as:
Figure BDA0002765520810000091
wherein o isiIs the output of the i-th layer,
Figure BDA0002765520810000092
is a high-level feature of the l-th layer,
Figure BDA0002765520810000093
is the result of fusing the characteristics of the ith layer and the ith layer.
Step S22: as shown in fig. 5, this step uses the CPT paradigm, and the intimacy between crowdsourcing workers and tasks, the movement period, the trend and the external influencing factors are input into the AutoST model, wherein the intimacy between crowdsourcing workers and tasks is found by the pearson correlation coefficient method. For an AutoST model with L layers, the structure parameters a and model parameters M may be defined as:
Figure BDA0002765520810000094
Figure BDA0002765520810000095
wherein n isc4 is the number of convolution kernels, n s2 represents the number of convolution kernels and connections respectively,
Figure BDA0002765520810000096
is the convolution parameter of the i-th layer,
Figure BDA0002765520810000097
is the convolution parameter for the s-th layer,
Figure BDA0002765520810000098
is the weight of the c unit in the l-th layer, θc1，θc2，θfcParameters of convolutional layer 1, convolutional layer 2 and fusion feature layer, respectively. And finally, the output of the space-time prediction search network is as follows:
o＝Relu(f(oL；θc2)+f(xe；θfc)) (6)
wherein o isLFor output of AutoST, xeRepresenting an external influencing factor.
Step S23: the model is further optimized by a search algorithm.
Training in AutoST includes two phases, a search phase and a verification phase. In the search phase, the data is first decomposed into a training set and a validation set, and then the training loss L is usedtrainAnd optimizing the common neural network training parameter theta. And using the verification loss LvalidTo optimize the structural parameter a. The updating process of θ and a is as follows:
θ'＝θ-βΔθθLtrain,a'＝a-γΔaLvalid (7)
where β and γ are the learning rates. The optimal convolution and join operations for each layer are calculated as follows:
Figure BDA0002765520810000101
wherein, clSum oflRepresents the optimal convolution and join operations, s, respectively, of the ith layercRepresenting the set of all candidate sets, ssRepresents the set of all the categories of the connection,
Figure BDA0002765520810000102
is the weight of the s unit at the l-th layer. In the training stage, the AutoST model can obtain different training structures in the training of the data set, and finally, the optimal structure is selected for carrying outAnd (5) training.
Step S24: the prediction accuracy was evaluated using the mean absolute error (RMSE) and the Mean Absolute Percentage Error (MAPE). The calculation for RMSE and MAPE is:
Figure BDA0002765520810000103
wherein n is a number of values,
Figure BDA0002765520810000104
is the predicted value, yiIs the true value.
Second embodiment
Unlike the first embodiment, we want to assign the crowd-sourced workers predicted by AutoST the most appropriate tasks for them in the acquired set of candidate tasks. The method comprises the following specific steps:
step S3: and acquiring all to-be-completed space-time crowdsourcing task sets in the current environment state, and numbering the sets in sequence to form a space-time crowdsourcing task set.
The spatiotemporal crowdsourcing task set is represented as: t ═ T1,t2,...,tn}。tnRepresenting the nth spatiotemporal crowdsourcing task. Furthermore, the nth spatial crowdsourcing task contains 4 attributes, namely: t is tn＝<ln,rn,dn,bn>Wherein l isnThe position of the nth crowdsourced task is represented and is taken as a two-dimensional geographic coordinate rnAnd dnRespectively representing the release time of the nth crowdsourcing task and the deadline of the task, taking the values as time points, bnRepresenting the benefit given to the platform when the nth crowdsourcing task is completed;
step S4: and acquiring a space-time crowdsourcing worker set predicted by the AutoST model, and numbering the set in sequence to form the space-time crowdsourcing worker set.
The spatiotemporal crowdsourcing worker set is represented as: w ═ W1,w2,...,wn}。wnRepresenting the nth spatiotemporal crowdsourcing worker. Furthermore, the nth spatiotemporal crowdsourcing worker contains 2 attributes, namely: w is an＝<pn,on>Wherein p isnThe position of the nth space-time crowdsourcing worker is represented, and the value is a two-dimensional geographic coordinate onThe order receiving state of the nth space-time crowdsourcing worker is shown, 0 represents that the order is not received, and 1 represents that the order is received;
step S5: and recommending an optimal task for each crowdsourcing worker by adopting an improved greedy algorithm, and implementing a global optimal task allocation strategy.
Constructing the acquired space-time crowdsourcing task set into a task queue, and finding a new task for a worker from the task queue by an algorithm when the worker arrives or the worker completes the last task of the worker; recommending a new task for crowdsourcing workers by using a formulated greedy function, comprehensively considering the effective time of the crowdsourcing task and the distance ratio between the workers and the task, and calculating as follows:
Figure BDA0002765520810000111
where Ben (w, t) represents the recommended quality of the task, dn-rnIndicating the effective time of the task, dis (p)n-ln) Representing the distance between crowdsourcing workers and the task, the larger the Ben (w, t) value, the more worthwhile the task is.
Third embodiment
In this example, to test the accuracy of the invention using the AutoST model to predict where the spatiotemporal crowdsourcing worker is likely to arrive at the next timestamp, the experimental dataset used by the invention was the check-in data in the new york city range from Gowalla2009, month 5 to 12 2010, containing 37385 pieces of user data, 940 spatiotemporal crowdsourcing task positions. For preprocessing of data, inactive user information is removed, and location information with check-in data of at least 10 times and access times of at least 10 times is reserved. The method comprises the steps of firstly dividing New York into 32 x 32 grids, then calculating inflow and outflow of each grid to serve as a statistical mode of movement period and trend of crowdsourcing workers, and finally inputting behavior data of the spatio-temporal crowdsourcing workers and external influence factors into a model. In the experimental evaluation, the first 70% of each user check-in record is used as a training set, the last 20% is used as a test set, and the rest 10% is used as a verification set for model parameter adjustment. In the search phase, the neural network structure is learned by using a validation set, in the training phase, a training model is built by using the training set, and the early-stopping strategy is executed by using the validation set.
In addition, in order to better test the prediction effect of the model proposed by the patent, a comparative experiment is also performed with the following algorithm, wherein AutoST is the model used by the patent.
An ENAS: and adopting reinforcement learning as a search strategy, and accelerating the search process by sharing parameters.
DARTS: discrete and non-differentiable search spaces are converted to continuous search spaces, allowing more efficient searches using gradient-based optimization strategies.
AutoST, including an optional volume block composed of multi-scale kernel, used for capturing different range of features under variable scale; the trainable connecting block is used for dynamically fusing multi-scale space characteristics and can automatically search and process a system structure of a multi-range and multi-scale prediction problem.
The results of comparing the performance with the calculation time are shown in fig. 6 and 7. The abscissa in the figure represents the ENAS algorithm, DARTS algorithm and AutoST model, respectively, and the ordinate represents the root mean square error and computation time of the algorithm or model, respectively. It is thus observed that the AutoST model takes the least amount of time to find the optimal structure and the error is smaller than the ENAS algorithm and DARTS algorithm. It can be seen that the prediction model used in the present invention is superior to the other two algorithms.
For example, as shown in fig. 8, we predict that at 9:00, a space-time crowd-sourced worker will arrive at a position w (6,3) in an area with four tasks to be solved, t1(0,2), t2(2,5), t3(10,0), and t4(8, 9). The distances used in the invention are all manhattan distances.
Table 1: information related to tasks to be distributed
Task numbering Task location Time of release Cut-off time Platform revenue
1 (0,2) 9:10 9:30 5
2 (2,5) 8:50 9:40 6
3 (10,0) 9:00 9:20 8
4 (8,9) 9:30 9:50 2
Table 1 shows the location, release time, deadline of the task to be solved and the relevant information of the benefit brought to the platform.
In an improved greedy-based algorithm, a task queue is used to store tasks that are to be assigned to spatio-temporal crowdsourcing workers, and the algorithm recommends an optimal task for a worker from the queue whenever a new worker arrives at a task area. The method comprises the steps of firstly sequencing tasks in a task queue according to platform profits, then defining two variables to respectively represent the effective time of the tasks and the distance between crowdsourcing workers and the tasks, finally inputting the variables into a greedy function, and recommending the tasks with the maximum Ben (w, t) values to the workers.
As shown in fig. 9, the recommended quality results of the four tasks by the optimization algorithm of this patent are as follows: ben (w, t)1＝2.5871、Ben(w,t)2＝8.3333、Ben(w,t)3＝2.2222、Ben(w,t)42.5000, the best to-solve task 2 is recommended to crowdsourcing workers.
In summary, according to the task allocation method based on the space-time crowdsourcing worker behavior prediction provided by the invention, from the global perspective, dynamic task allocation in a space-time environment is considered, meanwhile, the space-time crowdsourcing worker behavior trajectory prediction is considered, the use efficiency of a space-time crowdsourcing platform is improved, and under the condition that the constraint condition of time is met, the goal of recommending optimal tasks for the space-time crowdsourcing workers is realized, so that the quality and the efficiency of task allocation are improved. The method provided by the invention has strong implementability and can be used for multi-field research.
While the foregoing is directed to the preferred embodiment of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.

Claims (6)

1. A task allocation method based on crowd-sourced worker spatiotemporal behavior prediction is characterized by comprising the following steps:
step S1, dividing the search space into a series of continuous space-time environment according to space-time characteristics, dividing the area where the task is to be distributed into M multiplied by N grids along longitude and latitude, and controlling the resolution of grid division by M and N according to application requirements;
step S2, inputting the spatio-temporal data of crowdsourcing workers and external influence factors into a high-efficiency neural network learning model facing spatio-temporal prediction, and predicting the possible arrival place of the crowdsourcing workers at the next timestamp;
step S3, acquiring all to-be-completed space-time crowdsourcing task sets in the current environment state, and numbering the task sets in sequence to form a space-time crowdsourcing task set;
step S4, obtaining a space-time crowdsourcing worker set predicted by the AutoST model, and numbering the set in sequence to form a space-time crowdsourcing worker set;
and step S5, recommending the optimal task for each crowdsourcing worker by adopting an improved greedy algorithm, and implementing a global optimal task allocation strategy.
2. The task allocation method based on the crowd-sourced worker spatiotemporal behavior prediction as claimed in claim 1, wherein the specific steps of the step S2 are:
s21, designing an efficient space-time search space and dynamically capturing spatial correlation;
s22, inputting the intimacy of crowdsourcing workers and tasks, the movement period and trend of the crowdsourcing workers into an AutoST model together with external influence factors to perform space-time prediction;
s23, further optimizing the model through a search algorithm;
and S24, calculating accuracy, and selecting the average absolute error and the average absolute percentage error to measure the accuracy of prediction.
3. The method for task allocation based on crowd-sourced worker spatiotemporal behavior prediction as claimed in claim 1, wherein the specific steps of step S5 are as follows:
s51, constructing the space-time crowdsourcing task set acquired in the step S3 into a task queue, and finding a new task for a worker from the task queue by an algorithm when the worker arrives or finishes the last task of the worker;
s52, recommending a new task for crowdsourcing workers by using the formulated greedy function, comprehensively considering the effective time of the crowdsourcing task and the distance ratio between the workers and the task, and calculating as follows:
Figure FDA0002765520800000021
where Ben (w, t) represents the recommended quality of the task, dn-rnIndicating the effective time of the task, dis (p)n-ln) The distance between crowdsourcing workers and the task is represented, and under the condition that the upper limit time of the task is similar, the larger the Ben (w, t) value is, the more the task is worth recommending.
4. The method for task allocation based on crowd-sourced worker spatiotemporal behavior prediction as claimed in claim 2, wherein the specific steps of step S21 are as follows:
s211, processing a candidate unit, capturing local correlation by using low convolution and obtaining global correlation by using high convolution coding to predict crowdsourcing worker behaviors;
s212, processing by an operation module;
and S213, constructing a neural structure search network.
5. The method for task allocation based on crowd-sourced worker spatiotemporal behavior prediction as claimed in claim 2, wherein the specific steps of step S22 are as follows:
and inputting the intimacy of crowdsourcing workers and tasks, the movement period and trend of the crowdsourcing workers and external influence factors into an AutoST model by adopting a CPT paradigm, wherein the intimacy of the crowdsourcing workers and the tasks is calculated by a Pearson correlation coefficient method.
6. The method for task allocation based on crowd-sourced worker spatiotemporal behavior prediction as claimed in claim 2, wherein the specific steps of step S23 are as follows:
the training of the AutoST includes two stages, a search stage and a validation stage, in the search stage, the data is first decomposed into a training set and a validation set, and then the common neural network training parameters are optimized by using the training loss.
CN202011232008.2A 2020-11-06 2020-11-06 Task allocation method based on space-time crowdsourcing worker behavior prediction Pending CN112328914A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011232008.2A CN112328914A (en) 2020-11-06 2020-11-06 Task allocation method based on space-time crowdsourcing worker behavior prediction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011232008.2A CN112328914A (en) 2020-11-06 2020-11-06 Task allocation method based on space-time crowdsourcing worker behavior prediction

Publications (1)

Publication Number Publication Date
CN112328914A true CN112328914A (en) 2021-02-05

Family

ID=74315576

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011232008.2A Pending CN112328914A (en) 2020-11-06 2020-11-06 Task allocation method based on space-time crowdsourcing worker behavior prediction

Country Status (1)

Country Link
CN (1) CN112328914A (en)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113240189A (en) * 2021-06-01 2021-08-10 青岛科技大学 Reputation value-based dynamic vehicle task and calculation force matching method
CN113361912A (en) * 2021-06-04 2021-09-07 浙江工业大学 Service task scheduling method based on reinforcement learning
CN113393056A (en) * 2021-07-08 2021-09-14 山东大学 Crowdsourcing service supply and demand gap prediction method and system based on time sequence
CN113408738A (en) * 2021-05-17 2021-09-17 桂林电子科技大学 Method for designing ethical agent based on reinforcement learning
CN113627765A (en) * 2021-08-01 2021-11-09 湖南大学 User satisfaction-based distributed space crowdsourcing task allocation method and system
CN113792905A (en) * 2021-07-29 2021-12-14 长三角信息智能创新研究院 Accurate prompt grid social worker collection task prediction method based on block data
CN114282645A (en) * 2021-11-24 2022-04-05 杭州电子科技大学 DQN-based space-time crowdsourcing task allocation method
CN114372680A (en) * 2021-12-23 2022-04-19 电子科技大学(深圳)高等研究院 Spatial crowdsourcing task allocation method based on worker loss prediction
CN115269948A (en) * 2022-09-27 2022-11-01 北京科技大学 Variable-scale data analysis method and device supporting space-time data intelligent scale transformation
CN116976652A (en) * 2023-09-25 2023-10-31 烟台大学 Multi-target task allocation method based on space-time crowdsourcing
CN117455200A (en) * 2023-12-22 2024-01-26 烟台大学 Multi-stage task allocation method, system, equipment and medium in crowdsourcing environment

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113408738A (en) * 2021-05-17 2021-09-17 桂林电子科技大学 Method for designing ethical agent based on reinforcement learning
CN113240189A (en) * 2021-06-01 2021-08-10 青岛科技大学 Reputation value-based dynamic vehicle task and calculation force matching method
CN113240189B (en) * 2021-06-01 2022-10-14 青岛科技大学 Reputation value-based dynamic vehicle task and calculation force matching method
CN113361912B (en) * 2021-06-04 2022-05-27 浙江工业大学 Service task scheduling method based on reinforcement learning
CN113361912A (en) * 2021-06-04 2021-09-07 浙江工业大学 Service task scheduling method based on reinforcement learning
CN113393056A (en) * 2021-07-08 2021-09-14 山东大学 Crowdsourcing service supply and demand gap prediction method and system based on time sequence
CN113792905A (en) * 2021-07-29 2021-12-14 长三角信息智能创新研究院 Accurate prompt grid social worker collection task prediction method based on block data
CN113627765A (en) * 2021-08-01 2021-11-09 湖南大学 User satisfaction-based distributed space crowdsourcing task allocation method and system
CN113627765B (en) * 2021-08-01 2024-01-05 湖南大学 Distributed space crowdsourcing task distribution method and system based on user satisfaction
CN114282645A (en) * 2021-11-24 2022-04-05 杭州电子科技大学 DQN-based space-time crowdsourcing task allocation method
CN114372680A (en) * 2021-12-23 2022-04-19 电子科技大学(深圳)高等研究院 Spatial crowdsourcing task allocation method based on worker loss prediction
CN115269948A (en) * 2022-09-27 2022-11-01 北京科技大学 Variable-scale data analysis method and device supporting space-time data intelligent scale transformation
CN115269948B (en) * 2022-09-27 2023-10-13 北京科技大学 Variable-scale data analysis method and device supporting space-time data intelligent scale transformation
CN116976652A (en) * 2023-09-25 2023-10-31 烟台大学 Multi-target task allocation method based on space-time crowdsourcing
CN116976652B (en) * 2023-09-25 2023-12-29 烟台大学 Multi-target task allocation method based on space-time crowdsourcing
CN117455200A (en) * 2023-12-22 2024-01-26 烟台大学 Multi-stage task allocation method, system, equipment and medium in crowdsourcing environment
CN117455200B (en) * 2023-12-22 2024-03-29 烟台大学 Multi-stage task allocation method, system, equipment and medium in crowdsourcing environment

Similar Documents

Publication Publication Date Title
CN112328914A (en) Task allocation method based on space-time crowdsourcing worker behavior prediction
Li et al. Autost: Efficient neural architecture search for spatio-temporal prediction
CN111862579B (en) Taxi scheduling method and system based on deep reinforcement learning
Xu et al. Tripres: Traffic flow prediction driven resource reservation for multimedia iov with edge computing
Li et al. A deep learning method based on an attention mechanism for wireless network traffic prediction
Pasandideh et al. Genetic application in a facility location problem with random demand within queuing framework
US20210233196A1 (en) System and method for ride order dispatching
CN102256369B (en) Task scheduling method for wireless sensor grid based on energy and communication overhead
CN111626128A (en) Improved YOLOv 3-based pedestrian detection method in orchard environment
CN111402616B (en) Intelligent parking control method, system, storage medium and terminal
Chu et al. Travel demand prediction using deep multi-scale convolutional LSTM network
CN111639791A (en) Traffic flow prediction method, system, storage medium and terminal
CN114066503A (en) Method for forecasting taxi demand based on virtual service area construction
CN114492978A (en) Time-space sequence prediction method and device based on multi-layer attention mechanism
CN113362637A (en) Regional multi-field-point vacant parking space prediction method and system
CN112469116B (en) Positioning method, positioning device, electronic equipment and computer readable storage medium
Xi et al. Hmdrl: Hierarchical mixed deep reinforcement learning to balance vehicle supply and demand
CN114372830A (en) Network taxi booking demand prediction method based on space-time multi-graph neural network
JP2017107385A (en) Task assignment device, method, and program
CN112382081A (en) Traffic flow prediction method based on multiple tasks
CN114338416A (en) Space-time multi-index prediction method and device and storage medium
CN116957260A (en) Intelligent allocation method for carrier vehicles based on vehicle demand prediction
CN115456238A (en) Urban trip demand prediction method based on dynamic multi-view coupling graph convolution
US20230153742A1 (en) Methods for shared bicycle delivery and operation area planning in smart cities and internet of things (iot) systems thereof
CN117058235A (en) Visual positioning method crossing various indoor scenes

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination