Hydrogen Energy Cross-Regional Spatiotemporal Prediction Method and System Based on Transfer Learning

Abstract

A cross-regional spatio-temporal prediction method and system for hydrogen energy utilization based on transfer learning, including: extracting the time features and spatial features of hydrogen production data, hydrogen storage data, mobile hydrogen energy behavior data, and environmental data in the hydrogen energy system to form a source domain dataset; establishing a hydrogen energy spatio-temporal prediction model, including: a hydrogen production demand prediction model, a hydrogen storage optimization model, and a mobile hydrogen energy behavior prediction model; using the source domain dataset and the target domain dataset to perform transfer learning training and adversarial training on the hydrogen energy spatio-temporal prediction model; using the economic index, efficiency index, and environmental impact index of the hydrogen energy system in the target domain to form a reward function, and according to the change amount of the target domain data, when the reward function reaches the maximum value, the trained hydrogen energy spatio-temporal prediction model outputs the cross-regional spatio-temporal prediction result of hydrogen energy utilization, so as to accurately predict the hydrogen energy utilization amount in the regions with less hydrogen energy utilization.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy prediction, and particularly relates to a cross-regional spatio-temporal prediction and optimization method and system for a hydrogen production-storage-mobile hydrogen energy system based on transfer learning. Background Art

[0002] The wide application of hydrogen energy involves multiple links such as hydrogen production, hydrogen storage, and hydrogen utilization. Due to the randomness and volatility of the hydrogen energy usage behavior in terms of time and space, it increases the difficulty of power grid operation control.

[0003] In the prior art, in the planning and operation of the hydrogen energy system, there are data heterogeneity and regional differences. The data sources of the hydrogen energy systems in different regions are diverse, including the operating parameters of hydrogen production equipment, the distribution of hydrogen storage stations, the trajectories of hydrogen energy vehicles, etc. These data have significant differences in format, scale, and quality, making it difficult to directly integrate and analyze across regions. There are also differences in the economic development level, energy demand structure, infrastructure construction, etc. among regions, resulting in significant heterogeneity in the operating characteristics and optimization requirements of the hydrogen energy system among regions; in the prediction methods of hydrogen energy usage, little consideration is given to the influence of external factors such as hydrogen transportation and weather on hydrogen energy usage. The previous prediction methods of hydrogen energy usage mainly focus on the single-task prediction field. Since single-task prediction only divides each prediction task into simple and independent sub-problems, this not only ignores the potential coupling relationship between hydrogen energy usage and other influencing factors, but also has the problem of repeatedly extracting characteristic information, reducing the operation efficiency; the model generalization ability is insufficient. The traditional prediction and optimization models of the hydrogen energy system are mostly trained based on data from a single region and lack the ability to model cross-regional spatio-temporal characteristics. When directly applied to other regions, due to the differences in data distribution, the prediction accuracy and optimization effect of the model often drop significantly, making it difficult to meet the actual needs of different regions; dynamic collaborative optimization is difficult: the hydrogen production, hydrogen storage, and hydrogen utilization links are coupled in time and space, and global collaborative optimization is required to maximize the overall benefit. In the prior art, when dealing with the dynamic collaborative optimization of multiple links, phased or regional optimization strategies are often adopted, ignoring the mutual influence and dependence relationship between each link, resulting in sub-optimality and instability of the optimization results; the real-time adaptability and robustness are insufficient. The operating environment of the hydrogen energy system is complex and changeable, affected by various external factors such as traffic flow, weather conditions, and policies and regulations. In the face of emergencies and environmental changes, there is a lack of effective real-time learning and adaptive update mechanisms, making it difficult to timely adjust the optimization strategy to ensure the stable operation and efficient service of the system. Summary of the Invention

[0004] To address the deficiencies of existing technologies in hydrogen energy usage prediction, the present invention proposes a cross-regional spatio-temporal prediction and optimization method and system for a hydrogen production-storage-mobile hydrogen energy system based on transfer learning. By integrating multi-source heterogeneous data, constructing a source domain model, and using transfer learning techniques to transfer knowledge to the target domain, efficient cross-regional collaborative optimization is achieved. At the same time, an online learning and updating mechanism is introduced to enhance the dynamic adaptability and robustness of the system, providing innovative technical support for the intelligent management and operation of the hydrogen energy system, and achieving accurate prediction of the hydrogen energy usage in regions with less hydrogen energy usage.

[0005] The present invention adopts the following technical solutions.

[0006] The present invention proposes a cross-regional spatio-temporal prediction method for hydrogen energy usage based on transfer learning, including:

[0007] Extract the time features and spatial features of hydrogen production data, hydrogen storage data, mobile hydrogen energy behavior data, and environmental data in the hydrogen energy system to form a source domain dataset;

[0008] Establish a hydrogen energy spatio-temporal prediction model, including: a hydrogen production demand prediction model, a hydrogen storage optimization model, and a mobile hydrogen energy behavior prediction model; use the source domain dataset and the target domain dataset to perform transfer learning training and adversarial training on the hydrogen energy spatio-temporal prediction model;

[0009] Use the economic index, efficiency index, and environmental impact index of the hydrogen energy system in the target domain to form a reward function, and according to the change amount of the target domain data, when the reward function reaches the maximum value, the trained hydrogen energy spatio-temporal prediction model outputs the cross-regional spatio-temporal prediction result of hydrogen energy usage.

[0010] Preferably, the hydrogen production data includes: the operating power and operating efficiency of hydrogen production equipment, the hydrogen production method, and the carbon emission intensity;

[0011] The hydrogen storage data includes: the distribution of hydrogen storage facilities, the hydrogen storage capacity, the loss rate, and the current storage status;

[0012] The mobile hydrogen energy behavior data includes: the trajectories of hydrogen energy vehicles, hydrogen refueling behaviors, the distribution of hydrogen refueling stations, and the usage rate of hydrogen refueling stations;

[0013] The environmental data includes: the transportation network, traffic flow, weather conditions, policies and regulations, and the level of economic activities.

[0014] Preferably, according to the source domain dataset, the hydrogen production demand and load distribution are output by using the hydrogen production demand prediction model established based on the spatio-temporal cross-domain neural network; with the maximization of the usage efficiency of hydrogen storage facilities and hydrogen refueling stations as the objective function of the dynamic scheduling plan of the hydrogen storage station, according to the source domain dataset, hydrogen production demand and load distribution, the dynamic scheduling plan of the hydrogen storage facilities is output by using the hydrogen storage optimization model established with the objective function and hydrogen storage constraint conditions; according to the source domain dataset, load distribution and dynamic scheduling plan of the hydrogen storage facilities, the behavior and path of mobile hydrogen energy are output by using the mobile hydrogen energy behavior prediction model established by adopting the multi-agent reinforcement learning method.

[0015] Preferably, a hydrogen production demand prediction model is established based on the spatio-temporal cross-domain neural network, including multiple convolutional long short-term memory network units; the architecture of the hydrogen production demand prediction model includes: a feature extraction layer based on a spatial convolution kernel, a time series modeling layer based on a time bidirectional LSTM, and an adversarial transfer alignment layer;

[0016] Among them, an adversarial alignment module is embedded in the feature transfer layer of the spatio-temporal cross-domain neural network to form an adversarial transfer alignment layer.

[0017] Preferably, the hydrogen storage optimization model satisfies the following relational expression:

[0018]

[0019] In the formula, is the objective function of the dynamic scheduling plan of the hydrogen storage station, is the coefficient vector representing the maximization of the usage efficiency of hydrogen storage facilities and hydrogen refueling stations, is the dynamic scheduling variable of the hydrogen storage station, is the coefficient matrix in the hydrogen storage constraint condition, is the hydrogen storage constraint vector, and the superscript represents vector transpose.

[0020] Preferably, the same data in the source domain dataset and the target domain dataset are used to form transfer learning samples, and the hydrogen energy spatio-temporal prediction model is trained by using the transfer learning samples; the different data in the source domain dataset and the target domain dataset are used to form adversarial samples, and the hydrogen energy spatio-temporal prediction model trained by transfer learning is trained by using the adversarial samples.

[0021] Preferably, in the adversarial training, the adversarial transfer alignment layer in the hydrogen production demand prediction model drives the feature distributions of the source domain and the target domain to converge with the minimum loss function of the domain classifier; the domain classifier of the adversarial transfer alignment layer is a fully connected layer, including adversarial training and gradient reversal; the adversarial training is used to distinguish whether the features come from the source domain or the target domain, and the gradient reversal is used to reverse the gradient of the domain classifier during backpropagation. The expression of the gradient reversal is as follows:

[0022]

[0023] In the formula, is the gradient reversal, is the feature from the source domain during backpropagation for the gradient function of the domain classifier, is the feature from the target domain during backpropagation for the domain classifier gradient function, is the feature extractor.

[0024] Preferably, a graph convolution mapping is used to map the spatial features of the target domain dataset into a space unified with the spatial features of the source domain.

[0025] Preferably, an economic index, an efficiency index, and an environmental impact index of the hydrogen energy system in the target domain are used to form a reward function, satisfying the following relationship:

[0026]

[0027] In the formula, is the reward function of the hydrogen energy system in the target domain in state and action , , , are respectively the economic index, the efficiency index, and the environmental impact index of the hydrogen energy system in the target domain in state and action , , , are respectively the weights of the economic index, the efficiency index, and the environmental impact index.

[0028] Preferably, the reward function of the energy price response model is used as the economic index, satisfying the following relationship:

[0029]

[0030] In the formula, is the reward function of the energy price response model, is the profit, is the market share, is the inventory backlog, , , are respectively the weights of the profit, the market share, and the inventory backlog.

[0031] Preferably, according to the change amount of the target domain data, when the reward function reaches the maximum value, the trained hydrogen energy spatio-temporal prediction model outputs the cross-regional spatio-temporal prediction result of hydrogen energy usage, including:

[0032] Adopt reinforcement learning combined with the Pareto front optimization method to adjust the weights of the economic index, efficiency index, and environmental impact index to obtain the maximum value of the reward function;

[0033] According to the change amount of the target domain data, repeatedly calculate the reward function and adjust the weights. When the reward function reaches the maximum value, the trained hydrogen energy spatio-temporal prediction model outputs the cross-regional spatio-temporal prediction result of hydrogen energy usage.

[0034] Preferably, with the goal of minimizing the operating cost of the hydrogen energy system caused by the change amount of the target domain data, update the dynamic scheduling plan of the hydrogen storage facilities output by the hydrogen storage optimization model; where the goal satisfies the following relational expression:

[0035]

[0036] In the formula, is the input data of the hydrogen energy spatio-temporal prediction model based on the change amount of the target domain data, is the operating cost of the hydrogen energy system caused by the change amount of the target domain data, is the constraint condition, is the time The hydrogen storage volume of the hydrogen storage facility, is the time The hydrogen input volume, is the time The hydrogen output volume.

[0037] The present invention also proposes a cross-regional spatio-temporal prediction system for hydrogen energy usage based on transfer learning, including:

[0038] The data acquisition module is used to extract the time features and space features of the hydrogen production data, hydrogen storage data, mobile hydrogen energy behavior data, and environmental data in the hydrogen energy system to form the source domain data set;

[0039] The modeling and training module is used to establish a hydrogen energy spatio-temporal prediction model, including: a hydrogen production demand prediction model, a hydrogen storage optimization model, and a mobile hydrogen energy behavior prediction model; use the source domain data set and the target domain data set to perform transfer learning training and adversarial training on the hydrogen energy spatio-temporal prediction model;

[0040] The prediction module is used to form a reward function with the economic index, efficiency index, and environmental impact index of the hydrogen energy system in the target domain. According to the change amount of the target domain data, when the reward function reaches the maximum value, the trained hydrogen energy spatio-temporal prediction model outputs the cross-regional spatio-temporal prediction result of hydrogen energy usage.

[0041] The present invention is also a terminal, including a processor and a storage medium; the storage medium is used for storing instructions; the processor is used for operating according to the instructions to execute the steps of the method.

[0042] The present invention is also a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the method are implemented.

[0043] The beneficial effects of the present invention are at least as follows compared with the prior art. The method proposed by the present invention integrates multi-source heterogeneous data such as hydrogen production, hydrogen storage, mobile hydrogen energy, and the environment, constructs a prediction model in the source domain, and uses transfer learning technology to transfer knowledge to the target domain to achieve efficient cross-regional collaborative optimization. In regions with sufficient data, models such as hydrogen production demand prediction, hydrogen storage optimization, and mobile hydrogen energy behavior modeling are constructed to accurately predict the dynamics of each link. The transfer learning module transfers the knowledge of the model in the source domain to the target domain with scarce data to improve its prediction performance. Through joint prediction and optimization, the resource allocation is dynamically adjusted to achieve global optimization. During the operation of the system, data is collected in real time, the model parameters are updated through online learning, and anomaly detection is introduced to ensure that the system adapts to new environmental changes and operates stably and efficiently. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a schematic diagram of a cross-regional spatio-temporal prediction method for hydrogen energy use based on transfer learning proposed by the present invention; in the figure, STCNet is a spatio-temporal cross-domain neural network, is a spatio-temporal prediction model for hydrogen energy in the source domain, is a spatio-temporal prediction model for hydrogen energy in the target domain. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. The embodiments described in this application are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the spirit of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0046] In the field of hydrogen energy usage prediction, existing technologies mainly employ deep learning models such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). CNNs are good at extracting spatial features but have limited ability to mine temporal features of time series data. RNNs can capture time dependencies, but they suffer from issues such as inability to perform parallel computing, prone to vanishing and exploding gradients, and slow training speed. Long short-term memory networks (LSTMs) and gated recurrent units (GRUs), as improved RNNs, although alleviating the vanishing gradient problem, may still lose information when dealing with long sequence data, making it difficult to accurately model the complex structure between data, especially in regions with limited data volume, where the prediction accuracy is limited.

[0047] Transfer learning is a machine learning method that allows a model to apply the knowledge learned from one task (source task) to another related task (target task), especially suitable for situations with limited data volume. By leveraging the knowledge of other tasks, transfer learning can reduce the data requirements for the target task. Therefore, when the data volume in prediction region A is insufficient, the present invention enhances the data from region B where hydrogen energy is used more, and through two stages of data augmentation and hydrogen energy usage prediction, realizes transfer learning prediction of hydrogen energy usage in region A. Compared with traditional neural network parameter identification methods, this method has a faster training speed, lower computational burden, and higher accuracy, significantly improving the accuracy of parameter identification. In addition, by introducing the data of region B for data augmentation and adopting heterogeneous transfer learning, the data volume requirement for training the deep learning model for predicting hydrogen energy usage in region A is further reduced, and the accuracy of predicting hydrogen energy usage in region A is improved.

[0048] A cross-regional spatio-temporal prediction method for hydrogen energy usage based on transfer learning proposed by the present invention performs cross-regional and cross-spatio-temporal prediction and optimization on a hydrogen energy system including hydrogen production facilities, hydrogen storage facilities, hydrogen refueling stations, and mobile hydrogen energy, such as Figure 1 shown, including:

[0049] Step 1, extract the time features and spatial features of hydrogen production data, hydrogen storage data, mobile hydrogen energy behavior data, and environmental data in the hydrogen energy system to form a source domain dataset.

[0050] Specifically, taking the energy system as the source region, existing technologies do not consider the influence of external factors such as hydrogen energy transportation information and weather temperature information when predicting hydrogen energy usage. The present invention fully considers the time dimension information and spatial dimension information of hydrogen energy usage for hydrogen energy usage prediction, integrates the data layer of multi-source data information such as road network, weather, and price, and provides basic data support for model construction in the source region through the collection and standardization processing of multi-source heterogeneous data.

[0051] The collected data includes, but is not limited to: hydrogen production data, hydrogen storage data, mobile hydrogen energy data, and environmental data; hydrogen production data includes, but is not limited to: the operating power and efficiency of hydrogen production equipment, hydrogen production methods (such as electrolysis of water, natural gas reforming), and carbon emission intensity; hydrogen storage data includes, but is not limited to: hydrogen storage facility distribution, hydrogen storage capacity, loss rate, and current storage status; mobile hydrogen energy behavior data includes, but is not limited to: hydrogen energy vehicle trajectories, hydrogen refueling behavior, hydrogen refueling station distribution, and hydrogen refueling station utilization rate; environmental data includes, but is not limited to: transportation networks, traffic flow, weather conditions, policies and regulations, and economic activity levels.

[0052] Standardize and extract features from the collected data. Slice the time data according to time windows, and extract the trend and periodic features of the data as time features; encode the spatial data, and use graph structures to model spatial data such as transportation networks, hydrogen storage facility distributions, and hydrogen refueling station distributions. During the modeling process, use principal component analysis (PCA) or autoencoders to reduce the dimensionality of high-dimensional features and improve computational efficiency.

[0053] Take the hydrogen energy system as the source domain. Therefore, the time features and spatial features of hydrogen production data, hydrogen storage data, mobile hydrogen energy behavior data, and environmental data in the hydrogen energy system constitute the source domain dataset.

[0054] Step 2: Establish a hydrogen energy spatio-temporal prediction model, including: a hydrogen production demand prediction model, a hydrogen storage optimization model, and a mobile hydrogen energy behavior prediction model.

[0055] Specifically, establish a hydrogen production demand prediction model based on a spatio-temporal cross-domain neural network; take the maximization of the usage efficiency of hydrogen storage facilities and hydrogen refueling stations as the objective function of the dynamic scheduling plan of hydrogen storage stations, adopt a mixed-integer programming method, and use the objective function and hydrogen storage constraint conditions to establish a hydrogen storage optimization model; adopt a multi-agent reinforcement learning method to establish a mobile hydrogen energy behavior prediction model; according to the source domain dataset, use the hydrogen production demand prediction model to output the hydrogen production demand and load distribution; according to the source domain dataset, hydrogen production demand, and load distribution, use the hydrogen storage optimization model to obtain the dynamic scheduling plan of hydrogen storage facilities; according to the source domain dataset, load distribution, and dynamic scheduling plan of hydrogen storage facilities, use the mobile hydrogen energy behavior prediction model to obtain the behavior and path of mobile hydrogen energy.

[0056] In the source domain with sufficient data, construct a hydrogen energy spatio-temporal prediction model that can characterize the dynamic collaborative optimization of hydrogen production - hydrogen storage - hydrogen utilization, and use transfer learning technology to adjust the model parameters to complete feature alignment and knowledge transfer. Among them, the hydrogen energy spatio-temporal prediction model includes: a hydrogen production demand prediction model, a hydrogen storage optimization model, and a mobile hydrogen energy behavior prediction model.

[0057] Specifically, step 2 includes:

[0058] Step 2.1, establish a hydrogen production demand prediction model based on STCNet; according to the source domain dataset, use the hydrogen production demand prediction model to output the hydrogen production demand and load distribution;

[0059] In the embodiment, for the prediction of hydrogen production demand, STCNet is used to predict the future hydrogen production demand and load distribution, effectively extracting the complex spatio-temporal dependence relationships hidden in the data. STCNet uses a convolutional long short-term memory network as a sub-component and has extremely strong spatio-temporal dependence modeling capabilities.

[0060] In the prior art, a hierarchical transfer training method based on Bi-LSTM is adopted, which relies on the maximum mean discrepancy metric to divide the source domain data, and its optimization objective and real-time performance have obvious limitations. The present invention constructs a technical framework with the Spatial–Temporal Cross-domain neural Network (STCNet) as the core. On the basis of retaining its spatio-temporal feature extraction ability, it deeply integrates the adversarial transfer learning mechanism and the graph convolutional network. Specifically, the problem of model degradation caused by the difference in data distribution between regions is effectively alleviated through adversarial training, and the spatial correlation of hydrogen storage facilities is encoded by using the graph structure, achieving a balance between cross-regional knowledge transfer and local feature retention. In view of the dynamic characteristics of the hydrogen energy system, the present invention also introduces an incremental parameter update strategy to ensure that the model can respond quickly in the face of traffic emergencies or policy adjustments.

[0061] Using the hydrogen production demand prediction model to output the hydrogen production demand and load distribution includes:

[0062] The hydrogen production demand prediction model established based on STCNet includes multiple convolutional long short-term memory network units ; STCNet is a new type of deep learning prediction architecture that can effectively capture the complex patterns hidden in the data. STCNet uses a convolutional long short-term memory network as a sub-component and has strong spatio-temporal dependence modeling capabilities.

[0063] The before the target time interval hydrogen production data matrices , , ……, constitute the historical hydrogen production data sequence , which is input into the hydrogen production demand prediction model, where each target time interval takes a value of 10 minutes; in the embodiment, according to the historical hydrogen consumption data, historical hydrogen production data and environmental variable data in the source domain, the future hydrogen production demand and load distribution are predicted;

[0064] The three - level architecture of the hydrogen production demand prediction model includes: a feature extraction layer based on a spatial convolution kernel, a time - series modeling layer based on a bidirectional LSTM, and an adversarial transfer alignment layer.

[0065] Among them, the feature extraction layer based on the spatial convolution kernel includes convolution operations and pooling operations. In the embodiment, for the source - domain dataset, 3D convolution is used for convolution operations, and the size of the feature map is reduced through average pooling operations to enhance the robustness of the model.

[0066] The time - series modeling layer based on the bidirectional LSTM captures the long - term dependence relationships of data in the time series. The bidirectional LSTM contains two independent LSTMs, which process the time series in the forward and backward directions respectively. The forward LSTM is used to process the time series from t = 1 to t = T, and the backward LSTM is used to process the time series from t = T to t = 1. The outputs of the forward and backward LSTMs are concatenated in the time dimension to form the final time features.

[0067] To achieve accurate prediction and optimization of hydrogen production demand and load distribution in the hydrogen energy system, the present invention enhances the hydrogen production demand prediction model in multiple dimensions, including:

[0068] Taking STCNet as the basic architecture of the hydrogen production demand prediction model, a multi - channel data fusion method is adopted at the input end of the hydrogen production demand prediction model to process the operating power and efficiency of hydrogen production equipment, the distribution of hydrogen storage facilities, the trajectories of hydrogen - powered vehicles, and environmental data in parallel; the spatial convolution layer extracts the topological features of the transportation network through a 3×3 convolution kernel, and the time - bidirectional LSTM layer extracts the periodic fluctuation law of hydrogen production demand from the source - domain dataset.

[0069] The present invention embeds an adversarial alignment module in the feature transfer layer of STCNet to form an adversarial transfer alignment layer. Through adversarial training, the loss function of the domain classifier is minimized to drive the feature distributions of the source domain and the target domain to converge, solving the problem of the difference in feature distributions between the source domain and the target domain, achieving alignment between the source domain and the target domain, and significantly improving the generalization ability of the hydrogen production demand prediction model in the scenario where the target - domain data is scarce.

[0070] Step 2.2: Taking the maximum utilization efficiency of hydrogen storage facilities and hydrogen refueling stations as the objective function of the dynamic scheduling plan of the hydrogen storage station, adopting the mixed - integer programming method, and establishing a hydrogen storage optimization model by using the objective function and hydrogen storage constraint conditions; according to the source - domain dataset, hydrogen production demand, and load distribution, the dynamic scheduling plan of hydrogen storage facilities is obtained by using the hydrogen storage optimization model.

[0071] The hydrogen storage optimization model satisfies the following relationship:

[0072]

[0073] Wherein, is the objective function of the dynamic scheduling plan of the hydrogen storage station, is the coefficient vector characterizing the maximization of the utilization efficiency of hydrogen storage facilities and hydrogen refueling stations, is the dynamic scheduling variable of the hydrogen storage station, is the coefficient matrix in the hydrogen storage constraint condition, is the hydrogen storage constraint vector, and the superscript represents the vector transpose;

[0074] In the hydrogen storage optimization model, the mixed integer programming (MIP) method is used to optimize the dynamic scheduling plan of hydrogen storage facilities and achieve the balance of hydrogen storage and hydrogen release requirements.

[0075] Step 2.3, adopt the multi-agent reinforcement learning method to establish a mobile hydrogen energy behavior prediction model; according to the source domain dataset, load distribution and the dynamic scheduling plan of hydrogen storage facilities, use the mobile hydrogen energy behavior prediction model to obtain the behavior and path of mobile hydrogen energy.

[0076] In the mobile hydrogen energy behavior prediction model, the behavior of each mobile hydrogen energy is regarded as an independent agent, and learns the optimal strategy in a shared environment. Train multiple agents to learn to predictively optimize the behavior of hydrogen energy and select the path of hydrogen energy, reduce resource conflicts, improve efficiency and optimize the global performance.

[0077] Moreover, based on the spatio-temporal graph convolutional network (ST-GCN), a road traffic capacity prediction model is constructed. The input layer receives the traffic network topology map and real-time traffic flow data, and accurately predicts the traffic efficiency of each road section in the future period through inter-layer feature propagation, which is used as the input of the mobile hydrogen energy behavior prediction model.

[0078] The hydrogen energy spatio-temporal prediction model proposed by the present invention effectively improves the prediction accuracy and reduces the prediction time, and has great economic and practical value.

[0079] Step 3, use the source domain dataset and the target domain dataset to perform transfer learning training and adversarial training on the hydrogen energy spatio-temporal prediction model.

[0080] Specifically, the same data in the source domain dataset and the target domain dataset are used to form transfer learning samples, and the hydrogen energy spatio-temporal prediction model is trained with the transfer learning samples; the different data in the source domain dataset and the target domain dataset are used to form adversarial samples, and the hydrogen energy spatio-temporal prediction model trained by transfer learning is trained with the adversarial samples.

[0081] In the embodiment, based on steps 1 and 2, a transfer learning module is set up to transfer the knowledge of the hydrogen energy spatio-temporal prediction model in the source domain to the target domain. During the transfer learning training process, a small amount of labeled data in the target domain is used to fine-tune the parameters of the model to improve the prediction performance of the model.

[0082] In the adversarial transfer alignment layer of the hydrogen production demand prediction model, through adversarial training, the loss function of the domain classifier is minimized to drive the feature distributions of the source domain and the target domain to converge; among them, the domain classifier of the adversarial transfer alignment layer is a fully connected layer, including adversarial training and gradient reversal; adversarial training is used to distinguish whether the features come from the source domain or the target domain, and gradient reversal is used to reverse the gradient of the domain classifier during backpropagation, so as to achieve adversarial training; the expression of gradient reversal is as follows:

[0083]

[0084] In the formula, is the gradient reversal, is the gradient function of the domain classifier for the features from the source domain during backpropagation, is the gradient function of the domain classifier for the features from the target domain during backpropagation, and is the feature extractor.

[0085] The present invention first performs feature alignment of the transfer learning module. By performing adversarial training on the hydrogen energy spatio-temporal prediction model in the source domain, the feature distributions of the source domain and the target domain are adjusted so that the target domain can be effectively explained by the source domain.

[0086] The graph convolutional mapping is used to map the spatial features of the target domain dataset into a space unified with the spatial features of the source domain; in the embodiment, the graph convolutional (Graph Convolutional Networks, GCN) mapping is used to map the spatial features such as the traffic network and hydrogen storage stations in the target domain into a unified feature space.

[0087] Step 4, a reward function is constituted by the economic index, efficiency index and environmental impact index of the hydrogen energy system in the target domain. According to the change amount of the target domain data, when the reward function reaches the maximum value, the trained hydrogen energy spatio-temporal prediction model outputs the cross-regional spatio-temporal prediction result of hydrogen energy use.

[0088] Specifically, step 4 includes:

[0089] ​Step 4.1, construct a reward function with the economic index, efficiency index, and environmental impact index of the hydrogen energy system in the target field, satisfying the following relationship:

[0090]

[0091] In the formula, is the reward function of the hydrogen energy system in the target field in state and action , , , are respectively the economic index, efficiency index, and environmental impact index of the hydrogen energy system in the target field in state and action , , , are respectively the weights of the economic index, efficiency index, and environmental impact index.

[0092] Among them, based on the energy price response model, apply the dynamic game reinforcement learning framework to cope with the energy market price fluctuations, and use the reward function of the energy price response model as the economic index, satisfying the following relationship:

[0093]

[0094] In the formula, is the reward function of the energy price response model, is the profit, is the market share, is the inventory backlog, , , are respectively the weights of profit, market share, and inventory backlog.

[0095] Step 4.2, adopt the reinforcement learning combined with the Pareto front optimization method to adjust the weights of the economic index, efficiency index, and environmental impact index to obtain the maximum value of the reward function.

[0096] Integrate the prediction results of hydrogen production, hydrogen storage, and mobile hydrogen energy, and conduct global optimization. Adopt the reinforcement learning combined with the Pareto front optimization method. When training the reinforcement learning agent, optimize the scheduling strategy by learning the feedback in the environment. When the agent selects an action in each state, it not only considers a single reward function, but evaluates the effect of each action according to the Pareto optimality of multiple objectives, and uses the Pareto front method to transform it into a single comprehensive objective function. By combining the reinforcement learning algorithm, the weights of the objective function can be continuously adjusted , , , to explore the optimal resource allocation strategy, and then dynamically adjust the resource allocation and scheduling plan among various modules, the load distribution prediction and scheduling suggestions of hydrogen production facilities, the dynamic capacity planning and usage plan of hydrogen storage facilities, and the path planning and hydrogen refueling station access probability of hydrogen energy vehicles.

[0097] Based on Step 4.2, according to the resource allocation strategy under the maximum reward function, continuously optimize the hydrogen production strategy through the deep deterministic policy gradient algorithm, which can maximize the economic benefits of the hydrogen energy system in the target field while ensuring stable supply.

[0098] Step 4.3, according to the change amount of the target field data, repeat Steps 4.1 and 4.2, and when the reward function reaches the maximum value, the hydrogen energy spatio-temporal prediction model trained outputs the cross-regional spatio-temporal prediction results of hydrogen energy usage.

[0099] Through the online learning of the model, automatically identify and correct the output results. During the operation of the system, continuously optimize the model performance through real-time data feedback, including:

[0100] Real-time collect data such as changes in hydrogen energy demand and traffic flow fluctuations, and dynamically adjust the input of the prediction model;

[0101] In the embodiment, the real-time collected changes in hydrogen energy demand and traffic flow fluctuations , dynamically adjust the input of the prediction model is:

[0102]

[0103] In the formula, is the historical hydrogen production load at the kth nearest moment, is the current environmental factors such as electricity price.

[0104] Update the input of the hydrogen energy spatio-temporal prediction model, dynamically adjust the scheduling plans of hydrogen storage facilities and hydrogen refueling stations, and adaptively update the model;

[0105] In the embodiment, use the online learning algorithm (such as incremental stochastic gradient descent) to slightly update the parameters of each model, improve the adaptability of the model to the new environment, and the anomaly detection module automatically identifies and corrects the deviation in the prediction. Specifically as follows:

[0106] Taking the minimization of the operation cost of the hydrogen energy system caused by the change amount of the target field data as the goal, update the dynamic scheduling plan of the hydrogen storage facilities output by the hydrogen storage optimization model, where the goal satisfies the following relational formula:

[0107]

[0108] In the formula, is the input data for the hydrogen energy spatio-temporal prediction model based on the change amount of target field data, is the operating cost of the hydrogen energy system caused by the change amount of target field data, is the constraint condition, is the time instant the hydrogen storage amount of the hydrogen storage facility, is the time instant the hydrogen input amount, is the time instant the hydrogen output amount.

[0109] For the online learning algorithm, in the present invention, the incremental stochastic gradient descent (ISGD) is adopted to update the model parameters:

[0110]

[0111] In the formula, is the model parameter at time t, is the learning rate, is the loss function of the prediction model, and is selected as , , are the predicted output and the actual output at time t, respectively.

[0112] If , in the formula, is the set threshold. If the condition is satisfied, the predicted result is output. When the threshold condition is not satisfied, the model parameters are slightly updated, and the size of the loss function difference is recalculated until the condition is satisfied and the result is output.

[0113] The present invention also proposes a cross-regional spatio-temporal prediction system for hydrogen energy use based on transfer learning, including:

[0114] A data acquisition module, which is used to extract the time features and space features of hydrogen production data, hydrogen storage data, mobile hydrogen energy behavior data and environmental data in the hydrogen energy system to form a source domain data set;

[0115] A modeling and training module, which is used to establish a hydrogen energy spatio-temporal prediction model, including: a hydrogen production demand prediction model, a hydrogen storage optimization model, and a mobile hydrogen energy behavior prediction model; and perform transfer learning training and adversarial training on the hydrogen energy spatio-temporal prediction model by using the source domain data set and the target domain data set;

[0116] A prediction module, which is used to form a reward function with the economic index, efficiency index and environmental impact index of the hydrogen energy system in the target field, and output the cross-regional spatio-temporal prediction result of hydrogen energy use by the trained hydrogen energy spatio-temporal prediction model when the reward function reaches the maximum value according to the change amount of the target field data.

[0117] The working process of the whole system is divided into a data collection and processing part, a source domain model training part, a transfer learning part, a joint prediction and optimization part, and an online learning and updating part. The data collection and processing part is responsible for collecting and standardizing data on hydrogen production, hydrogen storage, mobile hydrogen energy, and the environment from multi-source heterogeneous data, providing high-quality input for the model. The source domain model training part constructs models such as hydrogen production demand prediction, hydrogen storage optimization, and mobile hydrogen energy behavior modeling in areas with sufficient data, and uses advanced algorithms to accurately predict the dynamics of each link. The transfer learning part addresses the problem of scarce target domain data by transferring the knowledge of the source domain model to the target domain through feature alignment and parameter transfer, improving its prediction performance and achieving cross-regional knowledge sharing. The joint prediction and optimization part integrates the prediction results of each link, constructs a multi-objective dynamic collaborative optimization model, comprehensively considers objectives such as cost, loss, and service coverage rate, and dynamically adjusts resource allocation to achieve global optimization. The online learning and updating part collects data in real time during the operation of the system, updates the model parameters, and introduces anomaly detection to ensure that the system adapts to new environmental changes and operates stably and efficiently.

[0118] The method proposed in the present invention is oriented to the full-life cycle management of the hydrogen energy system, integrates multi-source heterogeneous information such as hydrogen production, hydrogen storage, mobile hydrogen energy, and environmental data, and constructs a composite data layer covering traffic flow, weather conditions, and policies and regulations. In terms of the transfer learning mechanism, an adversarial transfer learning framework is introduced, and cross-regional feature distribution alignment is achieved through a domain classifier and a gradient reversal mechanism, and a graph convolutional network is combined to model the spatial topological relationship of hydrogen storage stations. In addition, breaking through the limitation of a single optimization target, a multi-objective dynamic collaborative optimization system for hydrogen production cost, hydrogen storage loss, and service coverage rate is designed, and real-time update of model parameters and anomaly adaptive adjustment are achieved through online learning.

[0119] Improve the spatio-temporal cross-domain neural network (STCNet), introduce the adversarial transfer learning mechanism into the STCNet architecture, and design a domain classifier with a gradient reversal function, effectively solving the problem of feature distribution deviation in cross-regional prediction of the hydrogen energy system. Secondly, a hydrogen storage scheduling optimization module based on a graph convolutional network is developed, expanding the traditional spatio-temporal prediction model into a decision-making system with the ability to optimize spatial resources. Finally, a multi-objective dynamic optimization system is constructed, and real-time balance between economic indicators and operation efficiency is achieved through reinforcement learning. This technical integration is not a simple superposition, but a deep adaptation transformation targeting the characteristics of the hydrogen energy system. Mitigate the impact of data heterogeneity through the adversarial transfer layer, and improve the utilization rate of spatial resources by using graph structure modeling, ultimately forming a technical solution with domain pertinence. Achieve better results in terms of prediction accuracy, system response speed, multi-objective coordination, etc.

[0120] The present disclosure may be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions thereon for causing a processor to implement aspects of the present disclosure.

[0121] A computer-readable storage medium may be a tangible device that can retain and store instructions for use by an instruction execution device. A computer-readable storage medium may be, for example—but not limited to—an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, a mechanically encoded device such as a punch card or raised structures in grooves having instructions stored thereon, and any suitable combination of the foregoing. The computer-readable storage medium as used herein is not construed to be a transitory signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through an optical fiber cable), or an electrical signal transmitted through a wire.

[0122] The computer-readable program instructions described herein may be downloaded to respective computing / processing devices from a computer-readable storage medium or may be downloaded to an external computer or an external storage device through a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing / processing device.

[0123] Computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine - related instructions, microcode, firmware instructions, state - setting data, or source code or object code written in any combination of one or more programming languages, including object - oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer - readable program instructions may be executed entirely on the user's computer, partially on the user's computer, executed as a stand - alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider). In some embodiments, by using the state information of the computer - readable program instructions to customize an electronic circuit, such as a programmable logic circuit, a field - programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute the computer - readable program instructions to implement various aspects of the present disclosure.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent substitutions can still be made to the specific embodiments of the present invention. Any modification or equivalent substitution that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning, characterized in that: include: Extract the temporal and spatial characteristics of hydrogen production data, hydrogen storage data, mobile hydrogen energy behavior data, and environmental data in the hydrogen energy system to form a source domain data set; Establish a hydrogen energy spatiotemporal prediction model, including: a hydrogen production demand prediction model, a hydrogen storage optimization model, and a mobile hydrogen energy behavior prediction model; among them, according to the source domain data set, the hydrogen production demand and load distribution are output by using the hydrogen production demand prediction model established based on the spatiotemporal cross-domain neural network; with the maximization of the utilization efficiency of hydrogen storage facilities and hydrogen refueling stations as the objective function of the dynamic scheduling plan of hydrogen storage stations, according to the source domain data set, hydrogen production demand and load distribution, the dynamic scheduling plan of hydrogen storage facilities is output by using the hydrogen storage optimization model established by the objective function and hydrogen storage constraints; according to the source domain data set, load distribution and dynamic scheduling plan of hydrogen storage facilities, the mobile hydrogen energy behavior prediction model established by the multi-agent reinforcement learning method is used to output the behavior and path of mobile hydrogen energy; the hydrogen energy spatiotemporal prediction model is transferred and trained by using the source domain data set and the target domain data set; The reward function is composed of the economic indicators, efficiency indicators and environmental impact indicators of the hydrogen energy system in the target field. According to the change of the target field data, the trained hydrogen energy spatiotemporal prediction model outputs the cross-regional spatiotemporal prediction results of hydrogen energy usage when the reward function reaches the maximum value.

2. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 1 is characterized in that: Hydrogen production data include: operating power and efficiency of hydrogen production equipment, hydrogen production method, and carbon emission intensity; Hydrogen storage data includes: distribution of hydrogen storage facilities, hydrogen storage capacity, loss rate, and current storage status; Mobile hydrogen energy behavior data includes: hydrogen energy vehicle trajectory, hydrogen filling behavior, hydrogen refueling station distribution, and hydrogen refueling station utilization rate; Environmental data include: transportation network, traffic flow, weather conditions, policies and regulations, and economic activity levels.

3. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 1 is characterized in that: A hydrogen production demand prediction model is established based on a spatiotemporal cross-domain neural network, including multiple convolutional long short-term memory network units; The architecture of the hydrogen production demand prediction model includes: a feature extraction layer based on spatial convolution kernels, a time series modeling layer based on temporal bidirectional LSTM, and an adversarial migration alignment layer; Among them, an adversarial alignment module is embedded in the feature migration layer of the spatiotemporal cross-domain neural network to form an adversarial migration alignment layer.

4. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 1 is characterized in that: The hydrogen storage optimization model satisfies the following relationship: In the formula, is the objective function of the dynamic scheduling plan of the hydrogen storage station, To characterize the coefficient vector that maximizes the efficiency of hydrogen storage facilities and hydrogen refueling stations, is the dynamic scheduling variable of the hydrogen storage station, is the coefficient matrix in the hydrogen storage constraint condition, is the hydrogen storage constraint vector, with superscript Represents the transpose of a vector.

5. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 3 is characterized in that: The same data in the source domain dataset and the target domain dataset constitute transfer learning samples, and the hydrogen energy spatiotemporal prediction model is trained using the transfer learning samples. The different data in the source domain dataset and the target domain dataset constitute adversarial samples, and the hydrogen energy spatiotemporal prediction model trained by transfer learning is trained using adversarial samples.

6. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 5 is characterized in that: In adversarial training, the adversarial transfer alignment layer in the hydrogen production demand prediction model drives the feature distribution of the source domain and the target domain to converge by minimizing the loss function of the domain classifier; The domain classifier of the adversarial transfer alignment layer is a fully connected layer, including adversarial training and gradient reversal. Adversarial training is used to distinguish whether the features come from the source domain or the target domain. Gradient reversal is used to reverse the gradient of the domain classifier during back propagation. The expression of gradient reversal is as follows: In the formula, is the gradient reversal, is from the source domain during back propagation Features The gradient function of the domain classifier, is from the target domain during back propagation Features Domain Classifier The gradient function of is a feature extractor.

7. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 5 is characterized in that: Graph convolution mapping is used to map the spatial features of the target domain dataset into a space that is unified with the spatial features of the source domain.

8. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 1 is characterized in that: The reward function is composed of the economic indicators, efficiency indicators and environmental impact indicators of the hydrogen energy system in the target field, which satisfies the following relationship: In the formula, For the hydrogen energy system in the target area and actions The reward function when , , The hydrogen energy system in the target area is in the state and actions Economic indicators, efficiency indicators and environmental impact indicators, , , They are the weights of economic index, efficiency index and environmental impact index respectively.

9. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 8 is characterized in that: Taking the reward function of the energy price response model as the economic indicator, the following relationship is satisfied: In the formula, is the reward function of the energy price response model, For profit, For market share, For inventory backlog, , , They are the weights of profit, market share and inventory backlog respectively.

10. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 8, characterized in that: According to the change in the target domain data, when the reward function reaches the maximum value, the trained hydrogen energy spatiotemporal prediction model outputs the cross-regional spatiotemporal prediction results of hydrogen energy usage, including: Reinforcement learning combined with Pareto frontier optimization method is used to adjust the weights of economic indicators, efficiency indicators and environmental impact indicators to obtain the maximum value of the reward function; According to the change in the target area data, the reward function is repeatedly calculated and the weights are adjusted. When the reward function reaches the maximum value, the trained hydrogen energy spatiotemporal prediction model outputs the cross-regional spatiotemporal prediction results of hydrogen energy usage.

11. The method for cross-regional spatiotemporal prediction of hydrogen energy use based on transfer learning according to claim 10, characterized in that: With the goal of minimizing the operating cost of the hydrogen energy system caused by the change in the target field data, the dynamic scheduling plan of the hydrogen storage facilities output by the hydrogen storage optimization model is updated; wherein the goal satisfies the following relationship: In the formula, It is the input data of the hydrogen energy spatiotemporal prediction model based on the variation of target area data. The operating cost of the hydrogen energy system caused by the change in the target area data, As constraints, For the moment The amount of hydrogen stored in the hydrogen storage facility, For the moment Hydrogen input, For the moment Hydrogen output.

12. A cross-regional spatiotemporal prediction system for hydrogen energy use based on transfer learning, characterized in that: include: The data acquisition module is used to extract the temporal and spatial characteristics of hydrogen production data, hydrogen storage data, mobile hydrogen energy behavior data, and environmental data in the hydrogen energy system to form a source domain data set; Modeling and training module, used to establish hydrogen energy spatiotemporal prediction model, including: hydrogen production demand prediction model, hydrogen storage optimization model, mobile hydrogen energy behavior prediction model; wherein, according to the source domain data set, the hydrogen production demand and load distribution are output by using the hydrogen production demand prediction model established based on the spatiotemporal cross-domain neural network; the objective function of the dynamic scheduling plan of the hydrogen storage station is to maximize the utilization efficiency of hydrogen storage facilities and hydrogen refueling stations, and according to the source domain data set, hydrogen production demand and load distribution, the dynamic scheduling plan of the hydrogen storage facility is output by using the hydrogen storage optimization model established by the objective function and hydrogen storage constraints; according to the source domain data set, load distribution and dynamic scheduling plan of the hydrogen storage facility, the mobile hydrogen energy behavior prediction model established by the multi-agent reinforcement learning method is used to output the behavior and path of the mobile hydrogen energy; the hydrogen energy spatiotemporal prediction model is transferred and trained by using the source domain data set and the target domain data set; The prediction module is used to form a reward function based on the economic indicators, efficiency indicators and environmental impact indicators of the hydrogen energy system in the target field. According to the change of the target field data, when the reward function reaches the maximum value, the trained hydrogen energy spatiotemporal prediction model outputs the cross-regional spatiotemporal prediction results of hydrogen energy use.

13. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 11 are implemented.

Images