Automobile service demand prediction method based on big data analysis

Abstract

The invention discloses an automobile service demand prediction method based on big data analysis, and the method comprises the steps: an image collection module which carries out the data cleaning and preprocessing through collecting the multi-dimensional data, such as vehicle owner behaviors, vehicle sensors, maintenance history, market trends, social media, weather and policy changes, and carries out the data cleaning and preprocessing through a machine learning algorithm; key features influencing automobile service requirements are identified and adjusted through deep learning, and new factors such as new automobile types, consumer behaviors and policy changes are dynamically tracked; in combination with various models such as regression analysis, a decision tree, a support vector machine and a deep neural network, the prediction accuracy is improved through integrated learning, and online learning and real-time updating are realized by using an incremental learning algorithm; an external event sensing module is introduced to monitor market, economic and policy changes in real time, and a prediction model is automatically adjusted; the model is continuously optimized through a feedback mechanism, demand changes are analyzed in combination with an interpretability analysis tool and causal inference, and simple decision support information is provided.

Description

Technical Field

[0001] This invention belongs to the field of automotive service demand forecasting based on big data analysis, and specifically relates to a method for forecasting automotive service demand based on big data analysis. Background Technology

[0002] The big data analytics-based automotive service demand forecasting method is a technical approach that predicts future automotive service demand by analyzing large amounts of automotive-related data. It collects relevant data from multiple sources, including vehicle maintenance records, driving behavior data, weather data, car sales data, owner feedback, market trends, and socioeconomic factors. The collected data undergoes cleaning and processing, including noise removal, missing value imputation, and data standardization, to ensure accuracy and reliability. By analyzing and mining the data, features related to automotive service demand are extracted; for example, factors such as vehicle age, mileage, usage frequency, and seasonal changes may influence service demand. Predictive models are built using machine learning and statistical modeling methods. The accuracy of the models is evaluated through cross-validation and error analysis, and the predictive capabilities are improved through parameter tuning and algorithm optimization. Based on the constructed model, future automotive service demand is predicted, helping companies develop strategies such as optimizing service arrangements, predicting maintenance demand, and managing inventory. Over time, service demand may change due to market conditions, technological advancements, and other factors; therefore, the predictive model needs continuous updates and adjustments to maintain high predictive accuracy.

[0003] However, most existing methods for forecasting automotive service demand rely on historical data for trend analysis and traditional statistical models for prediction. While these methods can reflect demand changes to some extent, they still cannot fully meet the automotive service industry's needs for accurate forecasting, dynamic adaptation, and adaptability to changes in user demand. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method for predicting automotive service demand based on big data analysis. It employs advanced technologies such as intelligent data processing, dynamic feature selection, external event perception, and incremental learning, thereby overcoming the deficiencies in existing technologies and providing a more accurate, flexible, and real-time responsive prediction solution.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] Methods for predicting automotive service demand based on big data analytics include:

[0007] Data is collected from multiple channels, including car owner behavior data, vehicle sensor data, vehicle maintenance history, market trends, social media sentiment, weather information, and policy changes. Machine learning algorithms are used to automatically clean and preprocess the data.

[0008] Based on deep learning algorithms, it automatically identifies and adjusts key features that affect the demand for automotive services, and dynamically identifies emerging factors, including new car models, changes in consumer behavior, and policy changes.

[0009] By combining multiple models such as regression analysis, decision trees, support vector machines, and deep neural networks, the accuracy and stability of predictions are improved through ensemble learning techniques. Incremental learning algorithms are also employed to enable the model to learn online and update in real time as data continues to flow in.

[0010] By introducing an external event perception module, changes in market, economic, and policy emergencies are monitored in real time, the impact of these external factors is automatically identified, and the prediction model is adjusted based on real-time information.

[0011] Establish a feedback mechanism based on real-time prediction results, continuously optimize the prediction model through car owner usage data and market response, and automatically adjust parameters based on feedback information to perform self-learning;

[0012] During the forecasting process, interpretability analysis tools are integrated to transform complex forecast results into concise decision support information, and causal inference techniques are combined to analyze the specific impact of different characteristics and external events on changes in demand.

[0013] As a preferred approach, data is collected from multiple channels, including vehicle owner behavior data, vehicle sensor data, vehicle maintenance history, market trends, social media sentiment, weather information, and policy changes. Machine learning algorithms are then used to automatically clean and preprocess the data.

[0014] For missing data, use the following formula for interpolation:

[0015] x imputed =μ+λ(x) nearest )

[0016] Among them, x imputed The missing values ​​are imputed, μ is the mean of the data, and x is the mean of the data. nearest The nearest neighbor value is the value near the missing value, and λ is the weighting coefficient.

[0017] Outliers are identified using a standard deviation-based test. Assuming the data follows a normal distribution, the formula for outliers is:

[0018]

[0019] Where z is the z-score, x is the data point, μ is the mean of the data, and σ is the standard deviation of the data. When |z|>3, the data point is considered an outlier.

[0020] To eliminate scale differences among different data features, the data is normalized. The Min-Max normalization formula is as follows:

[0021]

[0022] Among them, x norm For the normalized data, x min and x max These are the minimum and maximum values ​​of the data, respectively.

[0023] For categorical data, one-hot encoding is used to convert it into numerical form. If there is a category feature C, it is converted into a binary vector.

[0024] C one-hot =[if C=c1,then(1,0,0),if C=c2,then(0,1,0),...]

[0025] Among them, C one-hot The result of one-hot encoding for categorical data;

[0026] Smoothing the data to reduce noise:

[0027]

[0028] Among them, MA n Let x be the moving average over n time points. i This represents the value corresponding to time point i.

[0029] As a preferred approach, a method based on deep learning algorithms is used to automatically identify and adjust key features affecting automotive service demand, and to dynamically identify emerging factors, including new car models, changes in consumer behavior, and policy changes.

[0030] Assume the input data is X = [x1, x2, ..., x...]. n If the output is the predicted demand y, then the deep neural network makes the prediction using a multilayer perceptron, and its calculation formula is as follows:

[0031] y = f(W) L (f(W L1 (...f(W1X+b1)...)+b L1 )+b L )

[0032] Where f is the activation function, W i and b i Let L represent the weights and biases of the i-th layer, and L represent the number of layers in the network.

[0033] Based on the intermediate layer output after deep neural network training, feature importance analysis is used to automatically identify key features, and SHAP values ​​are used to evaluate the contribution of each input feature to the prediction result.

[0034]

[0035] Among them, SHAP(x i ) is a feature x i Contribution to the prediction results, x i To remove feature x i The input data afterwards;

[0036] Using dynamic feature learning methods (such as online learning) to identify emerging influencing factors, let X... new For datasets containing new features, the model is trained using incremental updates:

[0037]

[0038] Where, θ t Here are the model parameters after the t-th update, and η is the learning rate. Let X be the loss function. new and y new For inputting and labeling new data;

[0039] Based on time series data, recurrent neural networks or long short-term memory networks are used to learn from historical data, thereby dynamically capturing and predicting new influencing factors. The calculation formula of the LSTM model is as follows:

[0040] i t =σ(W i [h t 1,x t ]+b i ), f t =σ(W f [h t 1,x t ]+b f ), o t =σ(W o [h t 1,x t ]+b o )

[0041] c t =f t *c t 1+i t *tanh(W c [h t 1,x t ]+b c ), h t =ot *tanh(c t )

[0042] Among them, i i f t o t These are the input gate, forget gate, and output gate, respectively. t h is a memory unit. t For output, x t For the current input data, W i W f W o W c Let b be the weight matrix. i b f b o b c For bias terms;

[0043] The adaptive adjustments made by deep neural networks to new car models and external factors such as market changes can be achieved through transfer learning techniques, which involves fine-tuning existing models with new data.

[0044]

[0045] Where, θ new For the fine-tuned new model parameters, θ pre-trained X represents the parameters of the pre-trained model. new and y new For the new dataset.

[0046] As a preferred approach, combining multiple models such as regression analysis, decision trees, support vector machines, and deep neural networks, and using ensemble learning techniques to improve the accuracy and stability of predictions, and employing incremental learning algorithms to enable the model to learn online and update in real time as data continues to flow in, is as follows:

[0047] Regression analysis models are used to predict continuous demand values. The linear regression model is as follows:

[0048] y = β0 + β1x1 + β2x2 + ... + β n x n +∈

[0049] Where y is the predicted demand value, β0, β1, ..., β n Let x1, x2, ..., x be the regression coefficients. n The input features are ∈, and the error term is ∈.

[0050] Decision trees construct predictive models by recursively splitting datasets. For input data X, the output y of the decision tree is obtained through a series of decision rules:

[0051] y = DecisionTree(X) = f(X)

[0052] Where f(X) represents the prediction of input data X by the rules learned through the decision tree;

[0053] Support Vector Machines (SVMs) classify or regress by finding the hyperplane with the maximum margin. For regression problems, the formula for SVM is:

[0054] y =<w,x> +b

[0055] in<w,x> is the dot product of the input vector x and the weight vector w, and b is the bias term;

[0056] Deep neural networks perform non-linear mappings through multiple hidden layers, given an input X. DNN The predicted value is:

[0057] y = f(W) L (f(W L1 (...f(W1X+b1)...)+b L1 )+b L )

[0058] Among them, W i and b i Let f be the weight matrix and bias term of the i-th layer, and f be the activation function;

[0059] Ensemble learning techniques improve overall prediction performance by combining the predictions of multiple models. Using a weighted average method, the predictions of the ensemble model are as follows:

[0060]

[0061] Among them, y cnsemble For the final prediction result of the ensemble model, y i w represents the prediction result of the i-th base learner. i The weights of the base learner are, and

[0062] Incremental learning algorithms can update the model online when new data arrives without retraining the entire model. Assume the model parameters are θ. t The new training sample is (x t ,y t The update formula for incremental learning is:

[0063]

[0064] Where η is the learning rate. The loss function is applied to the model parameters θ t gradient, x t and y tFor newly arrived training data;

[0065] To combine incremental learning and ensemble learning, each base learner in the ensemble is updated online using incremental learning. Let the parameters of the current base learner be θ. i Then the online update of the i-th base learner is:

[0066]

[0067] Then, the output of the ensemble learning model is the weighted result of all the updated base learners:

[0068]

[0069] Among them, y i This represents the prediction result of the i-th base learner.

[0070] As a preferred approach, the method involves introducing an external event perception module to monitor changes in market, economic, and policy emergencies in real time, automatically identify the impact of these external factors, and adjust the prediction model based on real-time information.

[0071] The external event perception module is used to monitor emergencies in the market, economy, and policy fields in real time, and defines the impact factor E of external events. t This factor is evaluated based on the type and intensity of the event, as well as its potential impact on the prediction target.

[0072] Assume that the external event perception module automatically identifies and quantifies the impact intensity E of sudden events by analyzing real-time data from news, social media, policy releases, and market reports. t :

[0073] E t =f(event_type, event_intensity, event_relevance)

[0074] Among them, event_type represents the type of event, event_intensity represents the intensity of the event, and event_relevance represents the relevance of the event to the demand for car services;

[0075] Event impact factor E obtained through the external event perception module t We combine this as an additional input variable with the prediction model, denoted as f(X), where X is a data vector including historical data and other input features. The adjusted model can be expressed as:

[0076] y t =f(C t E t )

[0077] Among them, y t For the predicted demand at time t, X t E is an input vector that includes historical data and other input features. t The external event impact factor is calculated by the external event perception module;

[0078] The impact of external events on the prediction model is dynamic. Therefore, the prediction is updated in real time by adjusting the model's weights. In ensemble learning-based models, the weights of the base learners are adjusted to respond to the impact of external events.

[0079]

[0080] Among them, w i (t) represents the weights of the i-th base learner at time t, y i (t) represents the prediction result of the i-th base learner, with weights w i (t) Based on the external event impact factor E t The formula is dynamically adjusted and updated as follows:

[0081]

[0082] Among them, α i The adjustment coefficients for each base learner, Update the gradient for the weights of the base learner;

[0083] For incremental learning models, an external event influence factor E is introduced. t With each influx of new data, the learning rate or update strategy is adjusted based on real-time information. The update formula for the incremental learning model can be:

[0084]

[0085] Where, η t Let γ be the learning rate. t E is an adjustment factor for the impact of external events. t External event influencing factors The loss function is applied to the model parameters θ t The gradient;

[0086] The external event perception module also includes real-time monitoring and feedback mechanisms to ensure that the model can dynamically adapt to sudden events. The monitoring system continuously tracks changes after an event occurs and adjusts the model accordingly to optimize prediction performance. The core of the feedback mechanism is to adapt the model to external events by correcting errors.

[0087]

[0088] Where β and κ are adjustment coefficients, Δθ t This represents the adjustment amount for the model parameters. E is the gradient of the loss function. t External event influencing factors.

[0089] As a preferred approach, a feedback mechanism based on real-time prediction results is established. This mechanism continuously optimizes the prediction model using vehicle owner usage data and market feedback, and automatically adjusts parameters based on feedback information. This self-learning method is as follows:

[0090] Real-time forecast results and market feedback will become the inputs for dynamically adjusting the forecasting model, and the feedback error is defined as ∈ t To predict demand y pred (t) and actual demand y act The difference between (t):

[0091] ∈ t =y pred (t)-y act (t)

[0092] Among them, y pred (t) represents the demand predicted based on the current model, y act (t) represents actual market demand;

[0093] Car owner usage data influences future demand forecasts. This can be achieved by introducing car owner usage data (U...) t Compare it with market feedback. i Combine this with adjusting the model's parameters θ;

[0094] Set the objective of model parameter updates to minimize error ∈ t And it will adaptively adjust based on car owner usage data and market feedback:

[0095]

[0096] Where η is the learning rate and γ is the adjustment coefficient for the car owner data. For loss function, The loss function is applied to the model parameters θ t The gradient;

[0097] Through car owner usage data U t and market feedback ∈ t The model learns and optimizes its predictions. The self-learning process is based on the following formula:

[0098]

[0099] Where α is the learning rate and β is the self-learning adjustment coefficient. The gradient of the loss function;

[0100] Within the framework of incremental learning, the model parameters are updated by continuously receiving new user data and market feedback. The update formula for incremental learning is:

[0101]

[0102] Through a real-time feedback mechanism, market feedback error ∈ t Predictive models that will influence the future, combined with car owner usage data U t Adjust the model parameters and change the prediction formula to:

[0103] y pred (t+1)=f(X t U t θ t+1 )

[0104] Where, θ t+1 The model parameters are updated through the above feedback mechanism;

[0105] Whenever new market feedback is received t Or the car owner uses data U t The model updates itself using the following formula:

[0106]

[0107] Here, α and γ are adjustment coefficients to ensure a balance between the feedback mechanism and the self-learning process.

[0108] As a preferred approach, during the forecasting process, an interpretability analysis tool is integrated to transform complex forecast results into concise decision support information. This is combined with causal inference techniques to analyze the specific impacts of different characteristics and external events on changes in demand.

[0109] Complex forecasts are transformed into concise decision support information. Interpretability analysis tools are used to explain the model output, and SHAP values ​​are used to represent the contribution of each feature to the demand forecast results.

[0110]

[0111] Among them, SHAP i Let f(S) represent the contribution of the i-th feature to the prediction result, S be the feature subset excluding feature i, and f(S) be the prediction result in subset S. i The Shapley value, calculated incrementally, represents the marginal contribution of each feature to the final prediction.

[0112] Using a causal inference model, we analyze the impact of these characteristics on changes in demand, assuming external event E. t Demand y pred(t) has an impact, and the causal inference is derived through the model:

[0113] y pred (t)=f(X t E t )+∈ t

[0114] Where, f(X) t E t ) is an external event E t and feature X t The combined effects, ∈ t This is the error term, representing the portion not explained by the model.

[0115] Use causal effect estimation to calculate the marginal effect of external events on changes in demand:

[0116]

[0117] Where, τ cause Represents external event E t The causal effect of changes in demand;

[0118] By combining interpretability analysis tools and causal inference techniques, this study analyzes the contribution of different characteristics to changes in demand and further derives the causal impact of external events, assuming that the change in demand Δy(t) is related to the change in characteristics ΔX. t and external events E t The causal relationship is modeled using the following formula:

[0119]

[0120] Where, φ i Let ΔX be the SHAP value of the i-th feature, representing the contribution of that feature to the prediction result. i For feature X i The change, τ cause For external event E t causal effect, ΔE t For changes in external events, ∈ t This is the error term.

[0121] Another technical problem to be solved by the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the automotive service demand prediction method based on big data analysis as described above.

[0122] Another technical problem to be solved by the present invention is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for predicting automotive service demand based on big data analysis.

[0123] The beneficial effects of the present invention are:

[0124] By automatically cleaning and processing data from various sources, including car owner behavior, vehicle sensors, and vehicle maintenance history, machine learning algorithms significantly improve data processing efficiency and accuracy while reducing manual intervention. Deep learning algorithms can automatically identify and adjust key features affecting automotive service demand and promptly identify new factors, ensuring that the predictive model keeps pace with changes in the market and social environment. The integration of multiple machine learning models, such as regression analysis, decision trees, support vector machines, and deep neural networks, enhances the accuracy and stability of predictions. By introducing an external event perception module, changes in market, economic, and policy events can be monitored in real time. Based on real-time prediction results, the system collects car owner usage data and market reactions to continuously optimize the predictive model. By integrating interpretability analysis tools, complex prediction results become more intuitive and easier to understand. Through the integration of multiple predictive models, real-time monitoring of external events, and analysis of causal relationships, decision-makers can make more rational decisions based on accurate predictive information, optimizing resource allocation and market strategies for automotive services. Because this solution can automatically identify and adapt to new factors and process data from different dimensions, it is highly adaptable and can effectively cope with the complex and ever-changing market environment in automotive service demand. Detailed Implementation

[0125] The principles and features of the present invention are described below. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically by way of example in the following paragraphs. The advantages and features of the invention will become clearer from the following description and claims.

[0126] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0127] Example

[0128] The technical solution adopted by the present invention to solve its technical problem is:

[0129] Methods for predicting automotive service demand based on big data analytics include:

[0130] Data is collected from multiple channels, including car owner behavior data, vehicle sensor data, vehicle maintenance history, market trends, social media sentiment, weather information, and policy changes. Machine learning algorithms are used to automatically clean and preprocess the data.

[0131] Based on deep learning algorithms, it automatically identifies and adjusts key features that affect the demand for automotive services, and dynamically identifies emerging factors, including new car models, changes in consumer behavior, and policy changes.

[0132] By combining multiple models such as regression analysis, decision trees, support vector machines, and deep neural networks, the accuracy and stability of predictions are improved through ensemble learning techniques. Incremental learning algorithms are also employed to enable the model to learn online and update in real time as data continues to flow in.

[0133] By introducing an external event perception module, changes in market, economic, and policy emergencies are monitored in real time, the impact of these external factors is automatically identified, and the prediction model is adjusted based on real-time information.

[0134] Establish a feedback mechanism based on real-time prediction results, continuously optimize the prediction model through car owner usage data and market response, and automatically adjust parameters based on feedback information to perform self-learning;

[0135] During the forecasting process, interpretability analysis tools are integrated to transform complex forecast results into concise decision support information, and causal inference techniques are combined to analyze the specific impact of different characteristics and external events on changes in demand.

[0136] By comprehensively utilizing data such as car owner behavior, maintenance history, social media sentiment, and weather, the predictive model is provided with multi-dimensional information, improving its comprehensiveness and accuracy. Deep learning can automatically learn complex nonlinear relationships from the data and dynamically adjust when new factors emerge, ensuring the predictive model remains adaptable to new changes. The integration of multiple models, including regression analysis, decision trees, support vector machines, and deep neural networks, enhances the model's robustness and accuracy. These models can analyze from different perspectives and with different features, and their integration can compensate for the shortcomings of a single model. Incremental learning algorithms enable the predictive model to learn and update in real time with continuous data input, dynamically adapting to market changes. By monitoring changes in external events such as the market, economy, and policies in real time, the impact of these changes on automotive service demand can be identified in a timely manner, thereby adjusting the predictive model and improving its accuracy. Based on car owner usage data and market reactions, the predictive model is continuously optimized through a feedback mechanism. By integrating interpretability analysis tools, complex predictive results can be transformed into intuitive decision support information, helping decision-makers understand the basis of the predictions and enhancing trust.

[0137] The method of collecting data from multiple channels, including car owner behavior data, vehicle sensor data, vehicle maintenance history, market trends, social media sentiment, weather information, and policy changes, and then using machine learning algorithms for automatic data cleaning and preprocessing, is as follows:

[0138] For missing data, use the following formula for interpolation:

[0139] x imputed =μ+λ(x) nearest )

[0140] Among them, x imputed The missing values ​​are imputed, μ is the mean of the data, and x is the mean of the data. nearest The nearest neighbor value is the value near the missing value, and λ is the weighting coefficient.

[0141] Outliers are identified using a standard deviation-based test. Assuming the data follows a normal distribution, the formula for outliers is:

[0142]

[0143] Where z is the z-score, x is the data point, μ is the mean of the data, and σ is the standard deviation of the data. When |z|>3, the data point is considered an outlier.

[0144] To eliminate scale differences among different data features, the data is normalized. The Min-Max normalization formula is as follows:

[0145]

[0146] Among them, x norm For the normalized data, x min and x max These are the minimum and maximum values ​​of the data, respectively.

[0147] For categorical data, one-hot encoding is used to convert it into numerical form. If there is a category feature C, it is converted into a binary vector.

[0148] C one-hot =[if C=c1,then(1,0,0),if C=c2,then(0,1,0),...]

[0149] Among them, C one-hot The result of one-hot encoding for categorical data;

[0150] Smoothing the data to reduce noise:

[0151]

[0152] Among them, MA n Let x be the moving average over n time points. i This represents the value corresponding to time point i.

[0153] By combining mean and nearest neighbor imputation, missing values ​​can be more accurately predicted based on the data distribution pattern, avoiding the bias caused by simple imputation. Standard deviation testing can identify extreme values ​​(outliers), preventing them from misleading data analysis results. Data ranges for different features can vary significantly; normalization can eliminate scale differences between features, ensuring all features have the same influence on the model. Machine learning algorithms typically only handle numerical data; one-hot encoding converts categorical variables into numerical forms, making them suitable for model training. Noise and random fluctuations in the dataset can interfere with model training; moving averages can effectively smooth the data and reduce the impact of abnormal fluctuations on model training.

[0154] Based on deep learning algorithms, the method for automatically identifying and adjusting key features affecting automotive service demand, and dynamically identifying emerging factors, including new car models, changes in consumer behavior, and policy changes, is as follows:

[0155] Assume the input data is X = [x1, x2, ..., x...]. n If the output is the predicted demand y, then the deep neural network makes the prediction using a multilayer perceptron, and its calculation formula is as follows:

[0156] y = f(W) L (f(W L1 (...f(W1X+b1)...)+b L1 )+b L )

[0157] Where f is the activation function, W i and b i Let L represent the weights and biases of the i-th layer, and L represent the number of layers in the network.

[0158] Based on the intermediate layer output after deep neural network training, feature importance analysis is used to automatically identify key features, and SHAP values ​​are used to evaluate the contribution of each input feature to the prediction result.

[0159]

[0160] Among them, SHAP(x i ) is a feature x i Contribution to the prediction results, x i To remove feature x i The input data afterwards;

[0161] Dynamic feature learning methods (such as online learning) are used to identify emerging influencing factors, let X be... new For datasets containing new features, the model is trained using incremental updates:

[0162]

[0163] Where, θ t Here are the model parameters after the t-th update, and η is the learning rate. Let X be the loss function. new and y new For inputting and labeling new data;

[0164] Based on time series data, recurrent neural networks or long short-term memory networks are used to learn from historical data, thereby dynamically capturing and predicting new influencing factors. The calculation formula of the LSTM model is as follows:

[0165] i t =σ(W i [h t 1,x t ]+b i ),f t =σ(W f [h t 1,x t ]+b f ),o t =σ(W o [h t 1,x t ]+b o )

[0166] c t =f t *c t 1+i t *tanh(W c [h t 1,x t ]+b c ), h t =o t *tanh(c t )

[0167] Among them, i t f t o t These are the input gate, forget gate, and output gate, respectively. t h is a memory unit. t For output, x t For the current input data, W i W f W o W c Let b be the weight matrix. i b f b o b c For bias terms;

[0168] The adaptive adjustments made by deep neural networks to new car models and external factors such as market changes can be achieved through transfer learning techniques, which involves fine-tuning existing models with new data.

[0169]

[0170] Where, θ new For the fine-tuned new model parameters, θ pre-trained X represents the parameters of the pre-trained model. new and y new For the new dataset.

[0171] Deep neural networks, through their multilayer perceptron structure, can effectively capture the complex relationships in input data and output accurate demand forecasts. They can handle nonlinear features and adapt to various input patterns; by analyzing the contribution of features to the output, SHAP values ​​can help automatically identify and optimize the most critical features in the model, thereby improving predictive performance; through online learning methods, the model can be updated in real time to adapt to new data input, which is crucial for capturing changes in market or consumer behavior; LSTM excels at processing and predicting time series data, remembering long-term dependencies and helping to capture trends and cyclical changes in historical data; transfer learning allows the model to leverage existing knowledge (pre-trained models) to fine-tune new data, adapting to emerging influencing factors and reducing reliance on large amounts of new data.

[0172] By combining multiple models such as regression analysis, decision trees, support vector machines, and deep neural networks, ensemble learning techniques are used to improve the accuracy and stability of predictions. An incremental learning algorithm is employed to enable the model to learn online and update in real time as data continues to flow in.

[0173] Regression analysis models are used to predict continuous demand values. The linear regression model is as follows:

[0174] y = β0 + β1x1 + β2x2 + ... + β n x n +∈

[0175] Where y is the predicted demand value, β0, β1, ..., β n Let x1, x2, ..., x be the regression coefficients. n The input features are ∈, and the error term is ∈.

[0176] Decision trees construct predictive models by recursively splitting datasets. For input data X, the output y of the decision tree is obtained through a series of decision rules:

[0177] y = DecisionTree(X) = f(X)

[0178] Where f(X) represents the prediction of input data X by the rules learned through the decision tree;

[0179] Support Vector Machines (SVMs) classify or regress by finding the hyperplane with the maximum margin. For regression problems, the formula for SVM is:

[0180] y =<w,x> +b

[0181] in<w,x> is the dot product of the input vector x and the weight vector w, and b is the bias term;

[0182] Deep neural networks perform non-linear mappings through multiple hidden layers, given an input X. DNN The predicted value is:

[0183] y = f(W) L (f(W L 1(...f(W1X+b1)...)+b L 1)+b L )

[0184] Among them, W i and b i Let f be the weight matrix and bias term of the i-th layer, and f be the activation function;

[0185] Ensemble learning techniques improve overall prediction performance by combining the predictions of multiple models. Using a weighted average method, the predictions of the ensemble model are as follows:

[0186]

[0187] Among them, y cnsemble For the final prediction result of the ensemble model, y i w represents the prediction result of the i-th base learner. i The weights of the base learner are, and

[0188] Incremental learning algorithms can update the model online when new data arrives without retraining the entire model. Assume the model parameters are θ. t The new training sample is (x t ,y t The update formula for incremental learning is:

[0189]

[0190] Where η is the learning rate. The loss function is applied to the model parameters θ t gradient, x t and y t For newly arrived training data;

[0191] To combine incremental learning and ensemble learning, each base learner in the ensemble is updated online using incremental learning. Let the parameters of the current base learner be θ. i Then the online update of the i-th base learner is:

[0192]

[0193] Then, the output of the ensemble learning model is the weighted result of all the updated base learners:

[0194]

[0195] Among them, y i This represents the prediction result of the i-th base learner.

[0196] This approach improves prediction accuracy and stability by combining multiple models (regression analysis, decision trees, support vector machines, and deep neural networks) and utilizing ensemble learning techniques. Furthermore, the introduction of incremental learning algorithms ensures that the model can learn and update online as new data flows in, reducing training time and enabling rapid adaptation to new changes. Through ensemble learning and incremental updates, the overall prediction accuracy is improved. The combination of incremental learning and online updates allows the model to adapt to new data in real time, ensuring its timeliness. Combining different types of base learners better adapts to the complexity and changes in data.

[0197] The method of introducing an external event perception module to monitor changes in market, economic, and policy emergencies in real time, automatically identify the impact of these external factors, and adjust the prediction model based on real-time information is as follows:

[0198] The external event perception module is used to monitor emergencies in the market, economy, and policy fields in real time, and defines the impact factor E of external events. t This factor is evaluated based on the type and intensity of the event, as well as its potential impact on the prediction target.

[0199] Assume that the external event perception module automatically identifies and quantifies the impact intensity E of sudden events by analyzing real-time data from news, social media, policy releases, and market reports. t :

[0200] E t =f(event_type, event_intensity, event_relevance)

[0201] Among them, event_type represents the type of event, event_intensity represents the intensity of the event, and event_relevance represents the relevance of the event to the demand for car services;

[0202] Event impact factor E obtained through the external event perception module t We combine this as an additional input variable with the prediction model, denoted as f(X), where X is a data vector including historical data and other input features. The adjusted model can be expressed as:

[0203] y t =f(X) t E t )

[0204] Among them, y t For the predicted demand at time t, X t E is an input vector that includes historical data and other input features. t The external event impact factor is calculated by the external event perception module;

[0205] The impact of external events on the prediction model is dynamic. Therefore, the prediction is updated in real time by adjusting the model's weights. In ensemble learning-based models, the weights of the base learners are adjusted to respond to the impact of external events.

[0206]

[0207] Among them, w i (t) represents the weights of the i-th base learner at time t, y i (t) represents the prediction result of the i-th base learner, with weights w i (t) Based on the external event impact factor E t The formula is dynamically adjusted and updated as follows:

[0208]

[0209] Among them, α i The adjustment coefficients for each base learner, Update the gradient for the weights of the base learner;

[0210] For incremental learning models, an external event influence factor E is introduced. t With each influx of new data, the learning rate or update strategy is adjusted based on real-time information. The update formula for the incremental learning model can be:

[0211]

[0212] Where, η t Let γ be the learning rate. t E is an adjustment factor for the impact of external events. t External event influencing factors The loss function is applied to the model parameters θ t The gradient;

[0213] The external event perception module also includes real-time monitoring and feedback mechanisms to ensure that the model can dynamically adapt to sudden events. The monitoring system continuously tracks changes after an event occurs and adjusts the model accordingly to optimize prediction performance. The core of the feedback mechanism is to adapt the model to external events by correcting errors.

[0214]

[0215] Where β and κ are adjustment coefficients, Δθ t This represents the adjustment amount for the model parameters. E is the gradient of the loss function. t External event influencing factors.

[0216] This solution, by introducing an external event perception module, can monitor market, economic, and policy emergencies in real time and quantify their impact on forecasting objectives. By incorporating external event influencing factors into the forecasting model, the model can quickly adapt and adjust its predictions when emergencies occur. Furthermore, the dynamic adjustment mechanism based on ensemble learning and incremental learning gives the model extremely high flexibility and adaptability, enabling it to respond to changes in the external environment in real time and dynamically adjust its predictions according to the impact of external events, thus improving the model's flexibility. Real-time monitoring and optimization through error correction ensure that the model can adapt to long-term environmental changes. The introduction of external events enhances the model's sensitivity to emergencies, improving the accuracy and stability of predictions.

[0217] A feedback mechanism based on real-time prediction results is established to continuously optimize the prediction model through car owner usage data and market reactions, and to automatically adjust parameters based on feedback information. This self-learning method involves:

[0218] Real-time forecast results and market feedback will become the inputs for dynamically adjusting the forecasting model, and the feedback error is defined as ∈ t To predict demand y pred (t) and actual demand y act The difference between (t):

[0219] ∈ t =y pred (t)-y act (t)

[0220] Among them, y pred (t) represents the demand predicted based on the current model, y act (t) represents actual market demand;

[0221] Car owner usage data influences future demand forecasts. This can be achieved by introducing car owner usage data (U...) t Compare it with market feedback. t Combine this with adjusting the model's parameters θ;

[0222] Set the objective of model parameter updates to minimize error ∈ t And it will adaptively adjust based on car owner usage data and market feedback:

[0223]

[0224] Where η is the learning rate and γ is the adjustment coefficient for the car owner data. For loss function, The loss function is applied to the model parameters θ t The gradient;

[0225] Through car owner usage data U t and market feedback ∈ t The model learns and optimizes its predictions. The self-learning process is based on the following formula:

[0226]

[0227] Where α is the learning rate and β is the self-learning adjustment coefficient. The gradient of the loss function;

[0228] Within the framework of incremental learning, the model parameters are updated by continuously receiving new user data and market feedback. The update formula for incremental learning is:

[0229]

[0230] Through a real-time feedback mechanism, market feedback error ∈ t Predictive models that will influence the future, combined with car owner usage data U t Adjust the model parameters and change the prediction formula to:

[0231] y pred (t+1)=f(X t U t θ t+1 )

[0232] Where, θ t+1 The model parameters are updated through the above feedback mechanism;

[0233] Whenever new market feedback is received t Or the car owner uses data U t The model updates itself using the following formula:

[0234]

[0235] Here, α and γ are adjustment coefficients to ensure a balance between the feedback mechanism and the self-learning process.

[0236] This solution introduces a feedback mechanism based on real-time prediction results, combined with car owner usage data and market feedback, to continuously optimize the prediction model. Within a self-learning and incremental learning framework, the model dynamically adjusts based on real-time feedback. Through continuous learning and optimization, the model not only responds to market changes but also adaptively adjusts with each new data inflow, improving prediction accuracy and stability. The combination of feedback and self-learning allows the model to quickly optimize with each data change. Through continuous learning, the model adapts to constantly changing markets and user behavior, making more accurate predictions. By incorporating car owner data and market feedback, the model can precisely adjust in complex environments, providing higher-quality prediction results.

[0237] In the forecasting process, integrating interpretability analysis tools transforms complex forecast results into concise decision support information. Combined with causal inference techniques, the method for analyzing the specific impact of different characteristics and external events on changes in demand is as follows:

[0238] Complex forecasts are transformed into concise decision support information. Interpretability analysis tools are used to explain the model output, and SHAP values ​​are used to represent the contribution of each feature to the demand forecast results.

[0239]

[0240] Among them, SHAP i Let f(S) represent the contribution of the i-th feature to the prediction result, S be the feature subset excluding feature i, and f(S) be the prediction result in subset S. i The Shapley value, calculated incrementally, represents the marginal contribution of each feature to the final prediction.

[0241] Using a causal inference model, we analyze the impact of these characteristics on changes in demand, assuming external event E. t Demand y pred (t) has an impact, and the causal inference is derived through the model:

[0242] y pred (t)=f(X t E t )+∈ t

[0243] Where, f(X) t E t ) is an external event E t and feature X t The combined effects, ∈ t This is the error term, representing the portion not explained by the model.

[0244] Use causal effect estimation to calculate the marginal effect of external events on changes in demand:

[0245]

[0246] Where, τ cause Represents external event E t The causal effect of changes in demand;

[0247] By combining interpretability analysis tools and causal inference techniques, this study analyzes the contribution of different characteristics to changes in demand and further derives the causal impact of external events, assuming that the change in demand Δy(t) is related to the change in characteristics ΔX. t and external events E t The causal relationship is modeled using the following formula:

[0248]

[0249] Where φi is the SHAP value of the i-th feature, representing the contribution of that feature to the prediction result, ΔX i For feature X i The change, τ cause For external event E t causal effect, ΔE t For changes in external events, ∈ t This is the error term.

[0250] This solution integrates interpretability analysis tools (such as SHAP values) and causal inference techniques to make complex forecasting models more transparent and understandable, helping decision-makers to deeply analyze the causes of demand changes. Interpretability tools clearly demonstrate the contribution of each feature to the forecast results; while causal inference techniques allow the model to analyze the specific impact of external events on demand changes, revealing the causal relationships between factors. Ultimately, this combination not only enhances the model's transparency and operability but also provides decision-makers with more accurate decision support. SHAP values ​​make the model more transparent, enabling decision-makers to understand the contribution of each feature to the forecast; combined with causal inference, the causal impact of external events on demand can be identified, helping decision-makers understand the deeper dynamics of the market; and the combined analysis provides more comprehensive decision support, helping to better understand the multidimensional factors of demand changes, thereby making more accurate forecasts and decisions.

[0251] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-described method for predicting automotive service demand based on big data analysis.

[0252] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting automotive service demand based on big data analysis.

[0253] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0254] Those skilled in the art will understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0255] The above embodiments of the present invention are not intended to limit the scope of protection of the present invention. The implementation of the present invention is not limited thereto. All other modifications, substitutions or alterations made to the above structure of the present invention based on the above content of the present invention, in accordance with ordinary technical knowledge and common practice in the field, without departing from the basic technical idea of ​​the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A method for predicting automotive service demand based on big data analysis, characterized in that, Including: Data is collected from multiple channels, including car owner behavior data, vehicle sensor data, vehicle maintenance history, market trends, social media sentiment, weather information, and policy changes. Machine learning algorithms are used to automatically clean and preprocess the data. Based on deep learning algorithms, it automatically identifies and adjusts key features that affect the demand for automotive services, and dynamically identifies emerging factors, including new car models, changes in consumer behavior, and policy changes. By combining multiple models such as regression analysis, decision trees, support vector machines, and deep neural networks, the accuracy and stability of predictions are improved through ensemble learning techniques. Incremental learning algorithms are also employed to enable the model to learn online and update in real time as data continues to flow in. By introducing an external event perception module, changes in market, economic, and policy emergencies are monitored in real time, the impact of these external factors is automatically identified, and the prediction model is adjusted based on real-time information. Establish a feedback mechanism based on real-time prediction results, continuously optimize the prediction model through car owner usage data and market response, and automatically adjust parameters based on feedback information to perform self-learning; During the forecasting process, interpretability analysis tools are integrated to transform complex forecast results into concise decision support information, and causal inference techniques are combined to analyze the specific impact of different characteristics and external events on changes in demand.

2. The method for predicting automotive service demand based on big data analysis according to claim 1, characterized in that, The method of collecting data from multiple channels, including car owner behavior data, vehicle sensor data, vehicle maintenance history, market trends, social media sentiment, weather information, and policy changes, and then using machine learning algorithms for automatic data cleaning and preprocessing, is as follows: For missing data, use the following formula for interpolation: x imputcd =μ+λ(x nearcst ) Among them, x imputed The missing values ​​are imputed, μ is the mean of the data, and x is the mean of the data. nearest The nearest neighbor value is the value near the missing value, and λ is the weighting coefficient. Outliers are identified using a standard deviation-based test. Assuming the data follows a normal distribution, the formula for outliers is: Where z is the z-score, x is the data point, μ is the mean of the data, and σ is the standard deviation of the data. When |z|>3, the data point is considered an outlier. To eliminate scale differences among different data features, the data is normalized. The Min-Max normalization formula is as follows: Among them, x norm For the normalized data, x min and x max These are the minimum and maximum values ​​of the data, respectively. For categorical data, one-hot encoding is used to convert it into numerical form. If there is a category feature C, it is converted into a binary vector. C onc-hot =[if C=c1,then(1,0,0),if C=c2,then(0,1,0),...] Among them, C onc-hot The result of one-hot encoding for categorical data; Smoothing the data to reduce noise: Among them, MA n Let x be the moving average over n time points. i This represents the value corresponding to time point i.

3. The method for predicting automotive service demand based on big data analysis according to claim 2, characterized in that, Based on deep learning algorithms, the method for automatically identifying and adjusting key features affecting automotive service demand, and dynamically identifying emerging factors, including new car models, changes in consumer behavior, and policy changes, is as follows: Assume the input data is X = [x1, x2, ..., x...]. n If the output is the predicted demand y, then the deep neural network makes the prediction using a multilayer perceptron, and its calculation formula is as follows: y=f(W L (f(W L 1 (...f(W1X+b1)...)+b L 1 )+b L ) Where f is the activation function, W i and b i Let L represent the weights and biases of the i-th layer, and L represent the number of layers in the network. Based on the intermediate layer output after deep neural network training, feature importance analysis is used to automatically identify key features, and SHAP values ​​are used to evaluate the contribution of each input feature to the prediction result. Among them, SHAP(x i ) is a feature x i Contribution to the prediction results, x i To remove feature x i The input data afterwards; Using dynamic feature learning methods (such as online learning) to identify emerging influencing factors, let X... new For datasets containing new features, the model is trained using incremental updates: Where, θ t Here are the model parameters after the t-th update, and η is the learning rate. Let X be the loss function. new and y new For inputting and labeling new data; Based on time series data, recurrent neural networks or long short-term memory networks are used to learn from historical data, thereby dynamically capturing and predicting new influencing factors. The calculation formula of the LSTM model is as follows: i t =σ(W i [h t 1 ,x t ]+b i ),f t =σ(W f [h t 1 ,x t ]+b f ),o t =σ(W o [h t 1 ,x t ]+b o ) c i =f t *c t 1 +i t *tanh(W c [h t 1 ,x t ]+bc),h t =o t *fishy(c) t ) Among them, i t f t o t These are the input gate, forget gate, and output gate, respectively. t h is a memory unit. t For output, x t For the current input data, W i W f W o W c Let b be the weight matrix. i b f b o b c For bias terms; The adaptive adjustments made by deep neural networks to new car models and external factors such as market changes can be achieved through transfer learning techniques, which involves fine-tuning existing models with new data. Where, θ new For the fine-tuned new model parameters, θ prc-trained X represents the parameters of the pre-trained model. new and y new For the new dataset.

4. The method for predicting automotive service demand based on big data analysis according to claim 3, characterized in that, By combining multiple models such as regression analysis, decision trees, support vector machines, and deep neural networks, ensemble learning techniques are used to improve the accuracy and stability of predictions. An incremental learning algorithm is employed to enable the model to learn online and update in real time as data continues to flow in. Regression analysis models are used to predict continuous demand values. The linear regression model is as follows: y=β0+β1xx+β2x2+…+β n x n +∈ Where y is the predicted demand value, β0, β1, ..., β n Let x1, x2, ..., x be the regression coefficients. n The input features are ∈, and the error term is ∈. Decision trees construct predictive models by recursively splitting datasets. For input data X, the output y of the decision tree is obtained through a series of decision rules: y = DecisionTree(X) = f(X) Where f(X) represents the prediction of input data X by the rules learned through the decision tree; Support Vector Machines (SVMs) classify or regress by finding the hyperplane with the maximum margin. For regression problems, the formula for SVM is: y =<w,x> +b in<w,x> ω is the dot product of the input vector x and the weight vector ω, and b is the bias term; Deep neural networks perform non-linear mapping through multiple hidden layers. Given an input, the predicted value of an XDNN is: y=f(W L (f(W L 1 (...f(W1X+b1)...)+b L 1 )+b L ) Among them, W i and b i Let f be the weight matrix and bias term of the i-th layer, and f be the activation function; Ensemble learning techniques improve overall prediction performance by combining the predictions of multiple models. Using a weighted average method, the predictions of the ensemble model are as follows: Among them, y cnsemblc For the final prediction result of the ensemble model, y i w represents the prediction result of the i-th base learner. i The weights of the base learner are, and Incremental learning algorithms can update the model online when new data arrives without retraining the entire model. Assume the model parameters are θ. t The new training sample is (x t ,y t The update formula for incremental learning is: Where η is the learning rate. The loss function is applied to the model parameters θ t gradient, x t and y t For newly arrived training data; To combine incremental learning and ensemble learning, each base learner in the ensemble is updated online using incremental learning. Let the parameters of the current base learner be θ. i Then the online update of the i-th base learner is: Then, the output of the ensemble learning model is the weighted result of all the updated base learners: Among them, y i This represents the prediction result of the i-th base learner.

5. The method for predicting automotive service demand based on big data analysis according to claim 4, characterized in that, The method of introducing an external event perception module to monitor changes in market, economic, and policy emergencies in real time, automatically identify the impact of these external factors, and adjust the prediction model based on real-time information is as follows: The external event perception module is used to monitor emergencies in the market, economy, and policy fields in real time, and defines the impact factor E of external events. t This factor is evaluated based on the type and intensity of the event, as well as its potential impact on the prediction target. Assume that the external event perception module automatically identifies and quantifies the impact intensity E of sudden events by analyzing real-time data from news, social media, policy releases, and market reports. t : E t =f(event_type,event_intensity,event_relevance) Among them, event_type represents the type of event, event_intensity represents the intensity of the event, and event_relevance represents the relevance of the event to the demand for car services; Event impact factor E obtained through the external event perception module t We combine this as an additional input variable with the prediction model, denoted as f(X), where X is a data vector including historical data and other input features. The adjusted model can be expressed as: y t =f(X t ,E t ) Among them, y t For the predicted demand at time t, X t E is an input vector that includes historical data and other input features. t The external event impact factor is calculated by the external event perception module; The impact of external events on the prediction model is dynamic. Therefore, the prediction is updated in real time by adjusting the model's weights. In ensemble learning-based models, the weights of the base learners are adjusted to respond to the impact of external events. Among them, w i (t) represents the weights of the i-th base learner at time t, y i (t) represents the prediction result of the i-th base learner, with weights w i (t) Based on the external event impact factor E t The formula is dynamically adjusted and updated as follows: Among them, α i The adjustment coefficients for each base learner, Update the gradient for the weights of the base learner; For incremental learning models, an external event influence factor E is introduced. t With each influx of new data, the learning rate or update strategy is adjusted based on real-time information. The update formula for the incremental learning model can be: Where, η t Let γ be the learning rate. t E is an adjustment factor for the impact of external events. t External event influencing factors The loss function is applied to the model parameters θ t The gradient; The external event perception module also includes real-time monitoring and feedback mechanisms to ensure that the model can dynamically adapt to sudden events. The monitoring system continuously tracks changes after an event occurs and adjusts the model accordingly to optimize prediction performance. The core of the feedback mechanism is to adapt the model to external events by correcting errors. Where β and κ are adjustment coefficients, Δθ t This represents the adjustment amount for the model parameters. E is the gradient of the loss function. t External event influencing factors.

6. The method for predicting automotive service demand based on big data analysis according to claim 5, characterized in that, A feedback mechanism based on real-time prediction results is established to continuously optimize the prediction model through car owner usage data and market reactions, and to automatically adjust parameters based on feedback information. This self-learning method involves: Real-time forecast results and market feedback will become the inputs for dynamically adjusting the forecasting model, and the feedback error is defined as ∈ t To predict demand y pred (t) and actual demand y act The difference between (t): ∈ t =y pred (t)-y act (t) Among them, y pred (t) represents the demand predicted based on the current model, y act (t) represents actual market demand; Car owner usage data influences future demand forecasts. This can be achieved by introducing car owner usage data (U...) t Compare it with market feedback. t Combine this with adjusting the model's parameters θ; Set the objective of model parameter updates to minimize error ∈ t And it will adaptively adjust based on car owner usage data and market feedback: Where η is the learning rate and γ is the adjustment coefficient for the car owner data. For loss function, The loss function is applied to the model parameters θ t The gradient; Through car owner usage data U t and market feedback ∈ t The model learns and optimizes its predictions. The self-learning process is based on the following formula: Where α is the learning rate and β is the self-learning adjustment coefficient. The gradient of the loss function; Within the framework of incremental learning, the model parameters are updated by continuously receiving new user data and market feedback. The update formula for incremental learning is: Through a real-time feedback mechanism, market feedback error ∈ t Predictive models that will influence the future, combined with car owner usage data U t Adjust the model parameters and change the prediction formula to: y pred (t+1)=f(X t ,U t ,θ t 1 ) Where, θ t+1 The model parameters are updated through the above feedback mechanism; Whenever new market feedback is received t Or the car owner uses data U t The model updates itself using the following formula: Here, α and γ are adjustment coefficients to ensure a balance between the feedback mechanism and the self-learning process.

7. The method for predicting automotive service demand based on big data analysis according to claim 6, characterized in that, In the forecasting process, integrating interpretability analysis tools transforms complex forecast results into concise decision support information. Combined with causal inference techniques, the method for analyzing the specific impact of different characteristics and external events on changes in demand is as follows: Complex forecasts are transformed into concise decision support information. Interpretability analysis tools are used to explain the model output, and SHAP values ​​are used to represent the contribution of each feature to the demand forecast results. Among them, SHAP i Let f(S) represent the contribution of the i-th feature to the prediction result, S be the feature subset excluding feature i, and f(S) be the prediction result in subset S. i The Shapley value, calculated incrementally, represents the marginal contribution of each feature to the final prediction. Using a causal inference model, we analyze the impact of these characteristics on changes in demand, assuming external event E. t Demand y pred (t) has an impact, and the causal inference is derived through the model: y pred (t)=f(X t ,E t )+∈ t Where, f(X) t E t ) is an external event E t and feature X t The combined effects, ∈ t This is the error term, representing the portion not explained by the model. Use causal effect estimation to calculate the marginal effect of external events on changes in demand: Where, τ cause Represents external event E t The causal effect of changes in demand; By combining interpretability analysis tools and causal inference techniques, this study analyzes the contribution of different characteristics to changes in demand and further derives the causal impact of external events, assuming that the change in demand Δy(t) is related to the change in characteristics ΔX. t and external events E t The causal relationship is modeled using the following formula: Where, φ i Let ΔX be the SHAP value of the i-th feature, representing the contribution of that feature to the prediction result. i For feature X i The change, τ cause For external event E t causal effect, ΔE t For changes in external events, ∈ t This is the error term.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the automotive service demand forecasting method based on big data analysis as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the automotive service demand forecasting method based on big data analysis as described in any one of claims 1-7.

Images