Natural gas consumption and generating capacity prediction system
Through a system of data acquisition, preprocessing, multimodal data fusion modeling and human-computer interaction, combined with thermodynamic equations and machine learning models, the problem of time-consuming and inaccurate calculation of power generation based on natural gas consumption was solved, fast and accurate predictions were achieved, and the economic benefits of gas turbine power plants were improved.
Patent Information
- Application Number
- CN202510987064.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-17
- Publication Date
- 2025-09-16
AI Technical Summary
The existing method of calculating power generation based on natural gas consumption is time-consuming and inaccurate, resulting in reduced economic efficiency of unit operation.
A system that uses data collection, preprocessing, multimodal data fusion modeling, dynamic adjustment, human-computer interaction and hybrid prediction, combined with thermodynamic equations and machine learning models, can predict natural gas consumption and power generation in real time, and ensure system security and explainability through digital twin technology and blockchain.
It achieves fast and accurate forecasts of natural gas consumption and power generation, improving the profitability and operating economy of gas turbine power plants.
Smart Images
Figure CN120654898A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power generation, and in particular relates to a natural gas usage and power generation prediction system. Background Art
[0002] Kunshan Thermal Power has two combined cycle units: one operates continuously for power generation and heat production, while the other serves as a peak backup unit. Daily natural gas usage and power generation curves must be reported before 10:00 AM. Due to reduced government subsidies for gas turbine generators, daily power generation and natural gas usage must be strictly controlled. This necessitates rapid and accurate calculation of the next day's natural gas usage and the need to schedule a power generation curve based on the next day's natural gas consumption.
[0003] The existing method of calculating power generation based on natural gas consumption is mainly through manual calculation. However, manual calculation is time-consuming and may be inaccurate, resulting in reduced economic efficiency of unit operation. Summary of the Invention
[0004] In view of the problem that the existing method of calculating power generation based on natural gas consumption is time-consuming and inaccurate, resulting in reduced economic efficiency of unit operation, the present invention provides a natural gas consumption and power generation prediction system.
[0005] In order to achieve the above technical objectives, the technical solutions adopted by the present invention are as follows:
[0006] A natural gas consumption and power generation prediction system includes a data acquisition unit, a data preprocessing unit, a multimodal data fusion modeling unit, a dynamic adjustment unit, a human-computer interaction unit, and a hybrid prediction unit;
[0007] Data acquisition unit, used to collect natural gas calorific value, ambient temperature, power generation curve and heat supply;
[0008] The data preprocessing unit uses wavelet transform to denoise the time series data in the data acquisition unit;
[0009] Multimodal data fusion modeling unit, which builds a benchmark model based on thermodynamic equations;
[0010] The dynamic adjustment unit uses the FTRL-Proximal (Follow-the-Regularized-Leader) online optimizer to update model parameters in real time and automatically eliminate invalid features through L1 regularization;
[0011] The human-computer interaction unit screens the automatically generated next-day power generation curve and marks the uncertainty period (such as the variance of the Monte Carlo Dropout simulation >15%). The user adjusts the key nodes and refits the compliance curve through constrained optimization (CVXPY); human experience is embedded in the AI prediction in the form of hard constraints (such as the maximum climbing rate).
[0012] The hybrid prediction unit inputs the natural gas calorific value, ambient temperature, power generation curve and heat supply processed by the data preprocessing unit into the thermodynamic equation to build a benchmark model, and predicts the daily power generation and natural gas usage in real time.
[0013] Furthermore, the data preprocessing unit also includes a curve data dynamic adjustment module and a feature cross-enhancement module. The curve data dynamic adjustment module is used to perform dynamic time warping (DTW) alignment on the power generation / heating curve; and solve the phase offset problem caused by equipment start-up and shutdown.
[0014] The feature cross-enhancement module is used to quantify the lagged correlation between ambient temperature and power generation efficiency through Granger causality analysis, construct a three-dimensional tensor of calorific value-temperature-power generation, and extract latent features using Tucker decomposition.
[0015] Furthermore, the multimodal data fusion modeling unit includes a physical model layer, a data-driven layer, and a reinforcement learning layer;
[0016] Physical model layer: build a benchmark model based on thermodynamic equations;
[0017] Data-driven layer: uses an LSTM-Transformer dual-branch structure, with one branch capturing long-term trends and the other processing sudden changes (such as extreme temperatures).
[0018] Reinforcement learning layer: Dynamically adjust model weights (such as the confidence ratio of the physical model and the neural network) through DQN.
[0019] Furthermore, a benchmark model pre-training phase is constructed based on thermodynamic equations: Meta-Learning (such as the MAML algorithm) is trained on sufficient historical data;
[0020] During the fine-tuning phase, a baseline model is constructed based on thermodynamic equations. For new devices, a personalized model can be generated through few-shot learning with only 48 hours of data.
[0021] Furthermore, the dynamic adjustment unit also includes a virtual calibration module based on digital twins, which is used to build a digital twin of the equipment and simulate the theoretical operating status in real time;
[0022] When the deviation between the measured data and the simulation exceeds a threshold, Bayesian Optimization is triggered to calibrate the model parameters;
[0023] Output calibration report for manual review (interpretable design).
[0024] Furthermore, a counterfactual query module is also provided in the human-computer interaction unit, which generates a generative adversarial network (GAN) based on hypothetical questions set by the user to simulate data distribution under different working conditions.
[0025] Furthermore, the entire natural gas usage and power generation forecasting system is also equipped with a safety monitoring unit, including a data credibility verification module and an anti-interference module;
[0026] Data credibility verification module, which detects sensor anomalies based on Autoencoder reconstruction errors;
[0027] Use blockchain to store key operation logs (such as model parameter modifications) to ensure audit traceability.
[0028] The anti-interference module introduces adversarial sample generation (FGSM attack simulation) during training to improve anti-interference capabilities.
[0029] Compared with the prior art, the present invention has the following beneficial effects:
[0030] By using power generation consumption models at different ambient temperatures and load rates, we can accurately declare the amount of natural gas and arrange the power generation curve, rationally utilize each cubic meter of natural gas and accurately generate each kilowatt-hour of electricity, thereby improving the profitability of gas turbine power plants. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 The figure is a block diagram of the overall structure of a natural gas usage and power generation prediction system in an embodiment of the present invention. DETAILED DESCRIPTION
[0032] In order to facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and drawings. The contents mentioned in the embodiments are not intended to limit the present invention.
[0033] like Figure 1 As shown, this embodiment provides a natural gas consumption and power generation prediction system, including a data acquisition unit, a data preprocessing unit, a multimodal data fusion modeling unit, a dynamic adjustment unit, a human-computer interaction unit and a hybrid prediction unit;
[0034] Data acquisition unit, used to collect natural gas calorific value, ambient temperature, power generation curve and heat supply;
[0035] The data preprocessing unit uses wavelet transform to denoise the time series data in the data acquisition unit;
[0036] Multimodal data fusion modeling unit, which builds a benchmark model based on thermodynamic equations;
[0037] The dynamic adjustment unit uses the FTRL-Proximal (Follow-the-Regularized-Leader) online optimizer to update model parameters in real time, while automatically eliminating invalid features through L1 regularization; lightweight models (such as distilled TinyLSTM) are deployed on edge computing nodes to ensure low-latency response.
[0038] The human-computer interaction unit screens the automatically generated next-day power generation curve and marks the uncertainty period (such as the variance of the Monte Carlo Dropout simulation >15%). The user adjusts the key nodes and refits the compliance curve through constrained optimization (CVXPY); human experience is embedded in the AI prediction in the form of hard constraints (such as the maximum climbing rate).
[0039] The hybrid prediction unit inputs the natural gas calorific value, ambient temperature, power generation curve and heat supply processed by the data preprocessing unit into the thermodynamic equation to build a benchmark model, and predicts the daily power generation and natural gas usage in real time.
[0040] The data preprocessing unit also includes a curve data dynamic adjustment module and a feature cross-enhancement module. The curve data dynamic adjustment module is used to perform dynamic time warping (DTW) alignment on the power generation / heating curves and solve the phase offset problem caused by equipment start-up and shutdown.
[0041] The feature cross-enhancement module is used to quantify the lagged correlation between ambient temperature and power generation efficiency through Granger causality analysis, construct a three-dimensional tensor of calorific value-temperature-power generation, and extract latent features using Tucker decomposition.
[0042] The multimodal data fusion modeling unit includes a physical model layer, a data-driven layer, and a reinforcement learning layer;
[0043] Physical model layer: build a benchmark model based on thermodynamic equations;
[0044] Data-driven layer: uses an LSTM-Transformer dual-branch structure, with one branch capturing long-term trends and the other processing sudden changes (such as extreme temperatures).
[0045] Reinforcement learning layer: Dynamically adjust model weights (such as the confidence ratio of the physical model and the neural network) through DQN.
[0046] Pre-training phase: Building a benchmark model based on thermodynamic equations: Training Meta-Learning (such as the MAML algorithm) on sufficient historical data;
[0047] During the fine-tuning phase, a baseline model is constructed based on thermodynamic equations. For new devices, a personalized model can be generated through few-shot learning with only 48 hours of data.
[0048] The dynamic adjustment unit also includes a virtual calibration module based on digital twins. The virtual calibration module based on digital twins is used to build a digital twin of the equipment and simulate the theoretical operating status in real time.
[0049] When the deviation between the measured data and the simulation exceeds a threshold, Bayesian Optimization is triggered to calibrate the model parameters;
[0050] Output calibration report for manual review (interpretable design).
[0051] The human-computer interaction unit also has a counterfactual query module, which generates a generative adversarial network (GAN) based on hypothetical questions set by the user to simulate data distribution under different working conditions.
[0052] The entire natural gas usage and power generation forecasting system also has a safety monitoring unit, including a data credibility verification module and an anti-interference module;
[0053] Data credibility verification module, which detects sensor anomalies based on Autoencoder reconstruction errors;
[0054] Use blockchain to store key operation logs (such as model parameter modifications) to ensure audit traceability.
[0055] The anti-interference module introduces adversarial sample generation (FGSM attack simulation) during training to improve anti-interference capabilities.
[0056] Compared with the prior art, the present invention has the following beneficial effects:
[0057] By using power generation consumption models at different ambient temperatures and load rates, we can accurately declare the amount of natural gas and arrange the power generation curve, rationally utilize each cubic meter of natural gas and accurately generate each kilowatt-hour of electricity, thereby improving the profitability of gas turbine power plants.
[0058] The natural gas usage and power generation forecasting system automatically calculates natural gas usage based on the next day's power generation curve and can also schedule power generation based on the next day's natural gas availability. The natural gas usage and power generation forecasting system collects data on power generation curves, natural gas calorific value, heat supply, and ambient temperature. The system includes power generation consumption models for different ambient temperatures and load rates. It also uses recent natural gas calorific value and ambient temperature data to quickly and accurately estimate natural gas usage and schedule power generation.
[0059] The above provides a detailed introduction to the natural gas usage and power generation forecasting system provided by this application. The description of the specific embodiments is intended only to facilitate understanding of the method and core concepts of this application. It should be noted that those skilled in the art may make various improvements and modifications to this application without departing from the principles of this application, and such improvements and modifications are also within the scope of protection of the claims of this application.
Claims
1. A natural gas consumption and power generation prediction system, comprising a data acquisition unit, a data preprocessing unit, a multimodal data fusion modeling unit, a dynamic adjustment unit, a human-computer interaction unit, and a hybrid prediction unit; Data acquisition unit, used to collect natural gas calorific value, ambient temperature, power generation curve and heat supply; The data preprocessing unit uses wavelet transform to denoise the time series data in the data acquisition unit; Multimodal data fusion modeling unit, which builds a benchmark model based on thermodynamic equations; The dynamic adjustment unit uses the FTRL-Proximal online optimizer to update model parameters in real time and automatically eliminate invalid features through L1 regularization; The hybrid prediction unit inputs the natural gas calorific value, ambient temperature, power generation curve, and heat supply processed by the data preprocessing unit into the thermodynamic equation to build a benchmark model, and predicts the daily power generation and natural gas usage in real time; The human-computer interaction unit screens the automatically generated next-day power generation curve and marks the uncertainty period. The user adjusts the key nodes and refits the compliance curve through constrained optimization.
2. A natural gas consumption and power generation prediction system according to claim 1, characterized in that: The data preprocessing unit also includes a curve data dynamic adjustment module and a feature cross enhancement module. The curve data dynamic adjustment module is used to dynamically time-warp the power generation / heating curves; The feature cross-enhancement module is used to quantify the lagged correlation between ambient temperature and power generation efficiency through Granger causality analysis, construct a three-dimensional tensor of calorific value-temperature-power generation, and extract latent features using Tucker decomposition.
3. A natural gas consumption and power generation prediction system according to claim 2, characterized in that: The multimodal data fusion modeling unit includes a physical model layer, a data-driven layer, and a reinforcement learning layer; Physical model layer: build a benchmark model based on thermodynamic equations; Data-driven layer: uses an LSTM-Transformer dual-branch structure, with one branch capturing long-term trends and the other processing sudden changes. Reinforcement learning layer: Dynamically adjust model weights through DQN.
4. A method according to claim 3, characterized in that The natural gas consumption and power generation forecasting system builds a benchmark model based on thermodynamic equations. Pre-training phase: Meta-Learning is trained on sufficient historical data. Fine-tuning phase: Build a baseline model based on thermodynamic equations. For new devices, only 48 hours of data is needed to generate a personalized model through Few-shot Learning.
5. A natural gas consumption and power generation prediction system according to claim 4, characterized in that: The dynamic adjustment unit also includes a virtual calibration module based on digital twins. The virtual calibration module based on digital twins is used to build a digital twin of the equipment and simulate the theoretical operating status in real time. When the deviation between the measured data and the simulation exceeds a threshold, the Bayesian optimization calibration model parameters is triggered; Output calibration report for manual review.
6. A natural gas consumption and power generation prediction system according to claim 5, characterized in that: The human-computer interaction unit also has a counterfactual query module, which generates an adversarial network based on hypothetical questions set by the user to simulate data distribution under different working conditions.
7. A natural gas consumption and power generation prediction system according to claim 6, characterized in that: The entire natural gas usage and power generation forecasting system also has a safety monitoring unit, including a data credibility verification module and an anti-interference module; Data credibility verification module, which detects sensor anomalies based on Autoencoder reconstruction errors; Use blockchain to store key operation logs; The anti-interference module introduces adversarial sample generation during training to improve anti-interference capabilities.