Power plant carbon emission prediction method and system based on combined neural network model

Abstract

The invention discloses a power plant carbon emission prediction method and system based on a combined neural network model, and aims to solve the problems of weak data feature capture capability, insufficient prediction precision, incapability of self-adaptive learning and the like in the prior art. The method comprises the steps of acquiring data, preprocessing the data, dividing a data set into a training set and a verification set, constructing a combined neural network model, performing hyper-parameter optimization, training a carbon emission prediction model and realizing carbon emission prediction based on the carbon emission prediction model. The system comprises a data acquisition module, a data preprocessing module, a data set division module, a prediction model construction module, a carbon emission prediction module and a model updating optimization module. According to the method, the convolutional neural network model and the long-short-term memory network model are combined, the local time pattern and the long-term time sequence dependency relationship of the operation parameters and the carbon emission values are extracted respectively, and the model can focus on important parts in the time sequence in combination with an attention mechanism. The invention belongs to the technical field of carbon emission.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission, and in particular to a method and system for predicting carbon emission of a power plant based on a combined neural network model. Background Art

[0002] Thermal power plants generate electricity by burning fossil fuels such as coal, natural gas, and oil. While meeting society's demand for electricity, they also produce a large amount of greenhouse gases such as carbon dioxide (CO2). According to statistics, about 40% of the world's carbon emissions come from thermal power plants, posing a severe challenge to the sustainable development of human society. Many countries have introduced strict carbon emission control policies and carbon trading systems, prompting the power generation industry to find cleaner and more efficient technologies to reduce its carbon emissions. In this context, accurately predicting the carbon emissions of thermal power plants is of great practical significance for formulating carbon reduction strategies, optimizing energy production processes, and reducing environmental pollution. As the most important link in the "power-carbon emissions" system, thermal power companies have a large amount of actual operation data of power plants. However, due to the nonlinearity of carbon emission data and the numerous influencing parameters, how to build a high-precision prediction model in a data-driven way is still a major problem.

[0003] Traditional prediction models, such as time series models and support vector machine models, have various limitations. The time series model is one of the commonly used methods in the early days. It usually assumes that the data has a linear trend and is difficult to effectively deal with the characteristics of complex nonlinear data, resulting in low prediction accuracy. Although the support vector machine (SVM) model can handle a certain degree of nonlinear relationships, when the data scale is large and there are many variables, the training efficiency and prediction accuracy of the SVM model are often limited, making it difficult to adapt to actual carbon emission prediction scenarios.

[0004] Neural network models are also widely used in carbon emission prediction due to their powerful nonlinear mapping capabilities. Neural network models can adaptively learn complex relationships in data and improve the accuracy of prediction to a certain extent. However, general neural network models still face problems such as overfitting, insufficient model stability, and complex model parameter adjustment when processing high-dimensional, large-scale, and diverse power plant operation data. In addition, as the operating years of power plants increase, changes in the operating status and aging of equipment will also affect their carbon emission characteristics. General neural network models do not take into account the time characteristics of carbon emissions from power plants, and do not adaptively learn and update the latest mapping relationship between operating parameters and carbon emissions.

[0005] In summary, although a variety of models have been applied to predict carbon emissions from power plants, their accuracy and reliability are still limited. How to build a more accurate, generalized, and self-learning model to effectively predict carbon emissions from thermal power plants has become a key issue that needs to be solved. Summary of the invention

[0006] In order to solve the problems existing in the prior art such as weak data feature capture capability, insufficient prediction accuracy and inability to adaptively learn, the present invention proposes a method and system for predicting carbon emissions from power plants based on a combined neural network model.

[0007] The technical solution adopted by the present invention to solve the above problems is:

[0008] The present invention provides a method for predicting carbon emissions from power plants based on a combined upgraded network model, the method comprising the following steps:

[0009] Step 1, obtaining the historical operating parameters and corresponding carbon emission data of the target power plant;

[0010] Step 2, preprocessing the data obtained in step 1 to establish a data set;

[0011] Step 3, divide the data set created in step 2 into a training set and a validation set;

[0012] Step 4: Build a combined neural network model based on convolutional neural network, long short-term memory neural network and attention mechanism.

[0013] Step 5: Use the Bayesian hyperparameter optimization algorithm to find and determine the hyperparameter settings of the combined neural network model to obtain the final combined neural network model.

[0014] Step 6: Input the data divided in step 2 into the combined neural network model, train the model using the back propagation algorithm, and obtain the final carbon emission prediction model.

[0015] Step 7: Input the operating parameters of the prediction target into the carbon emission model in step 5 for prediction, obtain the carbon emission prediction value and output it.

[0016] A carbon emission prediction system for power plants based on a combined upgraded network model includes a data acquisition module, a data preprocessing module, a data set partitioning module, a prediction model building module, a carbon emission prediction module and a model update optimization module.

[0017] The data acquisition module is used to obtain the historical operating parameters and carbon emission data of the target power plant, and upload them to the data preprocessing module.

[0018] The data preprocessing module preprocesses the received data to obtain a data set. The preprocessing includes mean interpolation processing of abnormal data points and normalization processing of the data.

[0019] The data set division module divides the data set output by the data processing module into a training set and a verification set. The training set is used to train the combined neural network model, and the verification set is used to verify the prediction results of the combined neural network model.

[0020] The prediction model construction module is used to construct a combined neural network model based on the back propagation algorithm, use the Bayesian hyperparameter optimization algorithm to confirm the model hyperparameter settings, and use the training set output by the data set partitioning module as the model input to train the combined neural network model to obtain the final carbon emission prediction model.

[0021] The carbon emission prediction module is a final carbon emission prediction model based on the training output of the model building module, inputs the operating parameters of the prediction target, and outputs the corresponding carbon emission prediction value.

[0022] The model update optimization module means that in the subsequent actual operation process, if the error between the carbon emission prediction value output by the carbon emission prediction module and the actual true value exceeds the set value, the model update optimization module will pass the input value (operating parameter) and carbon emission value (actual true value) at this time as the latest data to the data preprocessing module, and add the latest preprocessed data to the training set in the data set partitioning module, and then use the updated training set to retrain the combined neural network model.

[0023] The beneficial effects of the present invention are:

[0024] 1. The present invention provides a carbon emission prediction method based on a combined neural network model, which combines a convolutional neural network model and a long short-term memory network model to extract the local time pattern and long-term time series dependency of operating parameters and carbon emission values, respectively, and combines the attention mechanism to enable the model to focus on the important parts of the time series. Finally, the model can achieve high-precision prediction of carbon emissions from power plants and has strong generalization ability.

[0025] 2. The present invention uses the Bayesian hyperparameter optimization algorithm to optimize the hyperparameters of the combined neural network model, which reduces the difficulty of model construction and shortens the model construction process and training cycle.

[0026] 3. The present invention provides a carbon emission prediction system based on a combined neural network model. The latest data in actual operation is added to the training set in the data set partitioning module through the model update optimization module, and then the combined neural network model is retrained with the updated training set, thereby realizing adaptive learning of the latest mapping relationship between operating parameters and carbon emissions, taking into account the time characteristics of carbon emissions from power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of a method for predicting carbon emissions from power plants based on a combined neural network model according to the present invention;

[0028] Figure 2 This is a structural diagram of a power plant carbon emission prediction system based on a combined neural network model of the present invention;

[0029] Figure 3 This is a schematic diagram of the structure of the combined neural network model proposed in the present invention. DETAILED DESCRIPTION

[0030] The power plant carbon emission prediction method based on the combined neural network model described in this embodiment is as follows: Figure 1 As shown, the method includes the following steps:

[0031] Step 1, obtaining the historical operating parameters and corresponding carbon emission data of the target power plant;

[0032] The operating parameters may include but are not limited to the following parameters: power generation load, heating load, fuel quantity, boiler main steam flow, boiler main steam pressure, boiler feed water flow, boiler feed water pressure and coal quality parameters, etc. The operating parameters used in this embodiment are power generation load, heating load, and fuel quantity. The operating parameters and corresponding carbon emission data are sampled from 4058 actual operating data of a 660MW unit in Hubei in March 2023, and the data sampling interval is 5 minutes.

[0033] Step 2, preprocessing the data obtained in step 1 to establish a data set;

[0034] The preprocessing includes mean interpolation processing of abnormal data points and normalization processing of data.

[0035] The abnormal data point refers to a data point when the target data is missing or exceeds a set range.

[0036] The data normalization process mentioned above means that since the model variables involve multiple physical parameters and the variables have different magnitudes, this will cause the model to update some feature weights too frequently or too slowly during the training process, affecting the performance of the model. Z-score normalization is used to ensure that the data has similar scale and distribution, which is conducive to better learning and fitting of the neural network model:

[0037]

[0038] In the above formula, X norm is the standardized data, X is the original data, μ is the mean of the data, and σ is the standard deviation of the data.

[0039] The establishment of the data set refers to generating time series segments from the historical operation data in chronological order according to a specific time step, so as to subsequently input them into the combined neural network model for sequence prediction under the time step.

[0040] The time step is used as one of the hyperparameters of the combined neural network model, and the optimal value is obtained by searching according to the Bayesian hyperparameter optimization algorithm.

[0041] Step 3, divide the data set in step 2 into a training set and a validation set;

[0042] The data set division refers to randomly dividing the time series segments in the data set established in step 3 into a training set and a validation set in a ratio of 4:1.

[0043] The training set is used to train the combined neural network model, and the validation set is used to validate the prediction results of the combined neural network model.

[0044] Step 4: Build a combined neural network model based on convolutional neural network, long short-term memory neural network and attention mechanism.

[0045] The Convolutional Neural Network (CNN) is a deep learning model that is good at extracting local features from data and was originally used for image processing tasks. It automatically extracts features from input data through a series of convolution operations, making it suitable for processing spatiotemporal data and inputs with complex local patterns. A convolutional neural network consists of a convolution layer, a pooling layer, and an activation function.

[0046] The core operation of the convolution layer is the convolution operation, and its output can be expressed by the following formula:

[0047]

[0048] In the above formula, X is the input data matrix, W is the convolution kernel matrix (filter), M and N are the height and width of the convolution kernel, b is the bias term, and Y[i,j] represents the value of the (i,j) position in the output feature map.

[0049] The pooling layer is used for dimensionality reduction and feature aggregation. The common maximum pooling operation can be expressed as:

[0050] Z[i,j]=max(X[i:i+M,j:j+N])

[0051] In the above formula, Z[i,j] represents the output after pooling, the pooling window size is M×N, and the maximum value is taken for the local area.

[0052] The activation function σ introduces nonlinearity into the model, such as the ReLU function, which is defined as:

[0053] σ(x)=max(0,x)

[0054] For the input time series, the CNN extracts the local change pattern within a period of time through the convolution kernel, and then uses these local features as the input of LSTM to further process the long-term dependency.

[0055] The Long Short-Term Memory (LSTM) network is a special recurrent neural network (RNN) designed specifically to handle long-term dependencies in sequence data. LSTM introduces a gating mechanism that enables it to effectively capture and memorize long-term and short-term dependencies in time series. LSTM controls the transmission of information through three gates in its internal structure (input gate, forget gate, and output gate). Input Gate: controls whether new input information is added to the current memory state; Forget Gate: determines which old information needs to be forgotten or retained; Output Gate: determines which information is output to the next moment. LSTM can dynamically adjust the information flow based on the current input and past states, retain useful information and ignore irrelevant information, thereby solving the problem of gradient disappearance in traditional RNN.

[0056] The core of the long short-term memory network is the introduction of three gating mechanisms: input gate, forget gate and output gate. The calculation formula of the LSTM unit at time step t is as follows:

[0057] Forget Gate t :

[0058] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0059] Input Gate i t and candidate memory cells

[0060] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0061]

[0062] Memory unit update C t :

[0063]

[0064] Output gate o t and hidden state h t :

[0065] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0066] h t =o t tanh(C t )

[0067] In the above formula, x t is the input of the current time step, h t-1 is the hidden state of the previous time step, C t-1 is the memory unit of the previous time step, W f , W i , W C and W o is a trainable weight matrix, b f , b i , b C and b o is the bias, σ is the sigmoid activation function, and tanh is the hyperbolic tangent activation function.

[0068] The long short-term memory network is used to process the feature sequence extracted by CNN. LSTM specifically captures the temporal dependencies in the data, especially the long-term dependencies. CNN extracts local features, while LSTM learns the global pattern of the entire time series through its recursive structure.

[0069] The core of the attention mechanism is to dynamically focus on key information by calculating the importance weight of each position in the input sequence to enhance important features and avoid irrelevant information from affecting the final result. Specifically, we first use the scoring function S t :

[0070] s t =tanh(Whh t +b h )

[0071] Then use the softmax function to normalize and get the weight α of each group of input vectors t :

[0072]

[0073] In the above formula, W y is the weight of Attention; b y is the deviation of Attention; v is the attention value; is the output of the Attention layer at time t; α t is the weight coefficient.

[0074] Step 5: Use the Bayesian hyperparameter optimization algorithm to find and determine the hyperparameter settings of the combined neural network model to obtain the final combined neural network model.

[0075] The Bayesian Hyperparameter Optimization algorithm is an intelligent search method for optimizing the hyperparameters of machine learning models. This algorithm integrates two key steps, sampling and optimization. Since the algorithm uses the gradient information of the optimization target, the search process of the hyperparameter is accelerated. In the sampling phase, the algorithm evaluates the observed value of the objective function under the current hyperparameter configuration according to the preset number of sampling times. Next, in the optimization phase, the algorithm searches for the next promising hyperparameter configuration based on the sampling information and the gradient information of the optimization target. The number of optimization rounds determines the number of searches, while the model performance index determines the optimization target. This method can not only search for hyperparameters more effectively in high-dimensional space, but also find a hyperparameter combination with good performance faster when resources are limited. In this embodiment, the Bayesian algorithm optimization target is the mean square error MSE, the number of sampling rounds is 30 rounds, the number of optimization rounds is 50 rounds, and the optimized hyperparameters include: the number of convolution kernels and the size of the convolution kernel of the CNN layer, the number of neural units, the number of layers and the forgetting (Dropout) rate of the LSTM layer. The search space of each hyperparameter is set based on experience, and its specific range and the final optimal hyperparameter combination can be seen in Table 1.

[0076] Table 1 Bayesian search space and optimal hyperparameter combination

[0077]

[0078] Step 6: Input the data divided in step 2 into the combined neural network model, train the model using the back propagation algorithm, and obtain the final carbon emission prediction model.

[0079] The structure diagram of the combined neural network model is as follows Figure 3 As shown in the figure. The input data first passes through the CNN layer, and the local features of the time series are extracted through convolution and pooling operations. The local features extracted by the CNN layer are used as the input of the LSTM layer. The LSTM layer uses the gating mechanism to process the long-term and short-term dependencies in the time series data and capture the global temporal dynamic characteristics. Then, the Attention mechanism calculates the attention weights and assigns different weights to different features to enhance the model's attention to important information. Finally, the fully connected layer summarizes the output results of the Attention layer to obtain the target output data of the model, that is, the carbon emission prediction value.

[0080] The back propagation algorithm is used to calculate the gradient of the loss function relative to the model parameters and provide the gradient information required by the optimizer. It propagates the error from the output layer to the input layer layer by layer through the chain rule and calculates the parameter gradient of each layer. In this implementation case, the loss function used is the mean square error MSE, and its formula is as follows:

[0081]

[0082] In the above formula, n is the number of samples, y is i is the true value of the i-th sample, is the model's predicted value for the i-th sample.

[0083] The optimizer uses the Adam optimizer. The Adam optimizer (Adaptive Moment Estimation) is a gradient descent optimization algorithm commonly used in deep learning. It combines the advantages of the momentum and RMSProp algorithms, can adaptively adjust the learning rate, and use different learning rates in different directions, making the optimization process more stable and efficient.

[0084] Step 7: Input the operating parameters of the prediction target into the carbon emission model in step 5 for prediction, obtain the carbon emission prediction value and output it.

[0085] In this implementation case, three evaluation indicators were used to evaluate and compare the performance of the neural network model, namely, root mean square error (RMSE), mean relative percentage error (MAPE), and coefficient of determination (R 2 ), and their calculation formulas are as follows:

[0086]

[0087]

[0088] In the above formula, y i Indicates the real carbon emission data, It represents the carbon emission data predicted by the model, and n represents the number of samples.

[0089] The proposed neural network model is compared with the CNN model and the LSTM model, and the error evaluation index results of each model are shown in Table 2. As can be seen from the table, the model proposed in this paper has the best accuracy in all aspects (R 2 =0.9531, RMSE=19.443, MAPE=3.545), compared with the CNN model, RMSE and MAPE decreased by 22.402% and 17.786%, respectively; compared with the LSTM model, RMSE and MAPE decreased by 28.845% and 25.541%, respectively.

[0090] Table 2 Error evaluation indicators of each model

[0091]

[0092] To achieve this goal, the present invention designs a power plant carbon emission prediction system based on a combined upgraded network model, which includes a data acquisition module, a data preprocessing module, a data set partitioning module, a prediction model building module, a carbon emission prediction module and a model update optimization module.

[0093] The data acquisition module is used to obtain the historical operating parameters and carbon emission data of the target power plant, and upload them to the data preprocessing module.

[0094] The data preprocessing module preprocesses the received data to obtain a data set. The preprocessing includes mean interpolation processing of abnormal data points and normalization processing of the data.

[0095] The data set division module divides the data set output by the data processing module into a training set and a verification set. The training set is used to train the combined neural network model, and the verification set is used to verify the verification result of the combined neural network model.

[0096] The prediction model construction module is used to construct a combined neural network model based on the back propagation algorithm, use the Bayesian hyperparameter optimization algorithm to confirm the model hyperparameter settings, and use the training set output by the data set partitioning module as the model input to train the combined neural network model to obtain the final carbon emission prediction model.

[0097] The carbon emission prediction module is a final carbon emission prediction model based on the training output of the model building module, inputs the operating parameters of the prediction target, and outputs the corresponding carbon emission prediction value.

[0098] The model update optimization module means that in the subsequent actual operation process, if the error between the carbon emission prediction value output by the carbon emission prediction module and the actual true value exceeds the set value, the model update optimization module will pass the input value (operating parameter) and carbon emission value (actual true value) at this time as the latest data to the data preprocessing module, and add the latest preprocessed data to the training set in the data set partitioning module, and then use the updated training set to retrain the combined neural network model.

[0099] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for predicting carbon emissions from power plants based on a combined neural network model, characterized in that: The method comprises the following steps: Step 1: Obtain historical operating parameters and corresponding carbon emission data of the target power plant; Step 2: pre-process the data obtained in step 1 to establish a data set; Step 3: Divide the data set created in step 2 into a training set and a validation set; Step 4: Construct a combined neural network model based on convolutional neural network, long short-term memory neural network and attention mechanism; Step 5: Use the Bayesian hyperparameter optimization algorithm to find and determine the hyperparameter settings of the combined neural network model to obtain the final combined neural network model; Step 6: Input the data divided in step 2 into the combined neural network model, train the model using the back propagation algorithm, and obtain the final carbon emission prediction model; Step 7: Input the operating parameters of the prediction target into the carbon emission model in step 5 for prediction, obtain the carbon emission prediction value and output it.

2. The method for predicting carbon emissions from power plants based on a combined neural network model according to claim 1, characterized in that: In step 1, the operating parameters may include but are not limited to the following parameters: power generation load, heating load, fuel quantity, boiler main steam flow, boiler main steam pressure, boiler feed water flow, boiler feed water pressure and coal quality parameters.

3. The method for predicting carbon emissions from power plants based on a combined neural network model according to claim 1, characterized in that: In step 2, the preprocessing includes mean interpolation processing of abnormal data points and normalization processing of data.

4. The method for predicting carbon emissions from power plants based on a combined neural network model according to claim 1, characterized in that: In step 3, the data set division refers to randomly dividing the time series segments in the data set established in step 3 into a training set and a validation set in a ratio of 4:

1.

5. A power plant carbon emission prediction system based on a combined neural network model, characterized in that: The system includes a data acquisition module, a data preprocessing module, a data set partitioning module, a prediction model building module, a carbon emission prediction module and a model update optimization module. The data acquisition module is used to acquire the historical operating parameters and carbon emission data of the target power plant and upload them to the data preprocessing module; The data preprocessing module preprocesses the data to obtain a data set according to the received data; the preprocessing includes mean interpolation processing of abnormal data points and normalization processing of the data; The data set division module divides the data set output by the data processing module into a training set and a verification set; The training set is used to train the combined neural network model, and the validation set is used to validate the prediction results of the combined neural network model; The model building module is used to build a combined neural network model based on the back propagation algorithm, use the Bayesian hyperparameter optimization algorithm to confirm the model hyperparameter settings, and use the training set output by the data set partitioning module as the model input to train the combined neural network model, and obtain the final carbon emission prediction model; The carbon emission prediction module is a final carbon emission prediction model based on the training output of the model building module, inputs the operating parameters of the prediction target, and outputs the corresponding carbon emission prediction value; The model update optimization module means that in the subsequent actual operation process, if the error between the carbon emission prediction value output by the carbon emission prediction module and the actual true value exceeds the set value, the model update optimization module will pass the input value and carbon emission value at this time as the latest data to the data preprocessing module, and add the latest preprocessed data to the training set in the data set partitioning module, and then use the updated training set to retrain the combined neural network model.

Images