Nuclear power plant net profit prediction system and method based on long short-term memory network
By constructing a nuclear power plant net profit forecasting system based on long short-term memory networks, the problem of relying on manual data collection for traditional nuclear power plant net profit calculation has been solved, and automated data processing and accurate prediction of future net profit have been achieved.
Patent Information
- Application Number
- CN202511460537.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-14
- Publication Date
- 2025-11-11
AI Technical Summary
Traditional nuclear power plant net profit calculations rely on manual data collection, which incurs significant labor costs, has limited data sources, and cannot predict future situations.
A deep learning technology based on long short-term memory networks is used to build a net profit prediction system for nuclear power plants. Through data acquisition, preprocessing, merging and prediction modules, data in production, operation and finance are processed automatically, and net profit is predicted using LSTM and random forgetting layers.
It enables automated aggregation of data from multiple fields and prediction of future net profit, saving labor costs, broadening data sources, and accurately predicting net profit for the coming year.
Smart Images
Figure CN120931318A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of deep learning technology in artificial intelligence, specifically relating to a nuclear power plant net profit prediction system and method based on long short-term memory networks. Background Technology
[0002] In the production and management of nuclear power plants, net profit is typically used as a key indicator of the plant's performance. Traditionally, net profit calculation for nuclear power plants is primarily financial, involving manual collection and calculation of revenue and cost data. This method suffers from drawbacks: high labor costs in data collection, limited data coverage, and the inability to predict future trends. However, advancements in information technology and the rise of deep learning in artificial intelligence offer new solutions to these problems. Therefore, combining information technology and deep learning can be used to aggregate and analyze data from relevant areas of nuclear power plant production and management, leading to future forecasts. Summary of the Invention
[0003] This invention utilizes information technology and deep learning technology, based on data from nuclear power plant production, operation, and cost, to provide a nuclear power plant net profit prediction system and method based on long short-term memory networks.
[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A net profit forecasting system for nuclear power plants based on long short-term memory networks: It includes a data acquisition module, a data preprocessing module, a data merging module, a data prediction module, and a results aggregation module; Data acquisition module: Acquires data on the impact on net profit from the nuclear power plant's production planning, overhaul, and financial sectors; Data preprocessing module: performs integrity checks and fills in the acquired data, cleans up dirty data, unifies statistical standards, and obtains preprocessed vector data, which is then passed to the data merging module; Data merging module: After obtaining the preprocessed data from the data preprocessing module, it merges the daily data for each category in chronological order, column by column, to obtain a dataset with 366 rows and 5 columns. The vector data of the columns is passed as input data to the data prediction module; Data prediction module: Input data is fed into a trained long short-term memory network model. The data in the window is processed through a network structure composed of several long short-term memory network layers and a random forgetting layer, and various matrix calculations are performed on the trained weight values to obtain the net profit data for the first day of the next year. Then the sliding window moves forward and continues to calculate until the prediction calculation of the net profit data for each day of the next year is completed. The results are then fed into the result summary module. Results summary module: The net profit amounts are summed to obtain the predicted total net profit for the next year. Then, the unit of the results is converted from yuan to 100 million yuan to obtain the final net profit forecast value displayed to the public.
[0005] The data acquisition modules are as follows: in the production planning field, annual power generation plan, annual plant power consumption rate, and annual electricity price data are included; in the overhaul field, annual overhaul plan for generating units is included; and in the financial field, other business income, income assessed under the two detailed rules, variable costs, and fixed costs are included.
[0006] The data acquisition module primarily relies on automatic data acquisition, supplemented by periodic data entry.
[0007] The aforementioned annual power generation plan: Based on the annual power generation plan information provided by the nuclear power plant for each unit, including the daily power generation of each unit in megawatts, and calculated using a year of 366 days, a set of positional dimensions is generated. vector , This indicates the number of units, where 1 represents 366 days. The vector will then be... The units are converted to kilowatt-hours. The data in the vector corresponds to the daily power generation of each unit, and this power value takes into account the impact of summer operating conditions.
[0008] The annual plant power consumption rate mentioned above: Based on the annual plant power consumption rate information provided by the nuclear power plant for each unit, including the daily plant power consumption rate data for each unit, and calculated over a year of 366 days, a set of bit dimensions is generated. vector , This represents the number of generating units, 1 represents 366 days, and the data in the vector corresponds to the daily power consumption rate of each generating unit, which is a decimal and only represents the proportion.
[0009] The annual electricity price data is based on average electricity price data provided by nuclear power plants, including monthly average electricity price data for each unit from January to December of a specified year, in yuan / kWh. The average electricity price data for each unit from January to December of a specified year is allocated to each day of the corresponding month. February is allocated to 29 days if it is a leap year; otherwise, it is allocated to 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0010] The aforementioned overhaul planning data: Based on the overhaul planning information provided by the nuclear power plant for each unit, including the planned start and end times of overhauls, and calculated using 366 days per year, a set of data with the following dimensions is generated. vector , This indicates the number of generating units; 1 represents 366 days; the vector value is 0 within the overhaul start and end time range; the vector value is [value missing] within the overhaul power reduction time before the overhaul begins and power increase time after the overhaul ends. The decimal values between these ranges correspond to the ratio of the power increase / decrease during overhaul to the rated power; the vector values for other time ranges are all 1.
[0011] Other operating revenue: Based on the other operating revenue data provided by the nuclear power plant, including monthly other operating revenue data for each unit from January to December of a specified year (in ten thousand yuan), the other operating revenue data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. February is distributed to 29 days if it is a leap year; otherwise, it is distributed to 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0012] The two detailed performance evaluation criteria are as follows: Based on the performance evaluation data provided by the nuclear power plant, including monthly performance evaluation data for each unit from January to December of a specified year (in ten thousand yuan), the performance evaluation data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. If February is a leap year, it is distributed among the 29 days; otherwise, it is distributed among the 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0013] The fixed costs mentioned above are based on the fixed cost data provided by the nuclear power plant, including monthly fixed cost data for each unit from January to December of a specified year, in units of ten thousand yuan. The fixed cost data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. If February is a leap year, it is distributed among the 29 days; otherwise, it is distributed among the 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0014] The variable costs mentioned: Based on the variable cost data provided by the nuclear power plant, including monthly variable cost data for each unit from January to December of a specified year (in ten thousand yuan), the variable cost data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. If February is a leap year, it is distributed to 29 days; otherwise, it is distributed to 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0015] The data prediction module described above: The model uses one dimension. Sliding window, Indicates the number of generating units.
[0016] The result aggregation module converts the unit of the results from yuan to 100 million yuan and retains two decimal places.
[0017] A method for predicting the net profit of nuclear power plants based on long short-term memory networks: Step 001: Obtain the annual power generation plan, annual plant power consumption rate, annual overhaul plan, other business revenue, revenue under the two detailed assessment rules, variable costs, and fixed costs from the nuclear power plant's production planning, overhaul, and financial management information systems to form the original dataset. ; Step 002: Transfer the original dataset Data from different fields is uniformly converted into dimensions. The vector, This indicates the number of generating units, 366 corresponds to one year, and 1 represents the date of 366 days. The annual power generation plan data corresponds to a vector. Annual plant power consumption rate data corresponding vector Average electricity price data corresponding to vector The vector corresponding to the overhaul planning data Other business revenue data corresponding to vectors The two detailed rules for assessing data correspond to vectors. Fixed cost data corresponding to vectors Variable cost data corresponding to vector ; Step 003: Merge the various data vectors obtained from data preprocessing to obtain the input set required by the model. ; Step 004: Input set As input, the Long Short-Term Memory (LSTM) deep learning algorithm is used to perform time series prediction based on the input data, predicting the daily time series for the next 366 days of the year. The value of net profit.
[0018] Step 003 specifically includes: Step 003-1: Convert the vector ,vector ,vector sum vector First, sort the data according to the date dimension, then remove the date dimension to obtain a vector. , , , Then, perform an inner product between the vectors, convert the units of the data to kilowatt-hours, and finally obtain the merged vector. ,Right now , dimension ; Step 003-2: Convert the vector , , , First, sort the data according to the date dimension, then remove the date dimension to obtain a vector. , , , , to obtain the merged vector ,Right now ,vector The dimension is ; Step 003-3: Convert the vector , Concatenate to obtain the input set ,Right now , dimension .
[0019] Step 004 specifically refers to: Step 004-1: Define a sliding window to display the input set. The data is taken from rows 1 to 366, each row Columns, to obtain input data ; Step 004-2: Normalize the input data, converting it into a unified format. Within the interval, the influence of different data units is eliminated, thus obtaining the normalized set. ; Step 004-3: Normalize the set The input is fed into the Long Short-Term Memory (LSTM) network algorithm, where a network structure consisting of LSM layers and random forgetting layers is used to obtain a set of feature sequences. ; Step 004-4: Then the feature sequence The output is obtained by processing through a fully connected layer. ; Steps 004-5: Then, concatenate the output with the input data to obtain the set. , dimension Then, by moving the sliding window one position to the right, we obtain the vector. Repeat steps 004-2 to 004-5 until... Dimensions This means that data forecasts for the next year have been completed.
[0020] Step 004 is followed by step 005: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] vectors in Extract the data and sum it up to obtain the projected net profit of the nuclear power plant for the next year.
[0021] The beneficial effects achieved by this invention are as follows: This invention is the first to construct a net profit forecasting model based on data from production, operation, finance, and overhaul. It also utilizes information technology and artificial intelligence algorithms to build a nuclear power plant net profit forecasting system that can automatically acquire and calculate data. By aggregating data from multiple fields within the nuclear power plant, it enables the prediction of future net profits, saving labor costs associated with data aggregation, broadening the range of data sources, and simultaneously realizing the function of predicting future scenarios. Attached Figure Description
[0022] Figure 1 This is a block diagram of a nuclear power plant net profit prediction system based on long short-term memory networks. Figure 2 This is a flowchart of a method for predicting the net profit of nuclear power plants based on long short-term memory networks. Detailed Implementation
[0023] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.
[0024] like Figure 1 As shown, a nuclear power plant net profit prediction system based on long short-term memory networks includes: a data acquisition module, a data preprocessing module, a data merging module, a data prediction module, and a result summarization module.
[0025] Data Acquisition Module: This module retrieves data impacting net profit from the nuclear power plant's production planning, overhaul, and financial sectors. The production planning sector includes annual power generation plans, annual plant power consumption rates, and annual electricity price data. The overhaul sector includes annual unit overhaul plans. The financial sector includes other business revenue, revenue assessed under two specific rules, variable costs, and fixed costs. This module utilizes computer technology, primarily for automatic data acquisition and supplemented by periodic reporting, reducing the workload of business departments.
[0026] The data preprocessing module, based on the acquired annual power generation plan, annual plant power consumption rate, overhaul plan, other business revenue, revenue assessed under the two detailed rules, variable costs, and fixed costs, performs data integrity checks and supplementation, cleans up dirty data, and standardizes statistical definitions to obtain preprocessed vector data, which is then input into the data merging module. This data cleaning provides high-quality data assurance for the accuracy of subsequent forecast calculations. Data preprocessing includes the following: Annual power generation plan: Based on the annual power generation plan information provided by the nuclear power plant for each unit, including the daily power generation of each unit in megawatts, and calculated using 366 days per year, a set of positional dimensions is generated. vector , This indicates the number of units, where 1 represents 366 days. The vector will then be... The units are converted to kilowatt-hours. The data in the vector corresponds to the daily power generation of each unit, and this power value takes into account the impact of summer operating conditions.
[0027] Annual plant power consumption rate: Based on the annual plant power consumption rate information provided by the nuclear power plant for each unit, including daily plant power consumption rate data for each unit, and calculated over a year of 366 days, a set of bit dimensions is generated. vector , This represents the number of generating units, with 1 indicating a 366-day period. The data in the vector corresponds to the daily power consumption rate of each generating unit, which is a decimal representing only the proportion.
[0028] Average electricity price data: Based on the average electricity price data provided by the nuclear power plant, including the monthly average electricity price data for each unit from January to December of a specified year, in yuan / kWh. The average electricity price data for each unit from January to December of a specified year is allocated to each day of the corresponding month. February is allocated to 29 days if it is a leap year; otherwise, it is allocated to 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0029] Overhaul planning data: Based on the overhaul planning information provided by the nuclear power plant for each unit, including the planned start and end times of overhauls, and calculated using 366 days per year, a set of data with the following dimensions is generated. vector , This represents the number of units, and 1 represents 366 days. The vector value is 0 within the start and end time range of the overhaul; the vector value is [value missing] within the time range of power reduction before the overhaul begins and power increase after the overhaul ends. The decimal values between these ranges correspond to the ratio of the power increase / decrease during overhaul to the rated power; the vector values for other time ranges are all 1.
[0030] Other operating revenue: Based on the other operating revenue data provided by the nuclear power plant, including monthly other operating revenue data for each unit from January to December of a specified year, in units of ten thousand yuan. The other operating revenue data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. February is distributed among the 29 days if it is a leap year; otherwise, it is distributed among the 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0031] Two detailed assessment criteria: Based on the two detailed assessment criteria provided by the nuclear power plant, including monthly assessment data for each unit from January to December of a specified year (unit: 10,000 yuan), the assessment data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. February is distributed among the 29 days if it is a leap year; otherwise, it is distributed among the 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0032] Fixed Costs: Based on fixed cost data provided by the nuclear power plant, including monthly fixed cost data for each unit from January to December of a specified year, in units of ten thousand yuan. The fixed cost data for each unit from January to December of a specified year is evenly distributed across each day of the corresponding month. February is distributed among the 29 days if it is a leap year; otherwise, it is distributed among the 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0033] Variable Costs: Based on variable cost data provided by the nuclear power plant, including monthly variable cost data for each unit from January to December of a specified year, in units of ten thousand yuan. The variable cost data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. February is distributed among the 29 days if it is a leap year; otherwise, it is distributed among the 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
[0034] Data merging module: After obtaining the preprocessed data from the data preprocessing module, it merges the daily data for each category in chronological order, column by column, to obtain a dataset with 366 rows and 5 columns. The vector data of the columns is passed as input data to the data prediction module.
[0035] The data prediction module obtains the merged input data, which can be used to calculate the prediction results, from the data preprocessing module. This input data is then fed into the trained Long Short-Term Memory (LSTM) network model, which uses a single dimension. The sliding window uses a network structure consisting of several Long Short-Term Memory (LSTM) layers and Dropout layers to perform various matrix calculations on the data within the window, along with trained weight values, to obtain the net profit data for the first day of the next year. The sliding window then moves forward and continues to calculate until the prediction calculation of the net profit data for each day of the next year is completed. The results are then fed into the result aggregation module.
[0036] Results Summary Module: The net profit amounts from the data prediction module are summed to obtain the total net profit for the entire next year. Then, the unit of the result is converted from yuan to 100 million yuan and two decimal places are retained to obtain the final net profit prediction value displayed to the public.
[0037] like Figure 2 As shown, the steps of a method for predicting the net profit of nuclear power plants based on long short-term memory networks are as follows: Step 001: Obtain the annual power generation plan, annual plant power consumption rate, annual overhaul plan, other business revenue, revenue under the two detailed assessment rules, variable costs, and fixed costs from the nuclear power plant's production planning, overhaul, and financial management information systems to form the original dataset. .
[0038] Step 002: Using the data preprocessing module, process the original dataset... Data from different fields is uniformly converted into dimensions. The vector, where This indicates the number of generating units; 366 corresponds to one year, and 1 represents 366 days. The annual power generation plan data corresponds to a vector. Annual plant power consumption rate data corresponding vector Average electricity price data corresponding to vector The vector corresponding to the overhaul planning data Other business revenue data corresponding to vectors The two detailed rules for assessing data correspond to vectors. Fixed cost data corresponding to vectors Variable cost data corresponding to vector .
[0039] Step 003: Using the data merging module, merge the various data vectors obtained from data preprocessing to obtain the input data set required by the model in the data prediction module.
[0040] Step 003-1: Convert the vector ,vector ,vector sum vector First, sort the data according to the date dimension, then remove the date dimension to obtain a vector. , , , Then, perform an inner product between the vectors, convert the units of the data to kilowatt-hours, and finally obtain the merged vector. ,Right now , dimension .
[0041] Step 003-2: Convert the vector , , , First, sort the data according to the date dimension, then remove the date dimension to obtain a vector. , , , , to obtain the merged vector ,Right now ,vector The dimension is .
[0042] Step 003-3: Convert the vector , The data is concatenated to obtain the input set for the data prediction module. ,Right now , dimension .
[0043] Step 004: Input set As input, the LSTM deep learning algorithm is used to perform time series prediction based on the input data, predicting the daily data for the next 366 days. The value of net profit.
[0044] Step 004-1: Define a sliding window to display the input set. The data is taken from rows 1 to 366, each row Columns, to obtain input data .
[0045] Step 004-2: Normalize the input data, converting it into a unified format. Within the interval, the influence of different data units is eliminated, thus obtaining the normalized set. .
[0046] Step 004-3: Normalize the set The input is fed into the LSTM algorithm, where a network structure consisting of Long Short-Term Memory (LSTM) layers and Dropout layers is used to obtain a set of feature sequences. .
[0047] Step 004-4: Then the feature sequence The output is obtained by processing through a fully connected layer. .
[0048] Steps 004-5: Then, concatenate the output with the input data to obtain the set. , dimension is Then, by moving the sliding window one position to the right, we obtain the vector. Repeat steps 004-2 to 004-5 until... Dimensions This means that data forecasts for the next year have been completed.
[0049] Step 005: Set vectors in Extract the data and sum it up to obtain the projected net profit of the nuclear power plant for the next year.
Claims
1. A nuclear power plant net profit prediction system based on long short-term memory networks, characterized in that: It includes a data acquisition module, a data preprocessing module, a data merging module, a data prediction module, and a results aggregation module; Data acquisition module: Acquires data on the impact on net profit from the nuclear power plant's production planning, overhaul, and financial sectors; Data preprocessing module: performs integrity checks and fills in the acquired data, cleans up dirty data, unifies statistical standards, and obtains preprocessed vector data, which is then passed to the data merging module; Data merging module: After obtaining the preprocessed data from the data preprocessing module, it merges the daily data for each category in chronological order, column by column, to obtain a dataset with 366 rows and 5 columns. The vector data of the columns is passed as input data to the data prediction module; Data prediction module: Input data is fed into a trained long short-term memory network model. The data in the window is processed through a network structure composed of several long short-term memory network layers and a random forgetting layer, and various matrix calculations are performed on the trained weight values to obtain the net profit data for the first day of the next year. Then the sliding window moves forward and continues to calculate until the prediction calculation of the net profit data for each day of the next year is completed. The results are then fed into the result summary module. Results summary module: The net profit amounts are summed to obtain the predicted total net profit for the next year. Then, the unit of the results is converted from yuan to 100 million yuan to obtain the final net profit forecast value displayed to the public.
2. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 1, characterized in that: The data acquisition modules are as follows: in the production planning field, annual power generation plan, annual plant power consumption rate, and annual electricity price data are included; in the overhaul field, annual overhaul plan for generating units is included; and in the financial field, other business income, income assessed under the two detailed rules, variable costs, and fixed costs are included.
3. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 1, characterized in that: The data acquisition module primarily relies on automatic data acquisition, supplemented by periodic data entry.
4. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 2, characterized in that: The aforementioned annual power generation plan: Based on the annual power generation plan information provided by the nuclear power plant for each unit, including the daily power generation of each unit in megawatts, and calculated using a year of 366 days, a set of positional dimensions is generated. vector , This indicates the number of units, where 1 represents 366 days. The vector will then be... The units are converted to kilowatt-hours. The data in the vector corresponds to the daily power generation of each unit, and this power value takes into account the impact of summer operating conditions.
5. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 2, characterized in that: The annual plant power consumption rate mentioned above: Based on the annual plant power consumption rate information provided by the nuclear power plant for each unit, including the daily plant power consumption rate data for each unit, and calculated over a year of 366 days, a set of bit dimensions is generated. vector , This represents the number of generating units, 1 represents 366 days, and the data in the vector corresponds to the daily power consumption rate of each generating unit, which is a decimal and only represents the proportion.
6. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 2, characterized in that: The annual electricity price data is based on average electricity price data provided by nuclear power plants, including monthly average electricity price data for each unit from January to December of a specified year, in yuan / kWh. The average electricity price data for each unit from January to December of a specified year is allocated to each day of the corresponding month. February is allocated to 29 days if it is a leap year; otherwise, it is allocated to 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
7. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 2, characterized in that: The aforementioned overhaul plan: Based on the overhaul planning information provided by the nuclear power plant for each unit, including the planned start and end times of the overhaul, and calculated using 366 days per year, a set of dimensions is generated. vector , This indicates the number of generating units; 1 represents a 366-day period; the vector value is 0 within the overhaul start and end time range; the vector value is [value missing] within the overhaul power reduction period before the overhaul begins and the overhaul power increase period after the overhaul ends. The decimal values between these ranges correspond to the ratio of the power increase / decrease during overhaul to the rated power; the vector values for other time ranges are all 1.
8. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 2, characterized in that: Other operating revenue: Based on the other operating revenue data provided by the nuclear power plant, including monthly other operating revenue data for each unit from January to December of a specified year (in ten thousand yuan), the other operating revenue data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. February is distributed among 29 days if it is a leap year; otherwise, it is distributed among 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
9. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 2, characterized in that: The two detailed performance evaluation criteria are as follows: Based on the performance evaluation data provided by the nuclear power plant, including monthly performance evaluation data for each unit from January to December of a specified year (in ten thousand yuan), the performance evaluation data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. If February is a leap year, it is distributed among the 29 days; otherwise, it is distributed among the 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
10. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 2, characterized in that: The fixed costs mentioned above are based on the fixed cost data provided by the nuclear power plant, including monthly fixed cost data for each unit from January to December of a specified year, in units of ten thousand yuan. The fixed cost data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. If February is a leap year, it is distributed among the 29 days; otherwise, it is distributed among the 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
11. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 2, characterized in that: The variable costs mentioned: Based on the variable cost data provided by the nuclear power plant, including monthly variable cost data for each unit from January to December of a specified year (in ten thousand yuan), the variable cost data for each unit from January to December of a specified year is evenly distributed to each day of the corresponding month. If February is a leap year, it is distributed to 29 days; otherwise, it is distributed to 28 days, with the 29th day set to 0. After processing, a set of bit dimensions is obtained. vector , This indicates the number of units, and 1 represents 366 days.
12. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 1, characterized in that: The data prediction module described above: The model uses one dimension. Sliding window, Indicates the number of generating units.
13. The nuclear power plant net profit prediction system based on long short-term memory networks according to claim 1, characterized in that: The result aggregation module converts the unit of the results from yuan to 100 million yuan and retains two decimal places.
14. A method for predicting the net profit of nuclear power plants based on long short-term memory networks, characterized in that: Step 001: Obtain the annual power generation plan, annual plant power consumption rate, annual overhaul plan, other business revenue, revenue under the two detailed assessment rules, variable costs, and fixed costs from the nuclear power plant's production planning, overhaul, and financial management information systems to form the original dataset. ; Step 002: Transfer the original dataset Data from different fields is uniformly converted into dimensions. The vector, This indicates the number of generating units, 366 corresponds to one year, and 1 represents the date of 366 days. The annual power generation plan data corresponds to a vector. Annual plant power consumption rate data corresponding vector Average electricity price data corresponding vector The vector corresponding to the overhaul planning data Other business revenue data corresponding to vectors The two detailed rules for assessing data correspond to vectors. Fixed cost data corresponding to vectors Variable cost data corresponding to vector ; Step 003: Merge the various data vectors obtained from data preprocessing to obtain the input set required by the model. ; Step 004: Input set As input, the Long Short-Term Memory (LSTM) deep learning algorithm is used to perform time series prediction based on the input data, predicting the daily time series for the next 366 days of the year. The value of net profit.
15. The method for predicting net profit of nuclear power plants based on long short-term memory networks according to claim 14, characterized in that: Step 003 specifically includes: Step 003-1: Convert the vector ,vector ,vector sum vector First, sort the data according to the date dimension, then remove the date dimension to obtain a vector. , , , Then, perform an inner product between the vectors, convert the units of the data to kilowatt-hours, and finally obtain the merged vector. ,Right now , dimension ; Step 003-2: Convert the vector , , , First, sort the data according to the date dimension, then remove the date dimension to obtain a vector. , , , , to obtain the merged vector ,Right now ,vector The dimension is ; Step 003-3: Convert the vector , Concatenate to obtain the input set ,Right now , dimension .
16. The method for predicting net profit of nuclear power plants based on long short-term memory networks according to claim 15, characterized in that: Step 004 specifically refers to: Step 004-1: Define a sliding window to display the input set. The data is taken from rows 1 to 366, each row Columns, to obtain input data ; Step 004-2: Normalize the input data, converting it into a unified format. Within the interval, the influence of different data units is eliminated, thus obtaining the normalized set. ; Step 004-3: Normalize the set The input is fed into the Long Short-Term Memory (LSTM) network algorithm, where a network structure consisting of LSM layers and random forgetting layers is used to obtain a set of feature sequences. ; Step 004-4: Then the feature sequence The output is obtained by processing through a fully connected layer. ; Steps 004-5: Then, concatenate the output with the input data to obtain the set. , dimension Then, by moving the sliding window one position to the right, we obtain the vector. Repeat steps 004-2 to 004-5 until... Dimensions This means that data forecasts for the next year have been completed.
17. The method for predicting net profit of nuclear power plants based on long short-term memory networks according to claim 16, characterized in that: Step 004 is followed by step 005: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] vectors in Extract the data and sum it up to obtain the projected net profit of the nuclear power plant for the next year.