(1) extracting an adjacency relationship among road segments in a target area road network to generate a road segment relationship adjacency matrix, wherein any one element of the road segment relationship adjacency matrix is A_(i,j)^R, A_(i,j)^R denotes an adjacency relationship between a road segment i and a road segment j, and when the road segment i and the road segment j are connected directly, A_(i,j)^R=1, and when the road segment i and the road segment j are connected indirectly, A_(i,j)^R is zero; a size of the road segment relationship adjacency matrix is M*M, any one column represents the adjacency relationship between a road segment represented by a current column and other road segments, any one row represents the adjacency relationship between a road segment represented by a current row and the other road segments, both the road segment i and the road segment j are any road segment in the target area road network, and M is a total number of the road segments in the target area;

(2) cleaning and calibrating vehicle trajectory data within a time range of the target area to obtain pre-processed data, wherein the time range at least comprises N time slots and N is a natural number greater than or equal to 10;

(3) constructing an auto trips quantity matrix and an auto arrival quantity matrix of the pre-processed data from an n^th time slot to an (n+K)^th time slot, wherein any one element in the auto trips quantity matrix is x(n+k,j), x(n+k,j) is an auto trips quantity generated on the road segment j in the (n+k)^th time slot, a size of the auto trips quantity matrix is (K+1) row*M column, any one column vector is a trips quantity sequence of the road segment represented by the current column and any one row vector is a trips quantity sequence of a time slot represented by the current row;

any one element in the auto arrival quantity matrix is y(n+k,j), y(n+k,j) is an auto arrival quantity generated on the road segment j in the (n+k)^th time slot, a size of the auto arrival quantity matrix is the (K+1) row*M column, any one column vector is an arrival quantity sequence of the road segment represented by the current column and any one row vector is an arrival quantity sequence of the time slot represented by the current row;

extracting an auto trips quantity vector and an auto arrival quantity vector of the pre-processed data in an (n+K+1)^th time slot, wherein any one element in the auto trips quantity vector is x(n+K+1, j), and x(n+K+1, j) is an auto trips quantity generated on the road segment j in the (n+K+1)^th time slot;

any one element in the auto arrival quantity vector is y(n+K+1, j), y(n+K+1,j) is an auto arrival quantity generated on the road segment j in the (n+K+1)^th time slot;

n is equal to 1, 2, . . . , (N−K−1), and n is initialized as 1;

k is equal to 1, 2, 3, . . . K, and K is a natural number greater than or equal to 7;

(4) calculating a similarity between any two road segment trips quantity sequences from the n^th time slot to the (n+K)^th time slot and a similarity between arrival quantity sequences to obtain a trips quantity similarity weight matrix and an arrival quantity similarity weight matrix,

wherein

any one element in the trips quantity similarity weight matrix is A_(i,j)^x, and A_(i,j)^x denotes a similarity between the any two road segment trips quantity sequences of the road segment i and the road segment j; a size of the trips quantity similarity weight matrix is M*M, any one row represents a trips quantity similarity among the road segment represented by the current row, the road segment and the other road segments, and any one column represents the trips quantity similarity among the road segment represented by the current column, the road segment and the other road segments;

any one element in the arrival quantity similarity weight matrix is A_(i,j)^y, and A_(i,j)^y is a similarity between the arrival quantity sequences of the road segment i and the road segment j; the size of the trips quantity similarity weight matrix is M*M, the any one row represents the trips quantity similarity among the road segment represented by the current row, the road segment and the other road segments, and the any one column represents the trips quantity similarity among the road segment represented by the current column, the road segment and the other road segments;

column vectors of above matrixes correspond to each other, and a column vector at a same sequence position corresponds to a same road segment;

(5) constructing a trips quantity local relationship graph, a trips quantity global relationship graph, an arrival quantity local relationship graph and an arrival quantity global relationship graph;

a characteristic matrix of the trips quantity local relationship graph is the auto trips quantity matrix constructed in the step (3) and an adjacency matrix of the trips quantity local relationship graph is the road segment relationship adjacency matrix based on a physical road network constructed in the step (1);

a characteristic matrix of the trips quantity global relationship graph is the auto trips quantity matrix constructed in the step (3) and an adjacency matrix of the trips quantity global relationship graph is the trips quantity similarity weight matrix constructed in the step (4);

a characteristic matrix of the arrival quantity local relationship graph is an auto arrival quantity matrix constructed in the step (3) and an adjacency matrix of the arrival quantity local relationship graph is the road segment relationship adjacency matrix based on a physical road network constructed in the step (1);

a characteristic matrix of the arrival quantity global relationship graph is the auto arrival quantity matrix constructed in the step (3) and an adjacency matrix of the arrival quantity global relationship graph is the arrival quantity similarity weight matrix constructed in the step (4);

(6) training or updating the prediction model by taking the trips quantity local relationship graph, the trips quantity global relationship graph, the arrival quantity local relationship graph and the arrival quantity global relationship graph constructed in the step (5), and taking the auto trips quantity vector and the auto arrival quantity vector of the (n+K+1)^th time slot as tags, wherein the prediction model is initialized as a Multi-task GCN-LSTM (MTGL) neural network;

(7) setting n=n+1, and executing the steps (3)-(6); and

(8) circularly executing the step (7) till n=N−K−1 to obtain a finalized prediction model.