A natural gas gas consumption prediction model training method, device, equipment and storage medium

Abstract

The application discloses a natural gas gas consumption prediction model training method and device, equipment and a storage medium. The method comprises the following steps: performing feature decomposition on the initial historical natural gas gas consumption data obtained and preprocessed to generate an initial IMF sequence and an initial residual sequence; performing quality verification on the initial IMF sequence and the initial residual sequence to generate a target IMF sequence and a target residual sequence; inputting the target IMF sequence and the target residual sequence into a pre-constructed natural gas gas consumption prediction model to obtain a predicted natural gas gas consumption and a predicted IMF sequence output by the model; and performing model training on the natural gas gas consumption prediction model based on a standard natural gas gas consumption and the target IMF sequence according to the predicted natural gas gas consumption and the predicted IMF sequence until a preset model training end condition is met, so that a trained natural gas gas consumption prediction model is obtained, which is used for predicting the natural gas gas consumption in a target time period.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, equipment and storage medium for training a natural gas consumption prediction model. Background Technology

[0002] Natural gas consumption forecasting is a critical task in energy management, dispatch, and planning, especially given the increasing prevalence of natural gas as a clean energy source. Accurate forecasting helps optimize supply and demand balance, reduce operating costs, and improve energy efficiency. However, natural gas consumption data often exhibits strong nonlinearity, seasonality, and random fluctuations, influenced by various factors such as weather, economic activity, and policy regulation, making forecasting more challenging.

[0003] Traditional time series forecasting methods have significant limitations in handling nonlinear patterns and complex seasonality. They cannot effectively capture the impact of nonlinear fluctuations and sudden events. For example, forecast accuracy often decreases when facing seasonal demand peaks and troughs or external shocks, and they are sensitive to data noise, failing to adapt to the dynamic characteristics of natural gas consumption. With the rise of deep learning technology, neural network-based models, such as Long Short-Term Memory (LSTM) networks and Gated Recurrent Units (GRUs), have been widely applied in the field of natural gas forecasting. These models perform well in capturing complex patterns, but still have many drawbacks. For example, they are prone to gradient vanishing or exploding problems, especially when the sequence length exceeds several hundred steps, significantly increasing training time and reducing model generalization ability.

[0004] Therefore, existing natural gas consumption prediction models cannot effectively improve the training accuracy and generalization ability of deep learning models, resulting in low accuracy in predicting natural gas consumption during actual use. Summary of the Invention

[0005] This invention provides a method, apparatus, equipment, and storage medium for training a natural gas consumption prediction model, in order to improve the training accuracy of the natural gas consumption prediction model and thus improve the prediction accuracy of the natural gas consumption prediction model.

[0006] According to one aspect of the present invention, a method for training a natural gas consumption prediction model is provided, the method comprising: Acquire historical natural gas consumption data and its corresponding standard natural gas consumption for historical time periods, and perform data preprocessing on the historical natural gas consumption data to obtain initial historical natural gas consumption data; The initial historical natural gas consumption data is subjected to feature decomposition to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. The initial IMF sequence and initial residual sequence are quality verified, and the target IMF sequence and target residual sequence are generated based on the quality verification results; The target IMF sequence and target residual sequence are input into a pre-built natural gas consumption prediction model to obtain the predicted natural gas consumption and predicted IMF sequence output by the model. Based on the predicted natural gas consumption and the predicted IMF sequence, and based on the standard natural gas consumption and the target IMF sequence, the natural gas consumption prediction model is trained until the preset model training termination condition is met, resulting in a completed natural gas consumption prediction model. The natural gas consumption prediction model is used to predict natural gas consumption within a target time period.

[0007] According to another aspect of the present invention, a training apparatus for a natural gas consumption prediction model is provided, the apparatus comprising: The historical data acquisition module is used to acquire historical natural gas consumption data and its corresponding standard natural gas consumption data for historical time periods, and to perform data preprocessing on the historical natural gas consumption data to obtain initial historical natural gas consumption data. The feature decomposition module is used to perform feature decomposition on the initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. The quality verification module is used to verify the quality of the initial IMF sequence and the initial residual sequence, and generate the target IMF sequence and the target residual sequence based on the quality verification results. The model prediction module is used to input the target IMF sequence and the target residual sequence into a pre-built natural gas consumption prediction model to obtain the predicted natural gas consumption and the predicted IMF sequence output by the model. The model training module is used to train the natural gas consumption prediction model based on the predicted natural gas consumption and the predicted IMF sequence, and on the standard natural gas consumption and the target IMF sequence, until the preset model training termination condition is met, and obtain the natural gas consumption prediction model that has been trained; the natural gas consumption prediction model is used to predict the natural gas consumption within a target time period.

[0008] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the natural gas consumption prediction model training method according to any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the natural gas consumption prediction model training method according to any embodiment of the present invention.

[0010] The technical solution of this invention performs feature decomposition on initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. The initial IMF and residual sequences are then quality-verified, and a target IMF and residual sequence are generated based on the verification results. Adaptive multi-scale decomposition is used to achieve mode separation and noise suppression of non-stationary sequences. Combined with AMamba's selective state-space mechanism, efficient long-sequence modeling is achieved. Through noise-assisted optimization iteration, mode aliasing is dynamically suppressed and nonlinear residual modes are captured, resulting in more refined multi-scale feature extraction. This provides highly orthogonal and low-redundancy modal inputs for subsequent predictions, improving the model training accuracy and robustness of the natural gas consumption prediction model, thereby achieving accurate prediction of natural gas consumption.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a natural gas consumption prediction model training method according to Embodiment 1 of the present invention; Figure 2A This is a flowchart of a natural gas consumption prediction model training method according to Embodiment 2 of the present invention; Figure 2B This is a schematic diagram of the loss curve during the training process of a natural gas consumption prediction model according to Embodiment 2 of the present invention; Figure 2CThis is a schematic diagram of the prediction results of a natural gas consumption prediction model provided in Embodiment 2 of the present invention on a test set; Figure 2D This is a schematic diagram showing the comparison of the absolute error of a natural gas consumption prediction model provided in Embodiment 2 of the present invention with that of an existing model on the test set. Figure 2E This is a schematic diagram showing the comparison of the mean square error of a natural gas consumption prediction model provided in Embodiment 2 of the present invention with that of an existing model on a test set. Figure 2F This is a schematic diagram showing the comparison of the coefficient of determination of a natural gas consumption prediction model provided in Embodiment 2 of the present invention with an existing model on a test set. Figure 3 This is a schematic diagram of a natural gas consumption prediction model training device according to Embodiment 3 of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device for implementing the natural gas consumption prediction model training method of this invention. Detailed Implementation

[0014] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0015] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0016] Example 1 Figure 1This is a flowchart of a natural gas consumption prediction model training method provided in Embodiment 1 of the present invention. This embodiment is applicable to the prediction of natural gas consumption in scenarios such as urban natural gas load scheduling. The method can be executed by a natural gas consumption prediction model training device, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes: S110. Obtain historical natural gas consumption data and its corresponding standard natural gas consumption for historical time periods, and perform data preprocessing on the historical natural gas consumption data to obtain initial historical natural gas consumption data.

[0017] S120. Perform feature decomposition on the initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence.

[0018] S130. Perform quality verification on the initial IMF sequence and the initial residual sequence, and generate the target IMF sequence and the target residual sequence based on the quality verification results.

[0019] S140. Input the target IMF sequence and the target residual sequence into the pre-built natural gas consumption prediction model to obtain the predicted natural gas consumption and the predicted IMF sequence output by the model.

[0020] S150. Based on the predicted natural gas consumption and the predicted IMF sequence, and based on the standard natural gas consumption and the target IMF sequence, train the natural gas consumption prediction model until the preset model training termination condition is met, and obtain the natural gas consumption prediction model that has been trained. The natural gas consumption prediction model is used to predict the natural gas consumption within the target time period.

[0021] Historical natural gas consumption data can be the natural gas consumption for different time periods within a historical time frame. These time periods can include days, weeks, months, and years, and can be presented as a time series of natural gas consumption. Standard natural gas consumption can be the actual natural gas consumption for a future time period predicted based on historical natural gas consumption data. For example, if the historical natural gas consumption data is from May 1, 2021 to October 1, 2021, then the corresponding standard natural gas consumption can be the actual natural gas consumption from November 1, 2021 to December 1, 2021. In other words, standard natural gas consumption serves as the label value for historical natural gas consumption data and is used in model training.

[0022] Historical natural gas consumption data undergoes data preprocessing, which can be performed from at least one data processing dimension. This preprocessing may include data standardization, outlier removal, and normalization to ensure data compliance and accuracy. In one optional embodiment, the data preprocessing to obtain initial historical natural gas consumption data includes: standardizing the historical natural gas consumption data to obtain standardized historical natural gas consumption data; performing outlier detection on the standardized historical natural gas consumption data to generate compliant historical natural gas consumption data after outlier removal; and normalizing the compliant historical natural gas consumption data to generate initial historical natural gas consumption data.

[0023] Specifically, the time data in the historical natural gas consumption data is formatted uniformly, for example, the date format is standardized to "YYYY-MM-DD", a continuous time series is generated, and periodic characteristics such as date (dow) and month (month) are extracted to obtain standardized historical natural gas consumption data after date standardization.

[0024] Outlier detection is performed on standardized historical natural gas consumption data. Specifically, this involves calculating the mean and standard deviation of the standardized historical natural gas consumption data to generate a consumption average. and standard deviation of gas consumption and adopt The rules trim standardized historical natural gas consumption data, limiting its values ​​to a certain range. By removing outliers such as extreme noise, compliant historical natural gas consumption data is obtained, thereby improving the robustness of the distribution of compliant historical natural gas consumption data.

[0025] The Min-Max standardization method can be used to normalize the compliant historical natural gas consumption data to the range of [0, 1], thereby avoiding the gradient explosion problem after inputting it into the model and obtaining the normalized initial historical natural gas consumption data.

[0026] The above technical solution eliminates noise in historical natural gas consumption data, standardizes the units of measurement, regulates the data structure, and ensures data compliance and accuracy.

[0027] Feature decomposition is performed on the initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. For example, Optimized Multivariate Empirical Mode Decomposition (OMEMD) can be used to adaptively decompose the normalized initial historical natural gas consumption time series, extracting multiple highly orthogonal intrinsically initialized IMFs and initial residual components. This effectively captures nonlinear and non-stationary residual modes and significantly reduces mode aliasing, providing focused and low-redundancy modal-level inputs for subsequent models.

[0028] In one optional embodiment, feature decomposition is performed on the initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. This includes: for the first iteration cycle of the inner and outer loops, generating multiple sets of auxiliary signals and the current residual sequence based on the initial historical natural gas consumption data and preset initial iteration parameters; performing inner loop screening on the multiple sets of auxiliary signals to generate candidate IMF sequences corresponding to each auxiliary signal; generating a global IMF sequence based on each candidate IMF sequence, and updating the current residual sequence based on the global IMF sequence and the initial historical natural gas consumption data; generating a global modal aliasing index (MMI) based on the global IMF sequence, and updating the initial iteration parameters based on the global MMI to generate iteration parameters for the next iteration cycle; generating the IMF sequence and residual sequence for the next iteration cycle based on the iteration parameters for the next iteration cycle, until a preset iteration termination condition is met, resulting in an initial IMF and initial residual sequence that meet the iteration termination condition.

[0029] Based on initial historical natural gas consumption data Generate the initial noise standard deviation ,in, The value can be set to 0.1~0.2 to achieve adaptive multi-scale smoothing, suppress endpoint effects, and weaken outlier interference, ensuring the robustness of the decomposition process to nonlinear and non-stationary sequences. The initial iteration parameters include the initial noise standard deviation. Based on the initial noise standard deviation Generate multiple sets of auxiliary signals ,in . It is independent and identically distributed Gaussian white noise. The current residual sequence... Where k represents the number of auxiliary signals generated, which is M.

[0030] For multiple sets of auxiliary signals Perform inner loop screening to generate candidate IMF sequences for the first iteration cycle. Specifically, using auxiliary signals Taking the inner loop screening as an example, the initialization operation is performed first: ,according to The instantaneous component of the extreme point generation upper envelope and lower envelope Specifically, this can be obtained through cubic spline interpolation fitting. According to... The upper envelope generated by the instantaneous component and lower envelope Determine the mean envelope : Update the instantaneous components of the current iteration The updated instantaneous components are obtained. : The update of the k-th instantaneous component for the j-th iteration update number is as follows: The instantaneous components are continuously updated until a pre-defined termination condition is met. The termination condition is a dual condition, meaning it is considered satisfied when both conditions are met. Condition 1 is an extreme point or zero-crossing point constraint, which guarantees single-component nature; Condition 2 is a mean-envelope-standard deviation constraint, which guarantees a local mean of 0. Specifically, the constraint for Condition 1 is: The absolute value of the difference between the number of local extrema and the number of zero crossovers is no greater than 1. Zero crossovers are defined as follows: Where j represents the number of instantaneous component iterations.

[0031] Condition 2 constraint is the mean envelope. The standard deviation is less than the mean envelope threshold in the initial iteration parameters. The standard deviation is calculated using the following formula: Where T is the total length of the signal.

[0032] When all of the above conditions are met, the iteration stops, and the current iteration ends. This refers to the candidate IMF sequence, namely: If any condition is not met, then let j = j + 1, and return to the above steps with a new... Continue iterative screening. For example, for... If the above two constraints are met, i.e., extreme value or zero crossover difference = 1, Less than the mean envelope threshold Then you can get candidates sequence.

[0033] Continue with the above steps to... Screening to generate candidates The sequence can be obtained based on the above steps. According to the candidates Sequence, Candidate Sequence, ..., Candidate Sequences can determine global IMF sequences. : For the current residual sequence Iterative updates are obtained Generate the Modal Mixing Index (MMI) based on the global IMF sequence: Since the above only generates one global IMF sequence, therefore Continue generating the second one. That is, i=2, update MMI. Specifically, for the updated residual sequence... Perform inner loop screening to obtain candidates at iteration number i=2. Sequence, Candidate Sequence, ..., Candidate The sequence, therefore, can be based on the above sequence. Determine the global IMF sequence According to the global IMF sequence and It can be obtained .according to and Generate a global MMI: The initial iteration parameters are dynamically adjusted with the goal of minimizing the global MMI. and The iteration continues until a preset termination condition is met, i.e., until convergence, such as when the residual is monotonic or the global MMI rate of change is less than a preset threshold. The initial IMF that satisfies the iteration termination condition is obtained: and initial residual sequence The initial IMF sequence and initial residual sequence are quality-verified, and the target IMF sequence and target residual sequence are generated based on the quality verification results.

[0034] In one optional embodiment, the initial IMF sequence and the initial residual sequence are quality verified, and a target IMF sequence and a target residual sequence are generated based on the quality verification results. This includes: performing local orthogonality verification based on the initial IMF sequence to generate local orthogonality verification results; performing reconstruction integrity verification based on the initial IMF sequence, the initial residual sequence, and the initial historical natural gas consumption data to generate reconstruction integrity verification results; and generating the target IMF sequence and the target residual sequence based on the local orthogonality verification results and the reconstruction integrity verification results.

[0035] Specifically, for any two adjacent IMF sequences in the initial IMF sequence, determine the IMF correlation coefficients corresponding to the adjacent IMF sequences: The reconstruction error was determined based on the initial IMF sequence, the initial residual sequence, and the initial historical natural gas consumption data. And based on the reconstruction error Perform reconstruction integrity verification; if the reconstruction error is less than a preset threshold... Perform elimination or screening: The above-mentioned verification technology can eliminate invalid data with a frequency lower than the daily cycle or an energy percentage lower than 0.5%, and finally output a verified set of selected multi-scale modal components as independent channel input for subsequent models, ensuring high interpretability, low aliasing and strong prediction specificity, thereby improving the overall prediction accuracy and generalization ability.

[0036] The target IMF sequence and target residual sequence are input into a pre-built natural gas consumption prediction model to obtain the model output predicted natural gas consumption and predicted IMF sequence. The natural gas consumption prediction model can be pre-built or selected by relevant technical personnel; for example, it can be a convolutional neural network model, preferably an Amamba (Attention Mamba) based model.

[0037] Based on the predicted natural gas consumption and the predicted IMF sequence, the natural gas consumption prediction model is trained using the standard natural gas consumption and the target IMF sequence until the preset model training termination condition is met, resulting in a completed natural gas consumption prediction model. The natural gas consumption prediction model is used to predict natural gas consumption within a target time period.

[0038] Based on the predicted natural gas consumption and the standard natural gas consumption, a first loss value is determined based on a preset loss function, such as mean square error. Based on the predicted IMF sequence and the target IMF sequence, a second loss value is determined based on a preset loss function. The natural gas consumption prediction model is trained based on the first and second loss values ​​until the preset model training termination condition is met, resulting in a completed natural gas consumption prediction model.

[0039] In a practical application scenario of a natural gas consumption prediction model, the predicted natural gas consumption data for the current time period is acquired, and feature decomposition is performed on the predicted natural gas consumption data to obtain the current IMF sequence and residual sequence. The current IMF sequence and residual sequence are then output to the natural gas consumption prediction model to obtain the natural gas consumption for the future predicted time period.

[0040] The technical solution of this invention performs feature decomposition on initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. The initial IMF and residual sequences are then quality-verified, and a target IMF and residual sequence are generated based on the verification results. Adaptive multi-scale decomposition is used to achieve mode separation and noise suppression of non-stationary sequences. Combined with AMamba's selective state-space mechanism, efficient long-sequence modeling is achieved. Through noise-assisted optimization iteration, mode aliasing is dynamically suppressed and nonlinear residual modes are captured, resulting in more refined multi-scale feature extraction. This provides highly orthogonal and low-redundancy modal inputs for subsequent predictions, improving the model training accuracy and robustness of the natural gas consumption prediction model, thereby achieving accurate prediction of natural gas consumption.

[0041] Example 2 Figure 2A This is a flowchart of a natural gas consumption prediction model training method provided in Embodiment 2 of the present invention. This embodiment is an optimization and improvement based on the above technical solutions.

[0042] Furthermore, the pre-built natural gas consumption prediction model includes a periodic time encoding module, a multimodal fusion module, and a modal-level prediction module. The step "inputting the target IMF sequence and the target residual sequence into the pre-built natural gas consumption prediction model to obtain the predicted natural gas consumption and the predicted IMF sequence output by the model" is refined to "inputting the target IMF sequence into the periodic time encoding module for periodic encoding, which encodes the time period corresponding to the historical natural gas consumption data to obtain a periodic encoding feature vector, and generating a core time feature encoding result based on the periodic encoding feature vector and the target IMF sequence; inputting the core time feature encoding result into the multimodal fusion module for multimodal fusion to output a hidden state sequence; and inputting the hidden state sequence and the target residual sequence into the modal-level prediction module for model prediction to obtain the predicted natural gas consumption and the predicted IMF sequence." This improves the generation method of the target fusion feature map.

[0043] It should be noted that for parts not described in detail in the embodiments of the present invention, please refer to the descriptions in other embodiments. For example... Figure 2A As shown, the method includes the following specific steps: S210. Obtain historical natural gas consumption data and its corresponding standard natural gas consumption data for historical time periods, and perform data preprocessing on the historical natural gas consumption data to obtain initial historical natural gas consumption data.

[0044] S220. Perform feature decomposition on the initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence.

[0045] S230. Perform quality verification on the initial IMF sequence and the initial residual sequence, and generate the target IMF sequence and the target residual sequence based on the quality verification results.

[0046] S240. Input the target IMF sequence into the periodic time coding module of the natural gas consumption prediction model for periodic coding. The periodic time coding module encodes the time period corresponding to the historical natural gas consumption data to obtain the periodic coding feature vector. Based on the periodic coding feature vector and the target IMF sequence, the core time feature coding result is generated.

[0047] S250. Input the core time feature encoding results into the multimodal fusion module of the natural gas consumption prediction model for multimodal fusion and output the hidden state sequence.

[0048] S260. Input the hidden state sequence and the target residual sequence into the modal prediction module of the natural gas consumption prediction model to perform model prediction, and obtain the predicted natural gas consumption and the predicted IMF sequence.

[0049] S270. Based on the predicted natural gas consumption and the predicted IMF sequence, and based on the standard natural gas consumption and the target IMF sequence, train the natural gas consumption prediction model until the preset model training termination condition is met, and obtain the natural gas consumption prediction model that has been trained. The natural gas consumption prediction model is used to predict the natural gas consumption within the target time period.

[0050] The target IMF sequence is input into the periodic time coding module for periodic coding. The periodic time coding module encodes the time period corresponding to the historical natural gas consumption data to obtain the periodic coding feature vector. The core time feature coding result is generated based on the periodic coding feature vector and the target IMF sequence.

[0051] Specifically, the target IMF sequence is treated as an independent channel and then concatenated with weekday or month-based periodic codes to obtain a four-dimensional periodic code. The weekday code is as follows: in, , where dow represents the weekday.

[0052] The month codes are as follows: in, ,in, Represents the month. The generated periodic encoded feature vector is: Let the feature vector obtained by concatenating the target IMF sequence be... Then, the core time feature encoding result X(t) is obtained by concatenating the target IMF sequence and the periodic encoded feature vector: Encoding results of core time features The input is fed into a multimodal fusion module for multimodal fusion, and the output is a hidden state sequence. In an optional embodiment, the core temporal feature encoding result is input into the multimodal fusion module for multimodal fusion, and the output is a hidden state sequence, including: linearly projecting the core temporal feature encoding result to generate a linear projection result; performing positional encoding on the linear projection result to generate a temporal position insertion encoding result; performing multi-head self-attention fusion on the temporal position insertion encoding to generate a fused global feature result; and generating a hidden state sequence based on the fused global feature result and the Selective State Space Model (SSM).

[0053] Encoding results of core time features Perform linear projection and generate linear projection results. : in, Where d represents the hidden dimension, and The weight parameters are used to train the model, and the model is continuously iterated and optimized during training.

[0054] For linear projection results Perform position encoding to generate temporal position insertion encoding results. First, sinusoidal position encoding is performed to obtain the position encoding vector. and Where pos is the location index, representing the specific location of a data point in the time series T; i is the encoding index, representing the i-th feature dimension in the location encoding vector.

[0055] Insert the position encoding vector into the linear projection result The timing position insertion coding result is obtained. : Inserting codes at timing positions Perform multi-head self-attention fusion to generate a fused global feature result. : Based on the fusion of global feature results Hidden state sequences are generated based on the Selective Spatial State Model (SSM) recursion. Define the learnable projective weight matrix: , and corresponding paranoia , Projection calculation: right Perform Sorfmax normalization to obtain .right Discretize the state matrix to obtain and : Here, A and B are both fixed state parameter matrices that can be learned throughout the model.

[0056] based on and and the results of fusing global features Hide Update to get the current hidden state : Generate hidden state sequence : Where C is a fixed state parameter matrix that the model can learn throughout the entire process.

[0057] The hidden state sequence and the target residual sequence are input into the modal-level prediction module for model prediction to obtain the predicted natural gas consumption and the predicted IMF sequence. In an optional embodiment, inputting the hidden state sequence and the target residual sequence into the modal-level prediction module for model prediction to obtain the predicted natural gas consumption and the predicted IMF sequence includes: generating a global context based on the hidden state sequence; generating a predicted IMF sequence based on the global context; generating a normalized spatial reconstruction result based on the predicted IMF sequence and the target residual sequence; and generating the predicted natural gas consumption based on the normalized spatial reconstruction result.

[0058] Specifically, take the hidden state sequence The final step in hiding the state And hide the last step. As a global context. Based on the global context. Generate predicted IMF sequences : Based on the predicted IMF sequence Normalized spatial reconstruction is performed on the target residual sequence Res to generate the normalized spatial reconstruction result. The mean residual value corresponding to the target residual sequence Res is determined. The normalized space reconstruction process is as follows: Perform a Min-Max inverse transform on the normalized spatial reconstruction results to generate predicted natural gas consumption. : The mean squared error (MSE) loss function can be used to minimize the modality prediction error, and an early stopping mechanism and a cosine annealing learning rate scheduler can be introduced to prevent overfitting. During the testing phase, the mean absolute error (MAE), mean squared error (MSE), and coefficient of determination (R²) can be calculated. 2The model's predictive performance is evaluated by using the IMF sequence and the complete prediction sequence reconstructed through residual compensation.

[0059] In a specific natural gas consumption scenario, historical natural gas consumption data from January 1, 2021 to October 15, 2025 for a certain region is retrieved from a data source. The data includes date and consumption columns to support subsequent time series operations and periodic analysis. The data is divided into the first 1700 time points, with an input sequence length of 7 and a prediction length of 2. The selective SSM hidden dimension of the Amamba model is 128, and the number of multi-head attention heads is 4. The maximum number of iterations for optimizing the multivariate empirical mode decomposition is 10, with 200 training iterations. 80% of the data is used as the training set, and the remainder as the test set. Figure 2B The loss curve of the proposed method is shown, with MSE as the ordinate. As can be seen from the figure, the loss value decreases rapidly in the initial stage, then stabilizes and remains at a low level in subsequent rounds. This indicates that the model achieves effective convergence after 200 training rounds and has good training stability, further validating the robustness and optimization efficiency of the model in handling natural gas consumption time series prediction tasks. Furthermore, Figure 2C The results of the natural gas consumption prediction model proposed in this invention on the test set are shown. As can be seen from the figure, the predicted curve and the actual curve are highly consistent throughout the time range, which verifies the high accuracy and reliability of the model in the task of predicting natural gas consumption. Figure 2D , Figure 2E and Figure 2F The mean absolute error (MAE), mean squared error (MSE), and coefficient of determination (R²) of the proposed natural gas consumption prediction model compared to existing models STL-Transformer, Standard Transformer, LSTM, and GRU on the test set are presented respectively. 2 The comparison results show that the model achieves optimal performance across all metrics, with a MAE of 8.97, an MSE of 182.65, and a coefficient of determination R0. 2 The coefficient of determination (R²) reached 0.972. These results demonstrate that the proposed model exhibits higher prediction accuracy and better fitting ability in natural gas consumption forecasting, and achieves leading performance across all evaluation metrics, with a MAE of 8.97, an MSE of 182.65, and a coefficient of determination (R²) of 0.972. 2 The efficiency score reached 0.9718. Compared to STL-Transformer, this model improved in MAE, MSE, and R... 2The performance improvements were 11.9%, 32.0%, and 1.4%, respectively; compared to the classic Transformer, the improvements were 14.3%, 44.6%, and 2.5%, respectively; and compared to LSTM and GRU, its MAE was reduced by more than 29%, MSE by more than 56%, and R... 2 This also represents an improvement of approximately 4.1%. The above results fully demonstrate that the model proposed in this embodiment has higher prediction accuracy and better fitting ability in natural gas load forecasting tasks, providing a new approach for natural gas consumption forecasting.

[0060] Example 3 Figure 3 This is a schematic diagram of a natural gas consumption prediction model training device provided in Embodiment 3 of the present invention. The natural gas consumption prediction model training device provided in this embodiment of the present invention is applicable to natural gas consumption prediction in scenarios such as urban natural gas load scheduling. This natural gas consumption prediction model training device can be implemented in hardware and / or software, such as... Figure 3 As shown, the device includes: a historical data acquisition module 301, a feature decomposition module 302, a quality verification module 303, a model prediction module 304, and a model training module 305. Among them, The historical data acquisition module 301 is used to acquire historical natural gas consumption data and its corresponding standard natural gas consumption data under historical time periods, and to perform data preprocessing on the historical natural gas consumption data to obtain initial historical natural gas consumption data. Feature decomposition module 302 is used to perform feature decomposition on the initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. The quality verification module 303 is used to perform quality verification on the initial IMF sequence and the initial residual sequence, and generate the target IMF sequence and the target residual sequence based on the quality verification results; The model prediction module 304 is used to input the target IMF sequence and the target residual sequence into a pre-built natural gas consumption prediction model to obtain the predicted natural gas consumption and the predicted IMF sequence output by the model. The model training module 305 is used to train the natural gas consumption prediction model based on the predicted natural gas consumption and the predicted IMF sequence, and on the standard natural gas consumption and the target IMF sequence, until the preset model training termination condition is met, and obtain the natural gas consumption prediction model that has been trained; the natural gas consumption prediction model is used to predict the natural gas consumption within a target time period.

[0061] The technical solution of this invention performs feature decomposition on initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. The initial IMF and residual sequences are then quality-verified, and a target IMF and residual sequence are generated based on the verification results. Adaptive multi-scale decomposition is used to achieve mode separation and noise suppression of non-stationary sequences. Combined with AMamba's selective state-space mechanism, efficient long-sequence modeling is achieved. Through noise-assisted optimization iteration, mode aliasing is dynamically suppressed and nonlinear residual modes are captured, resulting in more refined multi-scale feature extraction. This provides highly orthogonal and low-redundancy modal inputs for subsequent predictions, improving the model training accuracy and robustness of the natural gas consumption prediction model, thereby achieving accurate prediction of natural gas consumption.

[0062] Optionally, the feature decomposition module 302 is specifically used for: For the first iteration cycle of the internal and external circulation, based on the initial historical natural gas consumption data and preset initial iteration parameters, multiple sets of auxiliary signals and the current residual sequence are generated. The multiple sets of auxiliary signals are subjected to internal loop sieving to generate candidate IMF sequences corresponding to each of the auxiliary signals; A global IMF sequence is generated based on each of the candidate IMF sequences, and the current residual sequence is updated based on the global IMF sequence and the initial historical natural gas consumption data. A global modal aliasing index (MMI) is generated based on the global IFM sequence, and the initial iteration parameters are updated based on the global MMI to generate the iteration parameters for the next iteration cycle. Based on the iteration parameters of the next iteration cycle, the IMF sequence and residual sequence of the next iteration cycle are generated until the preset iteration termination condition is met, and the initial IMF and initial residual sequence that meet the iteration termination condition are obtained.

[0063] Optional, quality verification module 303, specifically used for: Based on the initial IMF sequence, local orthogonality verification is performed to generate local orthogonality verification results; and, Based on the initial IMF sequence, the initial residual sequence, and the initial historical natural gas consumption data, the reconstruction integrity verification is performed, and the reconstruction integrity verification result is generated. Based on the local orthogonality verification results and the reconstruction integrity verification results, the target IMF sequence and the target residual sequence are generated.

[0064] Optionally, the pre-built natural gas consumption prediction model includes a periodic time encoding module, a multimodal fusion module, and the modal-level prediction module. The model prediction module 304 includes: The feature vector encoding unit is used to input the target IMF sequence into the periodic time encoding module for periodic encoding. The periodic time encoding module encodes the time period corresponding to the historical natural gas consumption data to obtain the periodic encoding feature vector, and generates the core time feature encoding result based on the periodic encoding feature vector and the target IMF sequence. The multimodal fusion unit is used to input the core temporal feature encoding result into the multimodal fusion module for multimodal fusion and output the hidden state sequence; The model prediction unit is used to input the hidden state sequence and the target residual sequence into the modal-level prediction module for model prediction to obtain the predicted natural gas consumption and the predicted IMF sequence.

[0065] Optional, multimodal fusion unit, specifically used for: The core time feature encoding result is linearly projected to generate a linear projection result; The linear projection result is position-encoded to generate a temporal position insertion encoding result; Multi-head self-attention fusion is performed on the temporal position insertion encoding to generate a fused global feature result; Based on the fusion of global features, a hidden state sequence is generated using the Selective Spatial State Model (SSM).

[0066] Optional, model prediction unit, specifically used for: Generate a global context based on the hidden state sequence; Based on the global context, generate a predicted IMF sequence; Based on the predicted IMF sequence and the target residual sequence, a normalized spatial reconstruction result is generated; Based on the normalized spatial reconstruction results, a predicted natural gas consumption is generated.

[0067] Optional, the historical data acquisition module 301 is specifically used for: The historical natural gas consumption data is standardized to obtain standardized historical natural gas consumption data. Anomaly detection is performed on the standardized historical natural gas consumption data to generate compliant historical natural gas consumption data after removing outliers; The compliant historical natural gas consumption data is normalized to generate initial historical natural gas consumption data.

[0068] The natural gas consumption prediction model training device provided in this embodiment of the invention can execute the natural gas consumption prediction model training method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0069] Example 4 Figure 4 A schematic diagram of an electronic device 40 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0070] like Figure 4 As shown, the electronic device 40 includes at least one processor 41 and a memory, such as a read-only memory (ROM) 42 or a random access memory (RAM) 43, communicatively connected to the at least one processor 41. The memory stores computer programs executable by the at least one processor. The processor 41 can perform various appropriate actions and processes based on the computer program stored in the ROM 42 or loaded from storage unit 48 into the RAM 43. The RAM 43 may also store various programs and data required for the operation of the electronic device 40. The processor 41, ROM 42, and RAM 43 are interconnected via a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0071] Multiple components in electronic device 40 are connected to I / O interface 45, including: input unit 46, such as keyboard, mouse, etc.; output unit 47, such as various types of monitors, speakers, etc.; storage unit 48, such as disk, optical disk, etc.; and communication unit 49, such as network card, modem, wireless transceiver, etc. Communication unit 49 allows electronic device 40 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0072] Processor 41 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 41 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 41 performs the various methods and processes described above, such as the natural gas consumption prediction model training method.

[0073] In some embodiments, the natural gas consumption prediction model training method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 48. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 40 via ROM 42 and / or communication unit 49. When the computer program is loaded into RAM 43 and executed by processor 41, one or more steps of the natural gas consumption prediction model training method described above can be performed. Alternatively, in other embodiments, processor 41 can be configured to execute the natural gas consumption prediction model training method by any other suitable means (e.g., by means of firmware).

[0074] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0075] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0076] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0077] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0078] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0079] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0080] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0081] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for training a natural gas consumption prediction model, characterized in that, include: Acquire historical natural gas consumption data and its corresponding standard natural gas consumption for historical time periods, and perform data preprocessing on the historical natural gas consumption data to obtain initial historical natural gas consumption data; The initial historical natural gas consumption data is subjected to feature decomposition to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. The initial IMF sequence and initial residual sequence are quality verified, and the target IMF sequence and target residual sequence are generated based on the quality verification results; The target IMF sequence and target residual sequence are input into a pre-built natural gas consumption prediction model to obtain the predicted natural gas consumption and predicted IMF sequence output by the model. Based on the predicted natural gas consumption and the predicted IMF sequence, and based on the standard natural gas consumption and the target IMF sequence, the natural gas consumption prediction model is trained until the preset model training termination condition is met, resulting in a completed natural gas consumption prediction model. The natural gas consumption prediction model is used to predict natural gas consumption within a target time period.

2. The method according to claim 1, characterized in that, The step of performing feature decomposition on the initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence includes: For the first iteration cycle of the internal and external circulation, based on the initial historical natural gas consumption data and preset initial iteration parameters, multiple sets of auxiliary signals and the current residual sequence are generated. The multiple sets of auxiliary signals are subjected to internal loop sieving to generate candidate IMF sequences corresponding to each of the auxiliary signals; A global IMF sequence is generated based on each of the candidate IMF sequences, and the current residual sequence is updated based on the global IMF sequence and the initial historical natural gas consumption data. A global modal aliasing index (MMI) is generated based on the global IFM sequence, and the initial iteration parameters are updated based on the global MMI to generate the iteration parameters for the next iteration cycle. Based on the iteration parameters of the next iteration cycle, the IMF sequence and residual sequence of the next iteration cycle are generated until the preset iteration termination condition is met, and the initial IMF and initial residual sequence that meet the iteration termination condition are obtained.

3. The method according to claim 1, characterized in that, The step of performing quality verification on the initial IMF sequence and the initial residual sequence, and generating the target IMF sequence and the target residual sequence based on the quality verification results, includes: Based on the initial IMF sequence, local orthogonality verification is performed to generate local orthogonality verification results; and, Based on the initial IMF sequence, the initial residual sequence, and the initial historical natural gas consumption data, the reconstruction integrity verification is performed, and the reconstruction integrity verification result is generated. Based on the local orthogonality verification results and the reconstruction integrity verification results, the target IMF sequence and the target residual sequence are generated.

4. The method according to claim 1, characterized in that, The pre-built natural gas consumption prediction model includes a periodic time encoding module, a multimodal fusion module, and the modal-level prediction module. The step of inputting the target IMF sequence and the target residual sequence into the pre-built natural gas consumption prediction model to obtain the model output predicted natural gas consumption and predicted IMF sequence includes: The target IMF sequence is input into the periodic time encoding module for periodic encoding. The periodic time encoding module encodes the time period corresponding to the historical natural gas consumption data to obtain a periodic encoding feature vector. The core time feature encoding result is generated based on the periodic encoding feature vector and the target IMF sequence. The core temporal feature encoding result is input into the multimodal fusion module for multimodal fusion, and the hidden state sequence is output. The hidden state sequence and the target residual sequence are input into the modal-level prediction module for model prediction to obtain the predicted natural gas consumption and the predicted IMF sequence.

5. The method according to claim 4, characterized in that, The step of inputting the core temporal feature encoding result into the multimodal fusion module for multimodal fusion and outputting a hidden state sequence includes: The core time feature encoding result is linearly projected to generate a linear projection result; The linear projection result is position-encoded to generate a temporal position insertion encoding result; Multi-head self-attention fusion is performed on the temporal position insertion encoding to generate a fused global feature result; Based on the fusion of global features, a hidden state sequence is generated using the Selective Spatial State Model (SSM).

6. The method according to claim 4, characterized in that, The step of inputting the hidden state sequence and the target residual sequence into the modal-level prediction module for model prediction to obtain the predicted natural gas consumption and the predicted IMF sequence includes: Generate a global context based on the hidden state sequence; Based on the global context, generate a predicted IMF sequence; Based on the predicted IMF sequence and the target residual sequence, a normalized spatial reconstruction result is generated; Based on the normalized spatial reconstruction results, a predicted natural gas consumption is generated.

7. The method according to claim 1, characterized in that, The process of preprocessing the historical natural gas consumption data to obtain initial historical natural gas consumption data includes: The historical natural gas consumption data is standardized to obtain standardized historical natural gas consumption data. Anomaly detection is performed on the standardized historical natural gas consumption data to generate compliant historical natural gas consumption data after removing outliers; The compliant historical natural gas consumption data is normalized to generate initial historical natural gas consumption data.

8. A training device for a natural gas consumption prediction model, characterized in that, include: The historical data acquisition module is used to acquire historical natural gas consumption data and its corresponding standard natural gas consumption data for historical time periods, and to perform data preprocessing on the historical natural gas consumption data to obtain initial historical natural gas consumption data. The feature decomposition module is used to perform feature decomposition on the initial historical natural gas consumption data to generate an initial intrinsic mode function (IMF) sequence and an initial residual sequence. The quality verification module is used to verify the quality of the initial IMF sequence and the initial residual sequence, and generate the target IMF sequence and the target residual sequence based on the quality verification results. The model prediction module is used to input the target IMF sequence and the target residual sequence into a pre-built natural gas consumption prediction model to obtain the predicted natural gas consumption and the predicted IMF sequence output by the model. The model training module is used to train the natural gas consumption prediction model based on the predicted natural gas consumption and the predicted IMF sequence, and on the standard natural gas consumption and the target IMF sequence, until the preset model training termination condition is met, and obtain the natural gas consumption prediction model that has been trained; the natural gas consumption prediction model is used to predict the natural gas consumption within a target time period.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the natural gas consumption prediction model training method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the natural gas consumption prediction model training method according to any one of claims 1-7.

Images