Automatic data enhancement method and device for indicator diagram

Abstract

The invention relates to the technical field of artificial intelligence, and particularly provides an indicator diagram automatic data enhancement method and device, and the method comprises the following steps: S1, alternately updating the parameters of a target model and an enhancement model based on an automatic enhancement method of teacher knowledge, the enhancement method comprises a color amplification model and a geometric amplification model, the target model is set as a teacher model, and the teacher model is any one model and only needs to be different from the target model; s2, setting a search algorithm and a search space, and updating the structure of the search space by using a color amplification model and a geometric amplification model; s3, generating a semantic auxiliary field; s4, performing image and semantic multi-modal fusion; and S5, preparing a data set. Compared with the prior art, the working condition diagnosis accuracy can be improved while the sample generation efficiency is improved, and the enterprise cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically provides a method and a device for automatic data enhancement of a dynamometer diagram. Background Art

[0002] Working condition diagnosis is an important means of evaluating the operating status of mechanical equipment systems. As the data source of working condition diagnosis, the quality of the dynamometer diagrams will have a significant impact on the diagnostic results. At present, working condition diagnosis mostly uses machine learning methods. When training a machine learning model, what is really done is to adjust the parameters so that it can map the input (such as pictures) to the output (such as labels). Our optimization goal is to pursue the best point with low model loss. This happens when the parameters are adjusted in the right way. Obviously, if there are many parameters, the model needs to be given a sufficient proportion of samples. However, the probability of failure in the actual mechanical production process is small, and the number of samples of fault types is also small. Therefore, the number of samples between the types of dynamometer diagrams is obviously unbalanced.

[0003] Data augmentation has attracted the attention of many researchers as a method to increase the number of samples and balance the differences between classes. However, traditional data augmentation methods require manual design and the experimental results are not satisfactory. Summary of the invention

[0004] The present invention aims at the above-mentioned deficiencies of the prior art and provides a highly practical automatic data enhancement method for a dynamometer diagram.

[0005] A further technical task of the present invention is to provide a device for automatically enhancing data of a dynamometer diagram that is reasonably designed, safe and applicable.

[0006] The technical solution adopted by the present invention to solve its technical problem is: A method for automatic data enhancement of a dynamometer diagram comprises the following steps: S1. Automatic enhancement method based on teacher knowledge to update the target model alternately and enhanced models The enhancement method includes a color amplification model and geometric augmentation model ,set up is the teacher model, and the teacher model is any model that can be compared with the target model Just be different; S2. Set the search algorithm and search space, and use the color amplification model and geometric augmentation model To update the structure of the search space; S3, generate semantic auxiliary fields; S4, image and semantic multimodal fusion; S5. Dataset preparation.

[0007] Further, in step S1, two teacher models are used, one is a pre-trained model and the other is an EMA model; The automatic enhancement method of teacher knowledge is to update the enhancement model alternately by stochastic gradient descent and target model ; Assumptions represents the image sampled from the dataset X, Indicates that the parameter is The amplification function of ; set up To calculate the loss, the objective function is set as follows: .

[0008] Furthermore, in step S2, the main body of the search algorithm is reinforcement learning, which consists of two parts, namely a controller and a training algorithm, and the controller is played by RNN; Next, the generated enhancement strategy will be applied to the training of the sub-model. After the sub-model is trained, it will be evaluated on the validation set. The accuracy of the sub-model will be fed back to the RNN controller as a "reward signal". Finally, the RNN controller will be updated through the reward signal and under the optimization algorithm.

[0009] Furthermore, the search space consists of many sub-strategies, one of which is randomly selected for each image in each batch, the sub-strategy consists of two operations, each operation has an image processing function, and a search algorithm is used to find the best strategy so that the neural network produces the highest verification accuracy on the target dataset.

[0010] Furthermore, in step S3, the text descriptions of different working conditions in actual production are used as the source of text feature extraction, and accurate professional knowledge is used to semantically assist different types of indicator diagrams to obtain text data for subsequent feature fusion.

[0011] Furthermore, in step S4, an attention-based feature fusion mechanism is used to calculate a context vector of a word for each sub-region of the image. When generating the image, the image sub-region and the most relevant word are matched. For each part of the input image, the most relevant word vector is matched to constrain its generation, thereby increasing the fine-grained details of the image.

[0012] Furthermore, in step S5, the sample data points of the indicator diagram of the actual pumping unit are collected and drawn into an indicator diagram. The data in the indicator diagram is preprocessed, and the min-max normalization method is used to obtain the working condition samples. Then, the data enhancement method is used to increase the number of samples, and then the indicator diagram data set is prepared. The indicator diagram data set is divided into a training set and a test set in a ratio of 8:2.

[0013] An automatic data enhancement device for a dynamometer diagram, comprising: at least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is used to call the machine-readable program to execute a method for automatic data enhancement of a dynamometer diagram.

[0014] Compared with the prior art, the method and device for automatic data enhancement of a dynamometer diagram of the present invention have the following outstanding beneficial effects: The present invention uses teacher knowledge for automatic data enhancement and combines semantically assisted multimodal data for more accurate data amplification. It can automatically select the optimal data enhancement method according to different dynamometer diagram types and generate more accurate images. It proposes a solution to the problem of unbalanced number of dynamometer diagram samples, improves sample generation efficiency, increases the accuracy of working condition diagnosis, and reduces enterprise costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Attached Figure 1 It is a training flow chart of an automatic data enhancement method for a dynamometer diagram; Attached Figure 2 It is a schematic diagram of a mode of automatic data enhancement method of a dynamometer diagram; Attached Figure 3 It is a conceptual diagram of an automatic data enhancement method for dynamometer diagrams; Attached Figure 4 It is an error curve diagram of the automatic data enhancement method of the dynamometer diagram and other comparison methods in the test phase; Attached Figure 5 It is a curve chart of the accuracy of the automatic data enhancement method for dynamometer diagrams and other comparison methods in the testing phase. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention is further described in detail below in conjunction with specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] A best embodiment is given below: like Figure 1-3 As shown, a method for automatic data enhancement of a dynamometer diagram in this embodiment has the following steps: S1. Automatic enhancement method based on teacher knowledge to update the target model alternately and enhanced models The enhancement method includes a color amplification model and geometric augmentation model ,set up is the teacher model, and the teacher model is any model that can be compared with the target model Just be different; Automatic enhancement of teacher knowledge by alternating updates of the enhanced model via stochastic gradient descent and target model ; Assumptions represents the image sampled from the dataset X, Indicates that the parameter is The amplification function of ; set up To calculate the loss, the objective function is set as follows: .

[0019] For the target model, this objective has the same components as the adversarial data augmentation method. However, this augmentation function not only requires maximizing the loss of the objective function, but also minimizes the loss of the teacher model. Using this method can effectively avoid the disappearance of the inherent meaning of the image, because when the image becomes unrecognizable, the gradient of the teacher model will explode. This shows that the introduced teacher model loss requires the augmentation function to transform the image, so that the augmentation result is adversarial to the target model, but recognizable to the teacher model.

[0020] S2. Set the search algorithm and search space, and use the color amplification model and geometric augmentation model To update the structure of the search space; The main body of the search algorithm is reinforcement learning, which consists of two parts: the controller and the training algorithm. The controller is played by RNN (recurrent neural network). Next, the generated reinforcement strategy is applied to the training of the sub-model. After the sub-model is trained, it will be evaluated on the validation set, and the accuracy of the sub-model will be fed back to the RNN controller as a "reward signal". Finally, the controller will pass this signal and update it under the optimization algorithm.

[0021] In this paper, we focus on methods that update the policy in an online manner for two reasons: (1) they can directly search for data augmentation policies on the target network using all training data, and (2) this unifies the search and training processes and simplifies the framework.

[0022] The policy in the search space consists of many sub-policies, one of which is randomly selected for each image in each batch. A sub-policy consists of two operations, each of which has an image processing function such as translation, rotation or shear, and the probability and magnitude of these operations being taken. A search algorithm is used to find the best policy so that the neural network produces the highest validation accuracy on the target dataset.

[0023] For example: In the search space, the strategy consists of 5 sub-strategies, each of which consists of 2 image operations to be applied in sequence. In addition, each operation is associated with two hyperparameters: 1) the probability of applying the operation; 2) the magnitude of the operation.

[0024] S3, generate semantic auxiliary fields; The text descriptions of different working conditions in actual production are used as the source of text feature extraction, and the most accurate professional knowledge is used to provide semantic assistance for different types of indicator diagrams. In this way, several pieces of text data can be obtained for subsequent feature fusion.

[0025] S4, image and semantic multimodal fusion; Using an attention-based feature fusion mechanism, a word context vector is calculated for each sub-region of the image. When generating images, the image sub-regions and the most relevant words are matched. For each part of the input image, the most relevant word vector is matched to constrain its generation and increase the fine-grained details of the image.

[0026] S5, dataset preparation; Collect the data points of the actual pumping unit's dynamometer diagram from an oil field and draw them into dynamometer diagrams. Preprocess the data. Considering the different ranges of the working parameters of different pumping units, we use the min-max standardization method to ensure that this does not interfere with the working condition identification. In this way, 6 types of working condition samples can be obtained, with 700 samples in each type, and a total of 4200 dynamometer diagrams. The data enhancement method is used to uniformly increase the number of samples to 1000 per type, so that a total of 6000 samples are obtained.

[0027] Then we started to make the dynamometer data set. Here we refer to the format of the standard data set "cifar10", change the image size to 32 × 32, and make it into a standard data file. In terms of data set division, we divide each type of dynamometer into training set and test set in a ratio of 8:2, that is, the number of samples in the training set and test set of each type is 800 and 200 respectively.

[0028] For the model training problem, this automatic enhancement method based on teacher knowledge alternately updates the target model and enhanced models The training process of this method is as follows Figure 1 As shown. Figure 1 As shown in Figure 1, this method solves the problem by alternately updating the augmentation function and the target model through stochastic gradient descent. The model first updates the target network for n steps, and then updates the parameters of the augmentation function. The reason why the augmentation function can be updated by the gradient method is that the parameters of the proposed neural network are updated when the parameters are iterated. This method neither regularizes the augmentation function based on domain knowledge nor restricts the difficulty of the transformed images to ensure that the transformed images are recognizable.

[0029] This paper proposes to use parameters The data enhancement method of the neural network mainly includes two models, the color amplification model and geometric augmentation model The model can update parameters through gradient methods, using two functions to update the structure of the search space.

[0030] Comparative experiments have been conducted to verify the superiority of this method by comparing it with traditional data enhancement methods and original samples without data enhancement. The dynamometer diagram is generated for actual production.

[0031] like Figure 4-5As shown in the figure, it can be seen from the loss curve that the error loss drops very quickly at the beginning, slows down after twenty iterations, and finally stabilizes. The accuracy change curve is also very fast in the initial stage, and the rising speed slows down after reaching a certain value until it finally stabilizes. The following is a numerical analysis. The final loss of the traditional data enhancement method in the working condition diagnosis task is 0.000512, and the accuracy is 93.32%. The final loss of the original data on the test set is 0.00632, and the accuracy is 89.46%. Obviously, the experimental performance of the automatic data enhancement method based on teacher knowledge semantic assistance is better than the other two groups of comparative experiments.

[0032] The conceptual diagram of this method is shown in Figure 1 Adversarial data augmentation methods augment the target model by transforming the data Such enhanced image data is usually difficult to generate value (e.g. Figure 1 The black and noise parts in the image). The enhanced image is difficult to recognize without any constraints. For the problem of working condition diagnosis, providing directional guidance will help the model convergence speed of the indicator diagram classification. The method we proposed performs data transformation to make it adversarial to the target model, but to the teacher model is recognizable. Therefore, the enhanced image will be richer than ordinary adversarial data augmentation.

[0033] Based on the above method, a device for automatic data enhancement of a dynamometer diagram in this embodiment includes: at least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is used to call the machine-readable program to execute a method for automatic data enhancement of a dynamometer diagram.

[0034] The above-mentioned specific implementations are only specific cases of the present invention. The patent protection scope of the present invention includes but is not limited to the above-mentioned specific implementations. Any technical solutions that conform to the above-mentioned specific implementations of the present invention and any appropriate changes or substitutions made by ordinary technicians in the relevant technical field shall fall within the patent protection scope of the present invention.

[0035] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatic data enhancement of a dynamometer diagram, characterized in that: The steps are as follows: S1. Automatic enhancement method based on teacher knowledge to update the target model alternately and enhanced models The enhancement method includes a color amplification model and geometric augmentation model ,set up is the teacher model, and the teacher model is any model that can be compared with the target model Just be different; S2. Set the search algorithm and search space, and use the color amplification model and geometric augmentation model To update the structure of the search space; S3, generate semantic auxiliary fields; S4, image and semantic multimodal fusion; S5. Dataset preparation.

2. The method for automatic data enhancement of a dynamometer diagram according to claim 1, characterized in that: In step S1, two teacher models are used, one is a pre-trained model and the other is an EMA model; The automatic enhancement method of teacher knowledge is to update the enhancement model alternately by stochastic gradient descent and target model ; Assumptions represents the image sampled from the dataset X, Indicates that the parameter is The amplification function of ; set up To calculate the loss, the objective function is set as follows: 。 3. The method for automatic data enhancement of a dynamometer diagram according to claim 2, characterized in that: In step S2, the main body of the search algorithm is reinforcement learning, which consists of two parts, namely a controller and a training algorithm, and the controller is played by RNN; Next, the generated enhancement strategy will be applied to the training of the sub-model. After the sub-model is trained, it will be evaluated on the validation set. The accuracy of the sub-model will be fed back to the RNN controller as a "reward signal". Finally, the RNN controller will be updated under the optimization algorithm through the reward signal.

4. The method for automatic data enhancement of a dynamometer diagram according to claim 3, characterized in that: The search space consists of many sub-strategies, where one sub-strategy is randomly selected for each image in each batch. The sub-strategy consists of two operations, each with an image processing function. A search algorithm is used to find the best strategy so that the neural network produces the highest validation accuracy on the target dataset.

5. The method for automatic data enhancement of a dynamometer diagram according to claim 4, characterized in that: In step S3, the text descriptions of different working conditions in actual production are used as the source of text feature extraction, and accurate professional knowledge is used to semantically assist different types of indicator diagrams to obtain text data for subsequent feature fusion.

6. The method for automatic data enhancement of a dynamometer diagram according to claim 4, characterized in that: In step S4, an attention-based feature fusion mechanism is used to calculate a context vector of a word for each sub-region of the image. When generating the image, the image sub-region and the most relevant word are matched. For each part of the input image, the most relevant word vector is matched to constrain its generation and increase the fine-grained details of the image.

7. The method for automatic data enhancement of a dynamometer diagram according to claim 6, characterized in that: In step S5, the sample data points of the indicator diagram of the actual pumping unit are collected and drawn into an indicator diagram. The data in the indicator diagram is preprocessed, and the min-max normalization method is used to obtain the working condition samples. Then the data enhancement method is used to increase the number of samples. Then, the indicator diagram data set is prepared and divided into a training set and a test set in a ratio of 8:

2.

8. An automatic data enhancement device for a dynamometer diagram, characterized in that: include: at least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is configured to call the machine-readable program to execute the method according to any one of claims 1 to 7.

Images