Hoisting equipment carbon footprint model construction and optimization method based on machine learning

Abstract

The invention discloses a hoisting equipment carbon footprint model construction and optimization method based on machine learning, and the method comprises the steps: collecting historical multi-dimensional working condition time sequence sample data and real-time energy consumption sample data, and carrying out the training through a gradient lifting decision tree algorithm, thereby obtaining a dynamic carbon footprint intensity prediction model; a dynamic carbon emission factor corresponding to the current operation state is output in real time; a digital twinborn simulation environment is constructed based on a dynamic carbon footprint intensity prediction model, under a given operation constraint condition, an optimization search algorithm is adopted to generate a plurality of candidate operation strategies, carbon footprint forward simulation is carried out, and an optimal energy-saving operation strategy is selected by comparing total prediction carbon footprints. And finally outputting to an equipment man-machine interaction interface or a control system. According to the invention, the conversion of carbon footprint accounting from a static macroscopic factor to a dynamic equipment exclusive factor is realized, a closed loop from accurate perception to optimization decision is constructed, and the problems of low accounting precision and disjunction of monitoring and optimization caused by the use of a fixed carbon emission factor are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of energy conservation and emission reduction technology for industrial equipment, and in particular to a method for constructing and optimizing a carbon footprint model of lifting equipment based on machine learning. Background Technology

[0002] As key equipment in manufacturing and port logistics industries, the carbon emissions from the operation of lifting equipment cannot be ignored. Currently, carbon footprint assessments for such equipment mainly rely on calculation methods based on fixed carbon emission factors, such as using regional power grid average emission factors for energy consumption conversion. However, this method has significant limitations: it cannot reflect the dynamic changes in carbon emission intensity of the same equipment under specific operating conditions such as different load rates, operating speeds, and start-stop frequencies, leading to a large discrepancy between the calculation results and actual conditions, especially in complex and nonlinear industrial operating scenarios where its generalization ability is severely insufficient.

[0003] Furthermore, most existing carbon management systems focus on data monitoring and post-event statistics, forming an open-loop model that emphasizes monitoring but neglects optimization. Even when carbon footprint data is obtained, there is a lack of effective technical pathways to transform the data into real-time, actionable operational decisions, failing to achieve a closed loop from perception to control. Simultaneously, due to the variability of equipment operating states and the hidden nature of internal energy consumption, maintenance personnel struggle to quickly pinpoint the specific operational stages leading to high emissions, resulting in a lack of targeted energy-saving improvement measures and hindering further improvements in emission reduction. Therefore, there is an urgent need in this field for an intelligent management and control method that can achieve accurate carbon footprint accounting, dynamic adaptation, and directly drive emission reduction actions. Summary of the Invention

[0004] In view of this, the purpose of this invention is to propose a method for constructing and optimizing the carbon footprint model of lifting equipment based on machine learning. By constructing a dynamic carbon footprint intensity prediction model and optimizing the strategy based on digital twin simulation, this method solves the problem of inaccurate calculation and inability to achieve closed-loop optimization due to the use of fixed carbon emission factors.

[0005] To achieve the above-mentioned technical objectives, the technical solution adopted by this invention is: a method for constructing and optimizing a carbon footprint model of lifting equipment based on machine learning, comprising: Collect multi-dimensional operating condition time-series sample data and real-time energy consumption sample data of the lifting equipment during its historical operation; The gradient boosting decision tree algorithm is used, with multi-dimensional working condition time series sample data as features and instantaneous carbon emission rate calculated based on real-time energy consumption sample data and basic carbon emission factors as the supervision target, to train the model and obtain a dynamic carbon footprint intensity prediction model specific to lifting equipment. In addition, the real-time multi-dimensional operating condition time-series data of the lifting equipment is collected and input into the dynamic carbon footprint intensity prediction model, and the dynamic carbon emission factor corresponding to the current operating state is output. Obtain the operational constraints for the lifting task to be executed; A digital twin simulation environment is constructed based on a dynamic carbon footprint intensity prediction model, and an optimization search algorithm is used to generate multiple candidate operation strategies under operational constraints. In the digital twin simulation environment, carbon footprint forward simulation is performed on each generated candidate operation strategy, and the total predicted carbon footprint corresponding to each candidate operation strategy is calculated. By comparing the total predicted carbon footprint of all candidate operation strategies, the candidate operation strategy with the lowest total predicted carbon footprint is selected as the optimal energy-saving operation strategy. The optimal energy-saving operation strategy can be output to the equipment's human-machine interface, or the optimal energy-saving operation strategy can be output to the equipment control system.

[0006] In some embodiments, a gradient boosting decision tree algorithm is used, featuring multi-dimensional operating condition time-series sample data, with the instantaneous carbon emission rate calculated based on real-time energy consumption sample data and basic carbon emission factors as the supervision target, to train the model and obtain a dynamic carbon footprint intensity prediction model specific to lifting equipment, including: Data cleaning and feature construction processing are performed on multi-dimensional operating condition time series sample data to obtain a multi-dimensional feature vector including real-time load rate, operating speed time series, and acceleration statistical features. The acceleration statistical features include the acceleration standard deviation and acceleration extreme values ​​within the sliding time window. The multidimensional feature vector and the corresponding instantaneous carbon emission rate are combined to form a training sample set, which is then input into the gradient boosting decision tree algorithm for iterative training. The iterative training is configured to use the prediction residual generated in the previous iteration as the optimization objective. Multiple decision tree weak learners are constructed sequentially through an additive model. The goal of constructing each subsequent decision tree weak learner is to fit and correct the prediction residual of the previous decision tree weak learner. All decision tree weak learners built during training are linearly combined by weighted summation to form the final dynamic carbon footprint intensity prediction model. The weight of each decision tree weak learner is dynamically determined based on its training error. Feature importance analysis is performed based on the trained dynamic carbon footprint intensity prediction model, and the contribution ranking of each working condition parameter to carbon footprint intensity is output to obtain the feature importance analysis results. Based on the results of feature importance analysis, key impact features with a contribution rate higher than a preset threshold are selected to construct a simplified feature set; Establish a model performance monitoring and update mechanism. When the model prediction error of the dynamic carbon footprint intensity prediction model continues to exceed the preset error threshold, trigger the model retraining process based on newly added sample data, and use a simplified feature set to update the parameters and optimize the structure of the dynamic carbon footprint intensity prediction model.

[0007] In some embodiments, all decision tree weak learners built during training are linearly combined by weighted summation to form the final dynamic carbon footprint intensity prediction model, including: Calculate the initial weights based on the mean squared error of each decision tree weak learner on the validation dataset; The initial weights of each decision tree weak learner are dynamically adjusted based on their prediction accuracy on new samples to obtain dynamic weight coefficients. Among them, the weights of decision tree weak learners whose prediction accuracy continues to decline will be reduced. The prediction outputs of each decision tree weak learner are multiplied by their corresponding dynamic weight coefficients and then summed to obtain the final output of the dynamic carbon footprint intensity prediction model, resulting in a weighted summation prediction result. The weighted summation prediction results are compared and analyzed with the measured sample carbon footprint data, and an output calibration function is established to correct the prediction results. Furthermore, by setting constraints on the maximum number of decision trees and the number of leaf nodes, the overall complexity of the dynamic carbon footprint intensity prediction model is controlled. The final dynamic carbon footprint intensity prediction model is obtained.

[0008] In some embodiments, real-time multi-dimensional operating condition time-series data of the lifting equipment is collected and input into a dynamic carbon footprint intensity prediction model, outputting a dynamic carbon emission factor corresponding to the current operating state, including: The real-time acquired multi-dimensional operating condition time-series data is preprocessed to generate real-time feature vectors; The real-time feature vector is input into the trained dynamic carbon footprint intensity prediction model. Through forward propagation of the model, the predicted carbon footprint intensity value at the current time is output, including: The real-time feature vector is passed tree by tree in the decision tree set, and reaches the leaf node along the corresponding branch path in each decision tree according to the feature value; Obtain the output values ​​of the leaf nodes of each decision tree; The weighted sum of the output values ​​of all leaf nodes of the decision tree is used to obtain the predicted carbon footprint intensity at the current moment. The predicted carbon footprint intensity is used as a dynamic carbon emission factor and output.

[0009] In some embodiments, the method further includes: By combining the predicted carbon footprint intensity with real-time power data, the instantaneous carbon emission rate is obtained, which is expressed by formula (1), as follows: ; In formula (1), Instantaneous carbon emission rate, For real-time power, This is a predicted value for carbon footprint intensity. This is a time conversion factor; Based on the instantaneous carbon emission rate, the cumulative carbon emissions over a specified time period are calculated by time integration and expressed by formula (2), which is as follows: ; In formula (2), For cumulative carbon emissions, The starting point of a specified time period. The end point of a specified time period. This is the symbol for time integral.

[0010] In some embodiments, a digital twin simulation environment is constructed based on a dynamic carbon footprint intensity prediction model, and an optimization search algorithm is used to generate multiple candidate operation strategies under operational constraints, including: A digital twin of the crane operation is established and initialized based on a dynamic carbon footprint intensity prediction model. The digital twin can simulate the corresponding equipment operation status and carbon emission process according to the input operation strategy. The operational constraints are transformed into the constraint boundaries of the optimization problem. The constraint boundaries include the maximum allowable operation time, the safe operating speed range, and the maximum acceleration limit. The particle swarm optimization algorithm is used to search within the constraint boundary. By initializing the position and velocity of the particle swarm, the position vector of each particle represents a complete simulation operation strategy, which includes a simulation velocity control sequence and a simulation acceleration control sequence. In each iteration, candidate operation strategies are generated based on the current position of the particle. Each candidate operation strategy is input into the digital twin for forward simulation. The total predicted carbon footprint corresponding to each strategy is calculated through the dynamic carbon footprint intensity prediction model. The fitness of each particle is assessed based on the total predicted carbon footprint, the individual optimal position and the global optimal position are updated, and the search direction and step size of the particles are adjusted according to the velocity update formula. The iterative process is repeated until the termination condition is met, and the final output particle swarm positions constitute multiple candidate operation strategies.

[0011] In some embodiments, the velocity update formula is expressed by formula (3), which is as follows: ; In formula (3), For the first The particle in the first The velocity vector at the next iteration For the first The particle in the first The velocity vector at the next iteration This is the inertia weighting coefficient. For individual learning factors, As a social learning factor, For the first between Random numbers within, For the second one Random numbers within, For the first The particle in the first The optimal position of the individual in the next iteration. For the particle swarm in the th The global optimal position at the next iteration. For the first The particle in the first The current position at the next iteration.

[0012] In some embodiments, a particle swarm optimization algorithm is used to search within the constraint boundary. By initializing the positions and velocities of the particle swarm, the position vector of each particle represents a complete simulation operation strategy. The simulation operation strategy includes a simulation velocity control sequence and a simulation acceleration control sequence, including: Initialize the particle swarm and randomly generate the initial positions and velocities of multiple particles. The position vector of each particle consists of multiple dimensions, and each dimension corresponds to a velocity control value or acceleration control value for a time step. The position vector of the particle is constrained according to the constraint boundary to ensure that the values ​​of each dimension are within the allowable safe operating speed range and the maximum acceleration limit. The position vectors of each particle are decoded into a complete simulation operation strategy, which includes a simulated velocity control sequence and a simulated acceleration control sequence arranged in time sequence. Furthermore, the particle's velocity and position are updated based on the individual optimal position and the global optimal position during each iteration; When the particle swarm is detected to be trapped in a local optimum, the positions of some particles are randomly reset to expand the search range.

[0013] In some embodiments, within a digital twin simulation environment, a forward carbon footprint simulation is performed on each generated candidate operation strategy, and the total predicted carbon footprint corresponding to each candidate operation strategy is calculated, including: The candidate operation strategy is discretized into a sequence of operation instructions with multiple time steps. The sequence of operation instructions includes the device operating speed, acceleration and load status corresponding to each time step. The operating parameters corresponding to the operation instructions at each time step are input into the dynamic carbon footprint intensity prediction model to obtain the instantaneous carbon emission rate prediction value at each time step. Based on the instantaneous carbon emission rate prediction, the total predicted carbon footprint of the candidate operation strategy is calculated through time integration, including: The instantaneous carbon emission rate at each time step is multiplied by the corresponding time step and then summed.

[0014] In some embodiments, comparing the total predicted carbon footprint of all candidate operating strategies, the candidate operating strategy with the lowest total predicted carbon footprint is selected as the optimal energy-saving operating strategy, including: Establish a candidate strategy ranking queue and sort all candidate operation strategies in ascending order according to their corresponding total predicted carbon footprint values; The candidate operation strategy with the smallest total predicted carbon footprint value is extracted from the sorted queue as the preliminary optimal strategy. The feasibility of the preliminary optimal strategy is verified, including: Verify whether the operating parameters corresponding to the preliminary optimal strategy meet the equipment safety operation specifications and whether they exceed the equipment performance limits; When the preliminary optimal strategy passes the feasibility verification, it is determined as the final optimal energy-saving operation strategy; If the initial optimal strategy fails the feasibility verification, the next best candidate strategies are selected in order of the sorting queue for feasibility verification until the optimal energy-saving strategy that passes the verification is determined.

[0015] Compared with existing technologies, the present invention, employing the above technical solution, has the following advantages: By collecting historical multi-dimensional operating condition time-series sample data and real-time energy consumption sample data, a dynamic carbon footprint intensity prediction model is trained using a gradient boosting decision tree algorithm to output the dynamic carbon emission factor corresponding to the current operating state in real time; a digital twin simulation environment is constructed based on the dynamic carbon footprint intensity prediction model; under given operational constraints, an optimization search algorithm is used to generate multiple candidate operation strategies and perform forward carbon footprint simulation; the optimal energy-saving operation strategy is selected by comparing the total predicted carbon footprint, and finally output to the equipment human-machine interface or control system. This invention realizes the transformation of carbon footprint accounting from static macro factors to dynamic equipment-specific factors, and constructs a closed loop from accurate perception to optimized decision-making, effectively solving the problems of low accounting accuracy and disconnect between monitoring and optimization caused by using fixed carbon emission factors. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of steps S101 to S107 of the method described in the specific implementation embodiment; Figure 2 This is a schematic diagram of steps S201 to S206 of the method described in the specific implementation embodiment; Figure 3 This is a schematic diagram of steps S301 to S305 of the method described in the specific implementation. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This embodiment provides a method for constructing and optimizing a carbon footprint model of lifting equipment based on machine learning, including: S101. Collect multi-dimensional working condition time-series sample data and real-time energy consumption sample data of the lifting equipment during its historical operation. S102. Using the gradient boosting decision tree algorithm, with multi-dimensional working condition time series sample data as features, and with the instantaneous carbon emission rate calculated based on real-time energy consumption sample data and basic carbon emission factors as the supervision target, the model is trained to obtain a dynamic carbon footprint intensity prediction model specific to lifting equipment. In addition, the real-time multi-dimensional operating condition time-series data of the lifting equipment is collected and input into the dynamic carbon footprint intensity prediction model, and the dynamic carbon emission factor corresponding to the current operating state is output. S103. Obtain the operational constraints of the lifting task to be executed; S104. Construct a digital twin simulation environment based on a dynamic carbon footprint intensity prediction model, and use an optimization search algorithm to generate multiple candidate operation strategies under operational constraints. S105. In the digital twin simulation environment, perform forward carbon footprint simulation on each generated candidate operation strategy, and calculate the total predicted carbon footprint corresponding to each candidate operation strategy. S106. Compare the total predicted carbon footprint of all candidate operation strategies and select the candidate operation strategy with the lowest total predicted carbon footprint as the optimal energy-saving operation strategy. S107. Output the optimal energy-saving operation strategy to the equipment human-machine interface, or output the optimal energy-saving operation strategy to the equipment control system.

[0020] In step S101, the multi-dimensional operating condition time-series sample data consists of multi-dimensional operating parameters with time-series characteristics collected during the operation of the lifting equipment, including physical quantities such as load weight, operating speed, displacement, and acceleration. This data can be continuously acquired by sensors installed on key parts of the equipment at a fixed sampling frequency. The real-time energy consumption sample data is the equipment power or energy consumption value synchronously recorded by the power metering device, reflecting the energy consumption of the equipment under different operating conditions. This step provides a sufficient data foundation for establishing an accurate carbon footprint prediction model by constructing a time-aligned correspondence between the multi-dimensional operating data and the energy consumption data.

[0021] In step S102, the gradient boosting decision tree algorithm is an ensemble learning method that sequentially constructs multiple decision trees and combines their prediction results. The prediction residual of the previous decision tree is used as the optimization objective for subsequent trees, gradually improving the model's prediction accuracy. Multi-dimensional operating condition time-series sample data serves as the model input feature, containing complete information about the equipment's operating status. The instantaneous carbon emission rate, used as a supervised objective for model training, can be understood as the carbon emission per unit time obtained by multiplying real-time energy consumption data with a basic carbon emission factor and considering time unit conversion. After the dynamic carbon footprint intensity prediction model is trained, it can output a dynamic carbon emission factor reflecting the carbon emission efficiency under the current equipment operating status based on real-time collected multi-dimensional operating condition time-series data. This step establishes a non-linear mapping relationship from equipment operating conditions to carbon emission intensity through machine learning methods, overcoming the limitations of traditional fixed carbon emission factors.

[0022] In step S103, operational constraints refer to the technical limitations and safety requirements that must be followed during the execution of the lifting task, including boundary conditions such as maximum permissible operating time, safe operating speed range, and load limits. These constraints are typically derived from equipment technical specifications, operational safety procedures, and specific task requirements to ensure that the generated candidate operational strategies are feasible and safe in practical applications.

[0023] In step S104, the digital twin simulation environment is a virtual simulation system built based on a dynamic carbon footprint intensity prediction model, capable of simulating the operating status and carbon emission process of lifting equipment under different operating strategies. The optimization search algorithm generates multiple differentiated candidate operating strategies within this environment through systematic searching and evaluation, while satisfying operational constraints. This step, by constructing a virtual simulation environment, provides a reliable testing platform for subsequent carbon footprint assessment and strategy optimization.

[0024] In step S105, the forward carbon footprint simulation can be understood as simulating the complete operation process of the equipment from the start to the end of the task in a digital twin environment, according to the settings of the candidate operation strategy. By inputting the operating parameters of each time step in the simulation process into the dynamic carbon footprint intensity prediction model, the corresponding instantaneous carbon emission rate is obtained, and then the total predicted carbon footprint under the strategy is calculated through time integration. This step realizes the quantitative evaluation of the carbon emission effect of the candidate operation strategy.

[0025] In step S106, the system compares the total predicted carbon footprint values ​​of all candidate operating strategies and identifies the strategy with the lowest carbon emissions as the optimal energy-saving operating strategy. This selection process, based on quantitative evaluation results, ensures that the selected strategy achieves optimal carbon emission reduction while meeting operational constraints.

[0026] In step S107, the optimal energy-saving operation strategy is presented to the operator in a visual form through the equipment's human-machine interface, providing specific operational guidance. For equipment with a high degree of automation, the strategy can be directly converted into control commands and sent to the equipment control system. This step realizes the transformation from optimization results to practical application, forming a complete carbon emission reduction closed loop.

[0027] This embodiment constructs a dynamic carbon footprint intensity prediction model based on machine learning, combined with digital twin simulation and optimization algorithms, to achieve accurate prediction of the carbon footprint of lifting equipment and optimization of operating strategies. The dynamic carbon footprint intensity prediction model, established using a gradient boosting decision tree algorithm, accurately captures the complex nonlinear relationship between equipment operating conditions and carbon emission intensity, replacing the fixed carbon emission factor used in traditional accounting. Based on this, combined with digital twin simulation technology, it achieves forward simulation and quantitative evaluation of the carbon footprint of various candidate operating strategies, ultimately selecting the optimal energy-saving operating scheme through optimization search. This methodology elevates carbon footprint management from passive monitoring to proactive optimization, forming a data-driven, model-supported, and closed-loop optimization carbon reduction technology path.

[0028] Please see Figure 2In some embodiments, a gradient boosting decision tree algorithm is used, featuring multi-dimensional operating condition time-series sample data, with the instantaneous carbon emission rate calculated based on real-time energy consumption sample data and basic carbon emission factors as the supervision target, to train the model and obtain a dynamic carbon footprint intensity prediction model specific to lifting equipment, including: S201. Perform data cleaning and feature construction processing on the multi-dimensional working condition time series sample data to obtain a multi-dimensional feature vector including real-time load rate, running speed time series, and acceleration statistical features. The acceleration statistical features include the acceleration standard deviation and acceleration extreme values ​​within the sliding time window. S202. The multidimensional feature vector and the corresponding instantaneous carbon emission rate are combined to form a training sample set, which is then input into the gradient boosting decision tree algorithm for iterative training. The iterative training is configured to use the prediction residual generated in the previous iteration as the optimization objective. Multiple decision tree weak learners are constructed sequentially through an additive model. The construction objective of each subsequent decision tree weak learner is to fit and correct the prediction residual of the previous decision tree weak learner. S203. All decision tree weak learners built during the training process are linearly combined by weighted summation to form the final dynamic carbon footprint intensity prediction model. The weight of each decision tree weak learner is dynamically determined according to its training error. S204. Based on the trained dynamic carbon footprint intensity prediction model, perform feature importance analysis, output the contribution ranking of each working condition parameter to carbon footprint intensity, and obtain the feature importance analysis results. S205. Based on the results of the feature importance analysis, select key impact features with a contribution rate higher than the preset threshold and construct a simplified feature set; S206. Establish a model performance monitoring and update mechanism. When the model prediction error of the dynamic carbon footprint intensity prediction model continues to exceed the preset error threshold, trigger the model retraining process based on newly added sample data and use a simplified feature set to update the parameters and optimize the structure of the dynamic carbon footprint intensity prediction model.

[0029] In step S201, data cleaning includes addressing potential issues such as missing values, outliers, and signal interference in the sensor-acquired data to ensure data integrity and reliability. Feature construction involves extracting more representative feature parameters from the original time-series data. Preferably, the real-time load rate is calculated as the ratio of the current load weight to the rated load; the operating speed time-series sequence preserves the complete trajectory of the equipment's operating speed over time; and the acceleration statistical features characterize operational stability by setting a fixed-length sliding time window, calculating the standard deviation of the acceleration data within the window, and recording the maximum absolute value of acceleration within the window as the acceleration extreme value, reflecting the impact characteristics during equipment start-up and shutdown. This step, by constructing a multi-dimensional feature vector that comprehensively reflects the equipment's operating status, provides high-quality feature input for subsequent model training.

[0030] In step S202, the training sample set consists of multi-dimensional feature vectors as input features and the corresponding instantaneous carbon emission rate as the supervision target. In each iteration, the gradient boosting decision tree algorithm first calculates the prediction residual of the current model on the training samples, and then uses this residual as the new optimization target to construct the next decision tree weak learner. Through this additive model approach, each subsequent decision tree focuses on correcting the prediction error of the preceding model, gradually improving the overall model's fitting accuracy. This step achieves high-precision modeling of complex nonlinear relationships through the iterative optimization mechanism of residual fitting.

[0031] In step S203, the weighted summation linear combination process determines the weight allocation of each decision tree's weak learner by evaluating its predictive performance. Specifically, the initial weight of each decision tree is calculated based on its mean squared error on the validation dataset, with decision trees having smaller prediction errors assigned higher weights. During model usage, the system continuously monitors the prediction accuracy of each decision tree on new samples and dynamically adjusts its weight coefficients to ensure the model maintains optimal predictive performance. This step optimizes the overall predictive performance of the ensemble model through a dynamic weight adjustment mechanism.

[0032] In step S204, feature importance analysis is performed based on the internal structure of the trained model. By statistically analyzing the frequency of each feature's use during decision tree splitting and the magnitude of its information gain, the contribution of each operating parameter to carbon footprint intensity prediction is quantified. The analysis results are presented in a ranked format, clearly demonstrating the differences in the importance of different operating parameters to carbon emissions. This step provides data support for feature selection and model optimization.

[0033] In step S205, the preset threshold is set according to the actual application requirements, and features that rank in the top certain proportion in the feature importance ranking are usually selected as key influencing features. By filtering key features with a contribution higher than this threshold, the constructed simplified feature set significantly reduces computational complexity and feature acquisition costs while maintaining the model's prediction accuracy. This step achieves optimized compression of the feature space.

[0034] In step S206, model performance monitoring is achieved by periodically calculating the model's prediction error on newly added samples. A preset error threshold is determined based on the accuracy requirements of the actual application scenario. When the prediction error is detected to continuously exceed this threshold, the system automatically triggers the model retraining process. The retraining process uses newly added sample data and a simplified feature set to update the model parameters and optimize the model structure as necessary. This step ensures that the model can adapt to changes in equipment operating conditions and maintain long-term prediction accuracy.

[0035] This embodiment illustrates the construction and optimization process of a dynamic carbon footprint intensity prediction model. Through systematic feature engineering, iterative training, model integration, and continuous optimization, a carbon footprint prediction model that can accurately reflect the operating characteristics of lifting equipment and has adaptive capabilities has been established. This not only solves the problem of insufficient accuracy of traditional fixed carbon emission factor methods, but also provides personalized solutions for different equipment and operating conditions through feature importance analysis and model update mechanisms, forming a sustainable optimization technology system.

[0036] Please see Figure 3 In some embodiments, all decision tree weak learners built during training are linearly combined by weighted summation to form the final dynamic carbon footprint intensity prediction model, including: S301. Calculate the initial weights based on the mean squared error of each decision tree weak learner on the validation dataset; S302. Dynamically adjust the initial weights of each decision tree weak learner based on its prediction accuracy on new samples to obtain dynamic weight coefficients. Among them, the decision tree weak learners whose prediction accuracy continues to decline will have their weights reduced. S303. Multiply the prediction outputs of each decision tree weak learner by their corresponding dynamic weight coefficients and sum them up to obtain the final output of the dynamic carbon footprint intensity prediction model, and obtain the prediction result after weighted summation. S304. Compare and analyze the weighted summation prediction results with the measured sample carbon footprint data, and establish an output calibration function to correct the prediction results; Furthermore, by setting constraints on the maximum number of decision trees and the number of leaf nodes, the overall complexity of the dynamic carbon footprint intensity prediction model is controlled. S305. The final dynamic carbon footprint intensity prediction model is obtained.

[0037] In step S301, the initial weights are calculated based on the performance of each decision tree's weak learner on the independent validation dataset. Mean squared error (MSE), as an evaluation metric, reflects the average deviation between the predicted and actual values ​​of each decision tree. Specifically, the validation dataset is input into each decision tree to obtain its prediction results. The MSE value between this prediction and the measured carbon footprint data is calculated, and then the initial weights are determined using the inverse error normalization method; that is, the decision tree with the smaller the error is assigned a higher weight. This step provides a reasonable initial weight allocation for subsequent weighted combination by quantitatively evaluating the independent performance of each decision tree.

[0038] In step S302, the dynamic weight adjustment mechanism is implemented by continuously monitoring the prediction accuracy of each decision tree on new samples. The system periodically collects new operational data and calculates the trend of prediction error changes for each decision tree on new samples. For decision trees whose prediction accuracy continues to decline, the system gradually reduces their weight coefficients according to a preset ratio; while for decision trees that perform stably on new samples, their weights are maintained or appropriately increased. This step ensures that the model combination can adapt to changes in equipment operating status and maintain long-term prediction accuracy.

[0039] In step S303, the weighted summation process multiplies the predicted outputs of each decision tree by their corresponding dynamic weight coefficients and then sums them. In actual calculations, for a given input feature vector, each decision tree independently generates a predicted value. These predicted values ​​are multiplied by their respective dynamic weight coefficients and then summed to obtain the final output of the ensemble model. This step, through a dynamically weighted ensemble method, fully leverages the advantages of each decision tree, improving the robustness and accuracy of the overall model.

[0040] In step S304, the output calibration function corrects the model by establishing a mapping relationship between predicted and measured values. Specifically, a certain number of prediction results and corresponding measured carbon footprint data are collected, and a calibration function is established using linear regression or piecewise linear fitting methods to eliminate systematic prediction biases. Simultaneously, by setting limits on the maximum number of decision trees and leaf nodes, the model's complexity is controlled to avoid overfitting. This step further improves the model's prediction accuracy and generalization ability.

[0041] In step S305, the integrated model after weight optimization, output calibration, and complexity control is the final dynamic carbon footprint intensity prediction model. This model integrates the predictive capabilities of multiple decision trees, maintains adaptability to changes in equipment operating status through a dynamic weight adjustment mechanism, ensures the accuracy of prediction results through an output calibration function, and guarantees the model's practicality and efficiency through complexity control.

[0042] This embodiment illustrates the integrated optimization process of the dynamic carbon footprint intensity prediction model. Through system optimization of multiple stages such as initial weight calculation, dynamic weight adjustment, weighted summation, output calibration, and complexity control, a carbon footprint prediction model with high accuracy, strong adaptability, and good generalization ability is constructed. It can not only accurately reflect the carbon emission characteristics of lifting equipment under different operating conditions, but also adaptively adjust with changes in equipment status, providing reliable technical support for subsequent carbon footprint optimization.

[0043] In some embodiments, real-time multi-dimensional operating condition time-series data of the lifting equipment is collected and input into a dynamic carbon footprint intensity prediction model, outputting a dynamic carbon emission factor corresponding to the current operating state, including: The real-time acquired multi-dimensional operating condition time-series data is preprocessed to generate real-time feature vectors; The real-time feature vector is input into the trained dynamic carbon footprint intensity prediction model. Through forward propagation of the model, the predicted carbon footprint intensity value at the current time is output, including: The real-time feature vector is passed tree by tree in the decision tree set, and reaches the leaf node along the corresponding branch path in each decision tree according to the feature value; Obtain the output values ​​of the leaf nodes of each decision tree; The weighted sum of the output values ​​of all leaf nodes of the decision tree is used to obtain the predicted carbon footprint intensity at the current moment. The predicted carbon footprint intensity is used as a dynamic carbon emission factor and output.

[0044] In this embodiment, the preprocessing of real-time acquired multi-dimensional operating condition time-series data includes steps such as data format unification, outlier removal, and feature standardization. By converting raw data collected from different sensors into a unified numerical format, outlier data points exceeding reasonable ranges are identified and removed. Furthermore, each feature parameter is standardized to eliminate the influence of dimensional differences, ultimately generating a real-time feature vector that meets the model's input requirements. This preprocessing ensures the quality and consistency of the input data, providing a reliable data foundation for accurate model inference.

[0045] The transfer of real-time feature vectors within the decision tree ensemble follows the inference mechanism of the gradient boosting decision tree model. For each decision tree, the system traverses along the corresponding branch paths from the root node, based on the specific values ​​of each feature in the real-time feature vector, until a leaf node is reached. Each leaf node stores the output value corresponding to that node, which represents the model's basic prediction of carbon footprint intensity under the current feature combination. This process fully leverages the advantages of decision tree models in processing structured data, achieving efficient inference for complex operating conditions.

[0046] The output values ​​of each decision tree leaf node are obtained by recording the value corresponding to the leaf node finally reached during the traversal of each decision tree. Since the gradient boosting decision tree model consists of multiple decision trees, each decision tree produces an independent prediction output. These output values ​​together constitute the model's multi-faceted assessment of the carbon footprint intensity under the current operating conditions. This step provides the necessary intermediate results for the subsequent weighted summation.

[0047] The weighted summation process linearly combines the output values ​​of the leaf nodes of each decision tree according to their weights in the ensemble model. Specifically, the output value of each decision tree is multiplied by its corresponding dynamic weight coefficient and then summed to obtain the final predicted carbon footprint intensity value. These weight coefficients are determined during the model training phase and dynamically adjusted based on the performance of each decision tree during model use to ensure that the ensemble model always maintains optimal predictive performance. This step, through weighted ensemble, integrates the predictive capabilities of multiple decision trees, improving the accuracy and stability of the model.

[0048] By outputting the predicted carbon footprint intensity as a dynamic carbon emission factor, a real-time mapping from equipment operating conditions to carbon emission intensity is achieved. This dynamic carbon emission factor reflects the equipment's carbon emission efficiency under the current specific operating conditions and, compared to traditional fixed carbon emission factors, can more accurately characterize the actual carbon emission level of the equipment. This output provides precise data support for subsequent carbon footprint accounting and optimization decisions.

[0049] This embodiment applies a trained dynamic carbon footprint intensity prediction model to practical applications. Through a standardized data preprocessing process, raw operating data is transformed into feature vectors recognizable by the model. Utilizing the parallel inference mechanism of decision tree ensembles, the system can quickly generate multi-faceted intermediate prediction results. Then, a weighted summation strategy is used to fuse these scattered predictions into an accurate carbon footprint intensity value. Finally, a dynamic carbon emission factor is output, which can specifically reflect the carbon emission performance of each device under different operating conditions in real time. This dynamic perception capability lays a solid foundation for subsequent accurate carbon accounting and intelligent optimization control, enabling carbon footprint management to shift from static estimation to dynamic precision.

[0050] In some embodiments, the method further includes: By combining the predicted carbon footprint intensity with real-time power data, the instantaneous carbon emission rate is obtained, which is expressed by formula (1), as follows: ; In formula (1), Instantaneous carbon emission rate, For real-time power, This is a predicted value for carbon footprint intensity. This is a time conversion factor; Based on the instantaneous carbon emission rate, the cumulative carbon emissions over a specified time period are calculated by time integration and expressed by formula (2), which is as follows: ; In formula (2), For cumulative carbon emissions, The starting point of a specified time period. The end point of a specified time period. This is the symbol for time integral.

[0051] In this embodiment, the instantaneous carbon emission rate is calculated by combining the predicted carbon footprint intensity with real-time power data. The real-time power in formula (1) It can be directly measured by an electricity metering device, with the unit being kilowatts; predicted carbon footprint intensity value. Output from a dynamic carbon footprint intensity prediction model, in kilograms of CO2 equivalent per kilowatt-hour. Time conversion factor. Used to convert units from hours to seconds, ensuring instantaneous carbon emission rates The unit is kilograms of carbon dioxide equivalent per second. This calculation process organically combines the carbon emission efficiency predicted by the model with actual energy consumption data, achieving accurate quantification of the instantaneous carbon emissions of the equipment.

[0052] The calculation of cumulative carbon emissions is based on the time integral of the instantaneous carbon emission rate. The integral operation in formula (2) is performed over a specified time period. The carbon footprint is calculated by continuously summing the instantaneous carbon emission rates within a given time period. In practical applications, since data acquisition is discrete, this integral is usually calculated using numerical integration methods. For example, the time period is divided into several small intervals, and the product of the instantaneous carbon emission rate and the time interval is summed for each interval. This calculation method can accurately reflect the total carbon emissions of the equipment within any specified time period, providing a reliable data foundation for carbon footprint accounting.

[0053] This embodiment establishes a complete calculation system from instantaneous carbon emission rate to cumulative carbon emission. By combining the carbon emission efficiency output by the dynamic carbon footprint intensity prediction model with actual power data, it achieves real-time monitoring and accurate calculation of equipment carbon emissions. This hierarchical calculation method not only meets the needs of instantaneous carbon emission monitoring but also provides flexible cumulative carbon emission statistics capabilities, enabling unified technical management of carbon footprint at different granularities.

[0054] In some embodiments, a digital twin simulation environment is constructed based on a dynamic carbon footprint intensity prediction model, and an optimization search algorithm is used to generate multiple candidate operation strategies under operational constraints, including: A digital twin of the crane operation is established and initialized based on a dynamic carbon footprint intensity prediction model. The digital twin can simulate the corresponding equipment operation status and carbon emission process according to the input operation strategy. The operational constraints are transformed into the constraint boundaries of the optimization problem. The constraint boundaries include the maximum allowable operation time, the safe operating speed range, and the maximum acceleration limit. The particle swarm optimization algorithm is used to search within the constraint boundary. By initializing the position and velocity of the particle swarm, the position vector of each particle represents a complete simulation operation strategy, which includes a simulation velocity control sequence and a simulation acceleration control sequence. In each iteration, candidate operation strategies are generated based on the current position of the particle. Each candidate operation strategy is input into the digital twin for forward simulation. The total predicted carbon footprint corresponding to each strategy is calculated through the dynamic carbon footprint intensity prediction model. The fitness of each particle is assessed based on the total predicted carbon footprint, the individual optimal position and the global optimal position are updated, and the search direction and step size of the particles are adjusted according to the velocity update formula. The iterative process is repeated until the termination condition is met, and the final output particle swarm positions constitute multiple candidate operation strategies.

[0055] In this embodiment, the digital twin is built based on a dynamic carbon footprint intensity prediction model, simulating the actual operation process by constructing a virtual model corresponding to the physical device. This digital twin receives operational strategies as input, dynamically simulates the changes in the device's operating state under corresponding strategies, and calculates the corresponding carbon emission process in real time based on a built-in prediction model. This virtual-real mapping relationship provides a reliable simulation testing environment for subsequent strategy optimization.

[0056] The transformation process of operational constraints quantifies the actual operational requirements into the mathematical boundaries of the optimization problem. The maximum permissible operation time defines the upper limit of the task completion time, the safe operating speed range limits the speed values ​​for each motion stage, and the maximum acceleration limit ensures the stability and safety of equipment operation. These constraint boundaries together constitute the feasible solution space of the strategy search, guaranteeing the practicality and safety of the generated strategy.

[0057] The initialization process of the particle swarm optimization algorithm randomly generates multiple particles. Each particle's position vector encodes a complete operational strategy, including velocity and acceleration control values ​​arranged in a time sequence. The velocity vector determines the particle's direction and step size in the search space. This encoding method transforms the complex operational strategy optimization problem into a search problem in a multi-dimensional space.

[0058] During the iteration process, the system decodes the position vector of each particle into a specific operational strategy, which is then input into a digital twin for forward simulation. The total predicted carbon footprint under each strategy is calculated using a dynamic carbon footprint intensity prediction model, and this prediction is used as the evaluation criterion for particle fitness. Based on the fitness evaluation results, the system records the historical best position of each particle and the global best position of the entire swarm.

[0059] The velocity update process adjusts the particle search behavior based on individual and swarm experience. By balancing individual exploration and swarm collaboration, the particle swarm is guided to gradually converge to a high-quality solution region. Simulation evaluation and position updates are repeated until a preset number of iterations or convergence accuracy is reached. The final output particle positions constitute a series of candidate operation strategies with differences.

[0060] This embodiment combines digital twin technology with intelligent optimization algorithms. By constructing a virtual simulation environment, it achieves systematic evaluation and optimization of operation strategies. The particle swarm optimization algorithm can efficiently explore a vast strategy space and quickly identify feasible low-carbon emission operation schemes through directional search within the constraint boundary, enabling energy-saving operation of lifting equipment.

[0061] In some embodiments, the velocity update formula is expressed by formula (3), which is as follows: ; In formula (3), For the first The particle in the first The velocity vector at the next iteration For the first The particle in the first The velocity vector at the next iteration This is the inertia weighting coefficient. For individual learning factors, As a social learning factor, For the first between Random numbers within, For the second one Random numbers within, For the first The particle in the first The optimal position of the individual in the next iteration. For the particle swarm in the th The global optimal position at the next iteration. For the first The particle in the first The current position at the next iteration.

[0062] In this embodiment, the velocity update formula defines the update rule for the particle motion state in the particle swarm optimization algorithm. The particle motion state is updated through the synergistic effect of three key components: the inertial component maintains the original motion trend of the particle, the individual cognitive component guides the particle to move closer to its own historical optimal position, and the social cognitive component promotes the particle to converge towards the group's optimal position.

[0063] Inertia weighting coefficient in formula (3) The degree to which particle velocities are inherited controls the algorithm's global exploration and local development capabilities. Individual learning factor. Social learning factors Two random numbers are used to adjust the emphasis of particles on individual experience and group experience, respectively. and The introduction of these parameters injects necessary randomness into the search process, preventing the algorithm from getting trapped in local optima too early. The coordinated configuration of these parameters collectively determines the search efficiency and convergence performance of the optimization algorithm.

[0064] Individual optimal position Recorded the first The optimal position reached by each particle in each iteration reflects the individual learning experience of the particle; the global optimal position This represents the best solution discovered so far in the entire particle swarm. By continuously comparing the fitness of the current position with the historical best position, the particle can continuously optimize its search direction.

[0065] This embodiment defines in detail the core update mechanism of the particle swarm optimization algorithm. By quantitatively describing the dynamic adjustment process of particle velocities, it reveals the intrinsic working principle of intelligent optimization algorithms in solving high-dimensional complex problems. This swarm intelligence-based search strategy enables the system to navigate efficiently within a vast operational strategy space and quickly identify high-quality solutions that combine low carbon emissions and feasibility.

[0066] In some embodiments, a particle swarm optimization algorithm is used to search within the constraint boundary. By initializing the positions and velocities of the particle swarm, the position vector of each particle represents a complete simulation operation strategy. The simulation operation strategy includes a simulation velocity control sequence and a simulation acceleration control sequence, including: Initialize the particle swarm and randomly generate the initial positions and velocities of multiple particles. The position vector of each particle consists of multiple dimensions, and each dimension corresponds to a velocity control value or acceleration control value for a time step. The position vector of the particle is constrained according to the constraint boundary to ensure that the values ​​of each dimension are within the allowable safe operating speed range and the maximum acceleration limit. The position vectors of each particle are decoded into a complete simulation operation strategy, which includes a simulated velocity control sequence and a simulated acceleration control sequence arranged in time sequence. Furthermore, the particle's velocity and position are updated based on the individual optimal position and the global optimal position during each iteration; When the particle swarm is detected to be trapped in a local optimum, the positions of some particles are randomly reset to expand the search range.

[0067] In this embodiment, the particle swarm initialization process initiates the optimization search by randomly generating the positions and velocities of multiple particles. Each particle's position vector is a multi-dimensional array, the number of dimensions of which is determined by the total task duration and time step size. Each dimension stores the velocity or acceleration control value for the corresponding time step. The initial velocity vector determines the initial direction and magnitude of the particle's movement in the search space. This encoding method transforms the continuous operational strategy optimization problem into a discrete parameter optimization problem.

[0068] Boundary constraint handling is implemented through a truncation function. When a dimension of the particle's position vector exceeds the constraint boundary, the system automatically adjusts it to the nearest boundary value. For example, velocity control values ​​are limited to a safe operating speed range, and acceleration control values ​​must not exceed the maximum acceleration limit. This approach ensures that all generated candidate strategies meet the safety requirements of the actual operation.

[0069] The position vector decoding process transforms abstract numerical sequences into concrete operational instructions. By reading the values ​​of each dimension of the position vector in chronological order, the system constructs complete simulated velocity control sequences and simulated acceleration control sequences. These sequences specify in detail the specific operations that the equipment should take at each time step, providing clear input instructions for subsequent digital twin simulations.

[0070] The iterative update process is based on the core mechanism of the particle swarm optimization algorithm, adjusting the particle's motion state by continuously comparing the individual historical optimum with the swarm's global optimum. When the particle swarm is detected to be trapped in a local optimum, the system randomly selects some particles and resets their positions to a new region in the search space. This strategy effectively enhances the algorithm's global exploration capability and avoids premature convergence to a suboptimal solution.

[0071] In this embodiment, the specific implementation of the particle swarm optimization algorithm in operation strategy search is to transform the complex operation optimization problem into a computable parameter search process through the system's initialization, boundary constraints, decoding and update mechanisms. Through the optimization method based on swarm intelligence, it can efficiently explore a huge strategy space and find reliable low-carbon emission operation schemes while meeting safety constraints.

[0072] In some embodiments, within a digital twin simulation environment, a forward carbon footprint simulation is performed on each generated candidate operation strategy, and the total predicted carbon footprint corresponding to each candidate operation strategy is calculated, including: The candidate operation strategy is discretized into a sequence of operation instructions with multiple time steps. The sequence of operation instructions includes the device operating speed, acceleration and load status corresponding to each time step. The operating parameters corresponding to the operation instructions at each time step are input into the dynamic carbon footprint intensity prediction model to obtain the instantaneous carbon emission rate prediction value at each time step. Based on the instantaneous carbon emission rate prediction, the total predicted carbon footprint of the candidate operation strategy is calculated through time integration, including: The instantaneous carbon emission rate at each time step is multiplied by the corresponding time step and then summed.

[0073] In this embodiment, the discretization of candidate operation strategies transforms the continuous strategy curve into a discrete time-step sequence. Each time step corresponds to a fixed time interval, containing operational instructions such as the operating speed, acceleration, and load state that the device should achieve at that moment. Discretization transforms the complex continuous control problem into a computable discrete sequence, laying the foundation for subsequent stepwise simulation.

[0074] The process of inputting operating parameters at each time step transforms discretized operating commands into feature vectors recognizable by the dynamic carbon footprint intensity prediction model. For each time step, the system extracts parameters such as current velocity, acceleration, and load status, constructs a feature format that meets the model's input requirements, and then inputs it into the trained prediction model to obtain the instantaneous carbon emission rate prediction value at that moment. This step achieves real-time mapping from operating commands to carbon emission prediction.

[0075] The total predicted carbon footprint is calculated using a numerical integration method, which sums the instantaneous carbon emission rates at each time step. Specifically, the predicted instantaneous carbon emission rate for each time step is multiplied by the corresponding time step length to obtain the carbon emissions within that time interval. Then, the carbon emissions from all time steps are summed to obtain the total predicted carbon footprint of the candidate operational strategy. This discrete integration method improves computational efficiency while maintaining computational accuracy.

[0076] This embodiment achieves a precise quantitative assessment of the carbon emission effects of candidate operational strategies through discretization, stepwise simulation, and numerical integration. This time-step-based simulation method accurately reflects the differences in carbon emissions among different operational strategies throughout the entire operation process, providing a reliable quantitative basis for subsequent strategy selection and enabling carbon footprint optimization to move from qualitative analysis to quantitative decision-making.

[0077] In some embodiments, comparing the total predicted carbon footprint of all candidate operating strategies, the candidate operating strategy with the lowest total predicted carbon footprint is selected as the optimal energy-saving operating strategy, including: Establish a candidate strategy ranking queue and sort all candidate operation strategies in ascending order according to their corresponding total predicted carbon footprint values; The candidate operation strategy with the smallest total predicted carbon footprint value is extracted from the sorted queue as the preliminary optimal strategy. The feasibility of the preliminary optimal strategy is verified, including: Verify whether the operating parameters corresponding to the preliminary optimal strategy meet the equipment safety operation specifications and whether they exceed the equipment performance limits; When the preliminary optimal strategy passes the feasibility verification, it is determined as the final optimal energy-saving operation strategy; If the initial optimal strategy fails the feasibility verification, the next best candidate strategies are selected in order of the sorting queue for feasibility verification until the optimal energy-saving strategy that passes the verification is determined.

[0078] In this embodiment, the candidate strategy ranking queue is established by sorting the total predicted carbon footprint values ​​of all candidate operation strategies in ascending order. The system associates each candidate strategy with its corresponding total predicted carbon footprint value, forming an ordered list according to carbon emissions from low to high. This allows low-carbon emission strategies to enter the subsequent screening process first, improving optimization efficiency.

[0079] The initial optimal strategy extraction process directly selects the candidate operation strategy with the smallest total predicted carbon footprint from the top of the sorted queue. Since the queue is already sorted in ascending order of carbon emissions, the strategy at the top of the queue is the theoretically optimal low-carbon solution. This step ensures that the system first considers the potential solution with the lowest carbon emissions.

[0080] The feasibility verification process conducts comprehensive applicability checks on the preliminary optimal strategy. The system verifies whether the operating parameters in the strategy are within the allowable range of the equipment's safe operation specifications, including whether key parameters such as speed and acceleration exceed the equipment's technical performance limits. This verification mechanism ensures that the selected strategy not only has low carbon emissions but is also safe and feasible in actual operation.

[0081] The strategy determination mechanism adopts different processing paths based on the verification results. When the preliminary optimal strategy passes all feasibility checks, the system directly determines it as the final optimal energy-saving operation strategy; if the verification fails, the next best strategies are checked sequentially according to the sorting queue until the first strategy that meets all feasibility requirements is found. This progressive screening method ensures both optimal carbon emissions and the practicality and safety of the strategy.

[0082] This embodiment establishes a strategy optimization mechanism that balances carbon emission efficiency and practical feasibility through a systematic sorting, extraction, verification, and determination process. This progressive screening method, from theoretical optimality to practical applicability, fully leverages the search capabilities of the optimization algorithm while ensuring the engineering practicality of the final output strategy, providing reliable operational guidance for the low-carbon operation of lifting equipment.

[0083] By adopting the above technical solutions, this invention differs from existing technologies and possesses the following beneficial effects: By constructing a dynamic carbon footprint intensity prediction model based on a gradient boosting decision tree algorithm, accurate prediction of the carbon emission intensity of lifting equipment is achieved. This model can output a dynamic carbon emission factor reflecting the current operating state based on real-time collected multi-dimensional operating condition time-series data, effectively overcoming the problem of insufficient calculation accuracy of fixed carbon emission factor methods under complex operating conditions. Simultaneously, combined with a digital twin simulation environment and an optimization search algorithm, the system can generate multiple candidate operation strategies and perform forward carbon footprint simulation while meeting operational constraints. Through quantitative evaluation and a progressive screening mechanism, the optimal energy-saving operation strategy that combines low carbon emissions and practical feasibility is ultimately selected.

[0084] The aforementioned technical solution forms a complete closed loop from data acquisition and model training to strategy optimization, elevating carbon footprint management from passive monitoring to proactive optimization. The dynamic carbon footprint intensity prediction model, through feature importance analysis and continuous optimization mechanisms, can adapt to the personalized needs of different equipment and operating conditions; digital twin simulation technology provides a reliable testing platform for evaluating operating strategies; and the swarm intelligence-based optimization algorithm achieves efficient searching within a vast strategy space. This technical solution significantly improves the accuracy of carbon footprint accounting and provides a scientific basis and operable solutions for the energy-saving operation of lifting equipment, realizing the transformation of carbon footprint management from static estimation to dynamic precision.

[0085] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0086] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for constructing and optimizing a carbon footprint model of lifting equipment based on machine learning, characterized in that, include: Collect multi-dimensional operating condition time-series sample data and real-time energy consumption sample data of the lifting equipment during its historical operation; The gradient boosting decision tree algorithm is used to train the model, with the multi-dimensional working condition time series sample data as features and the instantaneous carbon emission rate calculated based on the real-time energy consumption sample data and the basic carbon emission factor as the supervision target, to obtain a dynamic carbon footprint intensity prediction model specific to lifting equipment. In addition, the real-time multi-dimensional operating condition time-series data of the lifting equipment is collected and input into the dynamic carbon footprint intensity prediction model, and the dynamic carbon emission factor corresponding to the current operating state is output. Obtain the operational constraints for the lifting task to be executed; A digital twin simulation environment is constructed based on the dynamic carbon footprint intensity prediction model, and an optimization search algorithm is used to generate multiple candidate operation strategies under the operational constraints. In the digital twin simulation environment, carbon footprint forward simulation is performed on each generated candidate operation strategy, and the total predicted carbon footprint corresponding to each candidate operation strategy is calculated. By comparing the total predicted carbon footprint of all candidate operation strategies, the candidate operation strategy with the lowest total predicted carbon footprint is selected as the optimal energy-saving operation strategy. The optimal energy-saving operation strategy can be output to the equipment's human-machine interface, or the optimal energy-saving operation strategy can be output to the equipment control system.

2. The method for constructing and optimizing the carbon footprint model of lifting equipment based on machine learning according to claim 1, characterized in that, A gradient boosting decision tree algorithm is used, with the multi-dimensional operating condition time-series sample data as features and the instantaneous carbon emission rate calculated based on the real-time energy consumption sample data and the basic carbon emission factor as the supervision target, to train the model and obtain a dynamic carbon footprint intensity prediction model specific to lifting equipment, including: Data cleaning and feature construction processing are performed on the multi-dimensional working condition time series sample data to obtain a multi-dimensional feature vector including real-time load rate, running speed time series, and acceleration statistical features. The acceleration statistical features include the acceleration standard deviation and acceleration extreme values ​​within the sliding time window. The multidimensional feature vector and the corresponding instantaneous carbon emission rate are combined to form a training sample set, which is then input into the gradient boosting decision tree algorithm for iterative training. The iterative training is configured to use the prediction residual generated in the previous iteration as the optimization target. Multiple decision tree weak learners are constructed sequentially through an additive model. The construction target of each subsequent decision tree weak learner is to fit and correct the prediction residual of the previous decision tree weak learner. All decision tree weak learners built during training are linearly combined by weighted summation to form the final dynamic carbon footprint intensity prediction model. The weight of each decision tree weak learner is dynamically determined based on its training error. Feature importance analysis is performed based on the trained dynamic carbon footprint intensity prediction model, and the contribution ranking of each working condition parameter to carbon footprint intensity is output to obtain the feature importance analysis results. Based on the results of the feature importance analysis, key impact features with a contribution rate higher than a preset threshold are selected to construct a simplified feature set; A model performance monitoring and update mechanism is established. When the model prediction error of the dynamic carbon footprint intensity prediction model continues to exceed the preset error threshold, a model retraining process based on newly added sample data is triggered, and the simplified feature set is used to update the parameters and optimize the structure of the dynamic carbon footprint intensity prediction model.

3. The method for constructing and optimizing the carbon footprint model of lifting equipment based on machine learning according to claim 2, characterized in that, All decision tree weak learners built during training are linearly combined through weighted summation to form the final dynamic carbon footprint intensity prediction model, including: Calculate the initial weights based on the mean squared error of each decision tree weak learner on the validation dataset; The initial weights of each decision tree weak learner are dynamically adjusted based on their prediction accuracy on new samples to obtain dynamic weight coefficients. Among them, the weights of decision tree weak learners whose prediction accuracy continues to decline will be reduced. The prediction outputs of each decision tree weak learner are multiplied by their corresponding dynamic weight coefficients and then summed to obtain the final output of the dynamic carbon footprint intensity prediction model, resulting in a weighted summation prediction result. The weighted summation prediction results are compared and analyzed with the measured sample carbon footprint data, and an output calibration function is established to correct the prediction results. Furthermore, by setting constraints on the maximum number of decision trees and the number of leaf nodes, the overall complexity of the dynamic carbon footprint intensity prediction model is controlled. The final dynamic carbon footprint intensity prediction model is obtained.

4. The method for constructing and optimizing the carbon footprint model of lifting equipment based on machine learning according to claim 1, characterized in that, The real-time multi-dimensional operating condition time-series data of the lifting equipment is collected and input into the dynamic carbon footprint intensity prediction model, which outputs a dynamic carbon emission factor corresponding to the current operating state, including: The real-time acquired multi-dimensional operating condition time-series data is preprocessed to generate real-time feature vectors; The real-time feature vector is input into the trained dynamic carbon footprint intensity prediction model. Through forward propagation calculation of the model, the predicted carbon footprint intensity value at the current moment is output, including: The real-time feature vector is passed tree by tree in the decision tree set, and reaches the leaf node along the corresponding branch path in each decision tree according to the feature value; Obtain the output values ​​of the leaf nodes of each decision tree; The weighted sum of the output values ​​of all leaf nodes of the decision tree is used to obtain the predicted carbon footprint intensity at the current moment. The predicted carbon footprint intensity is used as a dynamic carbon emission factor and output.

5. The method for constructing and optimizing the carbon footprint model of lifting equipment based on machine learning according to claim 4, characterized in that, The method further includes: The instantaneous carbon emission rate is obtained by combining the predicted carbon footprint intensity with real-time power data, and is expressed by formula (1), which is as follows: ； In formula (1), Instantaneous carbon emission rate, For real-time power, This is a predicted value for carbon footprint intensity. This is a time conversion factor; Based on the instantaneous carbon emission rate, the cumulative carbon emissions over a specified time period are calculated by time integration and expressed by formula (2), which is as follows: ； In formula (2), For cumulative carbon emissions, The starting point of a specified time period. The end point of a specified time period. This is the symbol for time integral.

6. The method for constructing and optimizing the carbon footprint model of lifting equipment based on machine learning according to claim 1, characterized in that, A digital twin simulation environment is constructed based on the dynamic carbon footprint intensity prediction model. An optimization search algorithm is used to generate multiple candidate operational strategies under the operational constraints, including: A digital twin of the crane operation is established and initialized based on the dynamic carbon footprint intensity prediction model. The digital twin can simulate the corresponding equipment operation status and carbon emission process according to the input operation strategy. The operational constraints are transformed into the constraint boundaries of an optimization problem, which include the maximum allowable operation time, the safe operating speed range, and the maximum acceleration limit. The particle swarm optimization algorithm is used to search within the constraint boundary. By initializing the position and velocity of the particle swarm, the position vector of each particle represents a complete simulation operation strategy, which includes a simulation velocity control sequence and a simulation acceleration control sequence. In each iteration, a corresponding candidate operation strategy is generated based on the current position of the particle. Each candidate operation strategy is input into the digital twin for forward simulation. The total predicted carbon footprint corresponding to each strategy is calculated through the dynamic carbon footprint intensity prediction model. The fitness of each particle is evaluated based on the total predicted carbon footprint, the individual optimal position and the global optimal position are updated, and the search direction and step size of the particles are adjusted according to the velocity update formula. The iterative process is repeated until the termination condition is met, and the final output particle swarm positions constitute multiple candidate operation strategies.

7. The method for constructing and optimizing a carbon footprint model of lifting equipment based on machine learning according to claim 6, characterized in that, The speed update formula is expressed by formula (3), which is as follows: ； In formula (3), For the first The particle in the first The velocity vector at the next iteration For the first The particle in the first The velocity vector at the next iteration This is the inertia weighting coefficient. For individual learning factors, As a social learning factor, For the first between Random numbers within, For the second one Random numbers within, For the first The particle in the first The optimal position of the individual in the next iteration. For the particle swarm in the th The global optimal position at the next iteration. For the first The particle in the first The current position at the next iteration.

8. The method for constructing and optimizing a carbon footprint model of lifting equipment based on machine learning according to claim 7, characterized in that, The particle swarm optimization algorithm is used to search within the constraint boundary. By initializing the particle swarm's position and velocity, the position vector of each particle represents a complete simulation operation strategy. This simulation operation strategy includes a simulated velocity control sequence and a simulated acceleration control sequence, including: Initialize the particle swarm and randomly generate the initial positions and velocities of multiple particles. The position vector of each particle consists of multiple dimensions, and each dimension corresponds to a velocity control value or acceleration control value for a time step. The position vector of the particle is constrained according to the constraint boundary to ensure that the values ​​of each dimension are within the allowable safe operating speed range and the maximum acceleration limit. The position vectors of each particle are decoded into a complete simulation operation strategy, which includes a simulated velocity control sequence and a simulated acceleration control sequence arranged in time sequence. Furthermore, the particle's velocity and position are updated based on the individual optimal position and the global optimal position during each iteration; When the particle swarm is detected to be trapped in a local optimum, the positions of some particles are randomly reset to expand the search range.

9. The method for constructing and optimizing a carbon footprint model of lifting equipment based on machine learning according to claim 1, characterized in that, In the digital twin simulation environment, a forward carbon footprint simulation is performed on each generated candidate operation strategy, and the total predicted carbon footprint corresponding to each candidate operation strategy is calculated, including: The candidate operation strategy is discretized into a sequence of operation instructions with multiple time steps, and the sequence of operation instructions includes the device operating speed, acceleration and load status corresponding to each time step; The operating parameters corresponding to the operation instructions at each time step are input into the dynamic carbon footprint intensity prediction model to obtain the instantaneous carbon emission rate prediction value at each time step. Based on the instantaneous carbon emission rate prediction, the total predicted carbon footprint of the candidate operation strategy is calculated through time integration, including: The instantaneous carbon emission rate at each time step is multiplied by the corresponding time step and then summed.

10. The method for constructing and optimizing a carbon footprint model of lifting equipment based on machine learning according to claim 1, characterized in that, By comparing the total predicted carbon footprint of all candidate operating strategies, the candidate operating strategy with the lowest total predicted carbon footprint is selected as the optimal energy-saving operating strategy, including: Establish a candidate strategy ranking queue and sort all candidate operation strategies in ascending order according to their corresponding total predicted carbon footprint values; The candidate operation strategy with the smallest total predicted carbon footprint value is extracted from the sorting queue as the preliminary optimal strategy. The feasibility of the preliminary optimal strategy is verified, including: Verify whether the operating parameters corresponding to the preliminary optimal strategy meet the equipment safety operation specifications and whether they exceed the equipment performance limits; When the preliminary optimal strategy passes the feasibility verification, it is determined as the final optimal energy-saving operation strategy; If the preliminary optimal strategy fails the feasibility verification, the next best candidate operation strategies are selected in sequence according to the sorting queue for feasibility verification until the optimal energy-saving operation strategy that passes the verification is determined.

Images