CN116610919A - Space time sequence prediction method based on cyclic graph algorithm sub-neural network - Google Patents

Abstract

The application discloses a space time sequence prediction method based on a cyclic graph computation sub-neural network, which comprises the following steps: modeling a traffic road network to obtain a graph structure model, and abstracting a sensor in the traffic road network into nodes in the graph structure model; carrying out standardization processing on data acquired by a sensor; constructing various graph calculation sub-networks as feature traps of data, wherein the feature traps respectively use different modes to aggregate information of each node in the graph structure model; constructing GGRU units, and respectively embedding various graph calculation sub-networks into GRU units of the neural network; aggregating the characteristic information captured by different graph sub-networks through an integrator; an encoder-decoder architecture is constructed using a plurality of integrators to enable prediction of spatial multi-temporal sequences. The method has the capability of understanding information in multiple angles, effectively improves the understanding of space multi-element time sequence data, and further improves the prediction accuracy.

Description

Space time sequence prediction method based on cyclic graph algorithm sub-neural network

Technical Field

The application relates to the technical field of space-time sequence prediction, in particular to a space-time sequence prediction method based on a cyclic graph algorithm sub-neural network.

Background

Space-time series prediction has very wide application requirements in modern times, such as: (1) weather forecast: weather prediction is an important application of spatial time series prediction. By modeling the meteorological data, the change trend of the meteorological variables such as the future temperature, rainfall, wind speed and the like can be predicted, and decision support is provided for weather forecast, disaster early warning, agricultural production and the like. (2) energy demand prediction: it is important for energy suppliers and consumers to understand the trend of future energy demand. By modeling the historical energy demand data, future energy demands can be predicted to adjust production planning and supply chain management. (3) traffic prediction: traffic prediction can help city planners and traffic managers to better manage traffic flow and improve traffic efficiency. By modeling the historical traffic data, future road congestion conditions, traffic accident occurrence rates and the like can be predicted so as to take corresponding measures. (4) population flow prediction: population flow is an important factor in urban planning and social policy formulation. By modeling historical population flow data, future population flow trends can be predicted to rationally plan urban infrastructure and social resource allocation. (5) stock price prediction: stock price prediction is an important application in the financial field. By modeling the historical stock price data, the change trend of the future stock price can be predicted, and decision support is provided for investors.

Currently, a deep learning method is mainly used for multi-element time series prediction. Such neural network-based models can be effectively used for multivariate time series prediction. For example: (1) recurrent neural network RNN: is a neural network that can process sequence data by capturing the timing relationship of the sequence with the output of the previous time as the input of the current time. In multivariate time series prediction, multiple RNNs may be used to process each sequence separately and then their outputs are connected to arrive at the final prediction result. RNNs are very widely used in multivariate time series prediction, for example, in the fields of traffic flow prediction, stock price prediction, etc. to achieve good results. (2) long and short term memory network LSTM: is a special RNN that can better handle long sequences and long-term dependencies. In multivariate time series prediction, multiple LSTMs may be used to process each series and then their outputs are connected to get the final prediction result. LSTM performs better than RNN in handling long sequences and long-term dependencies and may therefore be more suitable in certain applications. (3) convolutional neural network CNN: is a neural network mainly used for image processing, but can also be used for multivariate time series prediction in some cases. In multivariate time series prediction, multiple sequences can be considered as multiple channels, then each channel is convolved using multiple convolution checks, and finally the outputs of all channels are connected to obtain the final prediction result. CNNs have relatively few applications in multivariate time series prediction, but may be more suitable in certain applications. (4) Attention mechanism Attention: the method is a mechanism which can give different weights according to different parts of input data and can be used for multi-element time sequence prediction. In multivariate time series prediction, an attention mechanism is used to automatically learn the importance of different parts of the sequence and an attention weight is applied to the predictions for each sequence. The application of attention mechanisms in multivariate time series prediction is newer, but has achieved good results in some applications.

However, the above-described multivariate time series prediction either completely does not consider the spatial dependency between the sequences, or only considers the spatial dependency between the sequences from a single point of view. In fact, the underlying relationship between these sequences is complex. Thus, there is an urgent need for a new prediction method that ensures that these underlying relationships can be fully mined from a number of different dimensions. Just like human reading text, the metaphors of text tend to be more profound than the superficial meaning.

Disclosure of Invention

The application aims to provide a space time sequence prediction method based on a cyclic graph sub-neural network, which has the capability of understanding information at multiple angles, effectively improves the understanding of space multi-element time sequence data, and further improves the prediction precision.

In order to achieve the above object, the present application provides a space time sequence prediction method based on a cyclic graph algorithm sub-neural network, including:

modeling a traffic road network to obtain a graph structure model, and abstracting a sensor in the traffic road network into nodes in the graph structure model;

carrying out standardization processing on data acquired by a sensor;

constructing various graph calculation sub-networks as feature traps of data, wherein the feature traps respectively use different modes to aggregate information of each node in the graph structure model;

constructing GGRU units, and respectively embedding various graph calculation sub-networks into GRU units of the neural network;

aggregating the characteristic information captured by different graph sub-networks through an integrator;

a sequence-to-sequence encoder-decoder architecture is constructed using a plurality of integrators to implement prediction of spatial multi-element time series.

Further, use ofThe traffic flow data acquired by all the sensors at the moment r is represented, wherein N represents the number of the sensors, and P represents the number of traffic indexes detected by each sensor; use->Representing historical observations within T' time stamps, with +.>Predicted values representing T time stamps in the future; and then determining a function of the neural network learning graph structure model:

wherein ,is the neural network to be fitted.

Further, the specific mode for carrying out standardization processing on the data collected by the sensor is as follows:

wherein X represents training set sample, X ^* Representing normalized data, E (X) represents training set sample mean and D (X) represents training set sample variance.

Further, constructing various graph calculation sub-networks as feature traps of data, specifically:

constructing a graph diffusion convolution Operator DC Operator, which is a graph calculation sub-network based on a static graph;

a graph gating attention Operator GA Operator is constructed, which is a graph computation sub-network based on dynamic graphs.

Further, the specific way of constructing the graph diffusion convolution Operator DC Operator is as follows:

the graph diffusion convolution operator adopts a random walk strategy:

wherein X represents the input of the model and H represents the output of the model; the diffusion process uses limited S steps for cutting; d (D) _O ，D _I The out-degree matrix and in-degree matrix in the graph are represented,representing a forward state transition matrix and a reverse state transition matrix; alpha, beta E [0,1 ]]Respectively representing the restarting probability of random walk; />Represents model parameters, and Θ _O[q，p，s] ＝α(1-α) ^s ，Θ _I[q，p，k] ＝β(1-β) ^s ；

The critical matrix W is obtained using a distance-based gaussian kernel:

here dist (v) _i ，v _j ) Representing sensor v _i and v_j A distance therebetween; σ is the standard deviation of the distance set.

Further, the specific way to construct the graph gating attention Operator GA Operator is as follows:

using the K-head attention mechanism in the graph gating attention operator, there is one K-dimensional gating vector g for each node i _i ：

wherein ,/>Representing all neighbor node sets of node i; x is x _i ＝X _i，： A feature vector representing node i; />Reference vectors of all neighboring nodes of node i, and +.> Max represents taking the maximum value by element; />Representing the final result mapped to K-dimension and scaled to [0,1 ]]Between them;

dynamically acquiring an attention weight matrix among nodes:

wherein ,representing the parameter theta _xa Is a linear transformation of (2); />Representing the parameter theta _za Is a linear transformation of (2); i.e. < ->Representing a linear layer; θ represents various parameters.

Obtaining an output vector y of the node i _i ：

Wherein K represents the number of heads of attention, and K representsCurrently executing is the kth attention header;representing the parameter theta ₀ Is a linear transformation of (2); />The expression parameter is->Is a linear transformation of (a).

Further, for use in GGRU unitsRepresenting different graphic operators:

H ^(t) ＝u ^(t) ⊙H ^(t-1) +(1-u ^(t) )⊙C ^(t)

wherein ,X^(t) ，H ^(t) Input and output representing a t-th timestamp; r is (r) ^(t) ，u ^(t) A gating state of a Reset gate (Reset gate) and an Update gate (Update gate) representing a t-th time stamp; theta (theta) _r ，Θ _u ，Θ _C Is a different filter parameter; Γ -shaped structure _g Represented in a specified graph networkMiddle execution->An operator; as indicated by the letter Hadamard product.

As a further step, the integrator aggregates the feature information captured by the different graph sub-networks, specifically:

wherein ,representing GRU units embedded with graphic operators, I representing the number of categories of graphic operators,/-, etc.>Representing an output dimension d ₀ Feedforward neural network, τ, means using the tanh activation function in the last layer.

As a further step, the input to the encoder is a historical observationThe output of the decoder is the future prediction value +.>

Compared with the prior art, the technical scheme adopted by the application has the advantages that: according to the application, a deep learning model is adopted, a plurality of graph operator sub-networks are combined to carry out multi-angle modeling on the space dependency relationship among the nodes, and meanwhile, the information captured by a plurality of graph operators can be effectively aggregated through an integrator. The understanding capability of the model on the bottom layer dependency relationship among the multiple time sequences is enhanced, and the prediction accuracy is improved.

Drawings

FIG. 1 is a diagram of a single output architecture using parallel single inputs;

FIG. 2 is a diagram of a multiple output architecture using parallel multiple inputs;

FIG. 3 is a block diagram of a single output and class residual using parallel single inputs;

FIG. 4 is a diagram of a structure using parallel multiple inputs, multiple outputs, and class residuals;

FIG. 5 is a block diagram using a serial single input, single output;

FIG. 6 is a diagram of a multiple output architecture using multiple inputs in series;

FIG. 7 is a block diagram using serial single input, single output, and class residual;

FIG. 8 is a block diagram of a multi-output and residual-like structure using a series of multiple inputs;

FIG. 9 is a block diagram of a GGRU model;

fig. 10 is a diagram of an encoder/decoder architecture.

Detailed Description

The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the application, i.e., the embodiments described are merely some, but not all, of the embodiments of the application.

Thus, the following detailed description of the embodiments of the application, as presented in the figures, is not intended to limit the scope of the application, as claimed, but is merely representative of selected embodiments of the application. All other embodiments, which can be made by a person skilled in the art without making any inventive effort, are intended to be within the scope of the present application.

The application provides a space time sequence prediction method based on a cyclic graph algorithm sub-neural network, which specifically comprises the following steps:

step 1: modeling a traffic road network to obtain a graph structure model, and abstracting a sensor in the traffic road network into nodes in the graph structure model;

in particular, byRepresenting traffic flow data collected by all sensors at time t, wherein N represents the number of sensors and P represents the detection of each sensorThe number of traffic indicators measured; use->Representing historical observations within T' time stamps, with +.>Predicted values representing T time stamps in the future; and then determining a function of the neural network learning graph structure model:

wherein ,is the neural network to be fitted.

Step 2: and (3) carrying out standardization processing on the data acquired by the sensor:

wherein X represents training set sample, X ^* Representing normalized data, E (X) represents training set sample mean and D (X) represents training set sample variance.

Step 3: constructing various graph calculation sub-networks as feature traps of data, wherein the feature traps respectively use different modes to aggregate information of each node in the graph structure model;

specifically, the application includes two types of graph operators that aggregate information of each node in the graph structure model in different ways, respectively, and embed the information into the GRU unit as a substitute for the original linear unit.

Step 3.1: constructing a graph diffusion convolution Operator DC Operator, which is a graph calculation sub-network based on a static graph;

step 3.1.1: the graph diffusion convolution operator adopts a random walk strategy:

here, X represents the input of the model, and H represents the output of the model. The diffusion process uses a limited number of K-step truncations. D (D) _O ，D _I Representing the out-degree and in-degree matrices in the graph,representing a forward state transition matrix and a reverse state transition matrix. Alpha, beta E [0,1 ]]Respectively, the restart probabilities of the random walks. />Represents model parameters, and Θ _O[q，p，s] ＝α(1-α) ^s ，Θ _I[q，p，s] ＝β(1-β) ^s 。/>The probability value representing the S-step diffusion from the i-th node to its neighbor node, the former representing the forward propagation probability and the latter representing the backward propagation probability.

Step 3.1.2: the critical matrix W is obtained using a distance-based gaussian kernel:

here dist (v) _i ，v _j ) Representing sensor v _i and v_j A distance therebetween; σ is the standard deviation of the distance set.

Step 3.2: constructing a graph gating attention Operator GA Operator, which is a graph calculation sub-network based on a dynamic graph;

step 3.2.1: using the K-head attention mechanism in the graph gating attention operator, there is one K-dimensional gating vector g for each node i _i ：

wherein ,representing all neighbor node sets of node i; x is x _i ＝X _i，： A feature vector representing node i;reference vectors of all neighboring nodes of node i, and +.> Max represents taking the maximum value by element; />Representing the final result mapped to K-dimension and scaled to [0,1 ]]Between them;

step 3.2.2: dynamically acquiring an attention weight matrix among nodes:

wherein ,representing the parameter theta _xa Is a linear transformation of (2); />Representing the parameter theta _za Is a linear transformation of (2);

step 3.2.3: obtaining an output vector y of the node i _i ：

Step 4: constructing GGRU units, embedding multiple graph sub-networks into GRU units of neural network, see FIG. 9, whereRepresenting different graphic operators:

H ^(t) ＝u ^(t) ⊙H ^(t-1) +(1-u ^(t) )⊙C ^(t)

wherein ,X^(t) ，H ^(t) Input and output representing a t-th timestamp; r is (r) ^(t) ，u ^(t) A gating state of a Reset gate (Reset gate) and an Update gate (Update gate) representing a t-th time stamp; theta (theta) _r ，Θ _u ，Θ _C Is a different filter parameter;represented in the specified graph network +.>Middle execution->An operator; as indicated by the letter Hadamard product.

Step 5: aggregating the characteristic information captured by different graph sub-networks through an integrator;

wherein ,the GRU element embedded with the graph operator is represented. Fig. 1-8 show eight polymerizer structures.

Step 6: constructing a sequence-to-sequence encoder-decoder architecture using a plurality of integrators to implement prediction of spatial multi-element time series, see fig. 10;

specifically, the input to the encoder is a historical observationThe output of the decoder is the future prediction value +.>

In the embodiment, a Ubuntu system is used as a development environment, python is used as a development language, and Pytorch is used for building a frame. The space time sequence prediction method based on the cyclic graph computation sub-neural network is adopted to predict traffic flow: obtaining a graph structure model representation of a traffic network, and adopting a disclosed METR-LA data set; dividing the data set into a training set, a verification set and a test set, wherein the proportion is 7:1:2. to compare algorithm performance, it is compared with some commonly used multivariate time series prediction methods. Including statistical learning method ARIMA; a machine learning method LSVR; deep learning method FC-LSTM; and a graph-based deep learning method DCRNN, STGCN, gaAN, ASTGCN, GMAN. Meanwhile, various evaluation indexes are used for comprehensively evaluating the performance of the model, including MAE, RMSE, MAPE and the like.

wherein ,x_i Representing the actual data of the data set,data representing model predictions.

The comparison results are shown in Table 1: wherein the best evaluation index of all algorithms will be bolded.

Table 1: performance comparison of different traffic speed prediction models on METR-LA datasets

Table 1 compares the prediction results of the icornn model with other baseline models. By comparing the scores of each model over long-term and short-term time series predictions, it can be seen that the graph-based neural network model achieves better prediction accuracy. For the smoother sequence PeMS-BAY, some traditional models achieve good results even in short-term predictions (15 minutes), but perform poorly in long-term predictions (60 minutes). For the unstable sequence Metr-LA, the traditional model performs poorly in both short-term and long-term predictions. The GaAN model, on the other hand, performs well on the Metr-LA dataset. GMAN is more suitable for smoother sequences. All of these illustrate the importance of introducing graph structure in complex time series predictions. Furthermore, the model designed by the present application shows optimal or suboptimal performance in long-term and short-term predictions, especially for more complex time sequences, as it introduces the advantages of multiple graph structures at the same time.

In summary, compared with other advanced methods, the space time sequence prediction method based on the cyclic graph sub-neural network provided by the application has better performance and robustness, and can be suitable for the more complex multi-element time sequence prediction problem.

The foregoing descriptions of specific exemplary embodiments of the present application are presented for purposes of illustration and description. It is not intended to limit the application to the precise form disclosed, and obviously many modifications and variations are possible in light of the above teaching. The exemplary embodiments were chosen and described in order to explain the specific principles of the application and its practical application to thereby enable one skilled in the art to make and utilize the application in various exemplary embodiments and with various modifications as are suited to the particular use contemplated. It is intended that the scope of the application be defined by the claims and their equivalents.

Claims (9)

1. The space time sequence prediction method based on the cyclic graph algorithm sub-neural network is characterized by comprising the following steps of:

modeling a traffic road network to obtain a graph structure model, and abstracting a sensor in the traffic road network into nodes in the graph structure model;

carrying out standardization processing on data acquired by a sensor;

constructing various graph calculation sub-networks as feature traps of data, wherein the feature traps respectively use different modes to aggregate information of each node in the graph structure model;

constructing GGRU units, and respectively embedding various graph calculation sub-networks into GRU units of the neural network;

aggregating the characteristic information captured by different graph sub-networks through an integrator;

a sequence-to-sequence encoder-decoder architecture is constructed using a plurality of integrators to implement prediction of spatial multi-element time series.

2. The method for predicting the space-time sequence based on the cyclic graph sub-neural network according to claim 1, whereinThe traffic flow data acquired by all the sensors at the time t are represented, wherein N represents the number of the sensors, and P represents the number of traffic indexes detected by each sensor; use->Representing historical observations within T' time stamps, with +.>Predicted values representing T time stamps in the future; and then determining a function of the neural network learning graph structure model:

wherein ,is the neural network to be fitted.

3. The space-time sequence prediction method based on the cyclic graph sub-neural network according to claim 1, wherein the specific mode of carrying out normalization processing on the data acquired by the sensor is as follows:

wherein X represents training set sample, X ^* Representing normalized data, E (X) represents training set sample mean and D (X) represents training set sample variance.

4. The space-time sequence prediction method based on the cyclic graph sub-neural network according to claim 1, wherein a plurality of graph sub-networks are constructed as feature traps of data, specifically:

constructing a graph diffusion convolution Operator DC Operator, which is a graph calculation sub-network based on a static graph;

a graph gating attention Operator GA Operator is constructed, which is a graph computation sub-network based on dynamic graphs.

5. The space-time sequence prediction method based on the cyclic graph Operator neural network according to claim 4, wherein the specific mode of constructing the graph diffusion convolution Operator DC Operator is as follows:

the graph diffusion convolution operator adopts a random walk strategy:

wherein X represents the input of the model and H represents the output of the model; the diffusion process uses limited S steps for cutting; d (D) _O ,D _I The out-degree matrix and in-degree matrix in the graph are represented,representing a forward state transition matrix and a reverse state transition matrix; alpha, beta E [0,1 ]]Respectively representing the restarting probability of random walk; />Represents model parameters, and Θ _O[q,p,s] ＝α(1-α) ^s ,Θ _I[q,p,k] ＝β(1-β) ^s ；

The critical matrix W is obtained using a distance-based gaussian kernel:

here dist (v) _i ,v _j ) Representing sensor v _i and v_j A distance therebetween; σ is the standard deviation of the distance set.

6. The space-time sequence prediction method based on the cyclic graph Operator neural network according to claim 4, wherein the specific mode of constructing the graph gating attention Operator GA Operator is as follows:

using the K-head attention mechanism in the graph gating attention operator, there is one K-dimensional gating vector g for each node i _i ：

wherein ,representing all neighbor node sets of node i; x is x _i ＝X _i,: A feature vector representing node i; />Reference vectors of all neighboring nodes of node i, and z _i ＝/>Max represents taking the maximum value by element; />Representing the final result mapped to K-dimension and scaled to [0,1 ]]Between them;

dynamically acquiring an attention weight matrix among nodes:

wherein ,representing the parameter theta _xa Is a linear transformation of (2); />Representing the parameter theta _za Is a linear transformation of (2);

obtaining an output vector y of the node i _i :

Where K represents the number of attention heads and K represents the kth note currently being performedA force-imparting head;representing the parameter theta ₀ Is a linear transformation of (2); />The expression parameter is->Is a linear transformation of (a).

7. The method for spatial-temporal sequence prediction based on a cyclic graph sub-neural network of claim 1, wherein the GGRU unit is used inRepresenting different graphic operators:

H ^(t) ＝u ^(t) ⊙H ^(t-1) +(1-u ^(t) )⊙C ^(t)

wherein ,H^(t) ,H ^(t) Input and output representing a t-th timestamp; r is (r) ^(t) ,u ^(t) A reset gate and a gate state of the update gate representing a t-th timestamp; theta (theta) _r ,Θ _u ,Θ _C Is a different filter parameter;represented in the specified graph network +.>Middle execution->An operator; as indicated by the letter Hadamard product.

8. The space-time sequence prediction method based on cyclic graph sub-neural network according to claim 1, wherein the integrator is used for aggregating the characteristic information captured by different graph sub-networks, specifically:

wherein ,representing GRU units embedded with graphic operators, I representing the number of categories of graphic operators,/-, etc.>Representing an output dimension d ₀ Feedforward neural network, τ, means using the tanh activation function in the last layer.

9. The method of claim 1, wherein the encoder input is a historical observationThe output of the decoder is the future prediction value +.>