New energy project decommissioning cost prediction method and system based on machine model

Abstract

The invention relates to the technical field of cost prediction, in particular to a new energy project decommissioning cost prediction method and system based on a machine model, and the method comprises the steps: obtaining full-life-cycle basic data and multi-scene decommissioning related historical data of a new energy project; based on the full-life-cycle basic data and multi-scene decommissioning related historical data, static features and dynamic features related to decommissioning cost are extracted, and a multi-dimensional feature set is constructed; and dividing the multi-dimensional feature set into a training data set and a verification data set, constructing an initial machine prediction model based on the training data set, and optimizing parameters of the initial machine prediction model through iterative training. According to the method, the training data set and the verification data set are separated, and iterative optimization is carried out, so that the performance of the initial machine prediction model can be continuously improved, the problem of over-fitting or under-fitting in the model can be eliminated, and the model has higher generalization ability.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of cost prediction, in particular to a new energy project decommissioning cost prediction method and system based on a machine model. BACKGROUND

[0002] New energy projects refer to projects that use renewable resources or alternative energy sources to produce and supply energy. These projects aim to reduce dependence on traditional fossil fuels such as coal, oil, and natural gas, reduce carbon emissions, and promote the transformation of the energy structure.

[0003] Currently, traditional methods usually provide a single prediction result without considering the uncertainty of the prediction. There is no quantitative analysis of prediction bias and risk factors, which leads to a lack of basis for decision-makers to respond to possible changes. This makes traditional methods lack sufficient flexibility and security when facing complex and variable projects.

[0004] In addition, traditional methods may lack the ability to deeply mine historical data, especially when faced with complex multi-scenario data, often failing to fully consider the diversity and complexity of historical data. Therefore, traditional methods often lack flexible adaptability when facing different project decommissioning scenarios, and the prediction results are easily limited by relatively simple assumptions. SUMMARY

[0005] The main purpose of the present application is to provide a new energy project decommissioning cost prediction method and system based on a machine model to solve the problem of single prediction results in the prior art.

[0006] To achieve the above purpose, the present application provides the following technical solutions: The new energy project decommissioning cost prediction method based on a machine model comprises: Obtaining full life cycle basic data and multi-scenario decommissioning related historical data of a new energy project, wherein the full life cycle basic data includes project construction parameters, equipment operation records, and environmental adaptation information, and the multi-scenario decommissioning related historical data includes resource consumption records, compliance processing overhead, and residual value recovery data in the decommissioning process of the same type of project; Based on the full life cycle basic data and multi-scenario decommissioning related historical data, extracting static features and dynamic features related to decommissioning costs, and constructing a multi-dimensional feature set, wherein the static features are associated with inherent properties of the project, and the dynamic features are associated with state changes in the project operation process and external environmental influences; Dividing the multi-dimensional feature set into a training data set and a validation data set, constructing an initial machine prediction model based on the training data set, and optimizing the parameters of the initial machine prediction model through iterative training; inputting the verification data set into the trained initial machine prediction model to obtain an initial decommissioning cost prediction result, performing uncertainty analysis on the prediction result, and identifying key influence factors causing prediction deviation; Based on the uncertainty analysis result and the key influence factors, adjusting the feature weight in the multi-dimensional feature set, and re-inputting into the trained initial machine prediction model for secondary training and optimization to obtain a new energy project decommissioning cost prediction result.

[0007] Preferably, based on the full life cycle basic data and multi-scenario decommissioning related historical data, static features and dynamic features related to decommissioning cost are extracted, and a multi-dimensional feature set is constructed, including: From the full life cycle basic data and multi-scenario decommissioning related historical data, the basic data items directly related to decommissioning cost are screened out, the attribute types corresponding to each data item are determined, and static attributes and dynamic attributes are distinguished; For static attributes, static features representing inherent characteristics of the project are extracted, wherein the static features include device type adaptability, site geographical features, and initial construction standards; For dynamic attributes, the change law of data items over time is analyzed, and dynamic features representing state changes are extracted, wherein the dynamic features include equipment performance degradation trend, policy compliance requirement iteration, and market residual value fluctuation; The extracted static features and dynamic features are subjected to correlation analysis, redundant features are removed, and feature items with significant impact on decommissioning cost are retained to obtain a multi-dimensional feature set.

[0008] Preferably, the extracted static features and dynamic features are subjected to correlation analysis, redundant features are removed, and feature items with significant impact on decommissioning cost are retained to obtain a multi-dimensional feature set, including: The correlation coefficients between each static feature and dynamic feature are calculated to determine the correlation strength between the features, and the feature groups that may have information overlap are marked; Based on the correlation strength, the feature importance of the feature groups is evaluated, and the feature item with the highest explanation degree for decommissioning cost in each group is selected; The selected feature items are integrated to obtain a multi-dimensional feature set.

[0009] Preferably, the multi-dimensional feature set is divided into a training data set and a verification data set, an initial machine prediction model is constructed based on the training data set, and the parameters of the initial machine prediction model are optimized through iterative training, including: Based on the distribution characteristics of each feature item in the multi-dimensional feature set, the multi-dimensional feature set is divided into a training data set and a verification data set by hierarchical division; The integrated machine prediction model is constructed by integrating multiple basic prediction models, each of which is trained for different types of feature items. The training data set is input into the integrated machine prediction model, and the prediction value is calculated by forward propagation. The prediction error is calculated by comparing the prediction value with the actual retirement cost value. Based on the prediction error, the parameters of each basic prediction model and the fusion weight between models are adjusted, and the training process is iteratively performed until the prediction error is stable in the preset range, and the trained initial machine prediction model is obtained.

[0010] Preferably, the integrated machine prediction model is constructed by integrating multiple basic prediction models, each of which is trained for different types of feature items, comprising: Based on the difference between static features and dynamic features in the multi-dimensional feature set, a basic regression model suitable for static features and a time series prediction model suitable for dynamic features are selected as basic prediction models; A feature interaction mechanism is established between the basic prediction models, and the prediction results corresponding to the static features and the prediction results corresponding to the dynamic features are fused based on the correlation of the retirement link; An initial fusion weight is set, which is determined based on the prediction accuracy of each basic prediction model in the historical similar data set, and an integrated machine prediction model is obtained; The input and output interfaces of the integrated machine prediction model are standardized, and the retirement cost prediction value is output.

[0011] Preferably, based on the prediction error, the parameters of each basic prediction model and the fusion weight between models are adjusted, and the training process is iteratively performed until the prediction error is stable in the preset range, and the trained initial machine prediction model is obtained, comprising: The prediction error after each iteration training is calculated, and the prediction accuracy of the basic prediction model is evaluated; When the prediction accuracy does not meet the preset requirement, based on the error distribution, the basic prediction model that contributes most to the error is identified, and the parameters of the basic prediction model are preferentially adjusted; After adjusting the parameters, the training is re-executed, and the prediction accuracy changes before and after adjustment are compared. If the accuracy is improved, the adjusted parameters are retained, and if the accuracy is decreased, the parameters before adjustment are rolled back; The parameter adjustment and training process are repeated, and the fusion weight between each basic prediction model is dynamically optimized until the prediction error is stable in the preset range, and the training of the initial machine prediction model is completed.

[0012] Preferably, the verification data set is input into the trained initial machine prediction model to obtain an initial decommissioning cost prediction result, an uncertainty analysis is performed on the prediction result, and key influence factors causing prediction deviation are identified, including: The verification data set is input into the trained initial machine prediction model in the order of the decommissioning links to obtain initial decommissioning cost prediction values, and overall initial decommissioning cost prediction results are obtained by summarizing; Based on the actual decommissioning cost data in the verification data set, a deviation distribution of the initial prediction result and the actual result is calculated; The values of each feature item in the multi-dimensional feature set are adjusted to determine the change amplitude of the initial prediction result, and a feature item with a change amplitude exceeding a preset threshold is identified as a key influence factor; Based on the deviation distribution and the key influence factor, an uncertainty analysis report is obtained to determine the deviation source and improvement direction of the initial prediction model.

[0013] Preferably, the values of each feature item in the multi-dimensional feature set are adjusted to determine the change amplitude of the initial prediction result, and a feature item with a change amplitude exceeding a preset threshold is identified as a key influence factor, including: For each feature item in the multi-dimensional feature set, a plurality of adjustment gradients are set within a reasonable value range, and the value of the feature item is changed in turn according to the gradient; After changing the value of each feature item, the values of other feature items remain unchanged, and the initial machine prediction model is input to record the corresponding prediction result change; The ratio of the prediction result change corresponding to each feature item to the feature item adjustment amount is calculated as the sensitivity coefficient of the feature item; The sensitivity coefficient is compared with a preset threshold to screen out feature items with a sensitivity coefficient exceeding the threshold to obtain key influence factors.

[0014] Preferably, based on the uncertainty analysis result and the key influence factor, the weights of the features in the multi-dimensional feature set are adjusted, and the initial machine prediction model after training is re-input for secondary training and optimization to obtain a new energy project decommissioning cost prediction result, including: The weight of the key influence factor in the multi-dimensional feature set is increased, and the weight of the non-key feature item is reduced to obtain an adjusted feature set; The adjusted feature set is re-divided into a new training data set and a verification data set, and the initial machine prediction model is re-trained; The new verification data set is input into the initial machine prediction model after secondary training to obtain an optimized decommissioning cost prediction result, and the prediction error after optimization is calculated; If the optimized prediction error meets the preset accuracy requirement, the prediction result is output as the new energy project decommissioning cost prediction result; if not, the feature weight adjustment and model training process are repeated until the prediction error meets the requirement.

[0015] To achieve the above object, the application further provides the following technical solutions: a new energy project decommissioning cost prediction system based on a machine model, which is suitable for the new energy project decommissioning cost prediction method based on a machine model, comprising: a data acquisition unit configured to acquire full life cycle basic data and multi-scenario decommissioning related historical data of a new energy project, wherein the full life cycle basic data comprises project construction parameters, equipment operation records and environmental adaptation information, and the multi-scenario decommissioning related historical data comprises resource consumption records, compliance processing overhead and residual value recovery data in the decommissioning process of a same type project; a feature extraction unit configured to extract static features and dynamic features related to decommissioning cost based on the full life cycle basic data and the multi-scenario decommissioning related historical data, and construct a multi-dimensional feature set, wherein the static features are associated with inherent properties of the project, and the dynamic features are associated with state changes in the project operation process and external environmental influences; a model construction unit configured to divide the multi-dimensional feature set into a training data set and a verification data set, construct an initial machine prediction model based on the training data set, and optimize parameters of the initial machine prediction model through iterative training; a result analysis unit configured to input the verification data set into the trained initial machine prediction model, obtain an initial decommissioning cost prediction result, perform uncertainty analysis on the prediction result, and identify key influence factors causing prediction deviation; a cost prediction unit configured to adjust feature weights in the multi-dimensional feature set based on the uncertainty analysis result and the key influence factors, re-input the multi-dimensional feature set into the trained initial machine prediction model for secondary training and optimization, and obtain a new energy project decommissioning cost prediction result.

[0016] (1) The application can analyze a large amount of real historical data based on the extraction of static features and dynamic features from the full life cycle basic data and the multi-scenario decommissioning related historical data, reduce the deviation caused by subjective assumptions, improve the accuracy of decommissioning cost prediction, identify and quantify the source of prediction deviation through uncertainty analysis on the prediction result, help analyze which factors have the greatest impact on decommissioning cost, and thus optimize the parameters and prediction accuracy of the model; (2) By separating the training dataset and the validation dataset and performing iterative optimization, this application can continuously improve the performance of the initial machine prediction model, which helps to eliminate overfitting or underfitting problems in the model and make it more generalizable; and by analyzing the prediction bias, the key factors affecting the prediction of decommissioning costs can be identified. Adjusting the feature weights according to these key factors can further optimize the model and improve the reliability of the prediction. (3) By predicting decommissioning costs and their uncertainties, this application can identify and assess potential risks and cost fluctuations in advance, providing decision-makers with a basis for early prediction and response, and helping project managers to do a good job in risk prevention and control and fund arrangement. By building the model on historical decommissioning data in multiple scenarios, it can adapt to the characteristics of different projects, especially for new energy projects with different decommissioning scenarios, and can provide flexible and targeted decommissioning cost prediction. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of an embodiment of the machine model-based method for predicting the decommissioning costs of new energy projects according to this application. Figure 2 This is a schematic diagram of the system architecture of an embodiment of the machine model-based new energy project decommissioning cost prediction system of this application.

[0018] Figure labels: 1. Data acquisition unit; 2. Feature extraction unit; 3. Model building unit; 4. Result analysis unit; 5. Cost prediction unit. Detailed Implementation

[0019] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0020] like Figure 1 As shown, the new energy project decommissioning cost prediction method based on machine model proposed in this invention includes: S1. Obtain basic data of the entire life cycle of new energy projects and historical data related to retirement in multiple scenarios. The basic data of the entire life cycle includes project construction parameters, equipment operation records and environmental adaptation information. The historical data related to retirement in multiple scenarios includes resource consumption records, compliance processing expenses and residual value recovery data during the retirement process of similar projects. S2. Based on the basic data of the entire life cycle and historical data related to decommissioning in multiple scenarios, extract static and dynamic features related to decommissioning costs and construct a multi-dimensional feature set. Among them, static features are associated with the inherent attributes of the project, and dynamic features are associated with the state changes and external environmental influences during the operation of the project. S3, divide the multi-dimensional feature set into a training data set and a validation data set, construct an initial machine prediction model based on the training data set, and optimize the parameters of the initial machine prediction model through iterative training; S4, input the validation data set into the trained initial machine prediction model to obtain an initial decommissioning cost prediction result, perform uncertainty analysis on the prediction result, and identify key influence factors causing prediction deviation; S5, based on the uncertainty analysis result and the key influence factors, adjust the feature weights in the multi-dimensional feature set, and input them into the trained initial machine prediction model for secondary training and optimization to obtain a new energy project decommissioning cost prediction result.

[0021] In the present application, the full life cycle basic data includes initial parameters of project construction, operation records of equipment, and adaptability information of the environment where the project is located; for example, design capacity of the project, service life of the equipment, environmental conditions (such as temperature, humidity, wind force, etc.) and the like; multi-scenario decommissioning related historical data includes data collected in the decommissioning process of similar past projects, which includes: resource consumption records: materials, energy, labor and the like consumed in the decommissioning process; compliance processing overhead: legal and environmental compliance requirements that need to be met during decommissioning and the cost of processing these requirements; residual value recovery data: economic returns obtained through reuse, recycling or disassembly of equipment after decommissioning; static features: related to inherent properties of the project itself, usually factors determined in the project start-up stage; for example, design capacity of the project, geographical location, equipment type and the like; dynamic features: related to state changes occurring during project operation, equipment aging or failure, and influence of external environmental factors; for example, failure frequency during equipment operation, operating state, energy consumption change, influence of climate change on the project and the like; integrate static features and dynamic features together to form a multi-dimensional feature set; these feature sets will be used to train a machine learning model to predict the decommissioning cost of a new energy project; The training data set refers to the data set used to train the machine learning model, which will learn how to predict the retirement cost on this data set; the validation data set refers to the data set used to validate the prediction ability of the model; the data of the validation data set has not been seen by the model, and is used to test the generalization ability of the model; based on the training data set, a preliminary prediction model is constructed using machine learning algorithms such as regression analysis, random forest, support vector machine, etc.; the goal of the model is to predict the retirement cost by learning the characteristics of the input (such as device operating status, resource consumption, etc.); by continuously optimizing the parameters of the model (such as learning rate, regularization term, etc.), the prediction accuracy of the model is improved; the optimization process usually uses some algorithms (such as gradient descent, Bayesian optimization, etc.) to find the most suitable parameters; after preliminary training, the validation data set is input into the trained model to obtain the prediction result; at this time, uncertainty analysis needs to be performed on the prediction result to analyze the fluctuation range of the model's prediction result under different conditions; Uncertainty analysis is used to evaluate the stability and reliability of the model's prediction results; there may be multiple factors that cause prediction bias, such as incomplete feature selection, non-representative training data, etc.; through uncertainty analysis, it can be identified which factors have the greatest impact on the prediction result; according to the results of uncertainty analysis and the identified key influencing factors, adjust the feature weights; this means re-evaluating the importance of each feature and adjusting their roles in the model accordingly; then, input the adjusted features into the trained model for secondary training and optimization to obtain more accurate retirement cost prediction results; after secondary training and optimization, the model can output the final retirement cost prediction results; this prediction result takes into account more external factors and model adjustments, and has higher accuracy.

[0022] In an optional embodiment, based on the whole life cycle basic data and multi-scenario retirement related historical data, static features and dynamic features related to retirement cost are extracted, and a multi-dimensional feature set is constructed, including: From the whole life cycle basic data and multi-scenario retirement related historical data, the basic data items directly related to the retirement cost are filtered, the attribute types corresponding to each data item are determined, and static attributes and dynamic attributes are distinguished; For static attributes, static features representing inherent characteristics of the project are extracted, wherein the static features include device type adaptability, site geographical characteristics, and initial construction standards; For dynamic attributes, the change law of the data items over time is analyzed, and dynamic features representing state changes are extracted, wherein the dynamic features include device performance degradation trend, policy compliance requirement iteration, and market scrap value fluctuation; The extracted static features and dynamic features are subjected to correlation analysis, redundant features are eliminated, and feature items with significant impact on retirement cost are retained to obtain a multi-dimensional feature set.

[0023] Note that the basic data items are the data of each record in the project, such as the operation record of the equipment, construction parameters, decommissioning history, etc.; it is necessary to filter out the data items closely related to the decommissioning cost from these data; for example, the type of equipment, the running state, the resource recycling situation at the time of decommissioning, the overhead of compliance processing, etc., are all important factors affecting the decommissioning cost; the attribute type means that each data item may correspond to different types of attributes, for example: numerical attributes: such as equipment running time, resource consumption, etc.; category attributes: such as equipment type (wind turbine, solar panel, etc.), site type (mountainous area, plain, etc.); time attributes: such as project construction and decommissioning time; Static attributes refer to attributes that remain unchanged during the project life cycle, which are usually related to the inherent characteristics of the project; dynamic attributes: change over time and reflect changes that occur during the operation of the project, usually involving the use state of the equipment, changes in the external environment, etc.; equipment type adaptability refers to the evaluation of the specific technical requirements or special disassembly costs required during the decommissioning process of the equipment according to its type (such as wind power, photovoltaic, etc.); site geographical features refer to whether the equipment is located in an extreme climate area or a place with poor transportation, which will increase the transportation cost when the equipment is decommissioned; initial construction standards refer to the design standards at the initial stage of the project, which affect the maintenance cost of the equipment and the difficulty of resource recycling when decommissioning; if high-standard equipment is used during construction, the resource value when decommissioning may be higher; equipment performance degradation trend refers to analyzing the rate of performance decline through the operation record of the equipment; for example, the frequency of equipment failure, efficiency loss, etc., can be used to predict the number of components and materials that need to be handled when decommissioning; policy compliance requirement iteration refers to changes in policies and regulations over time, which may require early decommissioning or replacement of equipment; the impact of these regulatory changes on decommissioning costs needs to be analyzed, such as environmental protection requirements, new safety standards, etc.; market residual value fluctuation refers to the fact that the residual value of the same type of equipment on the market will fluctuate over time, and the actual value of the equipment when decommissioning needs to be considered; this may be affected by market demand, technological progress, or waste recycling policies; Correlation analysis is performed on all extracted features to see the degree of correlation between each feature; common methods include correlation coefficient analysis, principal component analysis (PCA), etc.; if two features are highly correlated, the information contained in the two features is almost the same; in order to reduce the complexity of the model and avoid overfitting, redundant features need to be removed; for example, the equipment performance degradation trend and the equipment service life may be highly correlated, so only one needs to be retained; after removing redundant features, those features that have a significant correlation with the decommissioning cost are retained; for example, some features may have a greater impact on the prediction of decommissioning cost in different projects, and need to be retained.

[0024] In an optional embodiment, the extracted static features and dynamic features are subjected to correlation analysis, redundant features are removed, and features that significantly affect the decommissioning cost are retained to obtain a multi-dimensional feature set, including: The correlation coefficient between each static feature and dynamic feature is calculated to determine the correlation strength between the features, and the feature groups that may have overlapping information are marked; Based on the correlation strength, the feature importance of the feature groups is evaluated, and the feature item with the highest explanatory degree for the decommissioning cost in each group is selected; The selected feature items are integrated to obtain a multi-dimensional feature set.

[0025] It should be noted that the correlation coefficient (usually Pearson correlation coefficient or Spearman rank correlation coefficient) is used to measure the strength of the linear relationship between two variables; its value ranges from -1 to 1, which means: 1 indicates a complete positive correlation, i.e. the increase of one variable is accompanied by the increase of another variable; -1 indicates a complete negative correlation, i.e. the increase of one variable is accompanied by the decrease of another variable; 0 indicates no linear correlation; The purpose of calculating the correlation coefficient between static features and dynamic features is to understand whether there is a certain relationship between them; for example, the type of equipment may be related to the performance degradation trend of the equipment, or the initial construction standard may be related to the fluctuation of the equipment market residual value; According to the calculated correlation coefficient, the relationship strength between each pair of features is determined; for example, if the correlation coefficient of two features is very close to 1 or -1, the correlation between them is very strong, and they may contain redundant information; If the correlation coefficients between a group of features are very high, it means that the information provided by these features may be overlapping, and there is a lot of redundant information; For these feature groups, further screening is usually required to retain the most representative or most influential features on the decommissioning cost; for example, the performance degradation trend of the equipment and the service life of the equipment may be highly correlated, and in some cases only one feature needs to be retained; Mark these feature groups with overlapping information, the main purpose is to reduce the redundancy of data and avoid affecting the accuracy of the model; After finding out which feature groups may have redundancy through correlation analysis, the importance of these features in decommissioning cost prediction needs to be evaluated next; this can be achieved in the following ways: statistical methods: such as analysis of variance (ANOVA), regression analysis, etc., to check the explanatory degree of each feature on the decommissioning cost; Machine learning models: use some methods such as decision tree, random forest, XGBoost, etc. to evaluate the importance of features; Many machine learning algorithms can automatically evaluate the importance of each feature to the prediction model and give a weight or score; The goal is to select the feature item with the highest explanatory degree for the decommissioning cost in each feature group through this evaluation; For example, the type of equipment may affect the initial construction standard, maintenance cost, etc. of the equipment, but they may exhibit different characteristics according to the use environment, and the specific importance needs to be determined according to the data; After the feature importance evaluation, the features that have the greatest impact on the retirement cost in each group are retained; for example, if the device type and device performance attenuation trend have a greater impact on the prediction of the retirement cost in some scenarios, while the site geographical features have a smaller impact on the retirement cost, the former can be selected to be retained; the core of this step is to select the most effective features by comparing the degree of explanation of the features to the target variable (retirement cost), to avoid including too much redundant information in the feature set; finally, after filtering out the most useful features, these features are integrated to form a multi-dimensional feature set; this feature set is the result of de-redundancy and importance filtering, which can best reflect the influencing factors of the retirement cost and retain the most representative features.

[0026] In an optional embodiment, the multi-dimensional feature set is divided into a training data set and a validation data set, an initial machine prediction model is constructed based on the training data set, and the parameters of the initial machine prediction model are optimized through iterative training, including: Based on the distribution characteristics of each feature item in the multi-dimensional feature set, the multi-dimensional feature set is divided into a training data set and a validation data set in a hierarchical division manner, to ensure that the feature distributions of the two types of data sets are consistent; An integrated machine prediction model is constructed, which integrates multiple basic prediction models, each of which is trained for different types of feature items; The training data set is input into the integrated machine prediction model, and the predicted value is calculated through forward propagation. The predicted value is compared with the actual retirement cost value, and the prediction error is calculated; The formula for calculating the retirement cost prediction value is as follows: ; Wherein, represents the retirement cost prediction value, represents the static model fusion weight, represents the static feature cost prediction value, represents the dynamic model fusion weight, represents the dynamic feature cost prediction value, represents the retirement residual value recovery amount, represents the model random error term; Based on the prediction error, the parameters of each basic prediction model and the fusion weight between models are adjusted, and the training process is iteratively executed until the prediction error is stable within a preset range, to obtain the trained initial machine prediction model.

[0027] It should be noted that in order to ensure that the training data set and the validation data set can fairly represent the overall data distribution, avoid overfitting or underfitting of the model due to uneven division, stratified sampling method is usually used; stratified sampling refers to stratifying data according to certain feature categories, and then extracting samples from each layer in proportion, which ensures that the feature distribution in the training data set and the validation data set is consistent; for example, if there are three categories A, B and C in the feature "device type", stratified division ensures that the proportion of the three categories in the training and validation data sets is similar; by this method of division, the training and validation data sets can better reflect the data characteristics in the actual scene, so that the trained model has better generalization ability when facing unknown data; ensemble learning is a technology that combines the prediction results of multiple base models to improve the accuracy of the final prediction; common ensemble learning methods include Bagging, Boosting, Stacking, etc. In this scenario, the integrated machine prediction model combines multiple base prediction models, each of which is trained for different types of feature items; each base model focuses on learning different features and may use different algorithms and methods to capture the complex relationships between features; for example: some base models may use linear regression to learn the impact of initial construction standards on the retirement cost of devices; other base models may use tree models (such as decision trees, random forests) to learn the non-linear relationship between the service life of devices and the retirement cost; there may also be base models using neural networks to learn the impact of complex dynamic features (such as device performance degradation trend) on the retirement cost; in this way, the integrated model can take advantage of different models to improve overall prediction performance; The training data set is input into the integrated model, and prediction is performed through the model; forward propagation refers to inputting data into the model, performing calculations at each layer, and finally obtaining the prediction result; for an integrated model, all base models participate in prediction, and the final output is the fusion of multiple model prediction results; the prediction error is the difference between the actual retirement cost and the model prediction value, which is usually measured by mean square error (MSE) or mean absolute error (MAE); the purpose of calculating the error is to evaluate the performance of the model on the current data set; if the error is too large, it means that the difference between the model prediction result and the actual value is large, and the model needs to be further adjusted; each base model usually has some adjustable hyperparameters, such as learning rate, tree depth, etc.; according to the calculated prediction error, the model adjusts these hyperparameters to better fit the data and reduce the error; the integrated model usually obtains the final prediction value by weighted averaging the prediction results of multiple base models; therefore, the model adjusts the weights of each base model according to their performance; if some base models perform well in learning certain features, their prediction results will be given higher weights; on the contrary, if a model performs poorly, its weight will be reduced; this process is achieved through gradient descent or similar optimization algorithms, with the goal of minimizing prediction error; The model will continue to be trained and optimized; in each iteration, the parameters and weights of the base models are adjusted, and then the prediction error is recalculated; this process continues until the prediction error converges to a preset range; this preset range is usually set according to actual needs, such as the error cannot exceed a certain threshold; when the prediction error reaches the set stopping criterion, the training process stops; at this time, the model has converged, meaning that the model's prediction ability has reached a stable level and there is no significant error fluctuation; through the above training and optimization process, an integrated machine prediction model is finally obtained; this model has learned the features and target (retirement cost) in the training data sufficiently and has high prediction accuracy, which can be used for actual prediction tasks.

[0028] In an optional embodiment, an integrated machine prediction model is constructed, which integrates multiple base prediction models, each of which is trained for different types of feature items, including: Based on the difference between static features and dynamic features in the multi-dimensional feature set, a base regression model suitable for static features and a time series prediction model suitable for dynamic features are selected as base prediction models; A feature interaction mechanism is established between the base prediction models, and the prediction results corresponding to the static features and the prediction results corresponding to the dynamic features are fused based on the correlation of the retirement link; An initial fusion weight is set, and the initial fusion weight is determined based on the prediction accuracy of each base prediction model in a historical similar data set, to obtain an integrated machine prediction model; The input interface and the output interface of the integrated machine prediction model are standardized, to ensure that the training data set and the verification data set can adapt to the model input, and an output of a retirement cost prediction value is obtained.

[0029] It should be noted that the static features use a regression model, and the dynamic features use a time series model, and the two respectively process different types of features, so as to more accurately predict the retirement cost; the feature interaction mechanism refers to the fact that the prediction results of the static feature model and the dynamic feature model in the integrated model are not simply juxtaposed, but are influenced and fused by a certain way; by designing a mechanism, the relationship between the static features and the dynamic features can be reflected in the model; each link in the retirement process may not only be independent, but also have certain correlation; for example, the service life of the equipment (dynamic feature) may affect the initial construction standard of the equipment (static feature); therefore, the static features and the dynamic features are not completely independent, and the prediction results of the two need to be combined in order to better predict the retirement cost; In this stage, the prediction results of two different base prediction models (regression model and time series model) will be combined; weighted average, model cascade or other methods can be used to fuse the outputs of the two models according to their prediction accuracy and contribution, to form the final retirement cost prediction; in the integrated machine prediction model, the results of multiple base prediction models need to be fused into a final prediction result in a certain way; this fusion process usually assigns a weight to each base model, and the weight value determines the influence of the model in the final prediction; in order to reasonably set the initial fusion weight, it is necessary to evaluate the performance of each base prediction model in the past tasks through a historical similar data set; the prediction accuracy (such as mean square error, R 2 index) of each base model can be used to determine its initial weight; the better the model performs, the higher the weight it is given, and the poorer the model performs, the lower the weight it is assigned; Finally, based on these initial fusion weights, the prediction results of the regression model and the time series model are fused to form an integrated machine prediction model that can consider both static and dynamic features to accurately predict the decommissioning cost. The features of the input data (static and dynamic features) can come from different sources and their dimensions and ranges may not be consistent. Standardization is a preprocessing step that scales all features to the same scale (e.g., by normalization or Z-score standardization), which can avoid the unfair influence of certain features due to their large or small scale on model training. The output, the predicted value of the decommissioning cost, may have different units or ranges. In the integrated model, it is necessary to ensure that the output of the decommissioning cost prediction value meets the expected standards and formats, for example, the decommissioning cost may be output in numerical form or within a certain range. During model training, input data (features) and output data (target variables) need to be standardized consistently to ensure that the training set and validation set are completely consistent in input and output formats. In this way, the trained model can be successfully applied to the validation set to ensure that the model can accurately predict unknown data.

[0030] In an optional embodiment, based on the prediction error, the parameters of each base prediction model and the fusion weights between models are adjusted, and the training process is iteratively performed until the prediction error stabilizes in the preset range, obtaining the trained initial machine prediction model, including: Calculate the prediction error after each iteration of training to evaluate the prediction accuracy of the base prediction model. When the prediction accuracy does not meet the preset requirements, based on the error distribution, identify the base prediction model that contributes most to the error and prioritize adjusting the parameters of the base prediction model. After adjusting the parameters, re-execute the training and compare the prediction accuracy before and after adjustment. If the accuracy improves, keep the adjusted parameters. If the accuracy decreases, revert to the parameters before adjustment. Repeat the parameter adjustment and training process while dynamically optimizing the fusion weights between the base prediction models until the prediction error stabilizes in the preset range, completing the training of the initial machine prediction model.

[0031] It should be noted that after each training iteration is completed, the prediction effect of the model needs to be evaluated; this is usually achieved by calculating the prediction error, which can be the difference between the model's predicted value and the actual value; through these errors, the prediction accuracy of the model can be obtained, helping to judge the performance of the model; if the prediction error of the model is too large, it means that the model has problems in processing the current data, and may need to be adjusted or optimized; if the prediction accuracy after a certain training does not meet the predetermined standard, improvement is needed; first, by analyzing the distribution of errors, it can be found out which basic prediction models contribute most to the error; for example, the predicted values of some basic models may deviate from the actual values far, thereby affecting the overall prediction accuracy; once the problem model is identified, the parameters of these models are adjusted first; adjusting the parameters of the model (such as learning rate, regularization parameter, etc.) can improve its fitting ability to data, thereby reducing errors; After adjusting the parameters, the training process needs to be re-executed, and the prediction error after training needs to be calculated; the goal of this step is to compare the changes in prediction accuracy before and after adjustment; if the accuracy after adjustment is higher, it means that the adjustment is effective, and these parameters can be retained; if the accuracy after adjustment is worse, it means that the adjustment is ineffective or counterproductive, and the model needs to be restored to its pre-adjustment state; this way ensures that the model's performance will not be degraded due to parameter adjustment; during optimization, parameter adjustment and the training process will be repeated constantly; this is because the model may need to be fine-tuned multiple times to reach its optimal state; after each adjustment, the prediction accuracy is checked to ensure that the model is being optimized in the right direction; in addition to adjusting the parameters of the basic prediction models, the fusion weights between the basic models are also dynamically optimized; the fusion weights between models determine the contribution of each basic model to the final prediction result; by optimizing these weights, the model can better integrate the strengths of each basic model and improve overall prediction accuracy; The training process will continue until the prediction error stabilizes within a pre-set range, indicating that the model has reached sufficient accuracy; the purpose of this process is to continuously adjust and optimize to ensure that the model can make accurate predictions on unknown data; once the prediction error stabilizes, the model has completed the initial training and can begin to be used for actual decommissioning cost prediction tasks.

[0032] In an optional embodiment, the validation data set is input into the trained initial machine prediction model to obtain an initial decommissioning cost prediction result, and uncertainty analysis is performed on the prediction result to identify key influence factors that cause prediction bias, including: The validation data set is input into the trained initial machine prediction model according to the order of the decommissioning link to obtain an initial decommissioning cost prediction value, and the overall initial decommissioning cost prediction result is obtained by summarizing; Based on the actual decommissioning cost data in the validation data set, the bias distribution of the initial prediction result and the actual result is calculated; Adjust the value of each feature item in the multi-dimensional feature set to determine the change range of the initial prediction result, and identify the feature item with a change range exceeding a preset threshold as a key influencing factor; Based on the bias distribution and the key influencing factor, an uncertainty analysis report is obtained to determine the bias source and improvement direction of the initial prediction model.

[0033] It should be noted that the validation dataset (usually a dataset not used for training) is input into the already trained initial prediction model in the order of the retirement stages; the retirement stages usually involve different stages of the process, and each stage may affect the change of the retirement cost; after inputting the data, the model will make predictions based on the rules learned previously to generate preliminary retirement cost prediction results; these results are the model's estimates of the actual situation and serve as the basis for subsequent improvements; the prediction results of each retirement stage are summarized to obtain the total cost prediction value of the entire retirement process; the purpose of this step is to integrate the prediction values of each stage to evaluate the overall retirement cost; next, the initial prediction results of the model are compared with the actual retirement costs in the validation dataset to calculate the differences or biases between them; the bias distribution reflects the size and distribution of the prediction errors of the model under different conditions; this analysis helps to understand the prediction errors of the model and determine under which conditions the model's predictions are inaccurate and have large biases; After obtaining the bias distribution, the feature items in the model (i.e., the variables in the input data) need to be adjusted; these feature items may include various factors that affect the retirement cost, such as equipment usage time, maintenance frequency, failure rate, etc.; by adjusting the values of these features, the change range of the model's prediction results is observed; feature items with a large change range are considered to be factors that have a significant impact on the prediction results, and are usually the most critical influencing factors; a threshold is set, and if the change range of a feature item exceeds this threshold, it is identified as a "key influencing factor"; these factors usually require special attention because they have a significant impact on the prediction results; Based on the bias distribution and the identified key influencing factors, an uncertainty analysis report is generated; the core of the report is to analyze the sources of prediction errors and identify factors that cause biases, and to provide directions for model improvement; this report reveals the weaknesses of the model under different conditions, such as the lack of certain features or the inaccuracy of certain assumptions; based on these findings, specific improvement suggestions can be made for the adjustment and optimization of the model, such as adding certain features, adjusting the model structure, or modifying the training process, etc.

[0034] In an optional embodiment, adjusting the value of each feature item in the multi-dimensional feature set to determine the change range of the initial prediction result, and identifying the feature item with a change range exceeding a preset threshold as a key influencing factor, comprises: For each feature item in the multi-dimensional feature set, a plurality of adjustment gradients are set within a reasonable value range, and the feature item value is sequentially changed according to the gradients; After each change of the feature item value, the other feature items remain unchanged, and are input into the initial machine prediction model to record the corresponding prediction result change; The ratio of the prediction result change corresponding to each feature item to the feature item adjustment is calculated as the sensitivity coefficient of the feature item; The sensitivity coefficient is compared with a preset threshold to screen out feature items with sensitivity coefficients exceeding the threshold to obtain key impact factors.

[0035] It should be noted that each feature item has a reasonable change range, for example, the equipment service life may be between 0 and 30 years, and the maintenance frequency may be between 0 and 10 times; within this range, several gradients, i.e. gradually changing value points, are set for each feature item; for example, the equipment service life can be gradually increased from 10 years to 15 years and then to 20 years; by gradually adjusting the feature value, the influence of the feature value on the model prediction result can be observed, rather than changing too much at one time to avoid distortion of the result; only one feature item is changed each time, and the other feature items remain unchanged, so that the observed result change is caused entirely by the feature; the new feature value is input into the model to obtain the prediction result, which is compared with the original prediction result to record the change; this process actually tests the response strength of the model to each feature, i.e. single-factor sensitivity test; For each feature item, the prediction result change is divided by the feature adjustment to obtain the sensitivity coefficient; the sensitivity coefficient measures the influence degree of unit feature change on the prediction result; the larger the value, the more significant the influence of the feature on the prediction result; the smaller the value, the less significant the influence of the feature on the result; a threshold is set, for example, "a sensitivity coefficient greater than a certain value is considered to have a significant influence"; the sensitivity coefficients of each feature are compared with the threshold, and the features exceeding the threshold are screened out and called key impact factors; these key impact factors are the main driving factors of the model prediction result and important targets for subsequent optimization or control.

[0036] In an optional embodiment, based on the uncertainty analysis result and the key impact factor, the feature weight in the multi-dimensional feature set is adjusted, and the initial machine prediction model after training is input for secondary training and optimization to obtain a new energy project decommissioning cost prediction result, including: The weight of the key impact factor in the multi-dimensional feature set is increased, and the weight of the non-key feature item is reduced to obtain an adjusted feature set; The calculation formula of the key feature weight adjustment is as follows: ; wherein, denotes the adjusted key feature weight, denotes the pre-adjusted key feature weight, denotes the key feature weight adjustment coefficient, , denotes the adjusted non-key feature weight, denotes the pre-adjusted non-key feature weight, denotes the non-key feature weight adjustment coefficient, ; re-divide the adjusted feature set into a new training data set and a validation data set, and perform secondary training on the initial machine prediction model; input the new validation data set into the secondary trained initial machine prediction model to obtain an optimized decommissioning cost prediction result, and calculate the optimized prediction error; wherein the calculation formula of the mean absolute error is as follows: ; wherein, denotes the mean absolute error, denotes the sample size of the validation data set, denotes the actual decommissioning cost of the i th sample, denotes the predicted decommissioning cost of the i th sample; If the optimized prediction error meets the preset accuracy requirement, output the prediction result as the new energy project decommissioning cost prediction result; if not, repeat the feature weight adjustment and model training process until the prediction error meets the requirement.

[0037] It should be noted that for key influencing factors, their weights are increased so that these features play a greater role in the model; and for non-key features that have less impact on the result, their weights are reduced so that their impact on the prediction result is reduced; through such adjustment, the model will focus more on the key factors affecting the decommissioning cost prediction, thereby improving the prediction accuracy; after the feature weight adjustment, the new feature set will become the basis for model training; in this step, the new feature set is re-divided into a training data set and a validation data set, usually the training data set is used to train the model, and the validation data set is used to test the performance of the model on unseen data; then, the initial machine prediction model is trained again using this new data set; the secondary training refers to further optimizing the model using the new data on the basis of the existing model, so that it can better adapt to the adjusted feature set; ​​After the secondary training is completed, the new verification data set is input into the optimized model to obtain a new prediction result, i.e., an optimized decommissioning cost prediction result; then, the error between the prediction result and the actual value is calculated to see how the prediction effect of the model is; the error is called the optimized prediction error; the smaller the prediction error, the more accurate the model, and vice versa, which means that the prediction accuracy of the model still needs to be improved; if the optimized prediction error meets the preset accuracy requirement, i.e., the error is small enough, the prediction result of the model is considered to be reliable; at this time, the optimized prediction result can be output as the decommissioning cost prediction result of the new energy project; but if the error is still large, it means that the prediction ability of the model is not good enough, so the feature weight needs to be adjusted or more model training is needed until the error meets the expected accuracy; this process is an iterative optimization process, and each adjustment will make the model more and more accurate until a satisfactory result is obtained.

[0038] As shown in Figure 2 The machine model-based new energy project decommissioning cost prediction system provided by the application is applicable to the machine model-based new energy project decommissioning cost prediction method described above and comprises: A data acquisition unit 1 is configured to acquire full life cycle basic data and multi-scenario decommissioning related historical data of a new energy project, wherein the full life cycle basic data comprises project construction parameters, equipment operation records and environmental adaptation information, and the multi-scenario decommissioning related historical data comprises resource consumption records, compliance processing overhead and residual value recovery data in the decommissioning process of a same type project; A feature extraction unit 2 is configured to extract static features and dynamic features related to decommissioning costs based on the full life cycle basic data and the multi-scenario decommissioning related historical data, and construct a multi-dimensional feature set, wherein the static features are associated with inherent properties of the project, and the dynamic features are associated with state changes in the operation process of the project and external environmental influences; A model construction unit 3 is configured to divide the multi-dimensional feature set into a training data set and a verification data set, construct an initial machine prediction model based on the training data set, and optimize parameters of the initial machine prediction model through iterative training; A result analysis unit 4 is configured to input the verification data set into the trained initial machine prediction model to obtain an initial decommissioning cost prediction result, perform uncertainty analysis on the prediction result, and identify key influence factors causing prediction deviation; A cost prediction unit 5 is configured to adjust feature weights in the multi-dimensional feature set based on the uncertainty analysis result and the key influence factors, re-input the multi-dimensional feature set into the trained initial machine prediction model for secondary training and optimization, and obtain a new energy project decommissioning cost prediction result.

[0039] The above detailed description of the application is merely exemplary in nature and, thus, does not limit the application. Any modification and the like that do not depart from the spirit and principles of the application are intended to be within the scope of the application.

Claims

1. A method for predicting the decommissioning cost of a new energy project based on a machine model, characterized in that, The method comprises the following steps: acquiring full-life-cycle basic data of a new energy project and multi-scenario retirement-related historical data, wherein the full-life-cycle basic data comprises project construction parameters, equipment operation records and environmental adaptation information, and the multi-scenario retirement-related historical data comprises resource consumption records, compliance processing overhead and residual value recovery data in the retirement process of a same type of project; based on the full-life-cycle basic data and the multi-scenario retirement-related historical data, extracting static features and dynamic features related to the retirement cost, and constructing a multi-dimensional feature set, wherein the static features are associated with inherent properties of the project, and the dynamic features are associated with state changes in the operation process of the project and external environmental influences; dividing the multi-dimensional feature set into a training data set and a verification data set, constructing an initial machine prediction model based on the training data set, and optimizing parameters of the initial machine prediction model through iterative training; inputting the verification data set into the trained initial machine prediction model to obtain an initial retirement cost prediction result, performing uncertainty analysis on the prediction result, and identifying key influence factors causing prediction deviation; based on the uncertainty analysis result and the key influence factors, adjusting the feature weights in the multi-dimensional feature set, and re-inputting the multi-dimensional feature set into the trained initial machine prediction model for secondary training and optimization to obtain a new energy project retirement cost prediction result.

2. The machine model-based decommissioning cost prediction method for new energy projects according to claim 1, characterized in that, Based on the full-life-cycle basic data and the multi-scenario retirement-related historical data, static features and dynamic features related to the retirement cost are extracted to construct a multi-dimensional feature set, including: filtering out basic data items directly related to the retirement cost from the full-life-cycle basic data and the multi-scenario retirement-related historical data, determining the attribute types corresponding to each data item, and distinguishing static attributes and dynamic attributes; for static attributes, static features representing inherent properties of the project are extracted, wherein the static features include device type adaptability, site geographical features and initial construction standards; for dynamic attributes, the change law of the data items over time is analyzed, and dynamic features representing state changes are extracted, wherein the dynamic features include equipment performance degradation trend, policy compliance requirement iteration and market residual value fluctuation; performing correlation analysis on the extracted static features and dynamic features, eliminating redundant features, and retaining feature items that have a significant impact on the retirement cost to obtain a multi-dimensional feature set.

3. The machine model-based decommissioning cost prediction method for new energy projects according to claim 2, characterized in that, Performing correlation analysis on the extracted static features and dynamic features, eliminating redundant features, and retaining feature items that have a significant impact on the retirement cost to obtain a multi-dimensional feature set, including: calculating the correlation coefficients between each static feature and dynamic feature to determine the correlation strength between the features, and marking feature groups that may have information overlap; based on the correlation strength, performing feature importance evaluation on the feature groups, and selecting the feature item with the highest explanation degree for the retirement cost in each group; integrating the selected feature items to obtain a multi-dimensional feature set.

4. The machine model-based decommissioning cost prediction method for new energy projects according to claim 3, characterized in that, The multi-dimensional feature set is divided into a training data set and a verification data set, an initial machine prediction model is constructed based on the training data set, and parameters of the initial machine prediction model are optimized through iterative training, including: Based on the distribution characteristics of each feature item in the multi-dimensional feature set, the multi-dimensional feature set is divided into a training data set and a validation data set by using a hierarchical division method; An integrated machine prediction model is constructed, which integrates multiple basic prediction models, each of which is trained for different types of feature items; The training data set is input into the integrated machine prediction model, and the predicted value is calculated by forward propagation. By comparing the predicted value with the actual retirement cost value, the prediction error is calculated; Based on the prediction error, the parameters of each basic prediction model and the fusion weight between models are adjusted, and the training process is iteratively executed until the prediction error stabilizes in the preset range, obtaining the trained initial machine prediction model.

5. The machine model-based decommissioning cost prediction method for new energy projects according to claim 4, characterized in that, An integrated machine prediction model is constructed, which integrates multiple basic prediction models, each of which is trained for different types of feature items, including: Based on the difference between static features and dynamic features in the multi-dimensional feature set, a basic regression model suitable for static features and a time series prediction model suitable for dynamic features are selected as basic prediction models; A feature interaction mechanism is established between the basic prediction models, and the prediction results corresponding to the static features and the prediction results corresponding to the dynamic features are fused based on the correlation of the retirement link; An initial fusion weight is set, which is determined based on the prediction accuracy of each basic prediction model in the historical similar data set, obtaining the integrated machine prediction model; The input and output interfaces of the integrated machine prediction model are standardized, and the retirement cost prediction value is output.

6. The machine model-based decommissioning cost prediction method for new energy projects according to claim 5, characterized in that, Based on the prediction error, the parameters of each basic prediction model and the fusion weight between models are adjusted, and the training process is iteratively executed until the prediction error stabilizes in the preset range, obtaining the trained initial machine prediction model, including: Calculate the prediction error after each iteration training, and evaluate the prediction accuracy of the basic prediction model; When the prediction accuracy does not meet the preset requirement, based on the error distribution, identify the basic prediction model that contributes most to the error, and preferentially adjust the parameters of the basic prediction model; After adjusting the parameters, re-execute the training, compare the prediction accuracy before and after adjusting, if the accuracy improves, keep the adjusted parameters, if the accuracy decreases, back to the parameters before adjusting; Repeat the parameter adjustment and training process, and dynamically optimize the fusion weight between each basic prediction model until the prediction error stabilizes in the preset range, completing the training of the initial machine prediction model.

7. The machine model-based decommissioning cost prediction method for new energy projects according to claim 6, characterized in that, The validation data set is input into the trained initial machine prediction model to obtain the initial retirement cost prediction result, and the key influence factors causing the prediction deviation are identified by uncertainty analysis, including: The validation data set is input into the trained initial machine prediction model according to the order of the retirement link to obtain the initial retirement cost prediction value, and the overall initial retirement cost prediction result is obtained by summarizing; Based on the actual retirement cost data in the validation data set, the deviation distribution of the initial prediction result and the actual result is calculated; Adjust the value of each feature item in the multi-dimensional feature set to determine the change range of the initial prediction result, and identify the feature item with a change range exceeding a preset threshold as a key influencing factor; Based on the bias distribution and the key influencing factor, an uncertainty analysis report is obtained to determine the bias source and improvement direction of the initial prediction model.

8. The machine model-based decommissioning cost prediction method for new energy projects according to claim 7, characterized in that, Adjusting the value of each feature item in the multi-dimensional feature set to determine the change range of the initial prediction result, and identifying the feature item with a change range exceeding a preset threshold as a key influencing factor, comprises: For each feature item in the multi-dimensional feature set, set multiple adjustment gradients within a reasonable value range, and change the feature item value in turn according to the gradient; After changing the value of each feature item, keep other feature items unchanged, input into the initial machine prediction model, and record the corresponding prediction result change; Calculate the ratio of the prediction result change corresponding to each feature item to the feature item adjustment amount as the sensitivity coefficient of the feature item; Compare the sensitivity coefficient with a preset threshold to filter out feature items with sensitivity coefficients exceeding the threshold to obtain the key influencing factor.

9. The machine model-based decommissioning cost prediction method for new energy projects according to claim 8, characterized in that, Based on the uncertainty analysis result and the key influencing factor, adjust the feature weight in the multi-dimensional feature set, and input it into the trained initial machine prediction model for secondary training and optimization to obtain the new energy project decommissioning cost prediction result, comprising: Increase the weight of the key influencing factor in the multi-dimensional feature set, and reduce the weight of non-key feature items to obtain an adjusted feature set; Divide the adjusted feature set into a new training data set and a validation data set, and perform secondary training on the initial machine prediction model; Input the new validation data set into the secondary trained initial machine prediction model to obtain the optimized decommissioning cost prediction result and calculate the optimized prediction error; If the optimized prediction error meets the preset accuracy requirement, output the prediction result as the new energy project decommissioning cost prediction result; if not, repeat the feature weight adjustment and model training process until the prediction error meets the requirement.

10. A machine model based decommissioning cost prediction system for new energy projects, adapted to the machine model based decommissioning cost prediction method of any one of claims 1-9, characterized in that, Comprise: A data acquisition unit is configured to acquire full life cycle basic data and multi-scenario decommissioning related historical data of a new energy project, wherein the full life cycle basic data includes project construction parameters, equipment operation records, and environmental adaptation information, and the multi-scenario decommissioning related historical data includes resource consumption records, compliance processing overhead, and residual value recovery data in the decommissioning process of the same type of project; A feature extraction unit is configured to extract static features and dynamic features related to decommissioning cost based on the full life cycle basic data and multi-scenario decommissioning related historical data, and construct a multi-dimensional feature set, wherein the static features are associated with inherent properties of the project, and the dynamic features are associated with state changes in the project operation process and external environmental influences; A model construction unit is configured to divide the multi-dimensional feature set into a training data set and a validation data set, construct an initial machine prediction model based on the training data set, and optimize the parameters of the initial machine prediction model through iterative training. The result analysis unit is configured to input the verification data set into the trained initial machine prediction model to obtain an initial decommissioning cost prediction result, perform uncertainty analysis on the prediction result, and identify key influence factors causing prediction deviation; The cost prediction unit is configured to adjust feature weights in the multi-dimensional feature set based on the uncertainty analysis result and the key influence factors, re-input the multi-dimensional feature set into the trained initial machine prediction model for secondary training and optimization, and obtain a new energy project decommissioning cost prediction result.

Images