70results about How to "Avoid fit problems" patented technology

The invention discloses an LLM capability enhancement method of a hybrid enhancement strategy, and particularly relates to the technical field of natural language processing. The method comprises the following steps: obtaining corpus data and external literature data of a target application field, carrying out denoising, word segmentation and semantic indexing processing, and establishing a searchable heterogeneous knowledge base; secondly, classifying the corpus data by adopting a topic analysis and word vector clustering method to form a basic corpus set; performing multi-round parameter updating on the large language model by using a domain fine tuning algorithm based on the basic corpus set, and setting a fusion threshold value between a new parameter and an original parameter; and retrieving related literature fragments from the heterogeneous knowledge base through a matching algorithm, screening content conforming to field correlation by utilizing semantic similarity evaluation, fusing the content with the fine-tuned model, processing a fusion result by adopting a context fusion algorithm and an entity linking method, completing syntactic analysis, and generating a precise solution oriented to the target application field.

The invention discloses an intelligent customer service question answering method, device and equipment based on a large language model and a knowledge graph, and a medium. The method comprises the following steps: constructing the knowledge graph of a target domain; performing LoRA fine tuning on the preset large language model; inputting the target question into a preset large language model to extract a keyword set corresponding to the target question; constructing a plurality of query statements according to the keyword set; performing retrieval in the knowledge graph of the target domain according to the plurality of query statements, obtaining filling content, combining the filling content with a target question of the target user, and constructing a question text of the target user; and inputting the question text of the target user into a preset large language model to obtain a target reply of the target question. The invention belongs to the field of intelligent reply. By constructing the question text of the target user, the model can be helped to better understand the question, and the model can accurately hit the question.

The invention relates to the technical field of network advertisements, in particular to an advertisement data management system based on market trend analysis, which comprises a data acquisition module, an analysis module, a prediction module, a conversion rate calculation module and an adjustment module. The advertisement effect is reflected by acquiring click and purchase data after advertisement putting, the change trend of user behaviors is accurately recognized through the analysis module, and the user change rate is calculated, so that data support is provided for an advertisement putting strategy, and the advertisement putting strategy is more accurate by analyzing historical data, predicting the future purchase trend and comparing with predicted data in real time. The conversion rate change of the advertisement is accurately evaluated, a basis is provided for optimizing the advertisement effect, whether the advertisement putting duration needs to be adjusted or not is judged according to the comprehensive analysis of the user change rate and the conversion rate change rate, efficient utilization of advertisement resources is ensured, the accuracy and flexibility of advertisement putting are remarkably improved, the conversion rate and the return on investment of the advertisement are facilitated, and the user experience is improved. And continuous optimization and improvement of the advertisement effect are realized.

The invention discloses a rolling bearing fault diagnosis method and system based on multi-channel adaptive fusion, and the method comprises the steps: obtaining a one-dimensional vibration acceleration signal, carrying out the normalization, and converting the one-dimensional vibration acceleration signal into a two-dimensional gray-scale map; inputting the two-dimensional grey-scale map into a convolutional neural network model for processing, specifically comprising the following steps: a multi-scale feature extraction module extracts feature information of the two-dimensional grey-scale map, adaptively weights the importance weight of multi-channel information, and outputs a first multi-scale feature map; a double-attention cavity convolution module extracts the spatial position and channel information of the first multi-scale feature map, and outputs a second feature map; and the residual dense module performs deep feature extraction on the second feature map through residual connection and dense connection, and outputs a rolling bearing fault diagnosis classification result. According to the method, the problems of low model fault diagnosis effect and poor generalization ability caused by strong noise interference and large load change of the rolling bearing can be solved.

The invention relates to the technical field of artificial intelligence, and discloses a multi-modal model training fine-tuning optimization method based on artificial intelligence, which comprises the following specific steps: preparing and preprocessing data, collecting a large-scale multi-modal data set, constructing an image text data set containing one million high-quality images and text description accurately corresponding to the images, and performing fine-tuning optimization on the image text data set. According to the method, data are subjected to comprehensive and meticulous cleaning operation, noise data and mistakenly labeled data in the data are strictly removed, so that the accuracy and reliability of the data are ensured, and the effective utilization rate of the data can be increased by at least 20% through data cleaning; wherein the image data is screened, blurred, low-resolution and unclear-content images are removed, it is ensured that each image can clearly convey specific visual information, and text description, grammar error checking, spelling error and semantic ambiguity parts are corrected or deleted.

The invention relates to a method, a device and equipment for constructing an elastic network regression model of a battery retention rate. The method comprises the following steps: acquiring sample data of a sample battery and an initial elastic network regression model; inputting the sample data into the initial elastic network regression model, and outputting a prediction result; determining the target loss of the initial elastic network regression model according to the prediction result; the target loss comprises an error term and a regularization term; and training the initial elastic network regression model according to the target loss to obtain a trained elastic network regression model. According to the method, penalty for the regression coefficient is introduced into the regularization item, the complexity of the elastic network regression model is controlled, and the elastic network regression model is prevented from being over-fitted, so that the training efficiency of the initial elastic network regression model is improved, and due to the fact that the number of sample data needed for training the regression model is small, the training efficiency of the regression model is improved. And the complexity of training the initial elastic network regression model is reduced.

The invention belongs to the technical field of artificial intelligence analysis, and particularly relates to a military target behavior analysis agent test method and equipment. The method comprises the following steps: step 1, acquiring multi-source data of a military target, and performing data preprocessing on each category of data in the multi-source data; 2, extracting tensor features of the input tensor; defining a multi-layer intention hierarchical structure based on the tensor features; 3, defining a likelihood function of each intention layer; modeling a prior probability using a recurrent neural network; calculating a posterior probability by using a Bayesian rule; when the number of times of iterative execution reaches the set number of times, executing the step 4: performing intention reasoning by using approximate posterior distribution, and calculating the uncertainty of intention reasoning. According to the method, the precision, the real-time performance, the interpretability and the reliability of military target behavior analysis are remarkably improved.

The invention discloses a forest fire detection and identification method, and the method comprises the steps: 1) replacing a Neck layer of a YOLOv8 model with a Gather-Distribate module, and replacing a detection head of the YOLOv8 model with an SE-ResNeXt detection head, and obtaining a forest fire detection and identification model; 2) training the forest fire detection and identification model by using a forest fire data set; and 3) inputting a to-be-identified forest fire image into the trained forest fire detection and identification model to obtain a detection and identification result. According to the invention, the Gather-Distribate module is introduced, so that the feature extraction capability of forest fires with different sizes is improved, small target features can be better captured, and the detection precision of the model is improved; an SE-ResNeXt detection head is introduced and an SE module is fused, so that the calculation cost is saved, and the forest fire feature extraction capability of the detection head is enhanced.

The invention relates to the technical field of machine learning and software development, in particular to a software development system based on a machine learning algorithm, which comprises an acquisition processing unit, a model training unit, an interaction prediction unit and an optimization feedback unit, and is characterized in that the acquisition processing unit constructs a database, extracts key elements such as variable names and converts the key elements into vectors; the model training unit adopts recurrent neural network hierarchical training and gradually introduces intermediate and complex structure codes and learns logic relation and semantic relation from code snippets related to a basic grammar structure, the interaction prediction unit screens and extracts the code snippets according to code function module labels and historical use frequency, the code snippets are input into a model after vector conversion, and the model is used for prediction. A probability weighting algorithm is used for predicting subsequent fragments, an optimization feedback unit calculates the similarity between predicted fragments and actual fragments, an optimization mechanism is automatically triggered if a threshold value is not reached, parameters are adjusted by utilizing reinforcement learning, and prediction accuracy is used as a key part of a reward function so as to improve the system performance.

The invention provides a self-supervised medical image segmentation method and system based on structural relation modeling, and the method comprises the steps: carrying out the modeling of a complex relation among nodes, sub-graphs and a whole graph, thereby comprehensively capturing the structural information of the graph; by constructing and optimizing the structural relation model of the graph, the performance of graph representation learning is improved, the method is particularly suitable for graph data with a complex topological structure and non-uniform distribution, and the segmentation effect of the medical image is improved by modeling the organization structure relation in the medical image.

The invention provides a wheel set anomaly detection method and system based on a generative adversarial network, and the method comprises the steps: obtaining wheel set data: obtaining the wheel set data; a data fusion and classification step of fusing the wheel set data to generate a data matrix and classifying the data matrix; a generative adversarial network construction step: according to the normal wheel set data set, training generator output of the generative adversarial network to obtain a pseudo normal wheel set data set, and inputting the normal wheel set data set and the pseudo normal wheel set data set into a discriminator of the generative adversarial network to train and output a similarity rate; a wheel set abnormity judgment step: inputting the real-time wheel set data into the generative adversarial network, and outputting wheel set data abnormity or wheel set data normality according to a similarity rate calculation and similarity threshold comparison result; the method has the advantages that the abnormal condition of train wheel set eccentric wear can be judged, the dependence on abnormal data samples is reduced, and the accuracy and the detection efficiency of train wheel set anomaly detection are improved.

The invention discloses a motor multi-objective optimization method and device based on a combined proxy model. The method comprises the following steps: training each meta-proxy model by adopting a one-leaving cross validation method; determining a weight factor of each element agent model based on the error correlation matrix, and combining the element agent models into an optimized combination model according to the weight factors of the element agent models; during optimization, processing the extracted data of each to-be-optimized variable by using the optimization combination model to obtain optimization target data, and performing Sobol sensitivity analysis on each to-be-optimized variable by using the optimization target data to obtain a first-order influence index corresponding to each to-be-optimized variable and each total effect index; on the basis of the first-order influence index and the total effect index, hierarchically dividing each to-be-optimized variable; optimizing each to-be-optimized variable layer by layer by using the combinatorial optimization model; and selecting a final optimization result of each to-be-optimized variable of the target motor from the optimization result of the last layer.

The invention discloses an attention neural network-based personnel ability automatic evaluation method and system. The method comprises the following steps of: dividing an ability evaluation data set # imgabs0 # into a training set # imgabs1 #, a verification set # imgabs2 # and a test set # imgabs3 #; setting a plurality of different neural network configurations to form a configuration set # imgabs4 #; the obtained hyper-parameter # imgabs5 # is selected, and a neural network model # imgabs6 # is initialized; using a training set # imgabs7 # to train a neural network model # imgabs8 # obtained through initialization; after a # imgabs9 # round is carried out, the neural network model trained in the # imgabs10 # round is recorded as # imgabs11 #, a verification set # imgabs12 # is used for evaluating the trained model, and the neural network model and the evaluation result after each round of training are stored; the model obtaining the highest evaluation index on the verification set in N rounds is selected to be tested and stored on a test set # imgabs13 #; and until all hyper-parameters are used for initializing the neural network models and tested by a test set # imgabs14 #, deploying the neural network model with the highest model evaluation result for personnel ability grade evaluation work. And the requirement on multi-dimensional personnel ability portraits in a personnel evaluation scene is met.

The invention belongs to the technical field of vehicle motor controllers, and particularly relates to a calibration method of a vehicle motor controller ampere meter, which comprises the following steps: S1, carrying out Id-Iq current combination traversal scanning at a rotating speed lower than a reference rotating speed to obtain torques and voltage utilization rates corresponding to various current combinations at each rotating speed lower than the reference rotating speed; s2, dividing the current combinations with the same torque into one group and performing fitting to obtain an equal-torque curve; s3, traversing the equal-torque curve of each torque at each rotating speed according to a preset traversing process, and obtaining a current combination capable of enabling the actual output current Is to be minimum or the voltage utilization rate to be maximum on each equal-torque curve; and S4, summarizing the obtained current combination of the equal torque curve of each torque at each rotating speed, and taking the current combination as a final obtained ampere meter. The method can improve the automation degree and accuracy of calibration of the ampere meter of the vehicle-mounted motor controller, reduces errors caused by manual operation, and ensures that the performance of the ampere meter is recovered to the optimal state.

The invention provides a photovoltaic output prediction method considering haze / dust factors. The method comprises the following steps: firstly, preprocessing data, screening out abnormal data and invalid data, and complementing missing data; the method comprises the following steps: firstly, analyzing haze / sand dust data, then classifying weather data by using a decision tree algorithm, then deeply understanding an internal rule and an operation mechanism by analyzing a haze / sand dust formation mechanism and a photovoltaic power generation principle, determining main characteristics, and then firstly establishing a haze / sand dust influence model 1 and a photovoltaic power generation power prediction model 2 for the haze / sand dust data; the model 1 obtains influence parameters of haze / dust factors on the photovoltaic power generation power, and the model 2 outputs a predicted value of the photovoltaic power generation power. The advantages of mechanism analysis and a data driving model are combined, and the accuracy and reliability of the model are improved.

The invention discloses a small sample face recognition method based on prototype calibration and adaptive fine tuning, which comprises the following steps: 1) pre-training an encoder on a large-scale face database, capturing rich semantic information and class distribution, constructing an evaluation set, and ensuring that training identity information is not overlapped with the constructed evaluation set; 2) initializing the weight of the classifier through a weight imprinting method, ensuring smooth transition, and inducing a discrimination structure of the encoder; 3) calculating the cosine similarity between the fine-tuning prototype and the pre-training prototype, fusing the pre-training information through weighted average, and calibrating the fine-tuning prototype with deviation; 4) introducing an adaptive interval loss function, adjusting the penalty intensity of the sample, and performing selective fine adjustment on the BatchNorm layer and the classifier weight; 5) using the domain perception similarity NAC to calibrate unknown category rejection, enhancing the similarity relation with the domain sample, and obtaining the final matching score. The method can improve the recognition precision and rejection capability of the model in a few-sample learning scene.

The invention provides a redried tobacco leaf mildewing detection method based on mixed attention mechanism enhancement and a 3D residual CNN network, and the method comprises the steps: collecting redried tobacco leaf samples, and preparing mildewed tobacco leaf samples in different periods; hyper-spectral image data of the mildewed and redried tobacco leaf samples are collected through hyper-spectral image collecting equipment; the collected hyperspectral image data is corrected; establishing a mildewed tobacco leaf data set according to the corrected hyperspectral image data; an MA3D-ResNet model is constructed; carrying out iterative training on the model by utilizing the training set and the verification set, and storing model parameters with the highest accuracy on the verification set within a specified number of iterations; model parameters which are obtained through training and have the highest accuracy rate are imported into the model, test set data are input, evaluation indexes are calculated according to a model output result, the corrected hyperspectral image of the redried tobacco leaves to be detected is input into the tested model, and a mildew detection result is obtained. The method can effectively improve the detection precision of the mildewing degree of tobacco leaves with extremely similar spectral characteristics, especially early mildewing tobacco leaves.

The application discloses a lining cloth dyeing control method and system based on artificial intelligence, and is used for the control field, and the method comprises the following steps: real-time monitoring of multi-source data in the lining cloth dye vat, capturing the video stream inside the lining cloth dye vat; using a Gaussian mixture variational autoencoder model to extract features from the multi-source data, obtaining multi-source feature data; obtaining the changing moment of the fabric state in the video stream, and extracting the fabric state change feature data according to the observed fabric color change and the heterogeneity of the dye distribution in the video stream; fusion of multi-source feature data and fabric state change feature data; introducing a model predictive control layer, predicting the result to adjust the dyeing parameters in real time; establishing an intelligent feedback mechanism. The Gaussian mixture variational autoencoder model is trained and optimized through the small-batch stochastic gradient descent and back propagation algorithm, important features are learned and captured from the data.

The utility model discloses an assembly for manufacturing a tire section mold, which comprises a pattern sheet assembly which is integrally formed by 3D printing; the pattern sheet assembly comprises a framework, a continuous sheet-shaped pattern sheet main body with a set thickness and a plurality of sheet-shaped blades with a set thickness, the pattern sheet main body extends upwards from the peripheral surface of the framework, and the plurality of sheet-shaped blades extend from the pattern sheet main body to the two sides and are of an integrated structure through 3D printing. The surface welding procedure of the matching surface of the combination part in the subsequent procedure can be omitted, and the subsequent welding procedure and polishing procedure can be reduced or omitted to a great extent; the problem of matching of the assembly parts of a plurality of original parts is avoided, and displacement caused by deformation of the assembly structures of the adjacent parts due to temperature change in the casting production process is reduced; in addition, corresponding procedures are reduced, so that the manufacturing period can be shortened to a great extent.

The present invention discloses a method for predicting boiler full-load NOx emissions using a channel-selective convolutional neural network, and belongs to the technical field of NOx emissions from coal-fired boilers. The historical data of the boiler combustion operation during the selected historical time period is extracted as needed, and then the boiler operation data in the historical data is preprocessed; the preprocessed boiler historical data is divided into a 4-D tensor by a sliding window method, and then the boiler historical data is input into a channel-selective convolutional neural network CS-CNN model in the form of a tensor for training. The CS-CNN model will automatically identify the channels ranked in the top 80% of importance according to the operating condition changes of the input boiler, release the hyperparameters of the channels with low importance, and maximize the use of the limited parameter resources of the CS-CNN model, so as to achieve the purpose of improving the prediction accuracy of the CS-CNN model. The method is simple, has high prediction accuracy, and fast prediction speed.

This invention discloses a method and system for large-scale urban village extraction based on a token mask mechanism, relating to the technical field of remote sensing geographic information systems. The method includes: acquiring a high-resolution remote sensing image dataset; labeling urban village areas to obtain an urban village labeled dataset; performing data augmentation on the urban village labeled dataset to obtain an augmented urban village labeled dataset; iteratively training a constructed urban village extraction model; setting training parameters; updating the model parameters of the urban village extraction model to obtain a trained urban village extraction model; acquiring a high-resolution remote sensing image of the area to be detected; inputting the trained urban village extraction model; and extracting the urban village areas within the area to be detected. This invention can quickly and accurately identify urban village areas from high-resolution remote sensing images, with high accuracy and efficiency, wide application range, and realizes high-precision and robust large-scale urban village mapping applications.

The application discloses a new energy station electric quantity prediction method and system based on a global-local double-layer prediction architecture, and aims at solving problems of insufficient monthly electric quantity prediction data of new energy stations, large medium and long term weather prediction errors, and low prediction precision of traditional experience methods, adopts a global-local double-layer model architecture, a XGBoost global prediction model is constructed by using all station data in the upper layer, common rules such as seasons, weather, installed capacity, shutdown plan and the like are mined, the global model output is taken as an enhanced feature, a XGBoost local model is trained in combination with single station data in the lower layer, and individual characteristics such as station equipment efficiency and operation and maintenance level are adapted, and a holiday adjustment factor is introduced to post-process and correct the prediction result. The application effectively solves the small sample modeling problem commonly existing in monthly electric quantity prediction, and significantly improves the monthly electric quantity prediction accuracy of wind power stations and photovoltaic stations.

The embodiments of the present disclosure disclose a behavior recognition method, device, and equipment based on user behavior prediction model training. A specific implementation of the method includes: obtaining a sample user browsing behavior information set, wherein the sample user browsing behavior information includes: a behavior response label, a behavior feedback label, and a user feature information set; training a behavior response label classification model and a behavior feedback label classification model based on the sample user browsing behavior information set; determining a user behavior causal feature set based on the user feature sequence corresponding to the behavior response label classification model and the user feature sequence corresponding to the behavior feedback label classification model; training an initial user behavior prediction model based on the user behavior causal feature set and the sample user browsing behavior information set to obtain a user behavior prediction model. This implementation avoids overfitting of the model and improves the accuracy of the recognition rate of the trained model.

The application discloses a kind of network type flexible direct system virtual impedance coordination's adaptive current differential protection method, system, equipment and medium, it is related to flexible direct current transmission relay protection technical field.At present, in network type flexible direct system, the problem that sensitivity is insufficient due to the limited output current of converter station.This application includes synchronously collecting the current phasor of the protected line both ends and the effective value of converter station output current;According to the effective value of output current, the virtual resistance parameter of converter station control inner loop is adaptively adjusted;Based on virtual resistance parameter, the adaptive braking coefficient is calculated, so that the braking coefficient is inversely proportional to virtual resistance;Differential current and braking current are calculated, whether differential current is greater than the product of braking coefficient and braking current and action threshold value at the same time is judged, if meet, then determine intra-zone fault and trip.This technical scheme realizes the adaptive improvement of sensitivity at fault, reliable and no misoperation at normal time, without additional judgment link, fast response and easy to realize.

The utility model provides a double-inlet full-specification cigarette stacking device which comprises a first cigarette inlet conveying line, a second cigarette inlet conveying line, a film wrapping machine and a thermal shrinkage furnace, a first cigarette inlet belt is arranged on the first cigarette inlet conveying line, and a first lifting cigarette inlet mechanism and a second lifting cigarette inlet mechanism are correspondingly arranged on the first cigarette inlet belt. A cigarette stacking device is correspondingly arranged at the position of a cigarette outlet of the first cigarette inlet belt, and a cigarette pushing device is arranged on a cigarette receiving structure of the cigarette stacking device. The cigarette stacking machine has the advantages that the problem that in the prior art, a cigarette stacking machine can place stacked cigarette boxes into a conveying belt to be conveyed only by depending on other equipment is solved. In the scheme, a plurality of cigarette receiving bin structures are adopted in the direct extraction type cigarette stacking device, the direct extraction type cigarette stacking device can be directly in butt joint with the conveying belt to arrange the cigarette boxes into regular cigarette box layers, and therefore the direct extraction type cigarette stacking device is suitable for the special-shaped cigarette boxes, and the problem that the cigarette boxes are damaged due to the fact that the cigarette boxes are not easy to arrange in an existing cigarette stacking machine and are clamped or fall off is solved.

The invention discloses an intelligent grade classification method based on ore image features, and belongs to the technical field of machine vision, and the method comprises the steps: constructing an ore image data set, dividing the ore image data set into a training set, a verification set and a test set, and carrying out the preprocessing operation of the ore image data in each subset; constructing a residual neural network model of a multi-branch structure as an ore grade detection model; training the ore grade detection model based on the ore image data set; and detecting the grade of the to-be-detected ore by using the trained ore grade detection model. According to the method, the ore grade classification model is constructed based on the multi-branch structure and the residual error structure, the intelligent degree of the mine in the mineral separation link can be improved, guidance is provided for production and ore blending, and the method has important practical application value and significance for improving the operation efficiency of the mine.

The invention relates to a dominant sample-based product appearance defect detection method, which comprises the following steps of: evaluating the importance degree of a sample on the basis of the intersection-to-union ratio IoU between the sample and a real target, and sequencing the importance degree of the sample; the loss function is adjusted according to the importance degree of the samples, and a higher loss function weight is distributed to the samples with the high importance degree. A high-IoU sample is used as a dominant sample. By modifying the loss function, a dominant sample with higher intersection-to-union ratio than IoU can obtain more attention of the model. Because the high IoU sample is closer to a real target, the contained information quality is higher, the noise is less, and the features learned from the samples by the model are more reliable. Therefore, compared with the prior art, more attention is paid to'difficult samples', attention is put on the high IoU samples, so that the model can learn important features more quickly and effectively, and the training efficiency is improved.

This application relates to the field of robotics, and in particular to a training data acquisition device for robot gripper control. The training data acquisition device mainly comprises four core modules: the acquisition device body, an image acquisition module (51), a motion detection module (52), and a data processing module (53). In use, the operator holds the device, aligns the mechanical finger (2) with the object to be grasped, and presses the handle (3) on the acquisition device body to control the mechanical finger (2) to close and grasp the object. During this process, the image acquisition module (51) records visual information, and the motion detection module (52) records the degree of finger opening and closing. The data processing module (53) aggregates these two types of heterogeneous data. Thus, the hand-eye coordination process of a real robot can be simulated at extremely low hardware cost, and key matching data is provided for robot imitation learning.

The present disclosure provides a dual-stream modulation signal classification method and system based on multi-feature fusion, which relates to the technical field of radio modulation signal classification, obtains I / Q data of the original modulation signal, and divides the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream; inputs the high signal-to-noise ratio stream and the low signal-to-noise ratio stream into a dual-stream CABLD model, the dual-stream CABLD model includes two CABLD classification channels, each CABLD classification channel includes a first CABLD model and a second CABLD model, after the high signal-to-noise ratio stream and the low signal-to-noise ratio stream are input into the dual-stream CABLD model, they enter a CABLD classification channel respectively, firstly perform multi-signal classification through the first CABLD model, obtain an AW-type signal synthesized by AM‑DSB and WBFM signals, then extract the frequency domain amplitude and phase features of the AW-type signal respectively and input them into the second CABLD model, perform binary classification on the AM‑DSB and WBFM signals in the AW-type signal, and obtain the classification results of the AM‑DSB and WBFM signals. The present disclosure improves the accuracy of signal classification.

This utility model belongs to the field of trolley technology, and specifically relates to a trolley that facilitates straight-line travel. It includes a balance scooter and a trolley body, connected by a connecting frame. The connecting frame is detachably connected to the trolley body and rotatably connected to the balance scooter. The balance scooter has a positioning groove, and the connecting frame has a positioning element and an elastic element. The elastic element drives the positioning element to engage with the positioning groove, fixing the balance scooter and the connecting frame relatively, thus allowing the trolley to maintain straight-line travel or maintain a certain angle of rotation. Under external force, the positioning element can move relative to the positioning groove, allowing the balance scooter and the connecting frame to rotate relative to each other, thus enabling the trolley to turn flexibly. The purpose of this utility model is to provide a user-friendly trolley that facilitates straight-line travel.