Wind turbine blade icing prediction method based on real-time rolling prediction and online modeling

Abstract

The present invention discloses a method for predicting wind turbine blade icing based on real-time rolling prediction and online modeling, comprising: step 1, collecting historical meteorological data, terrain data and wind turbine operation data information, and cleaning, integrating and standardizing the data; step 2, using a real-time rolling prediction method, continuously updating input data and prediction models to adapt to changes in weather and wind farm operation status, and using online modeling technology to dynamically adjust model parameters and structures according to new data to improve the real-time and accuracy of predictions; step 3, deploying the model in a wind farm monitoring system, and realizing icing risk prediction based on real-time data input; the method solves the technical problems that traditional icing monitoring methods mostly rely on on-site observations or simple meteorological condition judgments, have limited prediction accuracy, and are difficult to meet the needs of efficient operation and maintenance of wind farms.

Description

Technical Field

[0001] The present invention belongs to the technical field of prediction of icing on wind turbine blades; in particular, it relates to a method for predicting icing on wind turbine blades based on real-time rolling prediction and online modeling. Background Art

[0002] Wind turbine blade icing is a major challenge facing the wind power industry, especially in cold and humid climates. Ice accumulation on the blades can seriously affect the aerodynamic performance of the wind turbine, resulting in reduced power generation efficiency and even mechanical failures, causing significant economic losses. Traditional icing monitoring methods mostly rely on on-site observations or simple meteorological condition judgments, with limited prediction accuracy, and are difficult to meet the needs of efficient operation and maintenance of wind farms. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a wind turbine blade icing prediction method based on real-time rolling prediction and online modeling, so as to solve the technical problems that most traditional icing monitoring methods rely on on-site observations or simple meteorological condition judgments, have limited prediction accuracy, and are difficult to meet the needs of efficient operation and maintenance of wind farms.

[0004] The technical solution of the present invention is: A method for predicting wind turbine blade icing based on real-time rolling prediction and online modeling, the method comprising: Step 1: Collect historical meteorological data, terrain data and wind turbine operation data information, and clean, integrate and standardize the data; Step 2: Use a real-time rolling forecast method to continuously update input data and forecast models to adapt to changes in weather and wind farm operating conditions. Use online modeling technology to dynamically adjust model parameters and structures based on new data to improve the real-time and accuracy of forecasts. Step 3: Deploy the model in the wind farm monitoring system and realize icing risk prediction based on real-time data input.

[0005] The historical meteorological data include: ambient temperature, air humidity, wind speed and direction, liquid water content and duration; terrain data include altitude, terrain slope, terrain orientation and surface cover; wind turbine operation data include blade speed, power output, vibration monitoring data, blade surface temperature data and operation log data.

[0006] Data cleaning methods include: using clustering algorithms in unsupervised learning to identify and process outliers in the data set; for missing data, fill in with business knowledge or experience, fill in with the calculation results of the same field indicators, or fill in with the calculation results of different indicators; use the Laida criterion to filter out outliers in the data set, or calculate the correlation coefficient between the characteristic variable and the icing state through a loop calculation, and delete the characteristic variables with a small correlation coefficient; data integration methods include: extracting data related to wind turbine blade icing from various data sources; converting data in different formats into a unified format, including converting text data into numerical data, or converting data of different time granularities into data of the same time granularity; merging the converted data into a complete data set; standardization processing includes: converting data into a unified scale or range, converting data into values ​​between 0 and 1 to eliminate dimensional differences between different data; converting data into a standard normal distribution with a mean of 0 and a variance of 1.

[0007] Methods for continuously updating input data and forecast models using a real-time rolling forecast approach include: Step 3.1, use the time series diagram, autocorrelation diagram ACF and partial autocorrelation diagram PACF to observe the stationarity and randomness of the time series; Step 3.2: Determine the order of the ARIMA model based on the ACF and PACF graphs; Step 3.3, use the information criterion Akaike Information Criterion AIC or Schwarz Information Criterion BIC to select the optimal ARIMA model; Step 3.4: Estimate the parameters of the ARIMA model using maximum likelihood estimation or least squares method; Step 3.5: Perform residual analysis on the fitted ARIMA model to check whether the residual is white noise. If the residual is not white noise, reprocess it; Step 3.6: Use the fitted ARIMA model to predict the time series.

[0008] The method of using online modeling technology to dynamically adjust model parameters and structure according to new data includes: Step 3.7. Select an algorithm that supports incremental learning. Step 3.8, obtain new data in real time and pre-process the new data; Step 3.9: Use the partial_fit method to partially fit the new data and update the model parameters. Repeat this step and update the model every time there is new data. Step 3.10: Use the validation set or test set to evaluate the performance of the model and adjust the model parameters based on the evaluation results. Step 3.11. Save the updated model for later use.

[0009] Beneficial effects of the present invention: The present invention integrates multi-dimensional historical data, including meteorological data, terrain data and wind turbine operation data, to achieve accurate prediction of wind turbine blade icing conditions, thereby effectively preventing wind turbine operation failures caused by icing and improving the operating efficiency and safety of wind farms.

[0010] This invention uses machine learning technology and combines it with rich historical data resources to achieve an in-depth understanding and accurate prediction of the ice formation mechanism. At the same time, the prediction model is continuously optimized to improve the accuracy and reliability of the prediction by considering the influence of multiple factors such as meteorological conditions, terrain characteristics, and wind turbine operating status.

[0011] Advantages of the present invention: Improve prediction accuracy: By integrating multi-dimensional data and combining the powerful learning ability of machine learning algorithms, the accuracy and reliability of wind turbine blade icing prediction are significantly improved.

[0012] Enhanced operation and maintenance efficiency: The real-time warning mechanism enables operation and maintenance personnel to take measures in advance to effectively avoid or reduce the impact of icing on wind turbine operation and improve the overall operation and maintenance efficiency of the wind farm.

[0013] Reduce operating costs: Reduce downtime and maintenance costs caused by icing, optimize wind turbine operation strategies, improve power generation efficiency, and bring significant economic benefits to wind farms.

[0014] It solves the technical problems that traditional ice monitoring methods mostly rely on on-site observations or simple judgment of meteorological conditions, have limited prediction accuracy, and are difficult to meet the needs of efficient operation and maintenance of wind farms. DETAILED DESCRIPTION

[0015] A wind turbine blade icing prediction method based on real-time rolling prediction and online modeling, comprising: Step 1: Data collection and preprocessing: Collect multi-dimensional information including historical meteorological data, terrain data and wind turbine operation data. Clean, integrate and standardize these data to ensure data quality.

[0016] Historical weather data includes: Ambient temperature: This is one of the key factors affecting blade icing. When the ambient temperature drops below freezing, water vapor in the air easily condenses into ice on the blade surface. Especially when the blade surface temperature is below -5 degrees Celsius and the ambient temperature is also below 0 degrees Celsius, the possibility of icing increases significantly.

[0017] Air humidity: High humidity conditions are conducive to the condensation of water vapor on the blade surface. Generally, when the air humidity reaches more than 85%, the risk of blade icing will increase significantly.

[0018] Wind speed and direction: Wind speed and direction not only affect the airflow distribution on the blades, but may also affect the condensation process of water vapor on the blade surface. At the same time, strong winds may accelerate the shedding of ice on the blades, but may also cause new ice crystals to form on the blade surface.

[0019] Liquid Water Content (LWC): The higher the liquid water content in the air, the more water vapor will condense on the blade surface, increasing the likelihood of ice accumulation.

[0020] Duration: The longer the weather conditions that are conducive to icing, such as low temperature and high humidity, last, the greater the possibility of ice covering the blades.

[0021] Terrain data includes: Altitude: Altitude has a significant impact on temperature. Generally speaking, the higher the altitude, the lower the temperature. Therefore, wind farms in higher altitude areas may have a higher risk of wind turbine blade icing. At the same time, climatic conditions in high altitude areas may be more severe, such as higher wind speeds and lower temperatures, which further increases the possibility of blade icing.

[0022] Terrain slope: The slope of the terrain can affect the direction and speed of the wind, thereby affecting the airflow distribution on the blades. In areas with large slopes, the wind may be blocked or accelerated by the terrain, resulting in changes in the conditions for condensation of water vapor on the blades. In addition, the slope may also affect the accumulation and melting of snow on the blades, thereby affecting the formation of ice.

[0023] Terrain orientation: The orientation of the terrain (such as facing the ocean or inland) will also affect the operating environment of the wind turbine. For example, terrain facing the ocean may be more susceptible to the impact of the marine climate, resulting in higher water vapor content in the air, increasing the risk of blade icing. At the same time, terrain facing inland may be affected by the continental climate, with greater temperature fluctuations, which may also affect blade icing.

[0024] Ground cover: Ground cover such as vegetation, water, ice and snow will also affect the operating environment of the wind turbine. For example, the air humidity near water is usually higher, which is conducive to the condensation of water vapor on the blade surface. The ground covered with ice and snow may reflect more solar radiation, affecting the temperature distribution of the surrounding environment, thus affecting the ice covering of the blades.

[0025] Fan operation data includes: Blade speed: The blade speed affects the relative speed between its surface and the air, which in turn affects the condensation process of water vapor on the blade surface. Under icing conditions, a reduction in blade speed may cause more water vapor to condense on the blade surface, thereby increasing the risk of icing.

[0026] Power output: The power output of a fan reflects its operating efficiency and energy conversion capacity. When the blades are covered with ice, the ice on the blade surface will increase the resistance to air flow, causing the fan's power output to decrease. Therefore, the change in power output can be used as an important indicator to determine whether the blades are covered with ice.

[0027] Vibration monitoring data: After the wind turbine blades are covered with ice, their mass and shape will change due to the increase of ice layer, which will cause abnormal vibration of the blades during operation. By monitoring the vibration data of the wind turbine, the ice coverage of the blades can be discovered in time and corresponding measures can be taken to deal with it.

[0028] Temperature data: The temperature of the blade surface is one of the important factors affecting icing. When the blade surface temperature is below the freezing point, water vapor in the air easily condenses into ice on the blade surface. Therefore, monitoring the temperature data of the blade surface is of great significance for predicting and preventing blade icing.

[0029] Operation log data: The operation log data of the fan records various status information of the fan during operation, including active power average, active power minimum, active power maximum, wind speed average, etc. These data can be used to analyze the operating status of the fan under icing conditions.

[0030] Data cleaning: Data cleaning is to remove outliers, missing values, and duplicate values ​​in the data to ensure the accuracy and consistency of the data.

[0031] Predefined cleaning rules: Use big data algorithms or statistical methods to set data cleaning rules. For example, for data that is obviously beyond the reasonable range (such as excessive wind speed, low temperature, etc.), set a threshold to filter it.

[0032] Mathematical and statistical cleaning: Use clustering algorithms in unsupervised learning to identify and handle outliers in the data set.

[0033] Handling missing values: For missing data, different filling strategies are adopted according to the actual situation, such as filling with business knowledge or experience, filling with the calculation results of the same field indicators (such as mean, median, etc.), or filling with the calculation results of different indicators (such as inferring missing values ​​through other relevant meteorological data).

[0034] In the prediction of wind turbine blade icing, special attention should be paid to data cleaning. For example, the Raida criterion (3σ criterion) can be used to filter out outliers in the data set, or the correlation coefficient between the characteristic variable and the icing state can be calculated through cyclic calculation, and the characteristic variables with smaller correlation coefficients can be deleted.

[0035] Data Integration: Data integration is the process of combining and unifying data from different sources and in different formats to form a complete data set.

[0036] Data extraction: Extract data related to wind turbine blade icing from multiple data sources (such as weather stations, wind turbine monitoring systems, etc.).

[0037] Data conversion: converting data in different formats into a unified format, such as converting text data into numerical data, or converting data of different time granularities into data of the same time granularity.

[0038] Data merging: Merge the transformed data into a complete data set for subsequent analysis and prediction.

[0039] In the process of data integration, attention should also be paid to the relevance and consistency of data. For example, data from different data sources can be associated through timestamps or other identifiers to ensure the accuracy and consistency of data.

[0040] Data standardization: Data standardization is the process of converting data into a uniform scale or range for subsequent analysis and comparison.

[0041] Normalization: Convert the data to a value between 0 and 1 to eliminate the dimensional differences between different data. This is achieved by subtracting the minimum value from the original data and dividing it by the difference between the maximum and minimum values.

[0042] Standardization: Convert the data to a standard normal distribution with a mean of 0 and a variance of 1. This is done by subtracting the mean from the original data and dividing by the standard deviation.

[0043] In the prediction of wind turbine blade icing, standardizing the data can improve the prediction accuracy of the model. For example, the z-score standardization method can be used to transform the data into a standard normal distribution, or the min-max standardization method can be used to transform the data into a value between 0 and 1.

[0044] Step 2: Establish a model for real-time rolling prediction and online modeling.

[0045] The present invention adopts real-time rolling prediction and online modeling because: Real-time rolling prediction is a prediction method that continuously updates the prediction results as new data arrives, allowing the model to continuously learn and adapt to data changes during the prediction process. Online modeling is a method that uses new samples and new labels to update the model, allowing the model to reflect the latest trends of the data in a timely manner.

[0046] In the prediction of wind turbine blade icing of the present invention, a dynamic prediction system can be formed by combining real-time rolling prediction and online modeling. The system can continuously collect new environmental parameters (such as wind speed, temperature, humidity, etc.) and wind turbine operation data, and update the prediction model through the online modeling method, thereby realizing real-time rolling prediction of wind turbine blade icing, improving the prediction real-time performance and prediction accuracy, because: Data dynamics: The prediction of wind turbine blade icing is affected by a variety of environmental factors, which are dynamically changing. Therefore, a prediction model that can continuously learn and adapt to data changes is required.

[0047] Prediction accuracy: Real-time rolling prediction can use the latest data information to make predictions, improving the accuracy and timeliness of predictions. Online modeling can update the model in a timely manner to better reflect the latest trends in the data.

[0048] Model adaptability: Online modeling allows the model to continuously learn and adjust during the prediction process, thereby enhancing the adaptability and robustness of the model.

[0049] The present invention adopts the advantages of real-time rolling prediction and online modeling model: Improve forecasting accuracy: Through real-time rolling forecasting and online modeling, the model can continuously learn and adapt to changes in data, thereby improving the accuracy and precision of the forecast.

[0050] Enhanced timeliness: Real-time rolling forecasts can use the latest data information to make forecasts in a timely manner, making the forecast results closer to the actual situation.

[0051] Reduce maintenance costs: Online modeling avoids relearning of old samples, reduces storage and computing overhead, and reduces model maintenance costs.

[0052] Improve decision-making efficiency: Accurate real-time prediction results can provide strong decision-making support for the operation and maintenance of wind power plants, improving decision-making efficiency and safety.

[0053] The present invention adopts a real-time rolling forecasting method to continuously update input data and forecasting models to adapt to the rapid changes in weather and wind farm operating conditions. It also uses online modeling technology to dynamically adjust model parameters and structures based on new data to improve the real-time and accuracy of forecasts.

[0054] In real-time rolling forecasting, the ARIMA model based on time series analysis is used to capture the dynamic changing characteristics of time series data.

[0055] The reason why the present invention uses the ARIMA model (1) Model applicability: The ARIMA model is a commonly used statistical model in time series forecasting. It combines autoregression (AR), difference (I) and moving average (MA) to capture the linear relationship in the time series. For time series data such as wind turbine blade icing, which shows linear trends and seasonal changes, the ARIMA model is a good choice.

[0056] (2) Data requirements: The ARIMA model requires that the time series data is stationary or can become stationary after differentiation. In the prediction of wind turbine blade icing, the environmental parameters and wind turbine operation data can be differentiated to meet the stationarity requirements of the ARIMA model.

[0057] (3) The need for real-time rolling forecasting: Real-time rolling prediction requires the model to continuously update the prediction results as new data arrives. The ARIMA model can achieve real-time rolling prediction by using new samples and new labels to update model parameters through online modeling.

[0058] Advantages of using ARIMA model (1) Forecast accuracy: The ARIMA model can capture the linear relationship in the time series, and has a high prediction accuracy for data with linear trends and seasonal changes. In the prediction of wind turbine blade icing, the ARIMA model can extract useful information through the analysis of historical data, thereby achieving accurate prediction of future icing conditions.

[0059] (2) Flexibility: The ARIMA model is flexible and can capture different types of time series characteristics by adjusting parameters (p, d, q). In practical applications, appropriate ARIMA model parameters can be selected according to the characteristics of the data and forecasting requirements to achieve the best forecasting effect.

[0060] (3) Dynamic update: The ARIMA model can be dynamically updated through online modeling, so that the model can reflect the latest trends of the data in a timely manner. In real-time rolling forecasting, this dynamic update capability is very important because it ensures that the forecast results are always based on the latest data information.

[0061] ARIMA model, full name Autoregressive Integrated Moving Average Model, is a commonly used model in time series analysis. It combines autoregression (AR) and moving average (MA) and can process non-stationary time series through difference (I) operation. The specific steps of ARIMA model are as follows: 1) Observe the stationarity and randomness of the time series: Use tools such as time series plots, autocorrelation plots (ACF), and partial autocorrelation plots (PACF) to observe the stationarity and randomness of the time series.

[0062] If the time series is non-stationary, then differencing is required until the series becomes stationary.

[0063] 2) Model Identification: Determine the order of the ARIMA model based on the ACF and PACF plots.

[0064] The ACF plot is used to examine the short-term dependencies between series, whereas the PACF plot focuses on the long-term dependencies of the autoregressive part.

[0065] By observing the tailing and truncation characteristics of the ACF and PACF graphs, the orders of AR and MA can be preliminarily determined.

[0066] 3) Model selection: Use information criteria (such as Akaike Information Criterion AIC and Schwarz Information Criterion BIC) to select the optimal ARIMA model.

[0067] The smaller the information criterion value, the better the model fit.

[0068] 4) Model parameter estimation: Estimate the parameters of the ARIMA model using methods such as Maximum Likelihood Estimation (MLE) or the Method of Least Squares (OLS).

[0069] 5) Model verification: Perform residual analysis on the fitted ARIMA model to check whether the residual is white noise.

[0070] If the residuals are not white noise, you may need to reselect your model or do some other processing.

[0071] 6) Prediction: Use the fitted ARIMA model to predict time series.

[0072] Online Modeling: Incremental learning is used in online modeling. Incremental learning is a machine learning paradigm that allows the model to be continuously updated over time without retraining the entire model. This method is particularly suitable for processing large-scale data sets or real-time data streams. The present invention uses an incremental learning algorithm to update the parameters of the above ARIMA model.

[0073] The reason why the present invention adopts the incremental learning method is: (1) Data continues to grow: In the actual application of wind turbine blade icing prediction, data is continuously growing. New data points are constantly generated, and the model needs to be able to learn and adapt to these new data in real time. The incremental learning method allows the model to be fine-tuned with new data based on the original knowledge, thereby achieving continuous optimization of model performance.

[0074] (2) Limited computing resources: In online modeling, computing resources are usually limited. Retraining the entire model may require a lot of computing time and resources, which is not feasible in real-time applications. Incremental learning methods avoid the overhead of retraining the entire model by updating the model step by step, thus saving computing resources.

[0075] (3) Real-time requirements: Online modeling usually requires that the model can be updated and predicted in real time. Incremental learning methods enable the model to quickly adapt to data changes by learning new data in real time, meeting real-time requirements.

[0076] Advantages of using incremental learning methods (1) Efficiency: Incremental learning methods improve the efficiency of model updates by gradually updating the model, avoiding the overhead of retraining the entire model. As data continues to grow, incremental learning methods can continuously optimize model performance without interrupting the normal operation of the system.

[0077] (2) Flexibility: Incremental learning methods allow the model to be updated immediately when new data is received, so that it can flexibly adapt to changes in data. This flexibility enables the model to better cope with uncertainties in practical applications and improve the accuracy and robustness of predictions.

[0078] (3) Scalability: Incremental learning methods have good scalability and can handle large-scale data streams. As the amount of data increases, the model can gradually expand its knowledge base to maintain high performance on new data.

[0079] (4) Memory and forgetting management: The incremental learning method can balance the learning of new knowledge and the retention of old knowledge by introducing memory consolidation and forgetting mechanisms. This ensures that the model will not lose important historical knowledge during the continuous learning process, while being able to quickly adapt to new data changes.

[0080] (5) Real-time feedback and adjustment: Incremental learning methods allow the model to receive feedback and make adjustments in real-time applications. This enables the model to be dynamically optimized based on the performance of the actual application, improving the overall performance of the system and user experience.

[0081] Online modeling methods include: 1) Initialize the model: Select an algorithm that supports incremental learning: SGDClassifier (Stochastic Gradient Descent Classifier). Create an instance of the algorithm and set the necessary parameters.

[0082] 2) Data input: Get new data in real time, which can be streaming data or data arriving in batches. Preprocess the new data, such as data cleaning and feature extraction.

[0083] 3) Model update: Use the partial_fit method to partially fit the new data and update the model parameters. Repeat this step and update the model every time there is new data.

[0084] 4) Evaluate performance: Use the validation set or test set to evaluate the performance of the model, such as accuracy, recall, etc. Adjust the model parameters based on the evaluation results.

[0085] 5) Save model: Save the updated model for later use.

[0086] The present invention adopts online modeling technology as a technical means to realize real-time rolling prediction. It uses incremental learning algorithm to update the parameters of ARIMA model in real time to achieve the purpose of online modeling, and then realizes real-time rolling prediction.

[0087] Step 3: Real-time prediction and warning: The updated model is deployed in the wind farm monitoring system to achieve real-time data input and icing risk prediction. When the icing risk is predicted to be high, the early warning mechanism is automatically triggered to notify the operation and maintenance personnel to take preventive measures, such as starting the de-icing device and adjusting the wind turbine operation strategy.

Claims

1. A wind turbine blade icing prediction method based on real-time rolling prediction and online modeling, characterized in that: The method comprises: Step 1: Collect historical meteorological data, terrain data and wind turbine operation data information, and clean, integrate and standardize the data; Step 2: Use a real-time rolling forecast method to continuously update input data and forecast models to adapt to changes in weather and wind farm operating conditions. Use online modeling technology to dynamically adjust model parameters and structures based on new data to improve the real-time and accuracy of forecasts. Step 3: Deploy the model in the wind farm monitoring system and realize icing risk prediction based on real-time data input.

2. The wind turbine blade icing prediction method based on real-time rolling prediction and online modeling according to claim 1 is characterized by: The historical meteorological data include: ambient temperature, air humidity, wind speed and direction, liquid water content and duration; terrain data include altitude, terrain slope, terrain orientation and surface cover; wind turbine operation data include blade speed, power output, vibration monitoring data, blade surface temperature data and operation log data.

3. The wind turbine blade icing prediction method based on real-time rolling prediction and online modeling according to claim 1 is characterized by: Data cleaning methods include: using clustering algorithms in unsupervised learning to identify and process outliers in the data set; for missing data, fill in with business knowledge or experience, fill in with the calculation results of the same field indicators, or fill in with the calculation results of different indicators; use the Laida criterion to filter out outliers in the data set, or calculate the correlation coefficient between the characteristic variable and the icing state through a loop calculation, and delete the characteristic variables with a small correlation coefficient; data integration methods include: extracting data related to wind turbine blade icing from various data sources; converting data in different formats into a unified format, including converting text data into numerical data, or converting data of different time granularities into data of the same time granularity; merging the converted data into a complete data set; standardization processing includes: converting data into a unified scale or range, converting data into values ​​between 0 and 1 to eliminate dimensional differences between different data; converting data into a standard normal distribution with a mean of 0 and a variance of 1.

4. The wind turbine blade icing prediction method based on real-time rolling prediction and online modeling according to claim 1 is characterized by: Methods for continuously updating input data and forecast models using a real-time rolling forecast approach include: Step 3.1, use the time series diagram, autocorrelation diagram ACF and partial autocorrelation diagram PACF to observe the stationarity and randomness of the time series; Step 3.2: Determine the order of the ARIMA model based on the ACF and PACF graphs; Step 3.3, use the information criterion Akaike Information Criterion AIC or Schwarz Information Criterion BIC to select the optimal ARIMA model; Step 3.4: Estimate the parameters of the ARIMA model using maximum likelihood estimation or least squares method; Step 3.5: Perform residual analysis on the fitted ARIMA model to check whether the residual is white noise. If the residual is not white noise, reprocess it; Step 3.6: Use the fitted ARIMA model to predict the time series.

5. The wind turbine blade icing prediction method based on real-time rolling prediction and online modeling according to claim 4 is characterized by: The method of using online modeling technology to dynamically adjust model parameters and structure according to new data includes: Step 3.

7. Select an algorithm that supports incremental learning. Step 3.8, obtain new data in real time and pre-process the new data; Step 3.9: Use the partial_fit method to partially fit the new data and update the parameters of the ARIMA model. Repeat this step and update the model every time there is new data. Step 3.10: Use the validation set or test set to evaluate the performance of the ARIMA model and adjust the model parameters based on the evaluation results. Step 3.

11. Save the updated model for later use.