A method and system for predicting nuclear fuel production demand based on multi-model measurement

Abstract

The application discloses a kind of nuclear fuel production demand prediction method and system based on multi-model measurement, comprising the following steps: step S1. Obtain the basic data required for nuclear fuel production demand prediction;Step S2. From the basic data, extract the related parameters in the nuclear fuel production demand prediction;Step S3. Based on statistical model measurement and based on deep learning algorithm measurement, realize the prediction of nuclear fuel production demand;Step S4. The results of measurement are compared and analyzed using multi-criteria decision method, and the optimal solution is recommended according to the measurement parameter measurement scenario.The present application brings a brand-new idea and method for the prediction of nuclear fuel production demand, greatly improving the intelligent level of measurement.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of nuclear fuel production demand estimation, in particular to a nuclear fuel production demand prediction method and system based on multi-model estimation. BACKGROUND

[0002] Under the background of the accelerated adjustment of global energy structure, nuclear energy has become increasingly important in the energy system due to its clean and efficient advantages. The digital construction level of nuclear fuel production, as the core link of the nuclear energy industry, is directly related to the safe and efficient development of the nuclear energy industry. Nuclear fuel demand prediction is an important work for nuclear fuel production enterprises to manage production capacity planning. Building a flexible and intelligent prediction system to achieve intelligent and accurate prediction of demand can help optimize the production and distribution of nuclear fuel and ensure the sustainable development of the nuclear energy industry.

[0003] However, nuclear fuel production demand comes from nuclear power units that are running or being constructed in nuclear power enterprises. Different nuclear power units have different nuclear fuel demand and refueling cycles due to the use of different reactor technologies. Traditional nuclear fuel production demand prediction is based on the standard parameters selected during the design of all nuclear power units, including refueling cycle and fuel demand. Since the design parameters of nuclear power units lag behind market changes in the actual operation process, using fixed parameters for demand prediction will result in a large deviation between the calculated results and the actual demand, and the precision is not high. Nuclear fuel production demand prediction methods include statistical methods, system dynamics methods, machine learning methods, etc. The International Atomic Energy Agency has developed a nuclear fuel cycle simulation system model, and the World Nuclear Association publishes a nuclear fuel supply and demand report every two years, which predicts the supply and demand of different stages of the nuclear fuel cycle. However, there is no mature tool on the domestic market for nuclear fuel production demand prediction, and only relying on traditional manual prediction is tedious and has low accuracy. Medium and long-term prediction of nuclear fuel production demand needs to consider multiple factors such as policy and market, adjust different parameter variables, and realize demand prediction under different scenarios to find the optimal solution for company production capacity allocation. Traditional calculation methods highly depend on manual experience, and the final calculation scheme is determined through manual comparison and decision analysis, which has limitations in professionalism, accuracy, and sustainability. SUMMARY

[0004] The present application aims to overcome the shortcomings of the prior art and provide a nuclear fuel production demand prediction method and system based on multi-model estimation, which brings a new idea and method for nuclear fuel production demand prediction and greatly improves the intelligent level of calculation.

[0005] The purpose of the present application is achieved by the following technical solution: a nuclear fuel production demand prediction method based on multi-model estimation, comprising the following steps:

[0006] Step S1. Obtain the basic data required for nuclear fuel production demand prediction;

[0007] Step S2. Extract relevant parameters in nuclear fuel production demand prediction from the basic data;

[0008] Step S3. Realize the prediction of nuclear fuel production demand based on statistical model calculation and deep learning algorithm calculation;

[0009] Step S4. Compare and analyze the results of the calculation using the multi-criteria decision method, and recommend the optimal scheme according to the calculation parameter calculation scenario.

[0010] A nuclear fuel production demand prediction system based on multi-model calculation, comprising:

[0011] A data acquisition module for acquiring the basic data required for nuclear fuel production demand prediction;

[0012] A parameter extraction module for extracting relevant parameters in nuclear fuel production demand prediction from the basic data;

[0013] A demand prediction module for realizing the prediction of nuclear fuel production demand based on statistical model calculation and deep learning algorithm calculation;

[0014] A comparison and optimization module for comparing and analyzing the results of the calculation using the multi-criteria decision method, and recommending the optimal scheme according to the calculation parameter calculation scenario.

[0015] The beneficial effects of the present application are: based on a large amount of historical refueling demand data, according to the boundary conditions under different scenarios, using statistical regression model, deep learning model algorithm different from traditional experience algorithm to realize intelligent generation and comparison of calculation scheme, best scheme recommendation and optimization, break through the lagging effect brought by traditional fixed parameter prediction, break the professional limitations of relying on artificial experience, bring a new idea and method for the prediction of nuclear fuel production demand, greatly improve the intelligent level of calculation. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The method flowchart of the present application;

[0017] Figure 2 The system function module diagram in the embodiment. DETAILED DESCRIPTION

[0018] The technical solutions of the present application will be described in further detail below in conjunction with the drawings, but the protection scope of the present application is not limited to the following description.

[0019] As Figure 1 shown, a nuclear fuel production demand prediction method based on multi-model calculation, comprising the following steps:

[0020] Step S1. Obtain the basic data required for nuclear fuel production demand prediction;

[0021] Before nuclear fuel production demand prediction, the basic data affecting the prediction are prepared and processed. The basic data include two categories: one is nuclear power unit standard data, and the other is nuclear power unit actual refueling data.

[0022] Nuclear power unit standard data preparation: The reactor technology, state (in service, under construction, new unit), refueling period, refueling demand (product enrichment, tail enrichment, demand) of the nuclear power unit involved are entered and prepared.

[0023] Nuclear power unit actual refueling data preparation: The historical refueling data of the nuclear power unit involved are entered and prepared, including refueling demand (product enrichment, tail enrichment, demand), refueling period, and refueling batch.

[0024] Step S2. Extract relevant parameters in nuclear fuel production demand prediction from the basic data;

[0025] For the two categories of basic data in data preparation, relevant parameters in nuclear fuel production demand prediction are extracted. All parameters are based on nuclear power units, and standard, actual average, and actual latest parameters of each group of nuclear power units are extracted.

[0026] Standard parameters, i.e. nuclear power unit theoretical design parameters, are derived from the extraction in nuclear power unit standard data preparation;

[0027] Actual average parameters, i.e. average parameters obtained by calculating historical refueling data using time series analysis model;

[0028] Actual latest parameters, i.e. the last refueling data in nuclear power unit actual refueling data.

[0029] In each type of parameter, the refueling period is included.

[0030] Step S3. Based on statistical model estimation and deep learning algorithm estimation, the nuclear fuel production demand is predicted;

[0031] When developing nuclear fuel production demand prediction, the time and unit to be predicted need to be selected, and the assignment of refueling period parameters (i.e. standard, actual average, and actual latest) needs to be selected according to the estimation scenario, as follows:

[0032] Estimation time selection: According to the estimation needs, the estimation time range is selected, such as 2026 to 2030, and the starting point of the estimation time is the time of the last historical refueling.

[0033] Unit selection: According to the needs of the calculation, the nuclear power unit involved in the calculation is selected, and the data can be filtered from the unit in the data preparation, which needs to automatically bring out the stack technology, state and parameters of the unit.

[0034] Recommendation of refueling cycle parameters: According to the pre-set calculation scenario, the system will automatically recommend the applicable refueling cycle parameters for calculation. That is, if the scenario is theoretical calculation, the standard parameters are recommended; if the scenario is calculation according to the latest state, the actual latest parameters are recommended; if the scenario is based on model calculation, the actual average parameters are recommended.

[0035] Statistical model-based calculation:

[0036] Other parameter adjustment: that is, on the basis of selecting the refueling cycle parameters, using statistical model (multiple linear regression) and uranium enrichment material balance theory, based on the historical refueling data of nuclear fuel, the quantitative influence relationship between product enrichment, tailing enrichment and fuel demand is analyzed. Here, the variable relationship is derived by combining the uranium enrichment material balance formula, and the influence is quantified by regression analysis. The following are the detailed sub-steps:

[0037] In nuclear fuel production, natural uranium (U-238) =0.711%, that is, 0.00711) is enriched to obtain product (nuclear fuel) and tailings, which satisfies the material balance: (U-235 mass conservation); (total uranium mass conservation). Among them, F is the natural uranium feed quantity, P is the nuclear fuel product quantity (i.e. nuclear fuel demand), and T is the tailings quantity.

[0038] Solving the two equations eliminates T, and the relationship between nuclear fuel demand and product enrichment ( ) and tailing enrichment ( ) is derived:

[0039]

[0040] When the F natural uranium feed quantity is a fixed value, the P nuclear fuel demand and the product enrichment and tailing enrichment have an influence relationship.

[0041] 50 sets of refueling data are extracted from the historical refueling data of nuclear fuel to construct a multiple linear regression model. Each set of data includes nuclear fuel demand, product enrichment and tailing enrichment. The model is as follows:

[0042]

[0043] The coefficients are estimated by least squares method , , In the embodiments of this application, the accuracy of the model can be verified by mean squared error. The reliability of the model is ensured by significance testing, goodness-of-fit analysis, and residual analysis.

[0044] In the embodiments of this application, fuel demand can be calculated by adjusting the parameters of product enrichment and tailings enrichment based on the fitted model.

[0045] Scenario 1: Adjusting only the product enrichment ( ), fixed tailings enrichment ( For example, to Increased from 3.0% to 3.5%. The percentage remains unchanged at 0.25%. The product enrichment level is calculated based on the data. The product enrichment (P) is negatively correlated with the nuclear fuel demand P. The higher the value, the less nuclear fuel P is required.

[0046] Scenario 2: Adjust only the enrichment degree of tailings ( ), fixed product enrichment ( For example, to From 0.25% to 0.2%, The percentage remains unchanged at 3.0%. Calculations based on data show that the enrichment degree of the tailings (…) The enrichment of tailings is positively correlated with the nuclear fuel demand P. The higher the value, the greater the nuclear fuel demand (P).

[0047] Scenario 3: Collaboratively adjust the scenario and simultaneously adjust the product enrichment level. ) and tailings enrichment ( Sensitivity analysis was conducted to calculate the impact of product enrichment and tailings enrichment on fuel demand.

[0048] Through the above modeling process, based on historical refueling data, this invention enables flexible adjustment of product enrichment and tailings enrichment, thereby calculating changes in fuel demand.

[0049] Calculation based on deep learning model algorithm:

[0050] By using deep learning models for prediction, a deep neural network model containing at least one long short-term memory (LSTM) layer is constructed, which can capture more complex nonlinear relationships compared to multiple linear regression models.

[0051] C1. Data Acquisition and Preprocessing: Based on the selected nuclear power unit, extract multiple feature data from historical refueling data, including refueling time, product enrichment, tailings enrichment, and nuclear fuel demand.

[0052] Data cleaning and interpolation are performed, the refueling time is one-hot encoded or periodically encoded, and interpolation is performed to obtain feature data under different refueling times, and data normalization processing is carried out;

[0053] The normalization processing is to use the minimum-maximum normalization method to unify data features of different dimensions and different orders of magnitude to a specified scale space, and the formula is as follows:

[0054]

[0055] Among them:

[0056] X is the original feature value.

[0057] Xmin is the minimum value of the feature in the entire data set.

[0058] Xmax is the maximum value of the feature in the entire data set.

[0059] Xnorm is the normalized value, ranging between [0, 1].

[0060] C2, construct sequence, convert time data into supervised learning format:

[0061] Define a time step N (for example, N=12 months), and the normalized feature data at time t is taken as the input sequence , and the core fuel demand at t+1 is taken as the label ;

[0062] Thus, a sample is obtained.

[0063] C3, repeat C1 multiple times to obtain a sample set, and perform prediction model construction and training:

[0064] C301. A prediction model based on a deep learning algorithm is constructed using a stacked long short-term memory network architecture:

[0065] Input layer, input time step N input sequence ;

[0066] First LSTM layer, with neural units, activation function (such as relu or tanh), and set return_sequences=True to output its hidden state sequence to the next layer; (optional) first Dropout layer: randomly inactivate neurons by a certain proportion (such as 0.2), used to prevent overfitting; second LSTM layer: with one neural unit with return_sequences=False to output only the hidden state of the last time step of the sequence; (optional) a second Dropout layer: to further prevent overfitting. Output layer (fully connected layer): with 1 neural unit and linear activation function to output continuous regression prediction values ;

[0067] C302. Based on the regression prediction value and label , select mean squared error or mean absolute error as the regression loss function; select Adam or RMSprop as the adaptive learning rate optimizer to update the prediction model;

[0068] C303. Model training: divide the sample set into training set and validation set, use the samples in the training set to train the prediction model until the loss of the model on the validation set converges;

[0069] C4, demand prediction: feed the prepared input sequence of time step N into the trained LSTM model, and the model outputs the normalized prediction value , perform inverse transformation to obtain the physically meaningful nuclear fuel demand prediction value.

[0070] For different nuclear power units, the model needs to be constructed according to step S3 and the optimal scheme is recommended according to step S4;

[0071] Step S4. Compare and analyze the results measured by the multi-criteria decision method, and recommend the optimal scheme according to the measured parameters and the measured scene.

[0072] Based on the results of the schemes measured by different algorithm models, the system will record all the selected parameters of the scheme. In the case of fixed parameters, the results measured by different model algorithms are compared and analyzed by using the multi-criteria decision method (TOPSIS), and the optimal scheme is recommended according to the selection of measured parameters or the measured scene, including:

[0073] S401. Select performance indicators and evaluation indicators as statistical model measurement and deep learning algorithm measurement;

[0074] (1) Performance indicators include determination coefficient, root mean square error, and mean absolute error;

[0075] ​(1.1) For the prediction model of nuclear fuel production demand calculated based on a statistical model: extract the refueling demand quantity in multiple historical refueling data, including product enrichment, tail material enrichment and nuclear fuel demand quantity, input the product enrichment and tail material enrichment in each historical refueling data into the prediction model of nuclear fuel production demand calculated based on a statistical model, obtain multiple prediction results of nuclear fuel demand quantity, and combine the nuclear fuel demand quantity in each historical refueling data to calculate the determination coefficient, root mean square error and mean absolute error;

[0076] (1.2) For the prediction model calculated based on a deep learning algorithm, first obtain multiple samples according to steps C1-C2; then input the input sequence of each sample into the prediction model calculated based on a deep learning algorithm, and calculate the determination coefficient, root mean square error and mean absolute error of the output result of the model and the actual label of each sample;

[0077] The determination coefficient is , the closer to 1, the better the model fitting effect.

[0078] The root mean square error is , which reflects the average deviation of the predicted value from the true value.

[0079] The mean absolute error is , which represents the average absolute deviation between each predicted value and the true value.

[0080] When the prediction model of nuclear fuel production demand calculated based on a statistical model is calculated by using the above formula, since multiple historical refueling data are extracted, each data will obtain a nuclear fuel demand prediction result , the average value of each data is , and the true nuclear fuel demand of each data is ; wherein n represents the number of extracted historical refueling data, respectively represent the nuclear fuel demand prediction result and the true nuclear fuel demand of the i-th historical refueling data.

[0081] (2) The evaluation indexes include calculation cost and requirement for sample size, which are realized by self-definition:

[0082] The calculation cost score value: the higher the calculation cost, the lower the score value; the calculation cost score value is set for the prediction model of nuclear fuel production demand calculated based on a statistical model and the prediction model calculated based on a deep learning algorithm by self-definition, and the calculation cost score value of the prediction model of nuclear fuel production demand calculated based on a statistical model (for example, 0.8) is greater than that of the prediction model calculated based on a deep learning algorithm (for example, 0.2);

[0083] Sample size score value: the statistical model and neural network model sample needs to be valued: the higher the requirement for sample size, the lower the score value; by customizing the sample size score value, the sample size score value of the nuclear fuel production demand prediction model based on the statistical model and the prediction model based on the deep learning algorithm is set, and the sample size score value (for example, 0.8) of the nuclear fuel production demand prediction model based on the statistical model is greater than that (for example, 0.2) of the prediction model based on the deep learning algorithm;

[0084] S402. According to the scheme calculation scene, different models are selected to generate demand prediction, and the scheme calculation scene includes regular production plan and new fuel research and development;

[0085] (1) Let the self-defined weights of the determination coefficient, the root mean square error, and the average absolute error be The self-defined weights of the cost score value and the sample size score value are , and ;

[0086] For the regular production plan, it is necessary to set ; for example, ;

[0087] For the new fuel research and development, it is necessary to set ; for example, ;

[0088] (2) In any scheme calculation scene, the determination coefficient, the root mean square error, the average absolute error, the cost score value, and the sample size score value of the nuclear fuel production demand prediction model based on the statistical model are weighted and summed according to the set weights;

[0089] The determination coefficient, the root mean square error, the average absolute error, the cost score value, and the sample size score value of the prediction model based on the deep learning algorithm are weighted and summed according to the set weights;

[0090] The model corresponding to the calculation scheme with a higher weighted sum result is selected as the recommended prediction scheme.

[0091] In the embodiment of the present application, after the optimal calculation scheme is selected, the decomposition of the nuclear fuel demand covered by the scheme is generated, which can be split according to nuclear power units and years. And the time trend analysis and composition analysis of nuclear fuel production demand are displayed by using visual charts and other means to assist nuclear fuel enterprise capacity allocation decision-making.

[0092] A nuclear fuel production demand prediction system based on multi-model calculation, comprising:

[0093] A data acquisition module for acquiring basic data required for nuclear fuel production demand prediction;

[0094] a parameter extraction module configured to extract relevant parameters in the nuclear fuel production demand prediction from the basic data;

[0095] a demand prediction generation module configured to predict the nuclear fuel production demand based on statistical model calculation and deep learning algorithm calculation;

[0096] a comparison and optimization module configured to compare and analyze the results of the multi-criteria decision method calculation, and recommend an optimal scheme according to the selection of the calculation parameters or the calculation scenario.

[0097] In the embodiments of the present application, based on the system, the architecture in the actual application process is as shown in Figure 2 The modules mainly include a basic information management module, a refueling information management module, a calculation scheme formulation and management module, and a visual analysis module.

[0098] The basic information management module is deployed on the server side and is configured to realize preparation and processing of nuclear power unit basic data, and mainly provides entry, classification and retrieval of nuclear power enterprise, nuclear power station and nuclear power unit information, and extraction and display of parameters.

[0099] The refueling information management module is deployed on the server side and is configured to realize preparation and processing of nuclear power unit historical refueling data, and mainly provides entry, classification and retrieval of each refueling data of the nuclear power unit, and extraction and display of related parameters.

[0100] The calculation scheme formulation and management module receives nuclear power unit, refueling and parameter data from the basic information management module and the refueling information management module, and provides full-process function support for calculation scheme formulation. The main functions include calculation preparation, parameter adjustment and result confirmation. The calculation preparation supports selection of calculation time and unit range, and recommendation of selected unit parameters; the parameter adjustment realizes model selection, parameter adjustment and calculation; and the result confirmation forms a complete calculation scheme and realizes scheme comparison and optimization.

[0101] The visual analysis module receives data from the calculation scheme formulation and management module, realizes visual display and analysis of the calculation results, and mainly includes nuclear fuel production demand trend analysis and composition analysis to assist nuclear fuel enterprise management decision-making.

[0102] Workflow

[0103] The system can be deployed on a cloud server and interacts with users through a front-end interface such as a web page. The typical workflow is as follows:

[0104] User information entry: the user enters the basic information and the refueling information through the front-end interface. The system extracts the relevant parameters and displays and queries them on the front-end interface.

[0105] Calculation scheme formulation: the user carries out nuclear fuel production demand prediction through the front-end interface, selects the calculation time and the calculation range, and the system automatically recommends the relevant parameters of the selected unit. The user selects the calculation model according to the actual situation and adjusts the parameters, and the system automatically generates the relevant scheme. According to the data dependency, the applicable scene and other factors, the optimal scheme is recommended, the user can compare multiple schemes, the server side background system compares the differences between the schemes and calculates the rationality of the schemes, and the front-end interface directly displays the results to assist the user in selecting the final scheme.

[0106] Result presentation: after the selected calculation scheme is generated, the server side transmits the data to the corresponding visual display page, the front-end interface updates the calculation result data of the latest calculation scheme, the user can further click and view, and further interact.

[0107] In summary, on the basis of fully utilizing the historical refueling data, the statistical regression model and the deep learning model are innovatively adopted for nuclear fuel demand prediction, the model selection, parameter adjustment and scheme recommendation in the calculation process are comprehensively utilized by the system, the process of nuclear fuel production demand prediction is improved and promoted, the prediction accuracy is greatly improved, the regression model combines the physical nature of nuclear fuel production with statistical analysis methods, realizes the balance of "interpretability + practicality", breaks through the limitations of traditional experience prediction, solves the "nonlinearity, multi-coupling, extreme scene" problem in nuclear fuel demand prediction by using the neural network model, breaks through the fitting boundary of the traditional linear model, and realizes higher precision prediction.

Claims

1. A method for predicting the demand for nuclear fuel production based on multi-model estimation, characterized by: The method comprises the following steps: Step S1. Obtain the basic data required for nuclear fuel production demand prediction; Step S2. Extract relevant parameters in nuclear fuel production demand prediction from the basic data; Step S3. Realize the prediction of nuclear fuel production demand based on statistical model calculation and deep learning algorithm calculation; The step S3 comprises: A1, select the time and unit to be predicted, and select the assignment value of the refueling period parameter according to the calculation scenario; A2, select the calculation time range, and the starting point of the calculation time range is the time of the last historical refueling; A3, select the nuclear power unit to be calculated and filter the basic data of the unit; A4, refueling period parameter determination: according to the pre-set calculation scenario, recommend the refueling period parameter for calculation: If the scenario is theoretical calculation, recommend the refueling period in the standard parameter; If the scenario is model-based calculation, recommend the average value of the refueling period in the actual average parameter; If the scenario is the latest state calculation, recommend the refueling period of the last refueling in the actual latest parameter; A5, from the starting point of the calculation time range, every other refueling period in the calculation time range, the nuclear fuel demand is predicted; The prediction method includes statistical model calculation and deep learning algorithm calculation; Step S4. Compare and analyze the results of the calculation using the multi-criteria decision method, and recommend the optimal scheme according to the calculation parameter calculation scenario; The step S4 comprises: S401. Select performance indicators and evaluation indicators as statistical model calculation and deep learning algorithm calculation; (1) Performance indicators include determination coefficient, root mean square error, and mean absolute error; (1.1) For the nuclear fuel production demand prediction model based on statistical model calculation: extract the refueling demand in multiple historical refueling data, including product enrichment, tail enrichment and nuclear fuel demand, input the product enrichment and tail enrichment in each historical refueling data into the nuclear fuel production demand prediction model based on statistical model calculation, obtain multiple nuclear fuel demand prediction results, and calculate the determination coefficient, root mean square error and mean absolute error combined with the nuclear fuel demand in each historical refueling data; (1.2) For the prediction model based on deep learning algorithm calculation, first obtain multiple samples according to steps C1~C2; then input the input sequence of each sample into the prediction model based on deep learning algorithm calculation, and calculate the determination coefficient, root mean square error and mean absolute error of the output result of the model and the actual label of each sample; (2) Evaluation indicators include calculation cost and sample size requirement, which are realized by self-definition: The higher the calculation cost, the lower the score value; the calculation cost score value is set for the nuclear fuel production demand prediction model based on statistical model calculation and the prediction model based on deep learning algorithm calculation by self-defined calculation cost score value setting method, and the calculation cost score value of the nuclear fuel production demand prediction model based on statistical model calculation is greater than that of the prediction model based on deep learning algorithm calculation; Sample quantity score value: the statistical model and neural network model sample needs to be valued: the higher the requirement for sample quantity, the lower the score value; by setting the sample quantity score value in a self-defined manner, the sample quantity score value is set for the nuclear fuel production demand prediction model based on statistical model calculation and the prediction model based on deep learning algorithm calculation, and the sample quantity score value of the nuclear fuel production demand prediction model based on statistical model calculation is greater than that of the prediction model based on deep learning algorithm calculation; S402. According to the scheme calculation scene, different models are selected to generate demand prediction, and the scheme calculation scene includes regular production planning and new fuel research and development; (1) Set the self-defined weight of the coefficient of determination, the root mean square error, and the mean absolute error to be The self-defined weight of the cost score value and the sample size score value is , and ; For a regular production plan, it is necessary to set ; For the development of new fuels, it is necessary to set up ; (2) In any scheme calculation scene, the decision coefficient, root mean square error, average absolute error, cost score value and sample quantity score value of the nuclear fuel production demand prediction model based on statistical model calculation are weighted and summed according to the set weight; The decision coefficient, root mean square error, average absolute error, cost score value and sample quantity score value of the prediction model based on deep learning algorithm calculation are weighted and summed according to the set weight; The model corresponding to the calculation scheme with higher weighted sum result is selected as the recommended prediction scheme.

2. The nuclear fuel production demand prediction method based on multi-model calculation according to claim 1, characterized in that: There are multiple nuclear power unit groups, and for each nuclear power unit group, the basic data required for production demand prediction need to be obtained, including nuclear power unit standard data and nuclear power unit actual refueling data; The nuclear power unit standard data includes nuclear power unit standard reactor technology, state, refueling period and refueling demand; The nuclear power unit actual refueling data uses multiple historical refueling data of the nuclear power unit, and each historical refueling data includes refueling demand, refueling period and refueling batch.

3. The method for predicting the demand for nuclear fuel production based on multi-model estimation according to claim 2, characterized in that: The state includes in-service, under-construction and new unit; The refueling demand includes product enrichment, tailing enrichment and nuclear fuel demand.

4. The method for predicting the demand for nuclear fuel production based on multi-model estimation according to claim 3, characterized in that: In step S2, for each nuclear power unit group, the extracted relevant parameters include standard parameters, actual average parameters and actual latest parameters; The standard parameters include the standard refueling period of the nuclear power unit; The actual average parameter is obtained by calculating the average value of the refueling period from multiple historical refueling data; The actual latest parameter uses the refueling period of the last refueling in the historical refueling data.

5. The method for predicting the demand for nuclear fuel production based on multi-model estimation according to claim 4, characterized in that: The statistical model calculation is based on the historical refueling data of the nuclear fuel, which analyzes the quantitative influence relationship between the product enrichment, tailing enrichment and fuel demand by using the statistical model and the uranium concentrate material balance theory based on the determined refueling period parameter, including: B101. In nuclear fuel production, the enrichment of natural uranium product = 0.711%, after enrichment, nuclear fuel and tailings are obtained, satisfying the material balance, including: U-235 mass conservation: ; total uranium mass conservation ; wherein F is the amount of natural uranium feed, P is the amount of nuclear fuel product, i.e. the amount of nuclear fuel demand; T is the amount of tails, is the enrichment of the natural uranium product; The relationship between the amount of nuclear fuel required and the product enrichment and the tailings enrichment is derived by simultaneously eliminating T from both equations. ； When the F natural uranium feed quantity is a fixed value, there is an influence relationship between the P nuclear fuel demand and the product enrichment and tailing enrichment; B102, based on the selected nuclear power unit, the refueling demand in the multiple historical refueling data is extracted to construct a multiple linear regression model; Each set of data needs to include nuclear fuel demand, product enrichment and tailing enrichment; the model is as follows: ； coefficients are estimated by least squares , , , i.e. a prediction model for the demand of nuclear fuel production based on the statistical model The product enrichment degree and the tail enrichment degree are input into a prediction model of nuclear fuel production demand based on a statistical model calculation to obtain a calculation result.

6. The method for predicting the demand for nuclear fuel production based on multi-model estimation according to claim 5, characterized in that: The calculation based on the deep learning algorithm comprises: C1, data acquisition and preprocessing: based on the selected nuclear power unit, a plurality of characteristic data are extracted from historical refueling data, including refueling time, product enrichment degree, tail enrichment degree and nuclear fuel demand; Data cleaning and interpolation are performed, the refueling time is one-hot encoded or periodically encoded, and interpolation is performed to obtain characteristic data under different refueling times, and data normalization processing is performed; The normalization processing is to use a minimum-maximum normalization method to unify data characteristics of different dimensions and different orders of magnitude to a specified scale space; C2, sequence construction: converting time data into a supervised learning format; Define a time step N, take the normalized feature data at time t as the input sequence , and take the core fuel demand at time t+1 as the label . ; Thus, a sample is obtained ; C3, repeating C1 multiple times to obtain a sample set, and performing prediction model construction and training: C301. A stacked long short-term memory network architecture is used to construct a prediction model based on a deep learning algorithm calculation: input layer, input sequence of input time steps N ; The first LSTM layer has neurons with a relu or tanh activation function, and outputs its sequence of hidden states to the next layer. A first Dropout layer: randomly inactivating neurons according to a set proportion, for preventing overfitting; Second LSTM layer: with 128 neurons, set return_sequences=False to only output the hidden state of the last time step of the sequence; A second Dropout layer: randomly inactivating neurons according to a set proportion, for further preventing overfitting; Output layer, with a fully connected layer, with 1 neuron and linear activation function, to output continuous regression predictions ; C302. based on regression prediction values and labels The mean square error or mean absolute error is selected as the regression loss function; Adam or RMSprop is selected as the adaptive learning rate optimizer to update the prediction model; C303. Model training: dividing the sample set into a training set and a validation set, using samples in the training set to train the prediction model until the loss of the prediction model on the validation set converges; C4, demand prediction: feed the prepared input sequence of time step N into the trained LSTM model, and the model outputs the normalized predicted value , the perform inverse transformation to obtain the physically meaningful nuclear fuel demand prediction value.

7. A nuclear fuel production demand forecasting system based on multi-model estimation, based on the method of any one of claims 1-6, characterized by: Comprise: A data acquisition module for acquiring basic data required for nuclear fuel production demand prediction; A parameter extraction module for extracting related parameters in nuclear fuel production demand prediction from the basic data; A demand prediction generation module based on statistical model calculation and deep learning algorithm calculation to realize prediction of nuclear fuel production demand; A comparison and optimization module for recording all selected parameters of different calculation schemes, comparing and analyzing the results of multi-criteria decision-making under the condition that the parameters are fixed, and recommending the optimal scheme according to the selection of calculation parameters or the calculation scenario.

Images