New energy green electricity transaction quantity reporting strategy algorithm based on deep learning

Abstract

The invention provides a new energy green electricity transaction quantity reporting strategy algorithm based on deep learning. The new energy green electricity transaction quantity reporting strategy algorithm based on deep learning comprises the following steps: a, data collection: acquiring multi-source data from an electricity market, a weather forecast system, historical power generation data and a load prediction system, the data including historical electricity price, weather conditions, power generation power and power grid load data, and recording as X = {x1, x2,..., xn}, wherein xi is a time sequence data feature vector. According to the new energy green power transaction quantity reporting strategy algorithm based on deep learning, the problem of prediction errors caused by randomness and intermittency in new energy power generation is innovatively solved through a deep learning technology, and the prediction accuracy of the new energy green power transaction quantity reporting strategy algorithm based on deep learning is improved. Power generation fluctuation and weather change features are extracted through a multilayer convolutional neural network, and extreme weather events are automatically identified in combination with an attention mechanism, so that the algorithm can more accurately capture market fluctuation.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy green electricity transaction reporting strategy, and specifically to a new energy green electricity transaction reporting strategy algorithm based on deep learning. Background Art

[0002] The new energy green electricity trading quotation strategy algorithm based on deep learning mainly uses deep learning technology to model and predict multi-dimensional data such as supply and demand in the electricity market, historical price data, weather forecasts, load fluctuations, etc. Its basic process usually includes six stages: data collection, preprocessing, feature extraction, model training, prediction output, and strategy optimization. In the data collection stage, the system obtains historical electricity prices, weather conditions, power generation data, and market information from multiple data sources. In the preprocessing stage, these data are normalized and denoised. Next, through deep learning algorithms (such as LSTM, CNN, etc.), the model learns these features to predict future electricity demand, supply fluctuations, and price trends. Based on the prediction results, the algorithm will generate the optimal quotation strategy to ensure that new energy green electricity producers obtain maximum benefits in the market while meeting the requirements of grid stability.

[0003] Although the system has high prediction accuracy and flexibility, it also has some defects. Since renewable energy generation (such as wind power and solar power) is highly random and intermittent, deep learning models may have large prediction errors when dealing with extreme weather or emergencies, resulting in the failure of the reporting strategy. Deep learning models often require a large amount of historical data for training, and the performance of the model will drop significantly when the data is insufficient or of poor quality. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a new energy green electricity trading reporting strategy algorithm based on deep learning, which solves the problem that new energy power generation has strong randomness and intermittency, and deep learning models may have large prediction errors when dealing with extreme weather or emergencies; when data is insufficient or of poor quality, the performance of the model will drop significantly.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a new energy green electricity transaction reporting strategy algorithm based on deep learning, comprising:

[0006] a. Data collection: Multi-source data is obtained from the power market, weather forecast system, historical power generation data and load forecast system. The data includes historical electricity prices, weather conditions, power generation and grid load data, denoted as X = {x1, x2, ..., x n}, where x i is the time series data feature vector;

[0007] b. Data preprocessing: normalize, denoise, fill missing values ​​and align time of the collected data using the normalization formula:

[0008]

[0009] To standardize the data range and eliminate unit differences;

[0010] c. Feature extraction: A multi-layer convolutional neural network (CNN) is used to extract the characteristic information of renewable energy power generation, including the time series characteristics of historical power generation fluctuations, weather impacts, and power demand fluctuations. The CNN model extracts features through the convolution layer, and the calculation formula is:

[0011]

[0012] where w ij is the weight, b j is the bias, f is the activation function;

[0013] d. Extreme weather event recognition: The convolutional neural network is combined with the attention mechanism to automatically recognize weather data and extract the characteristics of extreme weather events. The calculation formula of the attention mechanism is:

[0014]

[0015] Where Q is the query matrix, K is the key matrix, V is the value matrix, and dk is the dimension parameter;

[0016] e. Generative adversarial network data enhancement: Generative adversarial networks are used to generate virtual historical data to make up for the problem of insufficient data or low quality. The generative adversarial network includes a generator G and a discriminator D, and its objective function is:

[0017]

[0018] where p data (x) is the real data distribution, p z (z) is the noise distribution;

[0019] f. Long Short-Term Memory Network (LSTM) Modeling: The enhanced data is input into the LSTM network to make multi-step predictions on future renewable energy power generation, load demand, and electricity prices. The update formula of the LSTM unit is:

[0020] i t =σ(W i [h t-1 , x t ]+b i ), f t =σ(W f [h t-1, x t ]+b f )

[0021] o t =σ(W o [h t-1 , x t ]+b o ), c t =f t ⊙c t-1 +i t ⊙tanh(W c [h t-1 , x t ]+b c )

[0022] where i t , f t , o t They are input, forget, and output gates, respectively, and c t is the memory unit state;

[0023] g. Model integration: Random forest and XGBoost machine learning models are introduced and combined with LSTM network to form an integrated model. The integrated model integrates the results through weighted voting method. The formula is:

[0024]

[0025] where w i For each model weight, y i is the prediction result of each model;

[0026] h. Uncertainty estimation: The uncertainty of the deep learning prediction model is estimated using the Bayesian neural network to generate a confidence interval. The update formula of the Bayesian neural network is:

[0027]

[0028] Where p(θ|D) is the posterior distribution of the parameter under a given data set D, and p(θ) is the prior distribution;

[0029] i. Reporting strategy generation: Based on the prediction results of the deep learning model, the reinforcement learning algorithm is used to generate the optimal reporting strategy. The objective function of reinforcement learning is:

[0030]

[0031] Where Q(s, a) is the value function of state s and action a, r is the reward, and γ is the discount factor;

[0032] j. Real-time adjustment mechanism: According to the real-time price fluctuations of the power market, actual power generation and grid load, the deep learning model is quickly updated to dynamically adjust the reporting strategy. The sliding window mechanism is used for real-time updates with a window size of W to ensure that the model responds to the latest data;

[0033] k. Risk control: To cope with extreme weather or emergencies, a risk control mechanism is established to automatically trigger adjustment strategies when the forecast error is large or the grid load pressure is high;

[0034] I. Output results: Output the final reporting strategy and its corresponding risk control suggestions, and apply them to the power trading market to maximize the profits of new energy power generators and maintain the balance of the power grid.

[0035] Preferably, the step a includes acquiring power grid operation status monitoring data, wherein the data preprocessing in step b uses principal component analysis to perform feature dimension reduction.

[0036] Preferably, the feature extraction in step C is combined with external factors such as seasonality and holiday data, and the attention mechanism in step d is used to weight the impact of different weather factors.

[0037] Preferably, the training of the generative adversarial network generator and discriminator in step e adopts an adversarial training method.

[0038] Preferably, the LSTM network in step f predicts the power generation and electricity price of multiple-step time series.

[0039] Preferably, the integrated model in step g adopts a weighted voting method to fuse the prediction results of multiple models.

[0040] Preferably, the Bayesian neural network in step h uses Monte Carlo sampling for uncertainty estimation, the reinforcement learning algorithm in step i uses a deep Q network to optimize the reporting strategy, and the real-time adjustment mechanism in step j includes a model update method based on a sliding window.

[0041] Preferably, the risk control mechanism described in step k includes emergency adjustment trigger conditions, and the output strategy described in step I takes into account the peaks and valleys of price fluctuations in the electricity market.

[0042] The present invention provides a new energy green electricity transaction reporting strategy algorithm based on deep learning. It has the following beneficial effects:

[0043] This new energy green electricity trading reporting strategy algorithm based on deep learning, through deep learning technology, innovatively solves the prediction error problem caused by randomness and intermittency in new energy power generation. It extracts power generation fluctuations and weather change characteristics through a multi-layer convolutional neural network, and combines the attention mechanism to automatically identify extreme weather events, so that the algorithm can more accurately capture market fluctuations. It further introduces a generative adversarial network for data enhancement, which makes up for the performance degradation caused by insufficient or poor data quality, thereby improving the stability and robustness of the model under complex conditions.

[0044] The present invention demonstrates excellent performance in uncertainty estimation and dynamic strategy optimization by combining an integrated model with a reinforcement learning algorithm. The Bayesian neural network is used to generate confidence intervals for prediction results, thereby effectively coping with the uncertainty and volatility of renewable energy power generation. Through a real-time adjustment mechanism, the system can automatically update and generate the optimal reporting strategy based on the dynamic changes in the power market, thereby maximizing the profits of renewable energy generators and ensuring the stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] like Figure 1 As shown, the embodiment of the present invention provides a new energy green electricity transaction quotation strategy algorithm based on deep learning, including: a. Data collection: acquiring multi-source data from the power market, weather forecast system, historical power generation data and load forecast system, the data including historical electricity prices, weather conditions, power generation and grid load data, denoted as X = {x1, x2, ..., x n}, where x i is a time series data feature vector, including obtaining power grid operation status monitoring data, wherein the data preprocessing in step b uses principal component analysis to perform feature dimension reduction;

[0048] b. Data preprocessing: normalize, denoise, fill missing values ​​and align time of the collected data using the normalization formula:

[0049]

[0050] To standardize the data range and eliminate unit differences;

[0051] c. Feature extraction: A multi-layer convolutional neural network (CNN) is used to extract the characteristic information of renewable energy power generation, including the time series characteristics of historical power generation fluctuations, weather impacts, and power demand fluctuations. The CNN model extracts features through the convolution layer, and the calculation formula is:

[0052]

[0053] where w ij is the weight, b j is the bias, f is the activation function, the feature extraction is combined with external factors such as seasonality and holiday data, and the attention mechanism described in step d is used to weight the impact of different weather factors;

[0054] d. Extreme weather event recognition: The convolutional neural network is combined with the attention mechanism to automatically recognize weather data and extract the characteristics of extreme weather events. The calculation formula of the attention mechanism is:

[0055]

[0056] Where Q is the query matrix, K is the key matrix, V is the value matrix, and dk is the dimension parameter;

[0057] e. Generative adversarial network data enhancement: Generative adversarial networks are used to generate virtual historical data to make up for the problem of insufficient data or low quality. The generative adversarial network includes a generator G and a discriminator D, and its objective function is:

[0058]

[0059] Where P data (x) is the real data distribution, p z (z) is a noise distribution, and the generator and discriminator of the generative adversarial network are trained using an adversarial training method;

[0060] f. Long Short-Term Memory Network (LSTM) Modeling: The enhanced data is input into the LSTM network to make multi-step predictions on future renewable energy power generation, load demand, and electricity prices. The update formula of the LSTM unit is:

[0061] i t =σ(W i [h t-1 , x t ]+b i ), f t =σ(W f [h t-1 , x t ]+b f )

[0062] ot =σ(W o [h t-1 , x t ]+b o ), c t =f t ⊙c t-1 +i t ⊙tanh(W c [h t-1 , x t ]+b c )

[0063] where i t , f t , o t They are input, forget, and output gates, respectively, and c t is the memory unit state, and the LSTM network predicts the power generation and electricity price of multiple-step time series;

[0064] g. Model integration: Random forest and XGBoost machine learning models are introduced and combined with LSTM network to form an integrated model. The integrated model integrates the results through weighted voting method. The formula is:

[0065]

[0066] where w i For each model weight, y i is the prediction result of each model, and the integrated model adopts a weighted voting method to fuse the prediction results of multiple models;

[0067] h. Uncertainty estimation: The uncertainty of the deep learning prediction model is estimated using the Bayesian neural network to generate a confidence interval. The update formula of the Bayesian neural network is:

[0068]

[0069] Wherein p(θ |D) is the posterior distribution of the parameter under a given data set D, p(θ) is the prior distribution, the Bayesian neural network uses Monte Carlo sampling for uncertainty estimation, the reinforcement learning algorithm in the step uses a deep network optimization reporting strategy, and the real-time adjustment mechanism in the step includes a model update method based on a sliding window;

[0070] i. Reporting strategy generation: Based on the prediction results of the deep learning model, the reinforcement learning algorithm is used to generate the optimal reporting strategy. The objective function of reinforcement learning is:

[0071]

[0072] Where Q(s, a) is the value function of state s and action a, r is the reward, and γ is the discount factor;

[0073] j. Real-time adjustment mechanism: According to the real-time price fluctuations of the power market, actual power generation and grid load, the deep learning model is quickly updated to dynamically adjust the reporting strategy. The sliding window mechanism is used for real-time updates with a window size of W to ensure that the model responds to the latest data;

[0074] k. Risk control: In order to cope with extreme weather or emergencies, a risk control mechanism is established to automatically trigger an adjustment strategy when the forecast error is large or the grid load pressure is large. The risk control mechanism includes emergency adjustment trigger conditions. The output strategy in step I takes into account the price fluctuation peaks and valleys in the power market;

[0075] I. Output results: Output the final reporting strategy and its corresponding risk control suggestions, and apply them to the power trading market to maximize the profits of new energy power generators and maintain the balance of the power grid.

[0076] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A new energy green electricity trading volume reporting strategy algorithm based on deep learning, characterized in that: include: a. Data collection: Multi-source data is obtained from the power market, weather forecast system, historical power generation data and load forecast system. The data includes historical electricity prices, weather conditions, power generation and grid load data, denoted as X = {x1, x2, …, x n }, where x i is the time series data feature vector; b. Data preprocessing: normalize, denoise, fill missing values ​​and align time of the collected data using the normalization formula: To standardize the data range and eliminate unit differences; c. Feature extraction: A multi-layer convolutional neural network (CNN) is used to extract the characteristic information of renewable energy power generation, including the time series characteristics of historical power generation fluctuations, weather impacts, and power demand fluctuations. The CNN model extracts features through the convolution layer, and the calculation formula is: where w ij is the weight, b j is the bias, f is the activation function; d. Extreme weather event recognition: The convolutional neural network is combined with the attention mechanism to automatically recognize weather data and extract the characteristics of extreme weather events. The calculation formula of the attention mechanism is: Where Q is the query matrix, K is the key matrix, V is the value matrix, and d k is the dimension parameter; e. Generative adversarial network data enhancement: Generative adversarial networks are used to generate virtual historical data to make up for the problem of insufficient data or low quality. The generative adversarial network includes a generator G and a discriminator D, and its objective function is: where p data (x) is the real data distribution, p z (z) is the noise distribution; f. Long Short-Term Memory Network (LSTM) Modeling: The enhanced data is input into the LSTM network to make multi-step predictions on future renewable energy power generation, load demand, and electricity prices. The update formula of the LSTM unit is: i t =σ(W i [h t-1 ,x t ]+b i ),f t =σ(W f [h t-1 ,x t ]+b f ) o t =σ(W o [h t-1 ,x t ]+b o ),c t =f t ⊙c t-1 +i t ⊙tanh(W c [j t-1 ,x t ]+b c ) where i t ,f t ,o t They are input, forget, and output gates, respectively, and c t is the memory unit state; g. Model integration: Random forest and XGBoost machine learning models are introduced and combined with LSTM network to form an integrated model. The integrated model integrates the results through weighted voting method. The formula is: where w i For each model weight, y i is the prediction result of each model; h. Uncertainty estimation: The uncertainty of the deep learning prediction model is estimated using the Bayesian neural network to generate a confidence interval. The update formula of the Bayesian neural network is: Where p(θ|D) is the posterior distribution of the parameters under a given data set D, and p(θ) is the prior distribution; i. Reporting strategy generation: Based on the prediction results of the deep learning model, the reinforcement learning algorithm is used to generate the optimal reporting strategy. The objective function of reinforcement learning is: Where Q(s,a) is the value function of state s and action a, r is the reward, and γ is the discount factor; j. Real-time adjustment mechanism: According to the real-time price fluctuations of the power market, actual power generation and grid load, the deep learning model is quickly updated to dynamically adjust the reporting strategy. The sliding window mechanism is used for real-time updates with a window size of W to ensure that the model responds to the latest data; k. Risk control: To cope with extreme weather or emergencies, a risk control mechanism is established to automatically trigger adjustment strategies when the forecast error is large or the grid load pressure is high; I. Output results: Output the final reporting strategy and its corresponding risk control suggestions, and apply them to the power trading market to maximize the profits of new energy power generators and maintain the balance of the power grid.

2. According to the deep learning-based new energy green electricity trading volume reporting strategy algorithm of claim 1, it is characterized by: The step a includes acquiring power grid operation status monitoring data, wherein the data preprocessing in step b uses principal component analysis to perform feature dimension reduction.

3. According to the deep learning-based new energy green electricity trading volume reporting strategy algorithm of claim 1, it is characterized by: The feature extraction described in step C is combined with external factors such as seasonality and holiday data, and the attention mechanism described in step d is used to weight the impact of different weather factors.

4. According to the deep learning-based new energy green electricity trading volume reporting strategy algorithm of claim 1, it is characterized by: The training of the generator and discriminator of the generative adversarial network described in step e adopts an adversarial training method.

5. According to the deep learning-based new energy green electricity trading volume reporting strategy algorithm of claim 1, it is characterized by: The LSTM network described in step f predicts the power generation and electricity price of multiple time series.

6. The new energy green electricity transaction reporting strategy algorithm based on deep learning according to claim 1 is characterized by: The integrated model described in step g adopts a weighted voting method to fuse the prediction results of multiple models.

7. The new energy green electricity transaction reporting strategy algorithm based on deep learning according to claim 1 is characterized by: The Bayesian neural network described in step h uses Monte Carlo sampling to perform uncertainty estimation, the reinforcement learning algorithm described in step i uses a deep Q network to optimize the reporting strategy, and the real-time adjustment mechanism described in step j includes a model update method based on a sliding window.

8. The new energy green electricity transaction reporting strategy algorithm based on deep learning according to claim 1 is characterized by: The risk control mechanism described in step k includes emergency adjustment trigger conditions, and the output strategy described in step I takes into account the price fluctuation peaks and valleys in the electricity market.

Images