Load model and system for power plant simulation system

Abstract

The invention discloses a load model and system for a power plant simulation system. The load model set comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is responsible for collecting load data, timestamps, weather data and regional data of load elements; the data analysis processing module classifies the load data into an active load and a reactive load; the feature data collection module selects key feature data; the data conversion module converts the feature data into a form suitable for LSTM model input; and the prediction module outputs a load prediction result in a future period of time based on the LSTM neural network, wherein the load prediction result comprises a load value. In addition, the method further comprises the steps of data preprocessing, data normalization, model verification and the like so as to ensure the accuracy and reliability of a prediction result. The load system can run on a server, interacts with an external system through a Web interface or an API interface, and supports multi-user concurrent access and data sharing.

Description

Technical Field

[0001] The present invention relates to the technical field of combining power systems with machine learning. More specifically, the present invention relates to a load model and system for a power plant simulation system, which uses a machine learning algorithm (especially a long short-term memory network LSTM) to model and predict the power plant load to support the simulation, optimization and scheduling decision of the power plant. Background Art

[0002] In the power system, accurate prediction of power plant load is of great significance to the stable operation of the power system, optimal resource allocation and economic benefit improvement. Traditional load forecasting methods mainly rely on statistical analysis and physical models, but these methods often do not work well when faced with complex and changeable load changes. With the rapid development of machine learning technology, especially the application of deep learning algorithms such as long short-term memory networks (LSTM), the accuracy and robustness of load forecasting have been significantly improved.

[0003] However, most of the load models in existing power plant simulation systems still use traditional prediction methods, which are difficult to adapt to the complexity and dynamics of modern power systems. Therefore, it is urgent to develop a load model that can fully utilize the advantages of machine learning algorithms and accurately predict power plant loads. Summary of the invention

[0004] An object of the present invention is to solve at least the above problems and to provide at least the advantages which will be described hereinafter.

[0005] In order to achieve these purposes and other advantages according to the present invention, a load model for a power plant simulation system is provided, comprising: A data acquisition module, which is used to collect load data, timestamps and weather data of each load element; A data analysis and processing module is used to classify load data into active load and reactive load according to load characteristics, and calculate the proportion and change trend of each characteristic category of load in the total load, the average load, the load fluctuation range, the load peak and valley time periods, and the load distribution; A characteristic data collection module is used to select characteristic data, wherein the characteristic data includes historical load values, timestamps, weather data, load category proportions, load change trends, average loads, load peak and valley periods, load fluctuation ranges, and load distribution conditions; A data conversion module, which is used to convert feature data into a form suitable for LSTM model input; The prediction module is used to output prediction results based on feature data. The prediction results include load value, load category proportion, load change trend, average load, load peak and valley time periods, load fluctuation range and load distribution in the future. The prediction module is established and trained based on the LSTM neural network.

[0006] Preferably, the data analysis and processing module is further configured to pre-process the collected data, including data cleaning and data interpolation.

[0007] Preferably, the data conversion module is further configured to perform data normalization processing on the feature data.

[0008] Preferably, the weather data includes temperature, humidity and wind speed.

[0009] Preferably, the load variation trend includes one or a combination of daily load curve, weekly load curve, monthly load curve, quarterly load curve and annual load curve.

[0010] Preferably, the load distribution condition specifically includes the standard deviation, median and load factor of the load curve; and the load fluctuation range is the difference between the maximum load and the minimum load in the load curve.

[0011] Preferably, the prediction result also includes a prediction confidence and an error range, wherein the prediction confidence indicates the reliability of the prediction result, and the error range is used to evaluate the accuracy of the prediction result.

[0012] Preferably, the trained LSTM neural network includes an input layer, two LSTM layers, a fully connected layer and an output layer, the number of neurons in each LSTM layer is 256, and the training is performed using an Adam optimizer and a mean square error loss function.

[0013] Preferably, the load model also includes a model verification module, which is used to verify the trained LSTM neural network and evaluate the prediction performance of the model by comparing the accuracy of historical data and predicted data.

[0014] The present invention provides a load system, which integrates the above-mentioned load model. The load system runs on a server, interacts with an external system through a Web interface or an API interface, and supports multi-user concurrent access and data sharing.

[0015] The present invention has at least the following beneficial effects: First, by establishing a load model, the load conditions in the future can be predicted, which will help thermal power plants make reasonable dispatch and arrangements to ensure the stability and reliability of power supply.

[0016] Second, by using the load model to accurately predict the load conditions in the future, we can statistically calculate key information such as the load change trend, the exact time and specific values ​​of the peak and valley values, the load fluctuation range, and the load distribution, providing strong support for the stable operation and efficient scheduling of the power system.

[0017] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of power load of a thermal power plant according to an embodiment of the present invention; Figure 2 This is an active power output curve of a unit according to an embodiment of the present invention; wherein the ordinate is Pg, in MW, and the abscissa is time, in s; Figure 3 For Figure 2 The load curve of the primary fan under the influence of the corresponding unit active output curve; the ordinate is Pg, in MW, and the abscissa is time, in seconds; Figure 4 For Figure 2 The load curve of the induced draft fan under the influence of the corresponding unit active output curve; the ordinate is Pg, in MW, and the abscissa is time, in s; Figure 5 For Figure 2 The load curve of the fan under the influence of the corresponding unit active output curve; the ordinate is Pg, the unit is MW, and the abscissa is time, the unit is s; Figure 6 For Figure 2 The load curve of the grinding machine under the influence of the corresponding unit active output curve; the ordinate is Pg, the unit is MW, and the abscissa is time, the unit is s; Figure 7 For Figure 2 The load curve of the circulating water pump under the influence of the corresponding unit active output curve; the ordinate is Pg, in MW, and the abscissa is time, in s; Figure 8 A schematic diagram showing data of one embodiment of the present invention; Fig. 9 This is a schematic diagram showing part of the collected data of one embodiment of the present invention; Fig.10 The average error of the load value predicted and output by the load model in one embodiment of the present invention; Fig.11A comparison diagram of the predicted value (simulation) and the measured value of the load model according to one embodiment of the present invention; Fig.12 This is a framework diagram of a load model of one of the technical solutions of the present invention. DETAILED DESCRIPTION

[0019] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0020] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials are commercially available unless otherwise specified; in the description of the present invention, the orientation or positional relationship indicated by the terms is based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0021] like Fig.12 As shown, the present invention provides a load model for a power plant simulation system, comprising: The data acquisition module is used to collect load data, timestamps, weather data and regional data of each load element in the power supply area; weather data, such as temperature, humidity, wind speed, etc., regional data, such as in and out traffic, regional altitude and regional latitude, all of which will affect the power load.

[0022] The data analysis and processing module is used to classify the load data into active load and reactive load according to the load characteristics; active load refers to the equipment or users that actually consume electricity, such as electric lights, motors, etc. The current and voltage of these equipment are in phase, and the load is composed of resistance and inductance. Active load is the main load type in the power system, which directly reflects the power consumption of the power system, and most active loads change according to cycles or seasons, and the changes are large. Reactive load refers to the current caused by inductance or capacitance in the power system, such as reactors, magnets, speed regulators, etc. The current and voltage of these equipment are out of phase, and the load is mainly absorbed by capacitance and inductance. Although reactive load does not directly consume electrical energy, it plays an important role in the power system, such as maintaining the voltage stability of the power system and providing the necessary magnetic field. The demand for reactive load is usually controlled according to the demand of the load. In a normal power system, the proportion of active load is usually much higher than that of reactive load.

[0023] A characteristic data collection module, which is used to select characteristic data, wherein the characteristic data includes historical load values, timestamps, weather data and regional data; A data conversion module, which is used to convert feature data into a form suitable for LSTM model input; The prediction module is used to output prediction results based on the feature data, and the prediction results include the load value for a period of time in the future. The prediction module is established and trained based on the LSTM neural network.

[0024] Specifically, the load value is the most direct and important forecast output, that is, the load size at one or several time points in the future, such as active power, reactive power or total load, etc. These forecast values ​​can help power plants and power grid dispatching centers understand future load demand in advance, so as to reasonably arrange power generation plans and power dispatching. According to the forecast value of the load, statistical methods can be used to calculate the change trend of the load, such as the increase and decrease trend of the load, the time of occurrence of peak and valley values, etc. This information is of great significance for formulating power dispatching strategies, optimizing energy configuration and ensuring stable operation of the power grid. The forecast value and proportion of various loads can help power plants understand the demand of various loads more accurately, so as to adjust the power generation plan and energy configuration in a targeted manner. Other load-related parameters, such as the fluctuation range of the load, the load distribution, etc., can provide more comprehensive load information, which can help power plants and power grid dispatching centers understand the future load situation more comprehensively.

[0025] In the above technical scheme, the LSTM neural network can capture the long-term dependency of time series, which is very useful for understanding the changes of power load over time and various factors (such as weather, region, and unit output). At the same time, the LSTM neural network also has the following advantages: it can process long-term series data, capture the long-term dependency and periodic pattern in the load series, thereby improving the prediction accuracy; in the prediction of plant power load, multiple factors such as time, weather, region and unit output can be considered at the same time to improve the accuracy of the prediction; to achieve rolling prediction, the LSTM neural network can update the model parameters in real time, and continuously adjust the prediction results according to the latest data, so as to meet the real-time requirements in practical applications. This is particularly important for power systems that require rapid response and dynamic adjustment. Therefore, the LSTM neural network has important application value in plant power load prediction. By taking advantage of the advantages of the LSTM neural network, a more accurate and reliable load model can be established to contribute to the stable operation and energy conservation and emission reduction of the power system. The accuracy of load prediction is crucial to optimize the combined dispatch of generator sets, and can avoid over-generation or insufficient power supply due to inaccurate load prediction.

[0026] In the above technical solution, the load model is used to directly and accurately output the load size, and calculate key information such as the change trend, the exact time of peak and valley values ​​and their specific values, the load fluctuation range, and the load distribution. This greatly improves the accuracy and timeliness of the prediction, and provides strong support for the stable operation and efficient dispatch of the power system.

[0027] In another technical solution, the data analysis and processing module is further configured to pre-process the collected data, including data cleaning, data interpolation and data aggregation. Specifically, data cleaning includes identifying and removing invalid data such as null values, duplicate values, and abnormal values ​​in the data; data interpolation includes inserting missing data using interpolation; data aggregation includes aggregating data at time intervals (such as hours, days, weeks, months, etc.) to calculate average values, maximum values, minimum values, etc.

[0028] In another technical solution, the data conversion module is further configured to perform data normalization processing on the feature data. Specifically, the feature data is normalized using Min-Max normalization so that its value range is between [0, 1], thereby improving the efficiency and accuracy of model training.

[0029] In another technical solution, the load change trend includes one or a combination of daily load curve, weekly load curve, monthly load curve, quarterly load curve and annual load curve.

[0030] In the above technical solution, by drawing the load curves of active power and reactive power, their changing trends in different time periods can be intuitively seen. The maximum value in the load curve is called the maximum load P max , the minimum value is called the minimum load P min , these two data are important data for analyzing the load characteristics of the power system. At the same time, the area under the load curve represents the electric energy consumed in a period of time.

[0031] The peak and valley periods of load refer to the periods of time with higher and lower loads in the load curve; the average load is the arithmetic mean of the load data, which reflects the central trend of the load data. In load characteristic analysis, it is used to indicate the average level of load over a period of time. By calculating the average load of different time periods (such as day, week, month, year), we can understand the long-term trend and seasonal characteristics of the load.

[0032] The load factor, also known as the load rate, is the ratio of the average load of the power load to its maximum load, reflecting the degree of fluctuation of the load curve. For active load and reactive load, their load factors α and β can be calculated respectively. The typical value range of the load factor can reflect the stability of the load.

[0033] In another technical solution, the load distribution condition specifically includes the standard deviation, median and load factor of the load curve; the load fluctuation range is the difference between the maximum load and the minimum load in the load curve.

[0034] In the above technical solution, the load fluctuation range refers to the difference between the maximum and minimum values ​​of the load within a certain period of time, reflecting the degree of load fluctuation. The calculation formula is: Load fluctuation range = maximum load value - minimum load value In order to more comprehensively describe the distribution of load, statistical indicators such as the standard deviation, median and load system are also calculated. The standard deviation is the square root of the variance and has the same dimension as the original data. The larger the standard deviation, the higher the dispersion of the load data, that is, the greater the load fluctuation; conversely, it means that the load data is relatively concentrated and the fluctuation is small. The median is the value in the middle position after sorting a set of load data from small to large. When there are extreme values ​​or skewed distribution in the load data, the median can better reflect the central trend of the data than the mean. The median can be used to evaluate the typical value or representative value of the load data, especially when the load data is unevenly distributed.

[0035] In another technical solution, the forecast result also includes forecast confidence and error range, where the forecast confidence indicates the reliability of the forecast result and the error range is used to evaluate the accuracy of the forecast result. In practical applications, in order to evaluate the reliability and accuracy of the forecast results, the load model can also output the forecast confidence and error range, which can help decision makers better understand the uncertainty of the forecast results and consider these uncertainties when making decisions.

[0036] In another technical solution, the trained LSTM neural network includes an input layer, two LSTM layers, a fully connected layer and an output layer. The number of neurons in each LSTM layer is 256, and the training is performed using the Adam optimizer and the mean square error loss function.

[0037] In the above technical solution, the trained LSTM neural network can deeply explore the temporal dependencies and potential rules in the load data through the stacking of two LSTM layers and the configuration of 256 neurons in each layer, so as to achieve high-precision load forecasting and provide a reliable basis for power dispatching and energy management. The Adam optimizer and mean square error loss function are used for training, making the training process of the LSTM neural network more efficient and stable. The Adam optimizer can adaptively adjust the learning rate, accelerate the training process and reduce the risk of overfitting; while the mean square error loss function can intuitively reflect the difference between the predicted value and the true value, which helps the model to continuously approach the optimal solution during the training process.

[0038] In another technical solution, the load model also includes a model verification module, which is used to verify the trained LSTM neural network and evaluate the prediction performance of the model by comparing the accuracy of historical data and predicted data. Specifically, the prediction performance of the trained LSTM neural network is verified by using mean square error (MSE) and mean absolute error (MAE); mean square error (MSE) refers to the average square difference between the predicted value and the actual value. The smaller the MSE, the closer the prediction result of the trained LSTM neural network is to the true value. The mean absolute error (MAE) refers to the average absolute difference between the predicted value and the true value, which is used to measure the average degree of prediction error of the trained LSTM neural network.

[0039] In the above technical solution, the model verification module can intuitively reflect the prediction error and deviation of the trained LSTM neural network by comparing historical data and predicted data. This helps to find that the prediction of the trained LSTM neural network is insufficient in a specific time period or under specific load conditions, so as to optimize and adjust the trained LSTM neural network in a targeted manner. Through continuous iteration and optimization, the prediction accuracy of the trained LSTM neural network can be significantly improved, providing more reliable decision support for power dispatching and energy management. The model verification module not only focuses on the performance of the trained LSTM neural network on historical data, but also focuses on the prediction performance of the trained LSTM neural network under new data or unknown conditions. This helps to evaluate the generalization ability of the trained LSTM neural network, that is, whether the trained LSTM neural network can maintain stable prediction performance when facing new data. By enhancing the generalization ability of the trained LSTM neural network, it can be ensured that the trained LSTM neural network can maintain a high prediction accuracy in different time periods, different load conditions or different regions.

[0040] Based on the same inventive concept, the present invention provides a load forecasting method, comprising the following steps: S1. Data collection: Collect load data (including load value, timestamp), weather data (including temperature, humidity and wind speed) and regional data (in and out traffic, regional altitude and regional latitude) and transmit them to the data center; Specifically, high-precision sensors and smart meters are installed at key locations in the power system, such as the inlet and outlet of the boiler, the bearing of the steam turbine, the stator winding of the generator, and the distribution box of the lighting system. These devices can monitor the current, voltage, power factor and other parameters of the load components in real time and transmit data via wired or wireless means. The collected data is stored in a relational database, and the database table structure is designed according to the data type and purpose, including load data tables, weather data tables, etc.

[0041] S2. Data analysis and processing: Classify the data into active load and reactive load according to load characteristics, and calculate the proportion of each characteristic category of load in the total load; draw load curves and load factor charts to show the load change trend; calculate statistical indicators such as average load, load fluctuation range, load peak and valley time periods; calculate the standard deviation and median of load distribution; Load change trends such as daily load curves, weekly load curves, monthly load curves, quarterly load curves, load fluctuation ranges, load peak and valley periods, etc. At the same time, the NumPy library can be used to calculate statistical indicators such as the standard deviation and median of the load distribution.

[0042] Specifically, before analysis and processing, you can use Python's Pandas library to preprocess the raw data, including data cleaning (such as removing null values ​​and duplicate values), data conversion (such as converting timestamps to a specific format), data aggregation (such as calculating averages by time intervals), etc.

[0043] S3. Feature data collection: Feature selection: Based on the requirements of load forecasting and the characteristics of the LSTM model, features that have a significant impact on the forecast results are selected as input features. These features include historical load values, timestamps, weather data (temperature, humidity, wind speed), and regional data (in and out traffic, regional altitude, and regional latitude).

[0044] Data integration: The selected feature data are integrated into a new dataset, which is organized in the form of time series, and each sample contains feature values ​​of multiple consecutive time steps.

[0045] S4. Data conversion: Converting time series data into supervised learning format, that is, each input sample contains feature values ​​of multiple time steps and corresponding labels (such as the load value of a future time step), can be achieved through the sliding window method. In order to improve the efficiency and accuracy of model training, the input features are normalized so that their value range is between [0,1].

[0046] S5. LSTM model construction: The LSTM model is constructed using the TensorFlow deep learning framework. The LSTM model structure includes an input layer, a multi-layer LSTM layer (two layers in this embodiment), a fully connected layer, and an output layer. Set the parameters of the LSTM layer, such as the number of neurons (256 in this embodiment), activation function (ReLU in this embodiment), and dropout rate (0.5 in this embodiment to prevent overfitting); The mean square error (MSE) is defined as the loss function and the Adam optimizer is used for model training.

[0047] S6. LSTM model training: Divide the converted data set into training set, validation set and test set, and the ratio can be 7:2:1; Use the training set data to train the LSTM model, set the learning rate (0.001 in this embodiment), batch size (32 in this embodiment), and number of training rounds (100 in this embodiment); During the training process, monitor the performance on the validation set, and adjust the model structure and parameters in a timely manner according to the changes in indicators such as the loss function value and accuracy on the validation set, such as increasing or decreasing the number of LSTM layers, adjusting the number of neurons or activation functions, etc. When the loss function value on the validation set no longer decreases, stop training to avoid overfitting. At the same time, use the learning rate decay strategy (such as reducing the learning rate after a certain number of training rounds) to optimize the training process.

[0048] S7, LSTM model prediction: Use the trained LSTM model to predict future loads. The prediction results include load size prediction confidence and error range.

[0049] S8. LSTM model evaluation: The prediction results are evaluated using the test set data, and indicators such as mean square error (MSE) and mean absolute error (MAE) are calculated to measure the accuracy of the prediction. At the same time, a comparison chart between the prediction results and the actual load is drawn to intuitively display the prediction effect.

[0050] Based on the prediction results, corresponding power generation plans and scheduling strategies are formulated to optimize the operating efficiency and stability of the power system, such as adjusting the output power of generators, optimizing load distribution, and adjusting the grid structure.

[0051] Specifically, high-efficiency units are dispatched preferentially: on the premise of meeting load demand, high-efficiency and low-energy-consuming generator sets are dispatched preferentially, and the energy consumption and emissions of the entire power generation system are reduced by optimizing the combined dispatch of generator sets; according to the performance parameters of each generator set and the current load demand, the load is reasonably distributed to avoid overload or underload operation of certain units; by balancing the load distribution, the operating efficiency of the generator set is improved and unnecessary energy consumption and emissions are reduced; the frequent start and stop times of the generator set are minimized to reduce energy consumption and emissions during the start and stop process; by reasonably arranging the start and stop plans of the units, the stable operation of the units in an efficient operating state is ensured.

[0052] Based on the same inventive concept, the present invention provides a load system, the load system comprising: The data acquisition module is used to collect load data (including load value, timestamp), weather data (including temperature and humidity) and regional data (in and out traffic, regional altitude and regional latitude) and transmit them to the data center.

[0053] Specifically, the data acquisition module includes an interface for communicating with sensors and smart meters to receive load data and weather data; it includes data cleaning and storage functions to perform preliminary processing on the data and store it in a database; the received data can be stored in a relational database or a distributed database (such as Hadoop, Spark).

[0054] The data analysis and processing module is used to classify the data into active load and reactive load according to the load characteristics, and calculate the proportion of the load of each characteristic category in the total load; draw the load curve and load factor chart to show the load change trend; calculate the average load, load fluctuation range, load peak and valley time periods and other statistical indicators; calculate the standard deviation and median of the load distribution; specifically, the Pandas and NumPy libraries in the Python programming language can be used to process and analyze the collected data.

[0055] The feature data collection module is used to select feature data and integrate the selected feature data into a new data set. The data set is organized in the form of a time series, and each sample contains feature values ​​of multiple consecutive time steps. Specifically, the feature data collection module integrates the processed and analyzed data into a new data set, including data screening and feature selection functions to ensure the data quality of the input model. Specifically, these features include historical load values, timestamps, weather data (temperature, humidity, wind speed), and regional data (in and out traffic, regional altitude and regional latitude).

[0056] The data conversion module is used to convert feature data into a form suitable for LSTM model input; specifically, the time series data is converted into a supervised learning format, that is, each input sample contains feature values ​​of multiple time steps and corresponding labels, and the converted feature data is normalized to improve the efficiency and accuracy of model training. Specifically, the Pandas library can be used to convert time series data into a supervised learning format.

[0057] LSTM model building module, which is used to build the structure of LSTM model. Specifically, the TensorFlow deep learning framework can be used to build the LSTM model. The LSTM model building module provides the structural design function of the LSTM model, including the design of the input layer, multi-layer LSTM layer, fully connected layer and output layer, as well as the setting function of the parameters such as the number of neurons, activation function, and discard rate of the LSTM layer, and defines the loss function and optimizer for evaluation and optimization during the model training process.

[0058] An LSTM model training module is used to train the LSTM model. Specifically, the LSTM model training module provides a data partitioning function to divide the data set into a training set, a validation set, and a test set. It is used to train the LSTM model using the training set data while monitoring the performance on the validation set. It also provides a function to adjust the model structure and parameters based on the performance on the validation set. It also provides an early stopping method and a learning rate decay strategy to optimize the training process and prevent overfitting.

[0059] The prediction and evaluation module is used to use the trained LSTM model to predict future loads and output prediction results. Specifically, the prediction results include load prediction confidence and error range. The prediction and evaluation module provides the evaluation function of the prediction results, calculates indicators such as MSE and MAE to measure the accuracy of the prediction. It also provides the function of generating a comparison chart between the prediction results and the actual load to intuitively display the prediction effect.

[0060] The load forecast results obtained through the load system forecast provide a data basis for power generation planning and dispatching strategies to optimize the operating efficiency and stability of the power system.

[0061] The load system of the present invention can be an integrated software system that runs on a server and interacts with an external system through a Web interface or an API interface. The load system has functions such as data collection, data analysis and processing, feature data collection, data conversion, LSTM model construction and training, prediction and evaluation, and can provide comprehensive decision support for the operation and management of the power system.

[0062] <Example> Taking a thermal power plant as an example, the schematic diagram of the power load of a thermal power plant is as follows: Figure 1 As shown in the figure, the largest loads in the boiler system are the primary fan, the forced draft fan and the induced draft fan. The loads of the primary fan, the forced draft fan and the induced draft fan are directly related to the amount of pulverized coal required by the unit, the air required for combustion, and the amount of flue gas generated. Take one of the units as an example. Figure 2 As shown in the figure, the load characteristics of the primary fan, induced draft fan and forced draft fan at different unit outputs are as follows Figures 3-5 As shown in Figure 2, in addition to the fan, the coal mill is also an important load in the boiler system. Its size depends on the amount of pulverized coal required by the boiler and is directly related to the operating conditions of the unit. The characteristics of the unit's active output and the corresponding coal mill load are shown in Figure 2. Figure 6 The medium-pressure auxiliary equipment of the steam turbine system is mainly circulating water pumps, condensing pumps and closed cooling pumps, etc. Their load size is usually independent of the unit operating conditions and has little fluctuation. The load characteristics of the circulating water pump at different unit outputs are shown in the figure below. Figure 7As shown in the figure, the circulating water pump load fluctuation is very small and has nothing to do with the active output of the unit, so it can be considered as a constant load. Figures 8 to 11 As shown, the collected data of the load model of the present invention, the average error of the predicted output load value, and the comparison between the predicted value (simulation) and the measured value are displayed, which shows that the load model of the present invention has high prediction accuracy.

[0063] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A load model for a power plant simulation system, characterized in that: include: A data acquisition module, which is used to collect load data, timestamps, weather data and regional data of each load element; A data analysis and processing module, which is used to classify load data into active load and reactive load according to load characteristics; A characteristic data collection module, which is used to select characteristic data, wherein the characteristic data includes historical load values, timestamps, weather and regional data; A data conversion module, which is used to convert feature data into a form suitable for LSTM model input; The prediction module is used to output prediction results based on the feature data, and the prediction results include the load value for a period of time in the future. The prediction module is established and trained based on the LSTM neural network.

2. The load model for a power plant simulation system according to claim 1, characterized in that: The data analysis and processing module is also configured to pre-process the collected data, including data cleaning and data interpolation.

3. The load model for a power plant simulation system according to claim 1, characterized in that: The data conversion module is also configured to perform data normalization processing on the feature data.

4. The load model for a power plant simulation system according to claim 1, characterized in that: The weather data includes temperature, humidity and wind speed, and the regional data includes inflow and outflow of people, regional altitude and regional latitude.

5. The load model for a power plant simulation system according to claim 1, characterized in that: It also includes calculating the load change trend using a statistical method based on the predicted load value for a period of time in the future, and the load change trend includes one or a combination of daily load curve, weekly load curve, monthly load curve, quarterly load curve and annual load curve.

6. The load model for a power plant simulation system according to claim 5, characterized in that: It also includes calculating the load distribution using statistical methods based on the predicted load values ​​for a period of time in the future. The load distribution specifically includes the standard deviation, median and load factor of the load curve; the load fluctuation range is the difference between the maximum load and the minimum load in the load curve.

7. The load model for a power plant simulation system according to claim 1, characterized in that: The prediction results also include prediction confidence and error range, where the prediction confidence indicates the reliability of the prediction results and the error range is used to evaluate the accuracy of the prediction results.

8. The load model for a power plant simulation system according to claim 1, characterized in that: The trained LSTM neural network includes an input layer, two LSTM layers, a fully connected layer and an output layer. The number of neurons in each LSTM layer is 256, and the training is performed using the Adam optimizer and the mean square error loss function.

9. The load model for a power plant simulation system according to claim 1, characterized in that: The load model also includes a model verification module, which is used to verify the trained LSTM neural network and evaluate the prediction performance of the model by comparing the accuracy of historical data and predicted data.

10. Load system, characterized in that, The load system integrates the load model described in any one of claims 1 to 9. The load system runs on a server, interacts with an external system through a Web interface or an API interface, and supports multi-user concurrent access and data sharing.

Images