Deep learning industrial time series data prediction algorithm based on spatial-temporal feature fusion

Abstract

The invention discloses a deep learning industrial time series data prediction algorithm based on spatial-temporal feature fusion, and the algorithm comprises the steps: input data preparation and graph construction: obtaining a historical state sequence and a control quantity sequence, constructing a graph structure representing the spatial relationship of measurement points, and defining the connection relationship of each measurement point through an edge index; extracting spatial features through GNN, including aggregation of future controlled variables and aggregation of historical controlled variables, and splicing a historical state sequence and aggregated historical controlled variable features to form input of the LSTM so as to capture time dependence; gradually generating a future prediction sequence by using the future control quantity # imgabs0 # after GNN aggregation and the final hidden state # imgabs1 # of the LSTM; the problem that the global prediction precision is affected due to the fact that a single model cannot comprehensively process the time features and the time sequence features at the same time in the prior art is solved, the spatial and time features of the industrial time sequence data are extracted in combination with the graph neural network GNN and the long and short term memory network LSTM, and the prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the field of industrial big data computing, and in particular to a deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion. Background Art

[0002] The operation of hydropower plants is influenced by numerous factors, including water flow, precipitation, temperature, and equipment status, all of which often exhibit significant time-series characteristics. Accurate time-series forecasting can help hydropower plants better schedule generators, optimize reservoir storage and release strategies, improve power generation efficiency and safety, ensure sufficient power supply during peak hours, and avoid excessive water consumption. Furthermore, hydropower plant equipment often operates in complex environments, and the operational status of key equipment such as turbines and generators is crucial to power generation efficiency. By monitoring equipment status in real time and analyzing historical data, time-series forecasting can predict equipment failure risks and schedule maintenance in advance, thereby avoiding downtime losses caused by unexpected failures and improving equipment reliability and availability. In terms of energy scheduling, hydropower plants often need to collaborate with other power supply sources. Especially in highly variable climates, predicting trends in precipitation and water flow is crucial for the optimal scheduling of grid resources. By deeply mining time-series data during hydropower generation, more accurate load and power demand forecasting can be achieved, providing a scientific basis for grid scheduling and load balancing, and reducing the instability and risks associated with power supply fluctuations. With the development of intelligent technology, hydropower plants can optimize water resource management, equipment maintenance, and energy scheduling through time-series prediction algorithms, improving power generation efficiency and promoting sustainable development. In short, the application of time-series prediction in hydropower generation can significantly improve energy efficiency, ensure stable power plant operation, and contribute to the utilization of green energy and environmental protection.

[0003] Among existing time series forecasting technologies, traditional LSTM methods, while effective in extracting temporal features, fall short in capturing the spatial relationships between different measurement points. Models that rely solely on GNNs struggle to fully process dynamic time series information. This technical shortcoming is particularly pronounced in industrial data forecasting, where the spatiotemporal correlations between different measurement points significantly impact forecast accuracy. Therefore, a deep learning industrial time series data forecasting algorithm based on the fusion of spatiotemporal features is needed to address this issue. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a deep learning industrial time series data prediction algorithm based on the fusion of spatiotemporal features, aiming to solve the problem that a single model in the existing technology cannot simultaneously perform comprehensive processing of time features and time series features, thereby affecting the global prediction accuracy. By combining the graph neural network GNN and the long short-term memory network LSTM to extract the spatial and temporal features of industrial time series data, GNN is used to capture the spatial relationship between different measuring points, while LSTM is used to extract time dependencies, so as to improve the prediction accuracy.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: The deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion includes the following steps: S1, input data preparation and graph construction: Get the historical status sequence ,in Dimensions representing state data; represents the set of real numbers, represents the number of time steps, represents the time step index; Get historical control volume and future control volume The control quantity sequence composed of the graph structure representing the spatial relationship of the measuring points is constructed, and the connection relationship of each measuring point is defined by the edge index edge_index; S2, extracts spatial features through GNN, including the aggregation of future control quantities and the aggregation of historical control quantities; S3, temporal feature extraction, concatenates the historical state sequence with the aggregated historical control quantity features to form the input of LSTM to capture temporal dependencies; S4, predict the future sequence step by step: Future control amount after GNN aggregation and the last hidden state of the LSTM , and gradually generate future forecast sequences.

[0006] Preferably, in step S2, the aggregation of future control quantities includes: Control the future Input graph neural network GNN to obtain spatial aggregation control quantity features , obtain the control quantity characteristics after spatial aggregation; the control quantity for each time step t For spatial aggregation, use the GATv2 convolutional layer: ; The output obtained ,in is the hidden dimension of GNN output; Indicates the dimension of the control quantity.

[0007] Preferably, in step S2, the aggregation of historical control quantities includes: The historical control volume Input GNN to obtain historical control quantity features : ; The output obtained .

[0008] Preferably, in step S3, LSTM extracts time features including concatenating historical state sequences and historical control quantity characteristics Forming the LSTM input sequence: ; in, is the input sequence of LSTM; Represents the feature dimension concatenation operation.

[0009] Preferably, extracting the time feature through LSTM further includes encoding the time feature through LSTM: Input the concatenated sequence into LSTM to obtain the encoded hidden state

[0010] ; in, is the final hidden state of LSTM, used for prediction tasks, Indicates that LSTM is at the final time step cell state.

[0011] Preferably, in step S4, gradually predicting the future sequence includes initializing the prediction state: Set the initial state is the last state of the history sequence: ; in is the final hidden state of the LSTM.

[0012] Preferably, step-by-step predicting of future sequences further comprises: Forecast multiple time steps into the future step by step. For each time step t∈[T+1,T+L], do the following: The current forecast status With the current control quantity characteristics Splicing to form input: Input LSTM single-step prediction and get the new state: ; Output the predicted value of the current time step through the fully connected layer: ; Iterate the above formula to get 、 … ,L represents the length of the time step; Stack prediction output: Combine the prediction results of all time steps into the final output sequence: ; Where, Indicates that LSTM is at time step The hidden state of Indicates that LSTM is at time step The cell state, LSTM represents the computing unit of the long short-term memory network, and Linear represents the fully connected layer.

[0013] Preferably, a deep learning industrial time series data prediction device based on spatiotemporal feature fusion includes: The data preprocessing and graph construction module is used to receive the historical state sequence, historical control amount and future control amount, output the edge index edge_index of the constructed graph structure, and output it to the spatial feature extraction module; The spatial feature extraction module includes a GNN processing unit, the first branch of which is used to perform GNN spatial aggregation on the future control quantity and output To the iterative prediction module; the second branch is used to perform GNN spatial aggregation on the historical control amount and output To the time feature extraction module; Temporal feature extraction module, including LSTM processing unit, for splicing and generate ; Extract time features through LSTM encoding and output the final hidden state To the iterative prediction module; Iterative prediction module, including step-by-step prediction unit, is used to utilize the future control quantity after GNN aggregation and the last hidden state of the LSTM , gradually generate future prediction sequences; The device is used to execute the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion.

[0014] Preferably, an electronic device includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion is implemented.

[0015] Preferably, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion.

[0016] The beneficial effects of the present invention are as follows: 1. Fusion of spatiotemporal features improves prediction accuracy: Traditional time series prediction models can only capture single features in time or space and are unable to effectively process complex industrial data that incorporates both temporal and spatial correlations. This algorithm combines a graph neural network (GNN) with a long short-term memory (LSTM) network to extract spatial features from control data and temporal features from status data, respectively, thereby achieving a deep fusion of spatiotemporal features. This innovative design enables the model to more accurately model the relationships and temporal dependencies between different measurement points, effectively improving prediction accuracy. It is particularly suitable for multi-point prediction in complex industrial scenarios.

[0017] 2. Parallel processing of spatial features to avoid error accumulation: Existing recursive multi-step prediction methods are often prone to error accumulation, which is particularly significant in long-term prediction tasks. This invention processes the spatial features of the control variables in parallel and incorporates future control variable information during the step-by-step prediction process, thereby reducing the impact of error accumulation on prediction accuracy and enhancing model stability. This structural design significantly improves the model's predictive performance and effectively addresses the error amplification problem common in recursive methods.

[0018] 3. Improved real-time performance and computational efficiency: By introducing a combined architecture of GNN and LSTM, the present invention enables efficient processing of spatiotemporal data while simplifying the structure. Compared to traditional pure LSTM or pure GNN models, the present invention achieves step-by-step processing of different features through a modular design, significantly reducing the model's computational burden. In industrial real-time prediction applications, this efficient structure not only meets the needs of high-frequency data processing, but is also suitable for deployment on edge devices with limited computing resources, providing a more practical solution for intelligent decision-making in industrial sites.

[0019] 4. Model Adaptability and Scalability: The modular design of this invention enables flexible adaptation to diverse measurement point configurations and control variable combinations in various industrial scenarios. Because the GNN and LSTM modules process spatial and temporal features, respectively, the model can be easily expanded to more measurement points or new data sources without redesigning the entire network structure. This adaptability significantly enhances the algorithm's application value in real-world industrial scenarios, reduces the cost of model adjustment and retraining, and enhances its potential for commercialization. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the overall process of the system of the present invention; Figure 2 It is a connection block diagram of the device part of the present invention; Figure 3 2 is a schematic diagram of a comparison curve diagram of an embodiment of the present invention and a separate GNN algorithm; Figure 4 It is a schematic diagram of a comparison curve diagram of an embodiment of the present invention and a separate LSTM algorithm. DETAILED DESCRIPTION

[0021] Example 1: like Figure 1 As shown in the figure, the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion includes the following steps: S1, input data preparation and graph construction: Get the historical status sequence ,in Dimensions representing state data; represents the set of real numbers, represents the number of time steps, represents the time step index; Get historical control volume and future control volume The control quantity sequence composed of the graph structure representing the spatial relationship of the measuring points is constructed, and the connection relationship of each measuring point is defined by the edge index edge_index; S2, extracts spatial features through GNN, including the aggregation of future control quantities and the aggregation of historical control quantities; S3, temporal feature extraction, concatenates the historical state sequence with the aggregated historical control quantity features to form the input of LSTM to capture temporal dependencies; S4, predict the future sequence step by step: Future control amount after GNN aggregation and the last hidden state of the LSTM , and gradually generate future forecast sequences.

[0022] Preferably, in step S2, the aggregation of future control quantities includes: Control the future Input graph neural network GNN to obtain spatial aggregation control quantity features , obtain the control quantity characteristics after spatial aggregation; the control quantity for each time step t For spatial aggregation, use the GATv2 convolutional layer: ; The output obtained ,in is the hidden dimension of GNN output; Indicates the dimension of the control quantity.

[0023] Preferably, in step S2, the aggregation of historical control quantities includes: The historical control volume Input GNN to obtain historical control quantity features : ; The output obtained .

[0024] Preferably, in step S3, LSTM extracts time features including concatenating historical state sequences and historical control quantity characteristics Forming the LSTM input sequence: ; in, is the input sequence of LSTM; Represents the feature dimension concatenation operation.

[0025] Preferably, extracting the time feature through LSTM further includes encoding the time feature through LSTM: Input the concatenated sequence into LSTM to obtain the encoded hidden state

[0026] ; in, is the final hidden state of LSTM, used for prediction tasks, Indicates that LSTM is at the final time step cell state.

[0027] Preferably, in step S4, gradually predicting the future sequence includes initializing the prediction state: Set the initial state is the last state of the history sequence: ; in is the final hidden state of the LSTM.

[0028] Preferably, step-by-step predicting of future sequences further comprises: Forecast multiple time steps into the future step by step. For each time step t∈[T+1,T+L], do the following: The current forecast status With the current control quantity characteristics Splicing to form input: Input LSTM single-step prediction and get the new state: ; Output the predicted value of the current time step through the fully connected layer: ; Iterate the above formula to get 、 … ,L represents the length of the time step; Stack prediction output: Combine the prediction results of all time steps into the final output sequence: ; Where, Indicates that LSTM is at time step The hidden state of Indicates that LSTM is at time step The cell state, LSTM represents the computing unit of the long short-term memory network, and Linear represents the fully connected layer.

[0029] like Figure 2 As shown, preferably, a deep learning industrial time series data prediction device based on spatiotemporal feature fusion includes: The data preprocessing and graph construction module is used to receive the historical state sequence, historical control amount and future control amount, output the edge index edge_index of the constructed graph structure, and output it to the spatial feature extraction module; The spatial feature extraction module includes a GNN processing unit, the first branch of which is used to perform GNN spatial aggregation on the future control quantity and output To the iterative prediction module; the second branch is used to perform GNN spatial aggregation on the historical control amount and output To the time feature extraction module; Temporal feature extraction module, including LSTM processing unit, for splicing and generate ; Extract time features through LSTM encoding and output the final hidden state To the iterative prediction module; Iterative prediction module, including step-by-step prediction unit, is used to utilize the future control quantity after GNN aggregation and the last hidden state of the LSTM , gradually generate future prediction sequences; The device is used to execute the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion.

[0030] Preferably, an electronic device includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion is implemented.

[0031] Preferably, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion.

[0032] Example 2: This embodiment provides an application of a deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion in a specific device. The algorithm of the present invention was put into use in a large hydropower plant in a prefecture-level city in central China to predict reservoir water levels and power generation for the next 24 hours to optimize unit scheduling. The specific process is as follows: Dataset: Historical state sequence : water level (m), water flow (m³ / s) and generator power (MW); Control quantity sequence : gate opening (%) and rainfall (mm); Spatial relationship: 12 key measurement points Time range: Data from January to June 2023; The prediction performance comparison table is shown in Table 1 below: Table 1: Comparison of prediction performance with GNN and LSTM methods:

[0033] From the above table 1 and Figure 3-Figure 4 It can be seen that the pure LSTM failed to capture the impact of upstream rainfall on turbine No. 3, resulting in a power generation prediction error of +18.7%; the pure GNN misjudged the linkage relationship between the spillway gates, resulting in a water level prediction error of +0.41m. The present invention uses GNN to capture the spatial association of "rainfall → upstream monitoring point → reservoir" and combines it with LSTM to learn the 2-hour delay pattern to accurately predict the water level jump.

Claims

1. A deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion, characterized by: The following steps are involved: S1, input data preparation and graph construction: Get the historical status sequence ,in Dimensions representing state data; represents the set of real numbers, represents the number of time steps, represents the time step index; Get historical control volume and future control volume The control quantity sequence composed of the graph structure representing the spatial relationship of the measuring points is constructed, and the connection relationship of each measuring point is defined by the edge index edge_index; S2, extracts spatial features through GNN, including the aggregation of future control quantities and the aggregation of historical control quantities; S3, temporal feature extraction, concatenates the historical state sequence with the aggregated historical control quantity features to form the input of LSTM to capture temporal dependencies; S4, predict the future sequence step by step: Future control amount after GNN aggregation and the last hidden state of the LSTM , and gradually generate future forecast sequences.

2. The deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion according to claim 1 is characterized in that: In step S2, the aggregation of future control quantities includes: Control the future Input graph neural network GNN to obtain spatial aggregation control quantity features , obtain the control quantity characteristics after spatial aggregation; the control quantity for each time step t For spatial aggregation, use the GATv2 convolutional layer: ； The output obtained ,in is the hidden dimension of GNN output; Indicates the dimension of the control quantity.

3. The deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion according to claim 2 is characterized in that: In step S2, the aggregation of historical control quantities includes: The historical control volume Input GNN to obtain historical control quantity features : ； The output obtained .

4. The deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion according to claim 3 is characterized in that: In step S3, LSTM extracts temporal features including concatenating historical state sequences and historical control quantity characteristics Forming the LSTM input sequence: ； in, is the input sequence of LSTM; Represents the feature dimension concatenation operation.

5. The deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion according to claim 4 is characterized in that: LSTM extracts time features and also encodes time features through LSTM: Input the concatenated sequence into LSTM to obtain the encoded hidden state ； in, is the final hidden state of LSTM, used for prediction tasks, Indicates that LSTM is at the final time step cell state.

6. The deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion according to claim 5 is characterized in that: In step S4, the step-by-step prediction of the future sequence includes initializing the prediction state: Set the initial state is the last state of the history sequence: ； in is the final hidden state of the LSTM.

7. The deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion according to claim 6 is characterized in that: Step-by-step forecasting of future series also includes: Forecast multiple time steps into the future step by step. For each time step t∈[T+1,T+L], do the following: The current forecast status With the current control quantity characteristics Splicing to form input: Input LSTM single-step prediction and get the new state: ； Where, Indicates that LSTM is at time step The hidden state of Indicates that LSTM is at time step The cell state; LSTM represents the computational unit of the long short-term memory network; Output the predicted value of the current time step through the fully connected layer: ； Iterate the above formula to get 、 … ,L represents the length of the time step; Stack prediction output: Combine the prediction results of all time steps into the final output sequence: ； Where Linear represents the fully connected layer.

8. A deep learning industrial time series data prediction device based on spatiotemporal feature fusion, characterized in that: include: The data preprocessing and graph construction module is used to receive the historical state sequence, historical control amount and future control amount, output the edge index edge_index of the constructed graph structure, and output it to the spatial feature extraction module; The spatial feature extraction module includes a GNN processing unit, the first branch of which is used to perform GNN spatial aggregation on the future control quantity and output To the iterative prediction module; the second branch is used to perform GNN spatial aggregation on the historical control amount and output To the time feature extraction module; Temporal feature extraction module, including LSTM processing unit, for splicing and generate ; Extract time features through LSTM encoding and output the final hidden state To the iterative prediction module; Iterative prediction module, including step-by-step prediction unit, is used to utilize the future control quantity after GNN aggregation and the last hidden state of the LSTM , gradually generate future prediction sequences; The device is used to execute the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion as described in any one of claims 1-7.

9. An electronic device, characterized in that: It includes a processor and a memory, the memory stores a computer program, and when the computer program is executed by the processor, it implements the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the steps of the deep learning industrial time series data prediction algorithm based on spatiotemporal feature fusion as described in any one of claims 1 to 7 are implemented.

Images