Variable frequency converter temperature rise prediction method and system, electronic device and computer program product

Abstract

The application provides a frequency converter temperature rise prediction method and system, an electronic device and a computer program product, wherein the method first acquires real-time operation data of the frequency converter; then inputs the real-time operation data into a plurality of preconfigured prediction models, receives a plurality of temperature rise prediction sequences output by the plurality of prediction models, wherein at least two prediction models in the plurality of prediction models are configured to predict temperature rises in different lengths of time, and each temperature rise prediction sequence corresponds to a length of time; and finally, based on the length of time corresponding to each temperature rise prediction sequence, the plurality of temperature rise prediction sequences are integrated according to a preset rule to generate a combined prediction result covering a plurality of lengths of time. Comprehensive temperature rise prediction can be provided for different decision-making needs from immediate operation to long-term planning.

Description

Technical Field

[0001] This application belongs to the field of industrial equipment management technology, and in particular relates to a method, system, electronic device and computer program product for predicting the temperature rise of a frequency converter. Background Technology

[0002] High-voltage frequency converters are key equipment in industrial production for realizing motor drive and energy consumption control. The temperature changes of their internal power devices and other core components during operation directly affect the reliability, lifespan, and stability of the entire production system. Effectively predicting the temperature rise of frequency converters helps to identify potential failure risks in advance and is an important means of achieving predictive maintenance and ensuring production continuity.

[0003] In related technologies, a common approach to predicting equipment status parameters is to collect historical operating data of the equipment and use this data to train a fixed prediction model. This model is then used to infer the trend of target parameters over a future period based on the data at the current moment.

[0004] However, these technical solutions based on a single prediction model have limitations in the form and application of their output prediction information. In real-world industrial scenarios, decision-making tasks ranging from real-time operation to long-term maintenance often require differentiated and multi-perspective prediction information. The prediction results provided by existing solutions are difficult to flexibly adapt to such diverse decision-making needs, resulting in the prediction system's practical value not being fully realized. Summary of the Invention

[0005] This application provides a method, system, electronic device, and computer program product for predicting the temperature rise of a frequency converter, which can provide comprehensive temperature rise prediction for different decision-making needs, from immediate operation to long-term planning.

[0006] The first aspect of this application provides a method for predicting the temperature rise of a frequency converter, comprising: acquiring real-time operating data of the frequency converter; inputting the real-time operating data into a plurality of pre-configured prediction models; receiving a plurality of temperature rise prediction sequences output by the plurality of prediction models, wherein at least two of the plurality of prediction models are configured to predict temperature rise at different time lengths, and each temperature rise prediction sequence corresponds to a time length; and integrating the plurality of temperature rise prediction sequences according to a preset rule based on the time length corresponding to each temperature rise prediction sequence to generate a combined prediction result covering multiple time lengths.

[0007] In the technical solution of this application, by inputting the real-time operating data of the frequency converter into multiple models that predict different time lengths, and integrating them based on the time lengths corresponding to each temperature rise prediction sequence, a combined prediction result covering multiple time lengths is generated, which can provide comprehensive temperature rise prediction for different decision-making needs from immediate operation to long-term planning.

[0008] Optionally, in one possible implementation of the first aspect, before inputting the real-time running data into the pre-configured multiple prediction models, the method further includes: configuring the multiple prediction models to handle prediction tasks of different time lengths, wherein at least one short-term prediction model among the multiple prediction models is configured to perform short-term prediction tasks, and at least one long-term prediction model among the multiple prediction models is configured to perform long-term prediction tasks, the time length corresponding to the long-term prediction tasks being longer than the time length corresponding to the short-term prediction tasks. Thus, by explicitly configuring the multiple prediction models to handle short-term and long-term prediction tasks respectively, with the long-term prediction task having a longer time frame than the short-term prediction task, different models can optimize predictions for their specific time lengths of expertise, forming a complementary division of prediction capabilities.

[0009] Optionally, in another possible implementation of the first aspect, the above-mentioned multiple prediction models are configured to handle prediction tasks of different time lengths, including: acquiring historical operating data of the frequency converter and constructing an initial sample set based on the historical operating data; dividing the initial sample set into a training dataset and a validation dataset; constructing a first hyperparameter search space for a preset first initial model, and using a Bayesian optimization algorithm to update model parameters with the training dataset and evaluate prediction errors with the validation dataset to search for a first optimal hyperparameter combination in the first hyperparameter search space; using the first optimal hyperparameter combination, training the first initial model on the training dataset, and determining that the first initial model has been trained successfully based on the validation dataset during the training process to obtain a short-term prediction model for short-term prediction tasks; constructing a second hyperparameter search space for a preset second initial model, and using a Bayesian optimization algorithm to update model parameters with the training dataset and evaluate prediction errors with the validation dataset to search for a second optimal hyperparameter combination in the second hyperparameter search space; using the second optimal hyperparameter combination, training the second initial model on the training dataset, and determining that the second initial model has been trained successfully based on the validation dataset during the training process to obtain a long-term prediction model for long-term prediction tasks. Therefore, by employing the Bayesian optimization algorithm, which updates parameters on the training set and evaluates errors on the validation set, the optimal hyperparameters for the first and second initial models are searched and trained, respectively. This automatically and efficiently yields the short-term and long-term prediction models that are optimal for their respective short-term and long-term tasks, improving the accuracy and efficiency of model configuration.

[0010] Optionally, in another possible implementation of the first aspect, before inputting the real-time running data into the pre-configured multiple prediction models, the method further includes: performing feature enhancement processing on the real-time running data to obtain enhanced feature data; and inputting the real-time running data into the pre-configured multiple prediction models, including: inputting the enhanced feature data into the multiple prediction models. Thus, by performing feature enhancement processing on the real-time running data to obtain enhanced feature data before inputting it into the prediction models, the models are provided with richer information and stronger representational capabilities, which helps to improve the accuracy of subsequent temperature rise predictions.

[0011] Optionally, in another possible implementation of the first aspect, the aforementioned feature enhancement processing of the real-time operating data to obtain enhanced feature data includes: calculating at least one physical derived feature based on the electrical parameters in the real-time operating data, wherein the physical derived feature includes at least one of input power, output power, inverter efficiency, unit load energy consumption, and load fluctuation rate; and combining the at least one physical derived feature with the real-time operating data to obtain the enhanced feature data. Thus, by calculating derived features with clear physical meaning, such as power and efficiency, based on electrical parameters and combining them with the original data, the enhanced feature data more directly reflects the physical mechanism of inverter heating, further improving the accuracy and generalization ability of the model prediction.

[0012] Optionally, in another possible implementation of the first aspect, the above-mentioned integration of multiple temperature rise prediction sequences according to preset rules based on the time length corresponding to each temperature rise prediction sequence to generate a combined prediction result covering multiple time lengths includes: extracting multiple candidate prediction values ​​corresponding to the target future time from all temperature rise prediction sequences for the target future time to be determined in the combined prediction result; and determining the candidate prediction value with the shortest corresponding time length as the prediction result for the target future time from the multiple candidate prediction values. Thus, by prioritizing the selection of the candidate value with the shortest corresponding prediction time length as the final result for a certain future time when integrating multi-scale prediction sequences, it is ensured that the prediction value of the combined prediction result in the near future is more likely to be provided by a short-term, highly sensitive model, thereby improving the timeliness of short-term predictions and its guiding value for immediate operations.

[0013] Optionally, in another possible implementation of the first aspect, the method further includes: determining the trigger period for each prediction model configured with a time length, wherein the trigger period for each prediction model is calculated based on the configured time length; and periodically triggering the steps of acquiring real-time running data and inputting it into the prediction model to receive the temperature rise prediction sequence according to the trigger period for each prediction model. Thus, by setting a trigger period for each prediction model based on its configured time length, and periodically triggering the prediction process accordingly, short-term prediction tasks can be updated more frequently, while long-term prediction tasks run at a reasonable frequency, achieving adaptive allocation and efficient utilization of the prediction system's computing resources.

[0014] A second aspect of this application provides a frequency converter temperature rise prediction system, comprising:

[0015] The data acquisition module is used to acquire the real-time operating data of the frequency converter.

[0016] The model invocation module is used to input real-time running data into multiple pre-configured prediction models and receive multiple temperature rise prediction sequences output by multiple prediction models. Among the multiple prediction models, at least two prediction models are configured to predict temperature rise at different time lengths, and each temperature rise prediction sequence corresponds to a time length.

[0017] The temperature rise prediction module is used to integrate multiple temperature rise prediction sequences according to preset rules based on the time length corresponding to each temperature rise prediction sequence, so as to generate a combined prediction result covering multiple time lengths.

[0018] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the inverter temperature rise prediction method of the first aspect described above.

[0019] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the inverter temperature rise prediction method of the first aspect described above.

[0020] The fifth aspect of this application provides a computer program product that, when run on an electronic device, causes the electronic device to execute the inverter temperature rise prediction method of the first aspect described above.

[0021] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a method for predicting the temperature rise of a frequency converter according to an embodiment of this application; Figure 2 This is a schematic diagram of the operation interface for device activation prediction provided in an embodiment of this application; Figure 3 This is a schematic diagram of the data table format provided in the embodiments of this application; Figure 4 This is a schematic diagram of the temperature rise prediction interface provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a frequency converter temperature rise prediction system provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0025] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0026] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0027] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0028] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0029] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0030] It should be understood that the sequence number of each step in this embodiment does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of this application embodiment.

[0031] In related technologies, a common approach to predicting equipment status parameters is to collect historical operating data and use this data to train a fixed prediction model. This model is then used to infer the future trend of target parameters based on current data. However, this type of solution, based on a single prediction model, has limitations in the form and application of its output prediction information. In real-world industrial scenarios, decision-making tasks ranging from real-time operation to long-term maintenance often require differentiated and multi-perspective prediction information. Existing solutions struggle to flexibly adapt to these diverse decision-making needs, thus failing to fully realize the practical value of the prediction system.

[0032] In view of this, embodiments of this application provide a method, system, electronic device, and computer program product for predicting the temperature rise of a frequency converter. The method first acquires real-time operating data of the frequency converter; then, it inputs the real-time operating data into multiple pre-configured prediction models, receiving multiple temperature rise prediction sequences output by these models. At least two of the prediction models are configured to predict temperature rises at different time lengths, with each temperature rise prediction sequence corresponding to a specific time length. Finally, based on the time length corresponding to each temperature rise prediction sequence, the multiple temperature rise prediction sequences are integrated according to preset rules to generate a combined prediction result covering multiple time lengths. Thus, by inputting the real-time operating data of the frequency converter into multiple models predicting different time lengths and integrating them based on the time lengths corresponding to each temperature rise prediction sequence, a combined prediction result covering multiple time lengths is generated, providing comprehensive temperature rise predictions for different decision-making needs, from immediate operation to long-term planning.

[0033] The following example illustrates the application scenario of the inverter temperature rise prediction method provided in this application. In industrial production scenarios where high-voltage inverters operate continuously, equipment is prone to abnormal temperature rise due to long-term high-load operation. Traditional single prediction models can only provide temperature trends for a fixed time window, failing to simultaneously meet the real-time monitoring needs of operators for short-term overheating risks (e.g., the next 30 minutes) and the planning needs of users for long-term thermal aging trends (e.g., the next 24 hours). Using the method provided in this application, the system simultaneously generates multi-scale temperature rise prediction curves ranging from minutes to hours by calling multiple pre-configured prediction models optimized for different time lengths in parallel and integrating their outputs. This allows users to adjust operating parameters or cooling strategies in a timely manner based on short-term predictions to avoid immediate risks, while also enabling users to scientifically schedule preventative maintenance windows based on long-term predictions. This achieves a comprehensive understanding of the equipment's thermal state and differentiated decision support, significantly improving the accuracy of predictive maintenance and the initiative in production management.

[0034] To illustrate the technical solution of this application, specific embodiments are described below.

[0035] Reference Figure 1 This diagram illustrates a flowchart of a frequency converter temperature rise prediction method provided in an embodiment of this application. This method can be executed by a frequency converter temperature rise prediction system. Figure 1 As shown, the inverter temperature rise prediction method may include the following steps: Step 101: Obtain the real-time operating data of the frequency converter.

[0036] In this embodiment, real-time operating data may include core electrical parameters and status variables that directly affect temperature rise, such as: bus voltage, given frequency, operating frequency, load, inverter status, cabinet ambient temperature, input current, input voltage, output current, output voltage, and cumulative operating time. Raw data can be obtained through an Application Programming Interface (API) request, and initial data structure integration and loading can be performed using Python's pandas library.

[0037] In one embodiment, after acquiring the real-time operating data of the frequency converter, a standardized preprocessing procedure can be executed. First, the original data stream can be resampled at different time intervals according to the needs of different subsequent prediction models to unify the data frequency. For example, high-frequency data (e.g., 2-minute intervals) can be prepared for short-term prediction models, and low-frequency data (e.g., 30-minute intervals) can be prepared for long-term prediction models. Next, missing values ​​and outliers are handled. For missing values, a forward imputation method can be used. Specifically, the missing value handling threshold can be dynamically set according to the resampling frequency: when the sampling interval is 2 minutes, forward imputation is performed if there are fewer than 30 consecutive missing values ​​in a field; when the interval is 6 minutes, the threshold is set to 10; when the interval is 12 minutes, the threshold is set to 5; and when the interval is 30 minutes, the threshold is set to 2. The number of missing values ​​is counted using `df[col].isna().sum()`, and `df[col].ffill()` is called to fill the missing values. For outliers, i.e., values ​​that deviate significantly from most data points, the interquartile range method can be used for identification: calculate the upper and lower quartiles of the data, and consider data points that exceed the interquartile range by a certain range (usually 1.5 times) as outliers, replacing them with the linear average of the data near the outlier. This preprocessing strategy, which combines dynamic thresholding with statistical methods, significantly enhances data quality and the robustness of subsequent models.

[0038] In one embodiment, after preprocessing, the data can be subjected to feature analysis and enhancement to construct an input feature set strongly correlated with the temperature rise physical process. Specifically, feature correlation analysis can be performed first: calculate the Pearson correlation coefficient and Spearman rank correlation coefficient between each input feature and the target variable (such as the inverter unit temperature), and combine this with a significance test (p-value < 0.05) to screen out the core feature set that is statistically significant and whose absolute correlation coefficient is greater than a set threshold (e.g., 0.4). At the same time, redundant variables with high multicollinearity among the features are checked and removed.

[0039] Feature enhancement can include both physical mechanism-based enhancement and time series statistics-based enhancement. For physical mechanism-based enhancement, new features with clear physical meaning can be constructed from the original features based on the inverter's operating principle and electrical formulas. For example, input power can be calculated using input voltage and current; output power and instantaneous power can be calculated using output voltage and current; inverter efficiency can be calculated using output power and input power; unit load energy consumption can be calculated using instantaneous power and load; and load volatility can be calculated using the standard deviation and average value of the load series. These derived features are directly related to energy conversion and heat loss processes. For time series statistics-based enhancement, the feature engineering concepts of the tsfresh (TimeSeries Feature Extraction on Basis of Scalable Hypothesis Tests) library can be used to calculate a series of statistical and morphological derived features for time series data to capture deep patterns in the data.

[0040] tsfresh is a Python library specifically designed for the automated extraction of numerous features from time series data. Its main function is to automatically extract hundreds of features from time series data, including statistical features (mean, variance, skewness, kurtosis), time-domain features (autocorrelation, partial autocorrelation), frequency-domain features (Fourier transform coefficients), information-theoretic features (entropy, complexity), physical features (fractal dimension, Lyapunov exponent), and model features (ARIMA parameters). It should be noted that the features obtained using the tsfresh library have some redundancy. This application's embodiments only refer to the idea of ​​feature expansion using the tsfresh library to calculate specific derived features for the data, including trend strength features (quantifying the linear trend strength, rate of change, and stability of time series data within a specific window), momentum features, volatility features, peak features, trough features, difference features, and time features. Using these derived features along with the original time series data as input can better improve the model's predictive performance.

[0041] Step 102: Input the real-time running data into multiple pre-configured prediction models and receive multiple temperature rise prediction sequences output by the multiple prediction models.

[0042] Among the multiple prediction models, at least two prediction models are configured to predict temperature rise at different time lengths, with each temperature rise prediction sequence corresponding to a time length.

[0043] In this embodiment, weight files of multiple pre-trained prediction models can be loaded. These models constitute a carefully designed heterogeneous model pool, containing various deep learning neural network models with complementary characteristics in the field of time series prediction. Specifically, they may include: Temporal Convolutional Network (TCN), which has causal convolution and dilated convolution structures, high parallel computation efficiency, sensitivity to local temporal patterns, and is good at capturing short-term periodic patterns; Long Short-Term Memory (LSTM), which can effectively learn medium- and long-term temporal dependencies through its gating mechanism; and the Informer model, which is an improved model of the traditional Transformer. Through the ProbSparse self-attention mechanism and distillation operation, it significantly optimizes the computational efficiency and memory consumption of long sequence prediction, and is particularly suitable for capturing long-term global trends under large time windows.

[0044] In one embodiment, to systematically implement the aforementioned multi-scale prediction division of labor, multiple prediction models need to be explicitly configured before inputting real-time running data into the models. Specifically, multiple prediction models can be configured to handle prediction tasks of different time lengths. At least one short-term prediction model is configured to perform short-term prediction tasks, and at least one long-term prediction model is configured to perform long-term prediction tasks, with the time length corresponding to the long-term prediction tasks being longer than that of the short-term prediction tasks. Thus, by explicitly configuring multiple prediction models to handle short-term and long-term prediction tasks respectively, and with the long-term prediction tasks having a longer time frame than the short-term prediction tasks, different models can optimize predictions for their specific time lengths of expertise, forming a complementary division of prediction capabilities.

[0045] Specifically, the models in the model pool are configured to handle prediction tasks of different time lengths. Short-term prediction models (such as TCN) have causal convolutional structures, are sensitive to local patterns, and are suitable for capturing short-term periodic patterns; long-term prediction tasks (such as Informer) optimize the computational efficiency of long-sequence prediction and are particularly suitable for capturing long-term global dependency trends. By explicitly assigning short-term and long-term tasks to model categories with matching characteristics, a complementary prediction model pool is formed.

[0046] Each prediction model underwent deep optimization during offline training using a multi-model training and Bayesian optimization module. For each model architecture, the system defines its own hyperparameter search space. For example, for the TCN model, its hyperparameter space includes the number of convolutional filters (num_filters), kernel size (kernel_size), number of residual block layers (num_layers), dilation base (dilation_base), dropout rate (dropout_rate), and learning rate (learning_rate); for the LSTM model, its hyperparameter space includes the number of hidden units (units), number of stacked layers (num_layers), recurrent dropout rate (recurrent_dropout_rate), dropout rate, and learning rate; for the Informer model, its hyperparameter space includes model dimension (d_model), number of attention heads (n_heads), number of encoder layers (e_layers), number of decoder layers (d_layers), feedforward network dimension (d_ff), dropout rate, and learning rate. A Bayesian optimization algorithm is employed to minimize prediction error metrics on the validation dataset, such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). With this as the optimization objective, the algorithm intelligently searches and determines the optimal combination of hyperparameters for each model within its respective hyperparameter space. Subsequently, these optimal hyperparameters are used to train the corresponding models on the complete training dataset, generating the final model weight file and storing it in the model repository. This automated optimization process completely replaces inefficient manual hyperparameter tuning, ensuring that each heterogeneous model in the model pool achieves optimal performance on its prediction task.

[0047] In one embodiment, configuring multiple prediction models to handle prediction tasks of different time lengths is achieved through an offline, automated, and optimized training process. Specifically, historical operating data of the frequency converter can be acquired, and an initial sample set can be constructed based on the historical operating data. The initial sample set is divided into a training dataset and a validation dataset. For a preset first initial model, a first hyperparameter search space is constructed, and a Bayesian optimization algorithm is used to search for a first optimal hyperparameter combination in the first hyperparameter search space by updating model parameters with the training dataset and evaluating prediction errors with the validation dataset. Using the first optimal hyperparameter combination, the first initial model is trained on the training dataset, and during the training process, the completion of the first initial model training is determined based on the validation dataset, resulting in a short-term prediction model for short-term prediction tasks. For a preset second initial model, a second hyperparameter search space is constructed, and a Bayesian optimization algorithm is used to search for a second optimal hyperparameter combination in the second hyperparameter search space by updating model parameters with the training dataset and evaluating prediction errors with the validation dataset. Using the second optimal hyperparameter combination, the second initial model is trained on the training dataset, and during the training process, the completion of the second initial model training is determined based on the validation dataset, resulting in a long-term prediction model for long-term prediction tasks. Therefore, by employing the Bayesian optimization algorithm, which updates parameters on the training set and evaluates errors on the validation set, the optimal hyperparameters for the first and second initial models are searched and trained, respectively. This automatically and efficiently yields the short-term and long-term prediction models that are optimal for their respective short-term and long-term tasks, improving the accuracy and efficiency of model configuration.

[0048] It should be noted that an initial sample set can be constructed by acquiring historical operating data of the frequency converter over a long period (e.g., the past year), and this data can be divided into a training dataset and a validation dataset. For a pre-defined first initial model (e.g., TCN) intended for short-term tasks, a first hyperparameter search space (e.g., kernel size, number of layers, learning rate) is constructed. A Bayesian optimization algorithm is used to intelligently optimize the search space to minimize the prediction errors (e.g., mean absolute error (MAE) and root mean square error (RMSE) on the validation dataset, obtaining the first optimal hyperparameter combination. Subsequently, this optimal combination is used to train the first initial model on the training dataset, and performance is monitored based on the validation dataset during training to determine completion, ultimately yielding a high-performance first-class model for short-term prediction tasks. In parallel, the above process can be repeated for a pre-defined second initial model (e.g., Informer) intended for long-term tasks: a second hyperparameter search space is constructed, a second optimal hyperparameter combination is obtained through Bayesian optimization, and a second-class model is trained. This process uses Bayesian optimization to replace inefficient manual hyperparameter tuning, automatically finding the optimal hyperparameter combination for each heterogeneous model, ensuring that the entire model pool can perform at its best on the corresponding task scale, thus guaranteeing the overall accuracy and stability of multi-scale prediction from the source.

[0049] In one embodiment, before inputting real-time operational data into multiple pre-configured prediction models, the system can pre-process the data with feature enhancement to improve the signal quality and prediction accuracy of the model input. Specifically, the real-time operational data can be feature-enhanced to obtain enhanced feature data; this enhanced feature data is then input into the multiple prediction models. Thus, by performing feature enhancement processing on the real-time operational data before inputting it into the prediction models to obtain enhanced feature data, the models are provided with richer and more representative inputs, which helps improve the accuracy of subsequent temperature rise predictions.

[0050] In this process, the raw real-time operational data obtained from step 101 undergoes feature enhancement processing. By calculating and fusing new information dimensions, enhanced feature data with richer information and stronger correlation to the physical process of temperature rise is obtained. Therefore, the actual input objects of multiple prediction models can be the enhanced feature data. Deeply processing the raw operational data provides the model with more direct and effective input information, which is an important means to improve the model's prediction accuracy and generalization ability.

[0051] In one embodiment, the core of the aforementioned feature enhancement processing is to construct derived features with clear physical meaning based on the electrical and thermodynamic physical principles of inverter heating. Specifically, at least one physical derived feature can be calculated based on electrical parameters in real-time operating data. This physical derived feature includes at least one of input power, output power, inverter efficiency, energy consumption per unit load, and load fluctuation rate. Combining this at least one physical derived feature with the real-time operating data yields enhanced feature data. Thus, by calculating derived features with clear physical meaning, such as power and efficiency, based on electrical parameters and combining them with the original data, the enhanced feature data more directly reflects the physical mechanism of inverter heating, further improving the accuracy and generalization ability of the model prediction.

[0052] For example, input power can be calculated using input voltage and current; output power and instantaneous power can be calculated using output voltage and current; inverter efficiency can be calculated using output power and input power; unit load energy consumption can be calculated using instantaneous power and load; and load fluctuation rate can be calculated using the standard deviation and average value of the load sequence. These features directly reflect the intrinsic factors causing temperature changes, such as losses and energy conversion efficiency. Subsequently, one or more calculated physical derivative features are combined with the original real-time operating data to form the final enhanced feature data. This end-to-end feature engineering strategy for industrial applications enables the model to learn physical laws strongly correlated with temperature rise, rather than merely fitting surface statistical relationships, thereby significantly improving the model's prediction accuracy and generalization ability to different operating conditions.

[0053] In this embodiment, the system can dynamically extract time-series data of a corresponding length from the data stream based on the pre-assigned prediction time scale (window size) for each prediction model (e.g., extracting the sequence of the most recent hours for a model predicting the next 30 minutes). This data, after standardization, is input in parallel into the corresponding prediction model. Each prediction model loads its optimal weights, performs forward inference independently, and outputs its corresponding scale's future temperature rise prediction sequence, i.e., the temperature rise prediction sequence. The configuration of models and scales follows explicit task mapping rules. For example, the TCN model is assigned prediction tasks for the next 30 minutes (H=30) and 60 minutes (H=60); the LSTM model is assigned a prediction task for the next 120 minutes (H=120); and the Informer model is assigned prediction tasks for the next 360 minutes, 720 minutes, and 1440 minutes (i.e., 6 hours, 12 hours, and 24 hours).

[0054] In one embodiment, to enable the multi-scale prediction system to operate continuously, efficiently, and automatically, and to adapt to the update frequency requirements of predictions at different scales, a trigger period can be independently set for each prediction model configured with a specific time length. Specifically, the trigger period for each prediction model with a configured time length can be determined, wherein the trigger period for each prediction model is calculated based on the configured time length; according to the trigger period for each prediction model, the steps of acquiring real-time running data and inputting it into the prediction model to receive the temperature rise prediction sequence are periodically triggered. Thus, by setting a trigger period for each prediction model based on its configured time length, and periodically triggering the prediction process accordingly, short-term prediction tasks can be updated more frequently, while long-term prediction tasks run at a reasonable frequency, achieving adaptive allocation and efficient utilization of the prediction system's computing resources.

[0055] It should be noted that the triggering period mentioned above is not fixed, but is dynamically calculated based on the length of time the prediction model is configured. The formula for calculating the triggering period is shown below: ; in, I For triggering period; The duration; k This is a scaling factor, for example, 0.2. Based on the trigger cycle calculated for each model, the system periodically and independently triggers the complete prediction process for that specific model through a background task scheduler. In other words, models responsible for short-term predictions (e.g., the next 30 minutes) will be started frequently at shorter intervals (e.g., 6 minutes) to ensure timely updates to the short-term prediction curves; while models responsible for long-term predictions (e.g., the next 24 hours) will run at longer intervals (e.g., 4.8 hours), balancing the timeliness of predictions with the consumption of system computing resources. This maintainable and scalable engineering design enables the system to operate stably for a long time and provides users with decision support at the required update frequency, making it highly practical.

[0056] In one possible implementation, the entire prediction task can be formally defined as a quadruple: ; in, For the model that performs inference, ; H The predicted time length (in minutes). I The trigger time interval for the reasoning task (in minutes). This is a data table for storing the prediction results of this task. The system executes the task according to a preset mapping relationship, for example: ; ; ; ; ; ; in, This indicates that the 30-minute prediction task is executed by the TCN model every 6 minutes, and the results are stored in a data table. Triggering cycle I Based on the predicted time length H The calculation is obtained using the aforementioned trigger cycle calculation formula.

[0057] In this embodiment, prediction results from different models and scales are aligned and associated according to the timestamp triggered by the current prediction (as the prediction result version number), and the database configuration module is invoked to store the results into pre-configured data tables corresponding to the tasks. For example, the 30-minute prediction sequence output by the TCN model is stored in the data table. 60-minute predicted sequence stored The 120-minute prediction sequence output by the LSTM model is stored in... The 360-minute, 720-minute, and 1440-minute prediction sequences output by the Informer model are stored in... , , Each data table's field structure includes metadata such as device identifier (ID), prediction target identifier, data time point, predicted temperature value, version number, model name, and the input sequence time period used for inference. This result storage architecture based on task mapping and version number management enables the orderly organization and efficient management of multi-scale prediction results, providing a solid foundation for subsequent combination and visualization, as well as the system's maintainability and traceability.

[0058] Step 103: Based on the time length corresponding to each temperature rise prediction sequence, integrate multiple temperature rise prediction sequences according to preset rules to generate a combined prediction result covering multiple time lengths.

[0059] In this embodiment, the multi-scale prediction sequences generated in parallel and scattered across different data tables obtained in the aforementioned embodiments can be fused into a unified and coherent prediction trend view according to preset rules, and presented to the user through a visualization and interactive interface.

[0060] In one embodiment, the integration process of multiple temperature rise prediction sequences is performed for each target future time that needs to be determined in the combined prediction result. Specifically, for the target future time to be determined in the combined prediction result, multiple candidate predicted values ​​corresponding to the target future time can be extracted from all temperature rise prediction sequences; from the multiple candidate predicted values, the candidate predicted value with the shortest corresponding time length is determined as the prediction result for the target future time. Thus, by prioritizing the selection of the candidate value with the shortest corresponding prediction time length as the final result when integrating multi-scale prediction sequences, it is ensured that the prediction value of the combined prediction result in the near future is more likely to be provided by a short-term, highly sensitive model, thereby improving the timeliness of short-term predictions and their guiding value for immediate operations.

[0061] It's important to note that all predicted values ​​for the target's future time can be extracted from the temperature rise prediction sequences output by all models, forming multiple candidate predicted values. Then, from these candidate predicted values, the one with the shortest corresponding time period is determined as the unique prediction for that target's future time in the final combined prediction result. This is because, for the same future time, models with shorter prediction scales rely on historical data windows closer to that time for training and inference, and their model architecture is usually more sensitive to recent patterns, thus their predictions for that time often have higher confidence. For example, for the time "25 minutes into the future," the 30-minute prediction model, the 1-hour prediction model, and the 2-hour prediction model might all output a value; the system will prioritize the value output by the 30-minute prediction model. This rule ensures that the combined prediction result presented to the user at each point in time is the most reliable and timely prediction among all currently available information, which is particularly beneficial for providing precise guidance for short-term operations.

[0062] For a given version number v Targeting the future All candidate predicted values ​​set It can be expressed by the following formula: ; in, Candidate predicted values; For the set of all prediction tasks; the target's future time. The model's optimal prediction (i.e., the prediction result for the target at a future time). It can be expressed by the following formula: ; Among them, it can be sorted by time length H Sort in ascending order, for the same H You can sort by version number vSort in ascending order and select the candidate predicted value that is ranked first. The predicted result as the ultimate goal at a future moment.

[0063] The inverter temperature rise prediction method disclosed in the above embodiments of this application first acquires the real-time operating data of the inverter; then, it inputs the real-time operating data into multiple pre-configured prediction models and receives multiple temperature rise prediction sequences output by these models. At least two of the prediction models are configured to predict temperature rise over different time lengths, with each temperature rise prediction sequence corresponding to a specific time length. Finally, based on the time length corresponding to each temperature rise prediction sequence, the multiple temperature rise prediction sequences are integrated according to preset rules to generate a combined prediction result covering multiple time lengths. Therefore, by inputting the real-time operating data of the inverter into multiple models predicting different time lengths and integrating them based on the time lengths corresponding to each temperature rise prediction sequence, a combined prediction result covering multiple time lengths is generated, providing comprehensive temperature rise prediction for different decision-making needs, from immediate operation to long-term planning.

[0064] To facilitate understanding of the above embodiments, the following detailed explanation of the inverter temperature rise prediction method described in the foregoing embodiments will be provided using a specific inverter temperature rise prediction example.

[0065] First, historical sensor data for the specified frequency converter over the past year is obtained in batches and divided into time blocks through a historical database interface (such as HBase). Data fields include bus voltage, given frequency, operating frequency, load, frequency converter status, cabinet ambient temperature, input current and voltage, output current and voltage, cumulative operating time, and the frequency converter unit temperature and transformer temperature used as prediction targets. Missing and outlier values ​​in all data are identified and processed: missing values ​​are filled using forward imputation, with dynamic thresholds set according to different resampling frequencies (e.g., 2 minutes, 6 minutes); outliers are identified using an interquartile range-based method and corrected using linear interpolation. Next, the processed data is uniformly resampled at 2-minute intervals to prepare regularized time-series data for subsequent steps.

[0066] Subsequently, feature engineering and feature enhancement were performed. Pearson correlation coefficient and Spearman rank correlation coefficient analyses were conducted on historical data to identify a core feature set significantly correlated with the target temperature rise. Furthermore, physically derived features, such as input power, output power, inverter efficiency, energy consumption per unit load, and load volatility, were calculated based on electrophysical principles. Simultaneously, drawing inspiration from the tsfresh library, statistically derived features of the time series were calculated, such as trend strength, momentum, and volatility. The original features, physically derived features, and statistically derived features were combined to construct an enhanced feature dataset for model training.

[0067] Next, we proceed to the multi-model training and optimization phase. A heterogeneous model pool is constructed, comprising Temporal Convolutional Networks (TCN), Long Short-Term Memory Networks (LSTM), and Informer models. A hyperparameter search space is defined for each model, as shown in Table 1 (hyperparameter space for TCN models), Table 2 (hyperparameter space for LSTM models), and Table 3 (hyperparameter space for Informer models). Table 1

[0068] Table 2

[0069] Table 3

[0070] A Bayesian optimization algorithm is employed to automatically find and determine the optimal hyperparameter combination for each model, aiming to minimize the mean absolute error (MAE) and root mean square error (RMSE) on the validation set. The optimal hyperparameters are then used to train each model on a historical augmentation feature dataset, and the trained model weight files are saved, completing the model pre-configuration.

[0071] See also Figure 2 The diagram shown illustrates the user interface for enabling prediction on the device. Figure 2 As shown, after model deployment, the system enters the online prediction phase. The "AI Prediction" function is activated for a specified device via the interface. Upon receiving the request, the server acquires the device's operational data stream in real time and performs resampling, preprocessing, and feature enhancement consistent with the offline phase. The online prediction service engine loads the optimal model weights for TCN, LSTM, and Informer in parallel, and inputs the processed real-time data into these models simultaneously according to the task mapping relationship. Specifically, the TCN model predicts the temperature rise for the next 30 minutes and 1 hour, the LSTM model predicts the temperature rise for the next 2 hours, and the Informer model predicts the temperature rise for the next 6 hours, 12 hours, and 24 hours. Each model infers in parallel, outputting its respective scale's predicted temperature rise sequence.

[0072] The prediction results are managed based on the version number generated at the prediction trigger time, and stored in different database tables according to the model and prediction scale. The specific data table format is as follows: Figure 3 As shown. The front-end visualization interface queries these data tables and, according to the "shortest scale priority" rule, optimizes and combines multi-scale prediction values ​​for the same future time, plotting historical temperature curves and combined prediction curves from short-term to long-term in the same view. See also... Figure 4The diagram shown illustrates the temperature rise prediction interface, allowing users to select and view trends for different future time intervals (e.g., 12 hours). The diagram uses dotted lines to separate historical / real-time data from future data. The system also sets different timed inference intervals for models with different prediction scales; for example, a 30-minute prediction task triggers automatically every 6 minutes, and a 24-hour prediction task triggers every 4.8 hours, thus achieving periodic automatic updates to the prediction results.

[0073] See Figure 5 The diagram shows a schematic of a frequency converter temperature rise prediction system provided in an embodiment of this application. For ease of explanation, only the parts related to the embodiment of this application are shown.

[0074] The inverter temperature rise prediction system may specifically include the following modules: The data acquisition module 501 is used to acquire the real-time operating data of the frequency converter.

[0075] The model calling module 502 is used to input real-time running data into multiple pre-configured prediction models and receive multiple temperature rise prediction sequences output by multiple prediction models. Among the multiple prediction models, at least two prediction models are configured to predict temperature rise at different time lengths, and each temperature rise prediction sequence corresponds to a time length.

[0076] The temperature rise prediction module 503 is used to integrate multiple temperature rise prediction sequences according to preset rules based on the time length corresponding to each temperature rise prediction sequence, so as to generate a combined prediction result covering multiple time lengths.

[0077] The inverter temperature rise prediction system disclosed in the above embodiments of this application first acquires the real-time operating data of the inverter; then, it inputs the real-time operating data into multiple pre-configured prediction models and receives multiple temperature rise prediction sequences output by these models. At least two of the prediction models are configured to predict temperature rises at different time lengths, with each temperature rise prediction sequence corresponding to a specific time length. Finally, based on the time length corresponding to each temperature rise prediction sequence, the multiple temperature rise prediction sequences are integrated according to preset rules to generate a combined prediction result covering multiple time lengths. Therefore, by inputting the real-time operating data of the inverter into multiple models predicting different time lengths and integrating them based on the time lengths corresponding to each temperature rise prediction sequence, a combined prediction result covering multiple time lengths is generated, providing comprehensive temperature rise predictions for different decision-making needs, from immediate operation to long-term planning.

[0078] Furthermore, in one possible implementation of this application embodiment, the above-mentioned inverter temperature rise prediction system may further include the following modules: The model configuration module is used to configure multiple prediction models to handle prediction tasks of different time lengths. At least one short-term prediction model among the multiple prediction models is configured to perform short-term prediction tasks, and at least one long-term prediction model among the multiple prediction models is configured to perform long-term prediction tasks. The time length corresponding to the long-term prediction task is longer than the time length corresponding to the short-term prediction task.

[0079] Thus, by explicitly configuring multiple prediction models to handle short-term and long-term prediction tasks respectively, with the long-term prediction task having a longer time frame than the short-term prediction task, different models can optimize predictions for their specific time lengths of expertise, forming a complementary division of prediction capabilities.

[0080] Furthermore, in another possible implementation of this application embodiment, the above-mentioned model configuration module may specifically include the following units: The first acquisition unit is used to acquire the historical operating data of the frequency converter and construct an initial sample set based on the historical operating data.

[0081] The first processing unit is used to divide the initial sample set into a training dataset and a validation dataset.

[0082] The first training unit is used to construct a first hyperparameter search space for a preset first initial model, and to search for the first optimal hyperparameter combination in the first hyperparameter search space by using a Bayesian optimization algorithm to update the model parameters with the training dataset and evaluate the prediction error with the verification dataset.

[0083] The second training unit is used to train the first initial model on the training dataset using the first optimal hyperparameter combination, and to determine the completion of the first initial model training based on the validation dataset during the training process, thereby obtaining a short-term prediction model for the short-term prediction task.

[0084] The third training unit is used to construct a second hyperparameter search space for a preset second initial model, and to search for the second optimal hyperparameter combination in the second hyperparameter search space by using a Bayesian optimization algorithm to update the model parameters with the training dataset and evaluate the prediction error with the validation dataset.

[0085] The fourth training unit is used to train the second initial model on the training dataset using the second optimal hyperparameter combination, and to determine the completion of the training of the second initial model based on the validation dataset during the training process, thereby obtaining a long-term prediction model for long-term prediction tasks.

[0086] Therefore, by employing the Bayesian optimization algorithm, which updates parameters on the training set and evaluates errors on the validation set, the optimal hyperparameters for the first and second initial models are searched and trained, respectively. This automatically and efficiently yields the short-term and long-term prediction models that are optimal for their respective short-term and long-term tasks, improving the accuracy and efficiency of model configuration.

[0087] Furthermore, in another possible implementation of this application embodiment, the above-mentioned model configuration module may specifically include the following modules: The feature enhancement module is used to perform feature enhancement processing on real-time running data to obtain enhanced feature data.

[0088] The aforementioned model calling module 502 can be used to input enhanced feature data into multiple prediction models.

[0089] Therefore, by performing feature enhancement processing on real-time running data before inputting it into the prediction model to obtain enhanced feature data, the model is provided with richer and more representative input, which helps to improve the accuracy of subsequent temperature rise prediction.

[0090] Furthermore, in another possible implementation of this application embodiment, the above-mentioned feature enhancement module may specifically include the following units: The first calculation unit is used to calculate at least one physical derived feature based on electrical parameters in real-time operating data, wherein the physical derived feature includes at least one of input power, output power, inverter efficiency, energy consumption per unit load, and load fluctuation rate.

[0091] The second processing unit is used to combine at least one physical derived feature with real-time running data to obtain enhanced feature data.

[0092] Therefore, by calculating derived features with clear physical meaning based on electrical parameters such as power and efficiency, and combining them with the original data, the enhanced feature data can more directly reflect the physical mechanism of inverter heating, thereby further improving the accuracy and generalization ability of model prediction.

[0093] Furthermore, in another possible implementation of this application embodiment, the temperature rise prediction module 503 may specifically include the following units: The third processing unit is used to extract multiple candidate predicted values ​​corresponding to the target future time from all temperature rise prediction sequences for the target future time to be determined in the combined prediction results.

[0094] The first determining unit is used to determine the candidate prediction value with the shortest corresponding time length from multiple candidate prediction values ​​as the prediction result for the target future time.

[0095] Therefore, by prioritizing the candidate value with the shortest prediction time for a given future moment when integrating multi-scale prediction sequences, the combined prediction results are more likely to be provided by short-term, highly sensitive models in the near future, thus improving the timeliness of short-term predictions and their guiding value for immediate operations.

[0096] Furthermore, in another possible implementation of this application embodiment, the above-mentioned inverter temperature rise prediction system may further include the following modules: The period setting module is used to determine the trigger period for each prediction model with a configured time length. The trigger period for each prediction model is calculated based on the configured time length.

[0097] The fourth processing unit is used to periodically trigger the steps of acquiring real-time running data and inputting it into the prediction model to receive the temperature rise prediction sequence, according to the triggering period corresponding to each prediction model.

[0098] Therefore, by setting a trigger cycle for each prediction model based on its configured time length, and triggering the prediction process periodically accordingly, short-term prediction tasks can be updated more frequently, while long-term prediction tasks run at a reasonable frequency, thus achieving adaptive allocation and efficient utilization of the prediction system's computing resources.

[0099] The inverter temperature rise prediction system provided in this application embodiment can be applied to the aforementioned method embodiment. For details, please refer to the description of the above method embodiment, which will not be repeated here.

[0100] Figure 6 This is a schematic diagram of the structure of the electronic device provided in an embodiment of this application. For example... Figure 6 As shown, the electronic device 600 of this embodiment includes: at least one processor 610 ( Figure 6 The diagram shows only one processor, a memory 620, and a computer program 621 stored in the memory 620 and executable on the at least one processor 610. When the processor 610 executes the computer program 621, it implements the steps in the above-described inverter temperature rise prediction method embodiment.

[0101] The electronic device 600 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. This electronic device may include, but is not limited to, a processor 610 and a memory 620. Those skilled in the art will understand that... Figure 6 This is merely an example of electronic device 600 and does not constitute a limitation on electronic device 600. It may include more or fewer components than shown, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0102] The processor 610 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0103] In some embodiments, the memory 620 may be an internal storage unit of the electronic device 600, such as a hard disk or memory of the electronic device 600. In other embodiments, the memory 620 may be an external storage device of the electronic device 600, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 600. Furthermore, the memory 620 may include both internal and external storage units of the electronic device 600. The memory 620 is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory 620 can also be used to temporarily store data that has been output or will be output.

[0104] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely 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 device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0105] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0106] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0107] In the embodiments provided in this application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0108] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0109] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0110] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0111] The implementation of all or part of the processes in the methods of the above embodiments can also be accomplished by a computer program product. When the computer program product is run on an electronic device, the electronic device can implement the steps in the various method embodiments described above.

[0112] The embodiments described above are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for predicting the temperature rise of a frequency converter, characterized in that, include: Obtain real-time operating data of the frequency converter; The real-time operating data is input into multiple pre-configured prediction models, and multiple temperature rise prediction sequences output by the multiple prediction models are received. Among the multiple prediction models, at least two prediction models are configured to predict temperature rise at different time lengths, and each temperature rise prediction sequence corresponds to a time length. Based on the time length corresponding to each of the temperature rise prediction sequences, the multiple temperature rise prediction sequences are integrated according to preset rules to generate a combined prediction result covering multiple time lengths.

2. The method according to claim 1, characterized in that, Before inputting the real-time running data into multiple pre-configured prediction models, the method further includes: The plurality of prediction models are configured to handle prediction tasks of different time lengths, wherein at least one short-term prediction model among the plurality of prediction models is configured to perform short-term prediction tasks, and at least one long-term prediction model among the plurality of prediction models is configured to perform long-term prediction tasks, wherein the time length corresponding to the long-term prediction tasks is greater than the time length corresponding to the short-term prediction tasks.

3. The method according to claim 2, characterized in that, The step of configuring the multiple prediction models to handle prediction tasks of different time lengths includes: Obtain historical operating data of the frequency converter, and construct an initial sample set based on the historical operating data; The initial sample set is divided into a training dataset and a validation dataset; For a preset first initial model, a first hyperparameter search space is constructed, and a Bayesian optimization algorithm is used to search for the first optimal hyperparameter combination in the first hyperparameter search space by updating the model parameters with the training dataset and evaluating the prediction error with the validation dataset. Using the first optimal hyperparameter combination, the first initial model is trained on the training dataset, and during the training process, the training of the first initial model is determined to be completed based on the validation dataset, thereby obtaining a short-term prediction model for short-term prediction tasks. For the preset second initial model, a second hyperparameter search space is constructed, and a Bayesian optimization algorithm is used to search for the second optimal hyperparameter combination in the second hyperparameter search space by updating the model parameters with the training dataset and evaluating the prediction error with the validation dataset. Using the second optimal hyperparameter combination, the second initial model is trained on the training dataset, and during the training process, the training of the second initial model is determined to be complete based on the validation dataset, thereby obtaining a long-term prediction model for long-term prediction tasks.

4. The method according to claim 1, characterized in that, Before inputting the real-time running data into multiple pre-configured prediction models, the method further includes: The real-time running data is subjected to feature enhancement processing to obtain enhanced feature data; The step of inputting the real-time running data into multiple pre-configured prediction models includes: The enhanced feature data is input into the multiple prediction models.

5. The method according to claim 4, characterized in that, The process of performing feature enhancement processing on the real-time running data to obtain enhanced feature data includes: Based on the electrical parameters in the real-time operating data, at least one physical derivative feature is calculated, wherein the physical derivative feature includes at least one of input power, output power, inverter efficiency, unit load energy consumption, and load fluctuation rate. The enhanced feature data is obtained by combining the at least one physical derived feature with the real-time running data.

6. The method according to claim 1, characterized in that, The step of integrating the multiple temperature rise prediction sequences according to a preset rule based on the time length corresponding to each of the temperature rise prediction sequences to generate a combined prediction result covering multiple time lengths includes: For the target future time to be determined in the combined prediction results, multiple candidate prediction values ​​corresponding to the target future time are extracted from all temperature rise prediction sequences; From the multiple candidate predicted values, the candidate predicted value with the shortest corresponding time length is determined as the prediction result for the target future time.

7. The method according to claim 1, characterized in that, The method further includes: Determine the trigger period for each prediction model configured with a time length, wherein the trigger period for each prediction model is calculated based on the configured time length; According to the triggering period corresponding to each prediction model, the steps of periodically triggering the acquisition of real-time operating data and inputting it into the prediction model to receive the temperature rise prediction sequence are triggered.

8. A frequency converter temperature rise prediction system, characterized in that, include: The data acquisition module is used to acquire the real-time operating data of the frequency converter; The model calling module is used to input the real-time running data into multiple pre-configured prediction models and receive multiple temperature rise prediction sequences output by the multiple prediction models. Among the multiple prediction models, at least two prediction models are configured to predict temperature rise at different time lengths, and each temperature rise prediction sequence corresponds to a time length. The temperature rise prediction module is used to integrate the multiple temperature rise prediction sequences according to preset rules based on the time length corresponding to each temperature rise prediction sequence, so as to generate a combined prediction result covering multiple time lengths.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when run on an electronic device, causes the electronic device to perform the method as described in any one of claims 1 to 7.

Images