CN-121983162-A - Sparse sensor data-based hydrogen concentration field space-time reconstruction and prediction method and system

CN121983162ACN 121983162 ACN121983162 ACN 121983162ACN-121983162-A

Abstract

The invention discloses a hydrogen concentration field space-time reconstruction and prediction method and system based on sparse sensor data, and relates to the technical field of hydrogen energy safety utilization. The method comprises the steps of simulating various hydrogen leakage working conditions through numerical simulation, generating time sequence data of a sparse sensor and complete two-dimensional concentration field data, preprocessing and standardizing the data, dividing the processed data into a training set, a verification set and a test set, training a neural network prediction model by using the training set and the verification set to obtain a trained neural network prediction model, inputting the time sequence data of the sparse sensor acquired in real time into the trained neural network prediction model, and outputting a current time concentration field reconstruction result and a future time concentration field prediction sequence through inverse standardization processing. The method can provide full space-time and prospective situation awareness for the hydrogen leakage risk of the closed space, and has important application value for guaranteeing the safe operation of the hydrogen energy transportation infrastructure.

Inventors

ZHOU WENJING
ZHAO JINPENG
WEI JINJIA

Assignees

西安交通大学

Dates

Publication Date: 20260505
Application Date: 20260114

Claims (10)

1. The hydrogen concentration field space-time reconstruction and prediction method based on sparse sensor data is characterized by comprising the following steps of: simulating various hydrogen leakage working conditions through numerical simulation, and generating time sequence data and complete two-dimensional concentration field data of the sparse sensor; Preprocessing and standardizing the time sequence data of the sparse sensor and the complete two-dimensional concentration field data, and dividing the processed data into a training set, a verification set and a test set; Training the neural network prediction model by using the training set and the verification set to obtain a trained neural network prediction model; And inputting the time sequence data of the sparse sensor acquired in real time into the trained neural network prediction model, and outputting a current time concentration field reconstruction result and a future time concentration field prediction sequence through inverse standardization processing.
2. The method for space-time reconstruction and prediction of a hydrogen concentration field based on sparse sensor data according to claim 1, wherein said simulating a plurality of hydrogen leakage conditions by numerical simulation comprises: Simulating multiple hydrogen leakage scenes with different leakage positions and leakage intensities in the comprehensive pipe rack; presetting sparse sensor position points corresponding to real deployment in a simulation model, and recording concentration time sequence data of the position points as sparse sensor time sequence data; And synchronously obtaining complete two-dimensional plane concentration field data of space-time evolution in the simulation area as complete two-dimensional concentration field data.
3. The method for hydrogen concentration field spatiotemporal reconstruction and prediction based on sparse sensor data of claim 1, wherein said preprocessing and normalizing comprises: using the time sequence data of the sparse hydrogen sensor obtained by simulation as input data of a neural network; Grid matrixing the complete hydrogen concentration field space-time data obtained by simulation, and generating structured matrix data as output data of a neural network; Constructing a data set according to the one-to-one correspondence of the input data and the output data according to the time sequence, and dividing the data set into a training set, a verification set and a test set according to the proportion; And calculating a standardized parameter based on the data of the training set, and adopting the parameter to synchronously standardize the training set, the verification set and the test set.
4. A method for hydrogen concentration field spatiotemporal reconstruction and prediction based on sparse sensor data according to claim 3, characterized in that said grid matrixing process comprises: Grid interpolation is carried out on discrete point concentration data obtained through simulation, an initial grid matrix is generated through linear interpolation, nearest neighbor interpolation filling is adopted on missing values, and a two-dimensional grid matrix is obtained; and carrying out threshold filtering on the two-dimensional grid matrix, and setting matrix element values with concentration values smaller than a preset threshold value to zero.
5. A method for the spatiotemporal reconstruction and prediction of hydrogen concentration fields based on sparse sensor data according to claim 3, characterized in that said normalization process is carried out according to the following formula: Wherein, the For the normalized sample value to be obtained, As the value of the original sample is set, As the mean value of the sample, Is the standard deviation of the samples.
6. The method for reconstructing and predicting hydrogen concentration fields based on sparse sensor data according to claim 1, wherein said training of said neural network prediction model comprises; Constructing a iTransformer architecture-based neural network prediction model, wherein the model comprises a iTransformer encoder and a grid cross attention decoder; Performing parameter training on the neural network prediction model based on the standardized training set, and performing performance verification and super-parameter tuning by using the standardized verification set; And performing performance evaluation on the optimized neural network prediction model by adopting a test set to determine a neural network prediction model finally used for reconstructing and predicting the hydrogen concentration field.
7. The method for hydrogen concentration field spatiotemporal reconstruction and prediction based on sparse sensor data of claim 6, wherein said iTransformer encoder comprises an inverted data embedding layer and a transducer coding layer; The grid cross attention decoder is configured to perform cross attention calculation on the learnable grid query parameters and the sensor characteristics output by the encoder, generate middle two-dimensional grid characteristics, process the middle two-dimensional grid characteristics through a neural network module comprising a convolution layer and an up-sampling layer, and decode to obtain a hydrogen concentration field with target spatial resolution.
8. The method for hydrogen concentration field spatiotemporal reconstruction and prediction based on sparse sensor data of claim 6, wherein said neural network predictive model is trained to minimize a complex loss function To that end, the composite loss function is defined as: Wherein, the And The weight coefficients of the reconstruction segment and the prediction segment respectively, In order to reconstruct the segment loss, To predict segment loss; the reconstruction segment loss And predicting segment loss All are obtained by weighted summation calculation: Wherein, the In order to weight Charbonnier the loss, In order to achieve a loss of structural similarity, In order to achieve a loss of the spatial gradient, 、、 The weight coefficients of Charbonnier loss, structural similarity loss, and spatial gradient loss, respectively.
9. The method for reconstructing and predicting the hydrogen concentration field based on sparse sensor data according to claim 1, wherein the specific steps of outputting the reconstruction result of the concentration field at the current moment and the prediction sequence of the concentration field at the future moment are as follows: acquiring hydrogen concentration time series data acquired by a sparse sensor in the comprehensive pipe rack at the current moment; inputting the time sequence data into the neural network prediction model to obtain a concentration field prediction sequence at a future target moment; and performing inverse standardization processing on the concentration field prediction sequence, and outputting the complete hydrogen concentration field reconstruction result at the current moment and the complete hydrogen concentration field prediction sequence at the future moment.
10. A hydrogen concentration field spatiotemporal reconstruction and prediction system based on sparse sensor data, comprising the following modules: the data acquisition and simulation module is used for simulating various hydrogen leakage working conditions through numerical simulation to generate time sequence data and complete two-dimensional concentration field data of the sparse sensor; the data preprocessing module is used for preprocessing and standardizing the time sequence data of the sparse sensor and the complete two-dimensional concentration field data, and dividing the processed data into a training set, a verification set and a test set; the model training module is used for training the neural network prediction model by utilizing the training set and the verification set to obtain a trained neural network prediction model; The concentration field prediction module is used for inputting the real-time acquired time sequence data of the sparse sensor into the trained neural network prediction model, and outputting a current time concentration field reconstruction result and a future time concentration field prediction sequence through inverse normalization processing.

Description

Sparse sensor data-based hydrogen concentration field space-time reconstruction and prediction method and system Technical Field The invention belongs to the technical field of hydrogen energy safety utilization, and particularly relates to a hydrogen concentration field space-time reconstruction and prediction method and system based on sparse sensor data. Background The hydrogen energy is used as a key component of a future clean energy system, and the large-scale and economic transportation of the hydrogen energy mainly depends on pipeline transportation, and specifically comprises two main paths of newly-built pure hydrogen pipelines and hydrogen-doped transportation by utilizing the existing natural gas pipe network. However, due to the characteristics of small molecular weight, high diffusion coefficient, wide flammable and explosive range and the like, if leakage occurs in the conveying process, the hydrogen is extremely easy to quickly accumulate and form explosive mixed gas in underground closed spaces such as utility tunnel and the like, and public safety and facility stability are seriously threatened. In the prior art, safety monitoring mainly relies on sparse distributed point sensor means, the technical means can only reflect local concentration conditions at sensor distribution points, serious 'monitoring blind areas' exist, leakage diffusion trend cannot be effectively predicted, early warning is delayed, and the requirements of real-time risk sensing and active prevention and control are difficult to meet. To overcome the limitations of point monitoring, complete concentration field information is obtained by employing Computational Fluid Dynamics (CFD) simulation methods. According to the method, a hydrodynamic equation set is solved through a numerical value, so that a diffusion process after hydrogen leakage can be simulated theoretically, and spatial concentration distribution can be obtained. However, the CFD simulation method is high in calculation cost, and single simulation takes very long time, and real-time monitoring and rapid prediction cannot be realized, so that the CFD simulation method is difficult to apply to online safety early warning. Although the data driving method provides a new thought for reconstructing and predicting the concentration field, how to use only a very small amount of sparse sensor data and simultaneously realize high-precision reconstruction of the concentration field and high-efficiency prediction at future time is still an unbroken technical problem. The model is required to have strong space-time feature mining and mapping capability so as to overcome the challenges of sparse data, noise interference, variable working conditions and the like. Disclosure of Invention The invention aims to solve the problems of space monitoring blind areas and time prediction lag in the conventional point type monitoring technology, and provides a hydrogen concentration field space-time reconstruction and prediction method and system based on sparse sensor data, so as to quickly and accurately realize the hydrogen concentration field space-time reconstruction and prediction. In order to achieve the above purpose, the invention adopts the following technical scheme: In a first aspect, the present invention provides a method for space-time reconstruction and prediction of a hydrogen concentration field based on sparse sensor data, comprising the steps of: simulating various hydrogen leakage working conditions through numerical simulation, and generating time sequence data and complete two-dimensional concentration field data of the sparse sensor; Preprocessing and standardizing the time sequence data of the sparse sensor and the complete two-dimensional concentration field data, and dividing the processed data into a training set, a verification set and a test set; Training the neural network prediction model by using the training set and the verification set to obtain a trained neural network prediction model; And inputting the time sequence data of the sparse sensor acquired in real time into the trained neural network prediction model, and outputting a current time concentration field reconstruction result and a future time concentration field prediction sequence through inverse standardization processing. The invention further improves that the simulation of various hydrogen leakage working conditions through numerical simulation comprises the following steps: Simulating multiple hydrogen leakage scenes with different leakage positions and leakage intensities in the comprehensive pipe rack; presetting sparse sensor position points corresponding to real deployment in a simulation model, and recording concentration time sequence data of the position points as sparse sensor time sequence data; And synchronously obtaining complete two-dimensional plane concentration field data of space-time evolution in the simulation area as complete two-dimensional concentration field