34 results about "Extreme learning machine" patented technology

This invention relates to the field of navigation and timing technology, specifically to a GNSS timing data prediction method and related equipment. The method involves acquiring Roland timing data when GNSS timing data is invalid and preprocessing it. The preprocessed data is then input into a trained single-hidden-layer extreme learning machine model. This model calculates the hidden-layer output matrix by randomly initializing the input layer weight matrix and bias vector, and solves for the output weights using regularized least squares. The model outputs a normalized GNSS timing data prediction value, which is then denormalized to obtain the final GNSS timing prediction result. This method combines the high stability of the Roland system with the efficient computational power of the extreme learning machine to achieve accurate GNSS timing data prediction. It is suitable for enhancing the resilience of positioning, navigation, and timing systems in critical infrastructure scenarios, ensuring the continuous operation of the system even when GNSS signals are interfered with or fail.

The invention discloses a rotor assembly method based on vision and artificial intelligence, and relates to the technical field of rotor assembly. The method comprises the following steps: acquiring a three-dimensional point cloud, a grayscale image and a polarization degree image of a rotor, generating rotor point cloud fusion data, and performing feature extraction on the rotor point cloud fusion data to obtain a rotor point cloud feature vector; inputting the rotor point cloud feature vector into an extreme learning machine model optimized based on an improved flower pollination method, and outputting a rotor risk category; constructing a rotor path optimization model, and generating a first rotor assembly path; extracting a rotor assembly first path feature vector, inputting the rotor assembly first path feature vector into the long-short-term memory network, outputting a predicted resistance sequence, and optimizing to obtain a rotor assembly second path; and actual assembly path data are collected in real time, and the error with the second path is reduced through a proportional integral differential control method, so that rotor assembly is realized.

The application discloses a wind power gear box fault diagnosis method based on time shift cosine similarity entropy, belongs to the technical field of wind turbine gear box fault diagnosis, and comprises the following steps: collecting vibration signals of a wind turbine gear box under different health states; key multi-scale features capable of reflecting different health states are mined from the vibration signals by using time shift cosine similarity entropy; the multi-scale features are proportionally divided into a training set and a test set; an intelligent fault diagnosis model based on an extreme learning machine is constructed, and the training set is input to perform model training; then, the test set is input into the trained extreme learning machine, and gear box fault diagnosis is completed. The application can accurately identify the specific fault type of the wind power gear box, provides a strong basis for discovering the position of the gear box fault and performing maintenance, and ensures reliable and stable operation of the wind turbine.

This invention discloses a method and system for detecting abnormal states of wind turbine generators based on Extreme Learning Machine (ELM). The method includes: real-time acquisition of multi-dimensional operational data from the wind turbine generator; data aggregation and timestamping; preprocessing of the multi-dimensional operational data including cleaning, feature engineering, and normalization to obtain a feature dataset; construction of an ELM model comprising an input layer, hidden layer, and output layer; construction of a training set and a validation set based on pre-set historical operational data of the wind turbine generator; training the ELM model using the training set and optimizing hyperparameters using the validation set; deployment of the trained ELM model at the edge of the wind turbine generator; real-time reading of the feature dataset and execution of inference to obtain anomaly judgment results, thus completing the detection of abnormal states of the wind turbine generator. This invention utilizes the ability to learn from large amounts of data to identify abnormal patterns such as excessively high temperatures and abnormal vibrations, improving detection accuracy and flexibility.

The application discloses a degradation fault output voltage ripple optimization method and system for a DC-DC converter, relates to data processing processes such as undersampling, feature extraction and feature selection, a fault diagnosis method based on an extreme learning machine and a reliability optimization method based on DC-DC converter model analysis, belongs to the field of power electronic system fault diagnosis, and comprises the following steps: undersampling technology for the output voltage of a DC-DC converter; time domain feature extraction technology for undersampling data of the output voltage of the DC-DC converter; recursive feature elimination technology based on a support vector machine for selecting the time domain features of the extracted undersampling data of the output voltage of the DC-DC converter; a support vector machine degradation fault diagnosis model based on a Gaussian kernel and a Bayesian optimization for learning multi-class degradation fault data and performing fault classification; and an analysis method based on a physical model of a power conversion device for optimizing the output voltage ripple of the power conversion device.

This invention proposes a high- and low-voltage line strong discharge monitoring system and method based on video image recognition, relating to the field of strong discharge monitoring. It includes a camera for acquiring video streams of high- and low-voltage lines and sending them to an image analyzer; the image analyzer extracts keyframe images from the video streams, extracts static and dynamic spark features based on the keyframe images, and inputs a feature vector composed of these features into an extreme learning machine for spark recognition, generating a discharge warning signal, which is then sent to a remote server; the remote server receives and displays the discharge warning signal. This invention improves the accuracy of video spark image recognition and can quickly and accurately determine the discharge location, ensuring the safe and stable operation of the power grid and reducing economic losses caused by transmission line discharge faults, thus having significant social and economic implications.

The present application relates to a kind of based on hyperspectral reconstruction sunglasses lens color classification method and system. It is related to optical quality inspection and spectral detection technical field, the method first constructs controlled reflection imaging light path, collects lens reflection image and extracts effective ROI area;Again, RGB image is converted to YCrCb space to realize luminance chroma decoupling, by YCrCb-MST multi-stage spectral Transformer network, RGB image is reconstructed as 400-700nm, 31 channel hyperspectral data;Finally, color classification is completed by particle swarm optimization extreme learning machine, and the category and confidence are output.The system is composed of imaging acquisition, hyperspectral reconstruction, spectral classification and man-machine interaction unit, can inhibit ambient light and brightness interference, accurately identify subtle color difference, detect fast, low cost, adapt to production line online detection, can realize color difference traceability and whole-process data traceability, suitable for sunglasses lens high-precision color classification and matching quality inspection.

The transformer fault early warning method based on multi-modal data fusion and dynamic weight relates to the technical field of transformer fault early warning, and solves the problems of single data, fixed weight, poor fusion generalization, and insufficient early warning adaptability and accuracy of traditional methods. The method first collects real-time operation, environment, and historical maintenance multi-modal heterogeneous data of the transformer and performs differential preprocessing accordingly; then constructs a multi-branch feature extraction architecture to extract fault-related features of the three types of data; then builds a dynamic weight distribution module based on an improved sparrow search algorithm to weight and fuse the features to obtain a multi-modal fusion matrix; finally, the matrix is input into an optimized extreme learning machine model to output the fault probability and specific type, and the weight is dynamically updated and the model is iterated when the working condition parameter changes exceed the threshold. The present application greatly improves the accuracy, reliability and adaptability of fault early warning, significantly reduces the false alarm and missed alarm rate, provides support for transformer operation and maintenance decision-making, and is suitable for intelligent substation equipment health management.

The application provides an image classification method based on a multiscale space spectrum extreme learning machine, which extracts classification features through hierarchical space spectrum feature extraction and cross-modal fusion to realize accurate classification of hyperspectral remote sensing images. A multiscale space feature extraction module is used to generate a space feature vector, a residual spectrum feature extraction module is used to simulate the receptive field response of the spectrum dimension, the shallow spectrum trend and the deep feature are combined, and a correlation-driven adaptive weighted fusion module is used to fuse the space feature and the spectrum feature, and finally classification is realized. For five hyperspectral remote sensing images with different resolutions, the SS-MSLRF-ELM classification model and other five models are compared in the experiment, and the overall accuracy OA and the Kappa coefficient are the highest among the six models. The experimental results show that the SS-MSLRF-ELM classification model can provide more accurate and stable classification results, and can improve the classification accuracy while considering the classification speed.

This invention discloses a method and device for diagnosing the health status of flow batteries based on particle swarm optimization (PSO), aiming to solve the problems of difficult modeling and poor real-time performance of traditional physical model-driven methods, and the unscientific feature selection and insufficient prediction stability of existing data-driven methods. The system is data-driven at its core, collecting multi-dimensional data such as voltage, current, and capacity of the flow battery during operation through sensors. After preprocessing, LASSO and grey relational analysis are used to jointly screen the optimal subset of health features. An extreme learning machine (ELM) is built as the main model for SOH prediction, using the ReLU activation function to improve nonlinear expression capabilities, and introducing a particle swarm optimization algorithm to optimize the weights and biases of the ELM, improving the instability caused by random initialization.

This disclosed method for multi-agent task understanding using an Extreme Learning Machine (ELM) initializes multi-agent parameters and environmental situational awareness information; it then formulates task understanding sample data for the ELM based on the multi-agent's task, parameters, and environmental situational awareness information; it determines the ELM's task understanding network structure based on the sample data; it trains the ELM's task understanding network structure using the sample data to obtain the ELM's task understanding model; and when the multi-agent receives a task instruction, it acquires the current environmental situational awareness information and multi-agent parameters, inputting these into the ELM's task understanding model to obtain the task understanding result. This method effectively utilizes battlefield situational information and multi-agent capability values ​​to generate understanding results that align with the commander's thinking, avoiding the subjective factors that rely to some extent on expert systems and ensuring the accuracy of task understanding.

This invention relates to the field of industrial automation testing technology, specifically disclosing a method for online intelligent detection and dynamic compensation of pH for high-concentration organic solvents. Based on only collecting the original pH reading and real-time temperature, the method reconstructs the solvent concentration, dielectric constant, and dynamic response factor, forming a five-dimensional feature vector with the aforementioned two measured values. This feature vector is input into a support vector regression or online sequential extreme learning machine model that dynamically switches according to the sample size, outputting the corrected true pH value to replace the original reading in closed-loop feedback control. Simultaneously, an incremental learning mechanism is built-in. When the residual between the model's predicted value and the offline test value exceeds the allowable range, the model parameters are automatically updated with new samples. Furthermore, by performing a cumulative sum test on the corrected residual sequence, the method actively senses electrode aging drift, requests offline testing in a targeted manner, and updates the model parameters online using a recursive least squares algorithm, achieving pH measurement without the need for additional hardware environmental sensing.

The present application relates to the technical field of concrete monitoring, in particular to a method for monitoring and predicting carbonation depth of concrete under dry-wet cycle conditions in salt-fresh water areas. The method comprises the following steps: by laying multiple source carbonation monitoring nodes, collecting carbonation related characteristic data, and using the maximum information coefficient algorithm to screen the dominant factor; using fuzzy C-means dynamic clustering algorithm to identify carbonation stage and spatial partition; constructing partition discriminant model through multiple discriminant analysis, and combining extreme learning machine to realize carbonation depth prediction and risk assessment, and finally generating carbonation risk early warning report. The present application realizes the whole process integration of multi-source data driving, feature screening, dynamic partition clustering, partition discrimination and intelligent prediction; improves the monitoring accuracy, risk identification rate and management response efficiency of concrete members, and can provide scientific and effective technical support for concrete durability evaluation and operation management in special environments such as ocean and estuary.

The present application belongs to the technical field of sensor fault diagnosis of railway traffic system, and particularly relates to a traction transmission system grid-side current transformer sampling loop fault diagnosis method, which combines physical feature extraction based on power balance principle and regularization extreme learning machine algorithm to establish a multi-working-condition grid-side current estimation model, uses a sliding window double statistical test method to perform abnormal detection on the residual error of actual values and estimated values of two-way grid-side current signals, establishes a Bayesian diagnosis model integrating hidden Markov model idea, defines a four-state space containing normal and various fault states, simultaneously introduces a state transition inertia mechanism to suppress random noise, adopts a logic lockout mechanism to exclude transient power failure interference caused by train passing through a neutral section, and in a non-lockout state, iteratively calculates the posterior probability of each state through Bayesian inference algorithm, and realizes diagnosis and positioning of the grid-side current transformer sampling loop fault according to the maximum posterior probability principle.

This invention relates to a method for intelligent optimization of desalination parameters in underground pipes based on multi-source data, belonging to the field of agricultural engineering technology. It includes the following steps: collecting multi-source environmental data and constructing a basic sample; achieving asynchronous temporal alignment between the recharge sequence and groundwater level through time-delay correlation analysis; enhancing salinity strip features using multi-scale Gabor filtering with underground pipe direction constraints; constructing a fused multi-source environmental feature vector; establishing an extreme learning machine process surrogate model to quickly predict desalination effects; combining engineering-mandated constraints to trim parameter search boundaries and construct a multi-objective fitness function; introducing drainage resistance gradient-guided particle swarm optimization; generating Pareto solutions by combining non-dominated sorting and adaptive stagnation perturbation; and outputting final construction parameters based on weighted sorting according to engineering risk preferences. This invention can improve the accuracy and convergence speed of parameter optimization, and output feasible implementation schemes that balance desalination effects, salt return risk, and improvement speed.

The present invention relates to the technical field of power grid dispatching optimization. Disclosed are a power grid forward-looking risk assessment method and device, and a medium, solving the problems in the prior art of unstable wind speed, low fan efficiency, difficult power grid dispatching and poor safety of wind power generation. In the present invention, a target fan power time series is constructed on the basis of fan power data of a historical time period, then an extreme learning machine model, a long short-term memory neural network model and a temporal convolutional network model are trained, and then a short-term wind power prediction result is obtained, improving the accuracy and the flexibility of the short-term wind power prediction result. In an offline phase of a power grid, a generative adversarial network model is constructed on the basis of a power grid section data set, and at least one power grid forward-looking scenario series is generated; and in an online operation stage of the power grid, a power grid forward-looking operation scenario series is obtained on the basis of preset risk assessment indexes and current power grid operation scenario information, enhancing the accuracy and timeliness of power grid operation risk assessment.

This invention discloses a meteorological signal spectral moment estimation method based on B-ELM. The method includes the following steps: acquiring meteorological pulse data; converting the meteorological pulse data into a meteorological power spectrum using the average corrected periodogram method; using the meteorological power spectrum as the input of the network training samples and the spectral moments as the output of the network training samples to construct the training samples; randomly setting the parameters of the hidden nodes, determining the number of input layer nodes and the number of output layer nodes, and selecting the activation function; constructing a prediction model based on the B-ELM algorithm to obtain the optimal meteorological signal radial velocity and spectral width prediction model; substituting the training samples, the set number of input layer nodes, and the number of output layer nodes into the obtained optimal meteorological signal radial velocity and spectral width prediction model to obtain the radial velocity and spectral width of the meteorological signal. This method has a fast convergence speed, small error, improves the accuracy of meteorological signal spectral moment estimation, and has low computational complexity, which is beneficial for engineering implementation.

This invention discloses a rapid stability assessment method for phosphogypsum slopes based on Extreme Learning Machine (ELM), belonging to the field of geotechnical engineering safety assessment technology. The method first constructs a t-Copula joint distribution model of phosphogypsum mechanical parameters, performs Bayesian updates and field adaptive calibration, and generates a synthetic training dataset through joint sampling. Next, it trains an ELM, using a small amount of field data to perform regularized incremental transfer correction on the output weights. Finally, based on the joint distribution model, the parameters are transformed to a standard normal space, and the HL-RF iteration driven by the ELM analytical gradient is used to solve for the most likely instability point, simultaneously outputting indicators such as the safety factor. This invention solves the problem that phosphogypsum stockpiles cannot quickly provide safety margin assessment results considering joint degradation paths when parameters are strongly correlated and data is scarce; the end-to-end computation time is less than 20 milliseconds.

This invention relates to the fields of machine learning and tooth surface waviness prediction. It discloses a method, apparatus, and device for predicting the waviness of tooth surfaces ground by worm gear grinding. The method includes: acquiring vibration signals from the grinding wheel spindle in three directions during worm gear grinding of helical gears and annotating them with semantic information to obtain an annotated dataset; constructing a tooth surface waviness prediction model based on the Goose optimization algorithm and optimizing an Extreme Learning Machine (ELM); inputting the vibration signals from the annotated dataset into the tooth surface waviness prediction model, and iteratively optimizing the input weights and biases of the ELM using a parameter optimization module to obtain a trained tooth surface waviness prediction model; and real-time acquiring the vibration signals of the workpiece being processed, extracting features from the real-time vibration signals using a feature extraction module, and then inputting the extracted features into the trained tooth surface waviness prediction model to obtain the ghost-order amplitude of the tooth surface waviness. This invention can improve the accuracy and efficiency of waviness prediction for worm gear grinding.

This application discloses a customer identification method, apparatus, and electronic device based on extreme domain adaptation. The method includes: acquiring first customer data corresponding to a first billing period and acquiring second customer data corresponding to a second billing period; wherein the first billing period is prior to the second billing period; constructing a target migration matrix based on Extreme Learning Machine (ELM), and updating the parameters of the target migration matrix according to the first and second customer data; acquiring third customer data corresponding to the second billing period, and determining the customer identification result corresponding to the third customer data through the parameter-updated target migration matrix. Implementing this application embodiment can improve the migration efficiency between customer identification models for new and old billing periods based on Extreme Learning Machine (ELM), thereby effectively and reliably achieving domain adaptation of the new model migrated and constructed under the new billing period, which is beneficial to improving the accuracy of customer identification based on the new model.

This invention discloses a data-driven adaptive control method for energy-saving traction of urban rail transit trains. First, based on the principle of a kernel limit learning machine autoencoder, a measurement data feature model is constructed by scaling the urban rail transit train operation data. Then, feature vectors of offline optimization results and online measurement data are obtained. Based on the measurement data feature model and model-free adaptive control, a recursive adaptive control law for online following of the train's energy-saving curve is established. Next, the adaptive control law for online following of the energy-saving traction curve based on data feature control is obtained. Finally, the process control variables for online following of the energy-saving traction curve are obtained. This invention introduces kernel limit learning machine and model-free adaptive control into the online following process of the energy-saving traction curve, using data feature control and transfer as the main line, and obtains the adaptive control law for online following of the energy-saving traction curve based on deep data-driven methods, thereby improving the overall stability, adaptability, and control accuracy of online following of the energy-saving traction curve.

The application discloses a kind of high density pool water treatment dosing quantity prediction method, in method, the water inflow of high density pool, water inflow turbidity, water inflow pH, water inflow temperature, raw water conductivity, raw water dissolved oxygen, raw water COD (chemical oxygen demand), raw water ammonia nitrogen, pre-coagulation dosing flow are collected and recorded, data are analyzed and pretreated to analyze the correlation of characteristic and target data Test and similarity test, predict via PCA_LSTM_ELM residual, and compare its error range with true value, long short memory neural network LSTM is used to predict extreme learning machine ELM residual value, use the first 50% of ELM residual value as test set, use the last 50% of ELM residual value as validation set, calculate and determine to obtain final prediction value, compare final prediction value with true value in this time interval.

This invention relates to the field of industrial automation testing technology, specifically disclosing a method for online intelligent detection and dynamic compensation of pH for high-concentration organic solvents. Based on only collecting the original pH reading and real-time temperature, the method reconstructs the solvent concentration, dielectric constant, and dynamic response factor, forming a five-dimensional feature vector with the aforementioned two measured values. This feature vector is input into a support vector regression or online sequential extreme learning machine model that dynamically switches according to the sample size, outputting the corrected true pH value to replace the original reading in closed-loop feedback control. Simultaneously, an incremental learning mechanism is built-in. When the residual between the model's predicted value and the offline test value exceeds the allowable range, the model parameters are automatically updated with new samples. Furthermore, by performing a cumulative sum test on the corrected residual sequence, the method actively senses electrode aging drift, requests offline testing in a targeted manner, and updates the model parameters online using a recursive least squares algorithm, achieving pH measurement without the need for additional hardware environmental sensing.

This invention relates to the fields of deep learning and public transportation technology, and discloses a method for differentiated and precise deployment of shared bicycles at various stations based on Extreme Learning Machine (ELM). First, an ELM prediction model is constructed. Then, data on the number of shared bicycles at each station in the target area within a specified time interval is collected, and the actual usage of shared bicycles at each station within that time interval is calculated to form a historical dataset. The historical dataset from day t (N+1 days prior) is normalized and formatted into a three-dimensional array {N+1,7,1}, which is then input into the ELM prediction model for training to predict the demand for shared bicycles at each station on day t+1 within the corresponding time interval. When the total predicted demand exceeds the threshold of the total number of schedulable bicycles in the target area, the total number to be reduced is calculated, resulting in the actual number of shared bicycles deployed at each station. This invention proposes using a deep learning model to accurately predict the demand for shared bicycles at each station from a time perspective, thereby scientifically guiding vehicle deployment.

This invention discloses an offline diagnostic method for nonlinear feature compression, initial screening, and hierarchical evaluation of sampling channels of substation voltage transformers. Addressing the offline risk assessment needs of substation voltage transformer sampling channels, this method first constructs a unified output health discriminant factor by combining kernel principal component analysis and entropy weighting, reducing redundancy and enhancing cross-site robustness. An extreme learning machine is introduced to perform offline initial screening of sample vectors, focusing on abnormal sampling channel segments that meet the anomaly judgment criteria to improve batch processing efficiency. Based on fuzzy membership functions and a monotonic mapping relationship between predefined risk levels and scores, multi-level risk classification of initial screening candidate samples is achieved. Furthermore, trend quantities are combined to correct risk evolution for risk decision-making, ensuring that the score remains stable near the boundary zone and is more sensitive to abrupt changes. This method effectively improves the offline data fault diagnosis and location capabilities of voltage measurement links under conditions of multiple disturbances and non-stationary operation.

The application is a lithium ion battery health state estimation method based on multi-objective optimization, first, the charging and discharging cycle data of the lithium ion battery is collected and pretreated; then, a plurality of health characteristics including the constant current charging time and the polynomial coefficient of the constant current charging voltage-time curve fitted by the Logit model are extracted from the charging and discharging cycle data; finally, the health state estimation model is established based on the extreme learning machine neural network, the health state estimation model is trained by using the charging and discharging cycle data of the lithium ion battery, in the training process, the health state estimation model is optimized by the multi-objective optimization algorithm based on the non-dominated sorting and the crowded distance, and the optimized health state estimation model is used for the estimation of the lithium ion battery health state. The polynomial coefficient extracted by the application reflects the nonlinear aging law in the constant current charging process, indirectly represents the battery degradation, the constant current charging time directly represents the battery degradation, and the model optimization considers the prediction accuracy and stability of the model.

This invention provides a method and system for online control of machining accuracy based on multi-sensor data fusion. The method includes the acquisition and preprocessing of a multimodal heterogeneous sensor dataset; weighted fusion using a weighted average method; extraction of aggregated feature vectors and time-frequency feature vectors using Local Feature Mean Aggregation (LFMA) and Wavelet Packet Transform (WPT); the introduction of a Fisher discriminant fusion algorithm to fuse the initial feature vectors; and prediction of machining error type and magnitude using an improved Extreme Learning Machine (ELM) model. Based on the error prediction results, an adaptive control strategy is constructed to control the machine tool online. This invention employs a multimodal data acquisition network using multiple types of sensors to acquire sensor parameters in real time, including dimensions, vibration, cutting force, temperature, and tool wear. Simultaneously, through a three-level hierarchical fusion architecture of data fusion, feature fusion, and decision fusion, the accuracy of error prediction is improved, thereby ensuring the accuracy of control.

The application provides a power grid bottom guarantee communication fault pre-judgment method based on an intelligent algorithm, and belongs to the technical field of power grid bottom guarantee communication.The application carries out wavelet packet decomposition and singular value decomposition noise reduction processing on an initial state parameter set through a dynamic topological graph structure, adopts principal component analysis and local linear embedding manifold learning for feature dimension reduction, inputs a compact feature vector into a graph attention time sequence fusion prediction model to output a fault risk score, uses an adaptive integrated extreme learning machine for secondary verification to obtain a refined fault probability distribution, establishes a double-layer game optimization model to solve an optimal communication resource allocation scheme, generates virtual samples to update model parameters according to the fault probability distribution and type judgment result, adjusts node transmission power and channel allocation strategies according to an early warning level, and triggers communication path reconstruction, so that the technical problem of low accuracy of power grid bottom guarantee communication fault pre-judgment is solved.

This invention discloses a UAV carrier aggregation resource allocation method based on Extreme Learning Machine (ELM), belonging to the field of UAV communication technology. The method includes the following steps: S1, preprocessing the sub-channels in the UAV communication system to filter out the available sub-channel set; S2, allocating sub-channels to users according to channel quality ranking; S3, performing power allocation for users in the initially allocated sub-channels based on an ELM network; S4, dynamically adjusting resources for users whose minimum rate requirements are not met after power allocation. This invention has clear logic, low computational complexity, and fast response speed, making it suitable for resource-constrained UAV communication systems and enabling efficient and fair resource allocation in dynamic environments.

This invention relates to the field of carbon dioxide emission estimation technology, and in particular to a method and system for estimating carbon dioxide emissions from non-road mobile machinery. The method is based on the NONROAD non-road mobile source emission model framework and combines it with the carbon balance method to construct an initial calculation model for carbon dioxide emission factors. It identifies potential influencing factors of carbon dioxide emissions from non-road mobile machinery and filters key influencing factors through correlation analysis and rough set theory to form a sample dataset. Furthermore, it constructs a deep extreme learning machine model based on an autoencoder, and after training on a training set and verification on a test set, it achieves the estimation of carbon dioxide emissions from non-road mobile machinery. By combining mechanistic analysis with a data-driven approach, it can improve the accuracy, stability, and adaptability of emission estimation, better meeting the actual needs of refined accounting of carbon dioxide emissions from non-road mobile machinery under complex operating conditions.