A desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm

Abstract

The present invention relates to the technical field of artificial intelligence research, and specifically to a desulfurization intelligent prediction system based on an FT-LSTM-XGBoost algorithm, comprising a data acquisition module, a data preprocessing module, a feature engineering module, a model training module, a model fusion module and a control execution module, wherein the data acquisition module monitors and collects multidimensional parameters in a desulfurization process in real time; the data preprocessing module preprocesses the collected multidimensional parameters; the feature engineering module forms a final feature set related to desulfurization; the model training module uses Fourier transform, long short-term memory network and extreme gradient boosting tree to train a desulfurization prediction model; the model fusion module obtains a fused prediction result; the control execution module adjusts the overall operating parameters of the desulfurization system in real time; the present invention has high-precision prediction capability, thereby ensuring accurate control of the desulfurization effect.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence research technology, and in particular to a desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm. Background Art

[0002] The energy structure and power production have undergone tremendous changes. In particular, the coal-fired power industry, as an emergency, supply guarantee, and regulatory power source, is in urgent need of transforming traditional operation, management, and regulation into data-driven, refined management, and intelligent regulation under conditions of variable loads and even start-stop peak regulation. This ensures that coal-fired power units can operate safely, controllably, greenly, low-carbon, economically, efficiently, and flexibly under deep peak regulation operation modes.

[0003] At present, the ultra-low emission facilities of coal-fired power plants are basically in the traditional manual control operation and management stage. They rely on the experience of front-line workers and the real-time collection of values ​​by sensors to operate and control environmental protection facilities. The existing operation and management methods are difficult to adapt to the new normal of continuous low load and rapid load change in the future. The wet limestone desulfurization system is complex and the operating conditions are harsh. The key parameters of the desulfurization system operation: gypsum slurry pH value, net flue gas SO2 online detection value and other parameters have the characteristics of large time lag, nonlinearity and strong coupling. In addition, the current operation process has a low level of automation, poor coordination, and a large maintenance workload. The precise control of slurry pH value and flue gas SO2 emissions is difficult. Massive data has been accumulated on the boiler side and the entire desulfurization process. Data mining is of great value. A large amount of historical data can effectively support the construction of an intelligent and autonomous operation model for environmental protection facilities, improve the operation level, achieve flexible operation and energy conservation and carbon reduction, so as to meet the emission standards requirements under deep peak-shaving low load sections and rapid load changes. Summary of the invention

[0004] The present invention provides a desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm.

[0005] A desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm includes a data acquisition module, a data preprocessing module, a feature engineering module, a model training module, a model fusion module and a control execution module, wherein:

[0006] The data acquisition module monitors and collects multi-dimensional parameters in the desulfurization process in real time, including flue gas flow, flue gas temperature, gypsum slurry pH, and slurry circulation pump current;

[0007] The data preprocessing module preprocesses the collected multidimensional parameters, including missing value processing, outlier processing, normalization, and training set, validation set, and test set division;

[0008] The feature engineering module performs feature selection and feature construction based on the preprocessed multidimensional parameters to form a final feature set related to desulfurization;

[0009] The model training module uses Fourier transform (FT), long short-term memory network (LSTM) and extreme gradient boosting tree (XGBoost) to train the desulfurization prediction model based on the final feature set related to desulfurization;

[0010] The model fusion module first inputs the final feature set through Fourier transform (FT) into the long short-term memory network (LSTM) for training, and the obtained prediction results are used as new features to input the extreme gradient boosting tree (XGBoost) for fusion training to obtain the fused prediction results;

[0011] The control execution module adjusts the overall operating parameters of the desulfurization system in real time according to the fused prediction results to achieve intelligent, fully automatic and autonomous operation.

[0012] Optionally, the data acquisition module includes:

[0013] Sensor installation: Various sensors are installed in the desulfurization tower process section, including flow sensors, temperature sensors, pH meters, pressure gauges and current sensors;

[0014] Data collection: Multi-dimensional parameters are collected in real time through a variety of sensors, and the collected multi-dimensional parameters are stored in the database.

[0015] Optionally, the data preprocessing module includes:

[0016] Missing value processing: Lagrange interpolation method is used to fill missing sensor data;

[0017] Outlier processing: Tukey's Test method was used to test outliers and eliminate error values;

[0018] Normalization: Min-Max normalization is used to normalize the collected multi-dimensional parameters;

[0019] Dataset division: Divide the normalized multidimensional parameters into training set, validation set and test set.

[0020] Optionally, the feature engineering module includes:

[0021] Feature selection: Calculate the mutual information between each feature and the target variable (SO2 in clean flue gas), and select the one with high mutual information as the modeling feature;

[0022] Feature construction: Based on the understanding and experience of the overall desulfurization process, new derived features are constructed by combining the multi-dimensional parameters collected by sensors, including the rate of change of flue gas flow and the rate of change of pH value;

[0023] Final feature set: The selected modeling features and new derived features are combined to form the final feature set related to desulfurization.

[0024] Optionally, the model training module includes:

[0025] Data smoothing and noise reduction: Perform Fourier transform on the final feature set, convert it into frequency domain signal and analyze the frequency and noise components therein, remove the noise through bandpass filter, retain the periodic components, and convert the data back to time domain signal after smoothing and noise reduction;

[0026] LSTM training: The SO2 in clean flue gas after Fourier transform denoising is input into the LSTM. The forget gate and output gate mechanism in the LSTM are used to capture the dependency of the time series, obtain the single sequence prediction value of SO2 in clean flue gas, and use cross-validation to optimize the hyperparameters of the LSTM on the validation set.

[0027] Extreme Gradient Boosting Tree (XGBoost) training: The final feature set after Fourier transform denoising and the single sequence prediction value predicted by the long short-term memory network (LSTM) are input into the Extreme Gradient Boosting Tree (XGBoost) for training to capture the nonlinear relationship between features and target variables, and use cross-validation to optimize the hyperparameters of the Extreme Gradient Boosting Tree (XGBoost) on the validation set.

[0028] Optionally, the Fourier transform is expressed as:

[0029] ;

[0030] in, yes The Fourier transform of represents the angular frequency, Is an imaginary unit.

[0031] Optionally, the model fusion module includes:

[0032] Metamodel training and integration: The prediction results of the long short-term memory network (LSTM) and the final feature set after denoising are input into the extreme gradient boosting tree (XGBoost) and integrated using the stacking method. The long short-term memory network (LSTM) is used as the base model and the extreme gradient boosting tree (XGBoost) is used as the metamodel. The results of the base model are used as the input of the metamodel to obtain a combined model, which can fully capture the dependencies and nonlinear relationships of the data.

[0033] Final prediction result output: The Extreme Gradient Boosting Tree (XGBoost) is trained based on the input features and the prediction results of the Long Short-Term Memory Network (LSTM), and the final desulfurization effect prediction results are output.

[0034] Optionally, the control execution module includes:

[0035] Generate control instructions: Generate control instructions based on the output desulfurization effect prediction results and the experience pump group expert database;

[0036] Adjust parameters: Execute control instructions through the PLC controller to adjust the facility operating parameters of the desulfurization system;

[0037] Real-time monitoring: Real-time monitoring of the operating status of the desulfurization system.

[0038] Beneficial effects of the present invention:

[0039] The present invention, by combining Fourier transform to smooth and reduce noise on data, and adopting stacking ensemble learning of LSTM algorithm and XGBoost algorithm, can more accurately predict the key parameters in the desulfurization process. Fourier transform can effectively remove noise and retain the main periodic components of the data, while the integrated model of LSTM and XGBoost can capture time series dependencies and nonlinear relationships, so that the system has high-precision prediction capabilities, thereby ensuring accurate control of the desulfurization effect.

[0040] The present invention, through the fusion of prediction results, the control execution module can adjust the operating parameters of the desulfurization system in real time, dynamically optimize the control strategy to adapt to changes in different working conditions, ensure the stable operation of the system, and at the same time, through the intelligent control algorithm, effectively reduce human intervention, improve the automation level of the desulfurization system, significantly enhance the intelligence of the system, and improve the overall operating efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 Schematic diagram of system function modules according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0044] It should be noted that the references to "one embodiment", "embodiment", "exemplary embodiments", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0045] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0046] like Figure 1 As shown, a desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm includes a data acquisition module, a data preprocessing module, a feature engineering module, a model training module, a model fusion module and a control execution module, wherein;

[0047] The data acquisition module monitors and collects various parameters in the desulfurization process in real time, including flue gas flow, flue gas temperature, gypsum slurry pH, circulation pump current, etc., including:

[0048] 1. Install various sensors at the key process sections of the desulfurization tower, such as flow sensors, temperature sensors, pH meters, pressure gauges, current sensors, etc.;

[0049] 2. The data collected by the sensor is transmitted to the data acquisition module through the data acquisition card, and the data acquisition module stores the collected data in the database for subsequent processing.

[0050] The data preprocessing module cleans and preprocesses the collected data, including missing value processing, outlier processing, normalization, and training set, validation set, and test set division, etc. Specifically, it includes:

[0051] 1. Use Lagrange interpolation or nearest neighbor interpolation to fill in the missing sensor data. Take Lagrange interpolation as an example. , , calculate the value that should be filled in for the missing value.

[0052] 2. Outliers can be identified by 3 The Tukey's Test method is used to test outliers and eliminate error values ​​caused by sensor measurement or other reasons. Take the Tukey's Test method as an example: calculate the upper quartile Q1, the lower quartile Q3, estimate the maximum value: Q3+K(Q3-Q1), and the estimated minimum value: Q1-K(Q3-Q1). When the data exceeds the estimated maximum or minimum value, it represents an anomaly, K=1.5 or K=3. When K=1.5, it represents moderate anomaly, and when K=3, it represents extreme anomaly. It is marked as an outlier and eliminated.

[0053] 3. If only the XGBoost algorithm is used, the data may not need to be normalized, but since the LSTM network is required in stacking, normalizing the data helps accelerate convergence, avoid gradient disappearance or explosion, improve numerical stability and thus enhance generalization ability. , where min(x) and max(x) are the minimum and maximum values ​​of the sample data, respectively. is the original data, is the normalized data.

[0054] 4. Divide the preprocessed dataset into training set, validation set and test set, with 80% of the training set used for model training, 10% of the validation set used for model hyperparameter adjustment, and 10% of the test set used for final model effect evaluation.

[0055] The feature engineering module selects and constructs features, extracts highly correlated features from all features, and constructs highly correlated features through original features and process experience, including:

[0056] 1. Use mutual information or random forest, gradient boosting tree model importance ranking, etc. to select features with strong correlation from all features. For example, calculate the mutual information between each feature and the target variable (SO2 in net flue gas), and select features with higher mutual information as modeling features.

[0057] 2. Through the understanding and experience of the overall desulfurization process, new features are constructed in combination with the original features collected by the sensor, such as calculating the rate of change of flue gas flow rate, the rate of change of pH value, etc.

[0058] 3. Combine the original features collected by the selected sensor and the derived features constructed under the guidance of experience to form the final feature set.

[0059] The model training module is based on three algorithms, namely Fourier transform (FT), long short-term memory network (LSTM) and extreme gradient boosting tree (XGBoost). It uses data smoothed by Fourier transform to train the desulfurization prediction model, including:

[0060] 1. By for continuous functions , and perform Fourier transform: ,in yes The Fourier transform of represents the angular frequency, Is an imaginary unit. The pre-processed time series data is Fourier transformed and converted into a frequency domain signal. The frequency domain signal is analyzed to identify the main frequency components and noise components in the signal. The noise components are removed using a bandpass filter or a low-pass filter, and the main periodic components are retained. The processed frequency domain signal is then converted back to a time domain signal, thereby completing smoothing and noise reduction.

[0061] 2. Input the time series data of SO2 in clean flue gas that has been denoised by Fourier transform into the LSTM model. By using the forget gate and output gate mechanism in the LSTM model, the model is trained to capture the long-term dependency of the time series data and obtain the single sequence prediction value of SO2 in clean flue gas. Then use cross-validation to optimize the model hyperparameters on the validation set to improve the generalization ability of the model.

[0062] 3. Input the feature set data (flue gas temperature, pressure, current of the circulating pump, etc.) that have been de-noised by Fourier transform and the predicted value of the single sequence target variable (SO2 in clean flue gas) predicted by the LSTM model into the XGBoost model for training. The training model captures the nonlinear relationship between features and variables. Then use cross-validation to optimize the model hyperparameters on the validation set to improve the generalization ability of the model.

[0063] The model fusion module first trains the Fourier transformed data using the LSTM model, and then inputs the obtained prediction results as new features to XGBoost for fusion training to obtain the final prediction results, including:

[0064] 1. Input the feature set data (flue gas temperature, pressure, current of circulating pump, etc.) that have been de-noised by Fourier transform and the predicted value of the single sequence target variable (SO2 in clean flue gas) predicted by the LSTM model into the XGBoost model;

[0065] 2. The two models are integrated using stacking, that is, the LSTM model is used as the base model and the XGBoost model is used as the meta-model. The results of the base model are used as the input of the meta-model to obtain the combined model, which can more effectively capture the long-term dependencies and nonlinear relationships of the data.

[0066] The control execution module adjusts the overall operating parameters of the desulfurization system in real time based on the fused prediction results to achieve intelligent, fully automatic and autonomous operation, including:

[0067] 1. Based on the prediction results of the LSTM model and the XGBoost model stacking integration, control instructions are generated by comparing the experience pump group expert library, such as adjusting the start and stop combination of the slurry circulation pump, adjusting the opening and closing of the slurry supply and control valve, etc.

[0068] 2. Execute control instructions through the PLC controller to adjust the facility operating parameters of the desulfurization system.

[0069] 3. Monitor the system's operating status in real time to ensure system response and operational stability.

[0070] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0071] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm, characterized in that: It includes data acquisition module, data preprocessing module, feature engineering module, model training module, model fusion module and control execution module, among which: The data acquisition module monitors and collects multi-dimensional parameters in the desulfurization process in real time, including flue gas flow, flue gas temperature, gypsum slurry pH, and slurry circulation pump current; The data preprocessing module preprocesses the collected multidimensional parameters, including missing value processing, outlier processing, normalization, and training set, validation set, and test set division; The feature engineering module performs feature selection and feature construction based on the preprocessed multidimensional parameters to form a final feature set related to desulfurization; The model training module uses Fourier transform, long short-term memory network and extreme gradient boosting tree to train the desulfurization prediction model based on the final feature set related to desulfurization; The model fusion module first inputs the final feature set through Fourier transformation into the long short-term memory network for training, and inputs the obtained prediction results as new features into the extreme gradient boosting tree for fusion training to obtain the fused prediction results; The control execution module adjusts the overall operating parameters of the desulfurization system in real time according to the fused prediction results to achieve intelligent, fully automatic and autonomous operation; The feature engineering module includes: Feature selection: Calculate the mutual information between each feature and the target variable, and select the one with high mutual information as the modeling feature; Feature construction: Based on the understanding and experience of the overall desulfurization process, new derived features are constructed by combining the multi-dimensional parameters collected by sensors, including the rate of change of flue gas flow and the rate of change of pH value; Final feature set: The selected modeling features and new derived features are combined to form the final feature set related to desulfurization; The model training module includes: Data smoothing and noise reduction: Perform Fourier transform on the final feature set, convert it into frequency domain signal and analyze the frequency and noise components therein, remove the noise through bandpass filter, retain the periodic components, and convert the data back to time domain signal after smoothing and noise reduction; LSTM network training: The SO2 in clean flue gas after Fourier transform denoising is input into the LSTM network. The forget gate and output gate mechanism in the LSTM network are used to capture the dependency of the time series, obtain the single sequence prediction value of SO2 in clean flue gas, and use cross-validation to optimize the hyperparameters of the LSTM network on the validation set. Extreme gradient boosting tree training: The final feature set after Fourier transform denoising and the single sequence prediction value predicted by the long short-term memory network are input into the extreme gradient boosting tree for training to capture the nonlinear relationship between the features and the target variable, and cross-validation is used to optimize the hyperparameters of the extreme gradient boosting tree on the validation set.

2. According to claim 1, a desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm is characterized in that: The data acquisition module comprises: Sensor installation: Various sensors are installed in the desulfurization tower process section, including flow sensors, temperature sensors, pH meters, pressure gauges and current sensors; Data collection: Multi-dimensional parameters are collected in real time through a variety of sensors, and the collected multi-dimensional parameters are stored in the database.

3. According to claim 2, a desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm is characterized in that: The data preprocessing module comprises: Missing value processing: Lagrange interpolation method is used to fill missing sensor data; Outlier processing: Tukey's Test method was used to test outliers and eliminate error values; Normalization: Min-Max normalization is used to normalize the collected multi-dimensional parameters; Dataset division: Divide the normalized multidimensional parameters into training set, validation set and test set.

4. According to claim 3, a desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm is characterized in that: The Fourier transform is expressed as: ； in, yes The Fourier transform of represents the angular frequency, Is an imaginary unit.

5. According to claim 4, a desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm is characterized in that: The model fusion module includes: Metamodel training and integration: The prediction results of the long short-term memory network and the final feature set after denoising are input into the extreme gradient boosting tree. The stacking method is used for integration. The long short-term memory network is used as the base model and the extreme gradient boosting tree is used as the metamodel. The results of the base model are used as the input of the metamodel to obtain the combined model, which can fully capture the dependency and nonlinear relationship of the data. Final prediction result output: The extreme gradient boosting tree is trained based on the input features and the prediction results of the long short-term memory network, and outputs the final desulfurization effect prediction results.

6. A desulfurization intelligent prediction system based on FT-LSTM-XGBoost algorithm according to claim 5, characterized in that: The control execution module includes: Generate control instructions: Generate control instructions based on the output desulfurization effect prediction results and the experience pump group expert database; Adjust parameters: Execute control instructions through the PLC controller to adjust the facility operating parameters of the desulfurization system; Real-time monitoring: Real-time monitoring of the operating status of the desulfurization system.

Images