Data center electric energy meter detection and identification model fine tuning training method

Abstract

The invention discloses a fine tuning training method for a data center electric energy meter detection and identification model. The method comprises the following steps: creating a virtual environment and setting a programming module; setting a text detection and recognition model framework; setting a marking tool module; entering a data annotation stage to establish a training data set; splitting the training data set into a training set, a test set and a verification set; respectively downloading a detection pre-training model and an identification pre-training model; carrying out detection model training on the detection pre-training model based on the training data set; performing identification model training on the identification pre-training model based on the training data set; converting the detection model and the recognition model to finish fine tuning training; according to the invention, fine tuning training is carried out, the performance of the model is optimized, and with the help of a refined parameter adjustment strategy, a deep model optimization technology and a multi-dimensional verification process, the bottlenecks of precision, efficiency and generalization ability in traditional model training are solved, and powerful support is provided for intelligent management and operation of a data center.

Description

Technical Field

[0001] The present invention relates to the technical field of data center operation and maintenance, and particularly to a method for fine-tuning and training a data center electricity meter detection and recognition model. Background Art

[0002] As the core infrastructure of the digital economy, data centers are undergoing profound changes and innovative developments. Driven by new-generation information technologies such as cloud computing, the Internet of Things, and artificial intelligence, modern data centers have evolved from traditional data storage carriers to intelligent hubs that support enterprise digital transformation, undertaking strategic functions such as massive data processing, computing resource scheduling, and business continuity guarantee.

[0003] In this context, high-precision electricity metering has become a key supporting link in data center infrastructure management. It should be noted that the current data center industry still generally adopts the traditional manual meter reading method, and the traditional manual meter reading mode has significant limitations: the manual record has a metering error rate of more than ±2%.

[0004] Therefore, using the OCR visual recognition algorithm to identify electricity parameters is an effective means, which can improve the recognition accuracy to more than 99.97% and eliminate the subjective error of manual transcription. However, when the existing detection and recognition models are applied to the identification of data center electricity parameters, they show obvious non-universality, mainly reflected in the following two aspects: First, the model has poor adaptability to the display style of the electricity meter. There are various types of electricity meters used in data centers, and their display styles are diverse, including but not limited to differences in font, number spacing, case, and unit identification. The existing general detection and recognition models can often only accurately identify specific styles of electricity meter displays. Second, the model is difficult to handle complex and variable electricity meter reading scenarios. In the actual data center environment, the electricity meter readings may be interfered by various factors, such as insufficient light, reflection, stains, angle problems, etc., resulting in unclear or incomplete electricity meter displays. The existing general models have limited processing capabilities for such complex scenarios and cannot effectively filter and correct these interference factors, thus affecting the accurate identification of electricity parameters.

[0005] In view of the above defects, the creator of the present invention finally obtained the present invention through long-term research and practice. Summary of the Invention

[0006] To solve the above technical defects, the technical solution adopted by the present invention is to provide a method for fine-tuning and training a data center electricity meter detection and recognition model, including the steps of:

[0007] S1, creating a virtual environment and setting programming modules;

[0008] S2, setting the text detection and recognition model framework;

[0009] S3, set up the annotation tool module;

[0010] S4, enter the data annotation stage to establish a training data set;

[0011] S5, split the training data set into a training set, a test set, and a validation set;

[0012] S6, download the pre-trained model; download the detection pre-trained model and the recognition pre-trained model respectively;

[0013] S7, perform detection model training on the detection pre-trained model based on the training data set;

[0014] S8, perform recognition model training on the recognition pre-trained model based on the training data set;

[0015] S9, perform the conversion of the detection model and the recognition model to complete the fine-tuning training.

[0016] Preferably, in step S4, the annotation tool module is used to accurately annotate the meter text information in the collected meter images, and the model training data set is synthesized according to the annotation content and rules in the meter images.

[0017] Preferably, in step S5, the data set is divided into the training set, the test set, and the validation set according to a ratio by using the stratified sampling method. The training set is used to train the model, the validation set is used to validate the model during the training process, and the test set is used to finally evaluate the performance of the model.

[0018] Preferably, in step S7, based on the annotated data set and the detection pre-trained model, start training the detection model; during the training process, adjust the hyperparameters and monitor the training metrics during the training process.

[0019] Preferably, in step S7, the hyperparameters include learning rate, batch size, number of iterations, number of training epochs, training data set path, pre-trained model path, model save path, and image size; the training metrics include loss function value and accuracy.

[0020] Preferably, in step S8, based on the annotated data set and the recognition pre-trained model, use distributed training or mixed-precision training to train the recognition model, and adjust the hyperparameters during the training process.

[0021] Preferably, in step S8, the hyperparameters include learning rate, batch size, number of iterations, number of training epochs, training data set path, pre-trained model path, model save path, and image size.

[0022] Preferably, in step S9, the detection model and the recognition model are respectively converted into inference models. During the model conversion process, the output result of the detection model is correctly recognized by the recognition model as the input content.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows: Through the present invention, the fine-tuning training of the data center electricity meter detection and recognition model can be quickly carried out, the performance of the model can be optimized, and by means of a refined parameter adjustment strategy, a deep model optimization technology and a multi-dimensional verification process, the bottlenecks of accuracy, efficiency and generalization ability in traditional model training are solved, providing strong support for the intelligent management and operation of the data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic flow chart of the method for fine-tuning and training the data center electricity meter detection and recognition model. DETAILED DESCRIPTION OF THE INVENTION

[0025] The following further describes in detail the above and other technical features and advantages of the present invention with reference to the accompanying drawings.

[0026] Embodiment 1

[0027] As Figure 1 shown, Figure 1 It is a schematic flow chart of the method for fine-tuning and training the data center electricity meter detection and recognition model.

[0028] The method for fine-tuning and training the data center electricity meter detection and recognition model of the present invention includes the steps of:

[0029] S1, creating a virtual environment and setting up programming modules. During the fine-tuning training of the data center electricity meter detection and recognition model, it is crucial to create an independent and stable development environment. When creating a virtual environment, the version of the Python programming module needs to be selected according to project requirements, such as Python 3.10, etc., to ensure that the subsequent installed OCR text detection and recognition model framework and other tools can run properly.

[0030] S2, setting up the text detection and recognition model framework. During the test, attention needs to be paid to indicators such as the response time and resource occupancy of the model to ensure that the framework can meet the requirements during actual operation.

[0031] S3, setting up the annotation tool module.

[0032] S4, entering the data annotation stage to establish a training data set.

[0033] Use the described annotation tool module to accurately annotate the meter text information in the collected electricity meter images, and synthesize the model training dataset according to the annotation content and rules in the electricity meter images.

[0034] S5. Split the training dataset into a training set, a test set, and a validation set.

[0035] Use stratified sampling to divide the dataset into the training set, the test set, and the validation set according to a ratio. The training set is used to train the model, the validation set is used to validate the model during the training process, and the test set is used to finally evaluate the performance of the model.

[0036] S6. Download the pre-trained models; download the detection pre-trained model and the recognition pre-trained model respectively.

[0037] S7. Perform detection model training on the detection pre-trained model based on the training dataset.

[0038] Based on the annotated dataset and the detection pre-trained model, start training the detection model; during the training process, it is necessary to adjust the hyperparameters to optimize the performance of the model, and monitor the training metrics during the training process to timely discover and solve possible problems during the training process, such as overfitting, etc.

[0039] The hyperparameters include learning rate, batch size, number of iterations, number of training epochs, training dataset path, pre-trained model path, model saving path, and image size; the training metrics include loss function value and accuracy.

[0040] To improve the detection accuracy of the model, data augmentation techniques such as random rotation, scaling, and translation can be used to increase the diversity of the data and enhance the adaptability of the model to different scenarios. Through continuous iterative training, the detection model can accurately locate the text area in the electricity meter image and provide accurate area information for subsequent text recognition.

[0041] S8. Perform recognition model training on the recognition pre-trained model based on the training dataset.

[0042] Based on the annotated dataset and the recognition pre-trained model, use distributed training or mixed-precision training to train the recognition model. During the training process, it is necessary to adjust the hyperparameters to optimize the performance of the model and improve the accuracy of text recognition. The hyperparameters include learning rate, batch size, number of iterations, number of training epochs, training dataset path, pre-trained model path, model saving path, and image size.

[0043] S9. Perform the conversion of the detection model and the recognition model to complete the fine-tuning training.

[0044] Convert the detection model and the recognition model into inference models respectively. During the model conversion process, the output result of the detection model is correctly recognized by the recognition model as the input content, forming a complete electricity meter detection and recognition system.

[0045] Through the present invention, the fine-tuning training of the electricity meter detection and recognition model for the data center can be quickly carried out, optimizing the performance of the model. With the help of refined parameter adjustment strategies, deep model optimization techniques and multi-dimensional verification processes, the bottlenecks of accuracy, efficiency and generalization ability in traditional model training are solved, providing strong support for the intelligent management and operation of the data center. The present invention is universal for the operation and maintenance management of the data center, has an important demonstration effect, and has certain economic and social benefits.

[0046] Embodiment 2

[0047] In this specific embodiment, the operation and maintenance team of a certain data center plans to develop a set of electricity meter automatic recognition and meter reading software to replace the manual meter reading method to eliminate the subjective error of manual transcription. During the development process, it is necessary to fine-tune and train the detection and recognition model according to the characteristics of the on-site meters through the present invention.

[0048] First, create a Conda virtual environment with Anaconda and install Python 3.10.10. The Conda virtual environment provides an isolated space for the project, effectively avoiding dependency conflicts between different projects and ensuring the consistency and stability of the environment.

[0049] Next, install the OCR text detection and recognition model framework. During the installation, it is necessary to follow the guidance of the official documentation and install it through tools such as pip, and ensure that all dependent packages are correctly installed. In this embodiment, the OCR is installed as version 2.6.2, and at the same time, the CUDA version is updated to be consistent with the version supported by the OCR. After the installation is completed, conduct a simple test, run the example test code, and upload pictures to verify whether the framework works properly.

[0050] Then, install the OCRLabel annotation tool. OCRLabel is a professional OCR annotation tool that can help users efficiently annotate the text information in the electricity meter images. Annotation is an important preliminary work for model training, and the annotation quality directly affects the performance of the model. During the installation process, attention should be paid to the compatibility with the Python version, and relevant parameters should be configured according to needs.

[0051] Next is the data annotation stage. During the training process of the data center electricity meter detection and recognition model, data annotation is a crucial step. The collected data center electricity meter images need to be strictly annotated to be used as valid data for training the model. Using the OCRLabel annotation tool, accurately annotate the electricity meter text information of the existing 100 pictures, including numbers, units, identifiers, etc. During the annotation process, it is necessary to follow unified annotation specifications to ensure the consistency and accuracy of the annotation. However, the quantity of the existing original image data is far lower than the data volume required for model training and cannot meet the needs of training the detection and recognition model. Therefore, the training dataset for the detection and recognition model can be synthesized according to the content and rules of the original image data annotation to increase data diversity and improve the model's adaptability to different scenarios.

[0052] Subsequently, perform operations to split the above dataset into a training set, a test set, and a validation set. A reasonable dataset split is crucial for model training and evaluation. Usually, the dataset is divided into a training set, a test set, and a validation set according to a certain ratio. For example, 80% of the data is used as the training set, 10% as the validation set, and 10% as the test set. The training set is used for the model training process, the validation set is used to validate the model during the training process to timely adjust model parameters and training strategies, and the test set is used to finally evaluate the performance of the model. When splitting the dataset, methods such as stratified sampling need to be adopted to ensure that the data distributions of the training set, the validation set, and the test set are as consistent as possible, avoiding inaccurate model evaluation results caused by uneven data distribution. At the same time, it is also necessary to consider data diversity and representativeness to ensure that each data subset can cover electricity meter images of different scenarios, different types, and different qualities, so as to improve the generalization ability and adaptability of the model.

[0053] Then download the pre-trained model. The pre-trained model is a model pre-trained on a large-scale dataset and contains rich feature information and knowledge. Downloading the pre-trained model corresponding to the selected OCR framework can provide a good initialization starting point for subsequent model fine-tuning. When downloading the pre-trained model, it is necessary to ensure that the model version matches the framework version and place it in the specified directory for correct loading during the model training process. The selection of the pre-trained model should be based on its performance and adaptability in similar tasks. For example, for the electricity meter detection and recognition task, a pre-trained model that performs well in similar scenarios should be preferentially selected. In this embodiment, the detection pre-trained model and the recognition pre-trained model are downloaded respectively.

[0054] Next is the training of the detection model. Based on the annotated dataset and the downloaded pre-trained model, start training the detection model. During the training process, hyperparameters such as the learning rate, batch size, and number of iterations need to be adjusted to optimize the performance of the model. The main hyperparameter modifications are

[0055] epoch_num: Number of training epochs

[0056] print_batch_step: Print training information every x steps

[0057] pretrained_model: Path to the pre-trained model

[0058] save_model_dir: Path to save the model

[0059] learning_rate: Learning rate

[0060] train dataset dir: Path to the training dataset

[0061] size: Image size

[0062] Then, train the recognition model. The training of the recognition model requires the use of the labeled dataset and the pre-trained model. During the training process, hyperparameters such as the learning rate, batch size, and number of iterations need to be adjusted according to the characteristics of the recognition task to optimize the performance of the model and improve the accuracy of text recognition. The hyperparameters for the training of the recognition model are similar to those of the detection model. The training of the recognition model requires more computing resources and time, so techniques such as distributed training and mixed-precision training can be considered to accelerate the training process.

[0063] Finally, convert the detection model and the recognition model. Convert the models into inference models and save them respectively to form a complete electricity meter detection and recognition system. During the model conversion process, it is necessary to ensure that the output results of the detection model can be correctly used as the input of the recognition model. In the code part, the detection model and the recognition model need to be replaced with the previously saved paths.

[0064] Ultimately, the fine-tuning training of the electricity meter detection and recognition model in the data center is completed through the above steps.

[0065] In summary, by using a method for fine-tuning training of an electricity meter detection and recognition model in a data center, the fine-tuning training of the electricity meter detection and recognition model in the data center can be quickly carried out, optimizing the performance of the model. With the help of a refined parameter adjustment strategy, deep model optimization technology, and multi-dimensional verification process, the bottlenecks in accuracy, efficiency, and generalization ability in traditional model training are solved.

[0066] This technological breakthrough not only significantly improves the efficiency of model training and the accuracy of model detection and recognition but also provides a solid and reliable foundation for the intelligent meter reading system in the data center, providing strong support for the intelligent management and operation of the data center.

[0067] The above are only the preferred embodiments of the present invention, which are illustrative rather than restrictive to the present invention. Those skilled in the art understand that many changes, modifications, and even equivalents can be made within the spirit and scope defined by the claims of the present invention, but all will fall within the protection scope of the present invention.

Claims

1. A data center electric energy meter detection and recognition model fine-tuning training method, characterized in that: Includes steps: S1, create a virtual environment and set up programming modules; S2, setting up the text detection and recognition model framework; S3, setting the annotation tool module; S4, enter the data annotation phase to establish a training data set; S5, splitting the training data set into a training set, a test set and a validation set; S6, downloading the detection pre-training model and the recognition pre-training model respectively; S7, performing detection model training on the detection pre-training model based on the training data set; S8, performing recognition model training on the recognition pre-training model based on the training data set; S9, converting the detection model and the recognition model to complete fine-tuning training.

2. The data center electric energy meter detection and recognition model fine-tuning training method according to claim 1 is characterized in that: In the step S4, the annotation tool module is used to annotate the electric energy meter text information in the collected electric energy meter image, and the model training data set is synthesized according to the annotation content and rules in the electric energy meter image.

3. The data center electric energy meter detection and recognition model fine-tuning training method according to claim 1 is characterized in that: In step S5, the data set is divided into the training set, the test set and the validation set in proportion by stratified sampling. The training set is used to train the model, the validation set is used to validate the model during the training process, and the test set is used to finally evaluate the performance of the model.

4. The data center electric energy meter detection and recognition model fine-tuning training method according to claim 1, characterized in that: In the step S7, the detection model is trained based on the labeled data set and the detection pre-training model; During the training process, the hyperparameters are adjusted and the training indicators are monitored during the training process.

5. The data center electric energy meter detection and recognition model fine-tuning training method according to claim 4 is characterized in that: In step S7, the hyperparameters include learning rate, batch size, number of iterations, number of training rounds, training data set path, pre-trained model path, model save path, and image size; The training indicators include loss function value and accuracy.

6. The data center electric energy meter detection and recognition model fine-tuning training method according to claim 1, characterized in that: In the step S8, based on the labeled data set and the recognition pre-training model, the recognition model is trained by adopting distributed training or mixed precision training, and during the training process, the hyper parameters are adjusted.

7. The data center electric energy meter detection and recognition model fine-tuning training method according to claim 6, characterized in that: In step S8, the hyperparameters include learning rate, batch size, number of iterations, number of training rounds, training data set path, pre-trained model path, model save path, and image size.

8. The data center electric energy meter detection and recognition model fine-tuning training method according to claim 1, characterized in that: In the step S9, the detection model and the recognition model are converted into inference models respectively. During the model conversion process, the output result of the detection model is correctly recognized by the recognition model as input content.

Images