Photovoltaic booster station meter reading identification method and system based on transfer learning enhancement, and product

Abstract

The invention discloses a photovoltaic booster station meter reading identification method and system based on transfer learning enhancement and a product, and the method comprises the steps: obtaining a new meter identification model of a target domain through training of a small number of target domain annotation pictures based on a built meter identification model of a source domain by employing a transfer learning method; inputting the target domain picture to the new meter recognition model to obtain a new recognition annotation picture, and performing enhancement training on the new meter recognition model to obtain an enhanced new meter recognition model; a fusion scene meter reading recognition model is obtained through training by adopting a transfer learning method and scene labeling pictures of a source domain and a target domain; and based on the fused scene meter reading identification model, realizing meter reading identification of a source field and a target field. According to the method, the transfer learning technology and the deep learning technology are utilized, meter reading recognition of source and target field scenes is achieved at the same time, the model generalization ability is enhanced, and the method has the transfer learning and application ability of a new scene.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric power, and particularly relates to a method, system and product for identifying meter readings of a photovoltaic booster station based on enhanced transfer learning. Background Art

[0002] Currently, the recognition accuracy of meter readings of photovoltaic booster station equipment is insufficient, and the recognition accuracy of most meter readings is lower than 80%.

[0003] Due to insufficient training samples, as well as the influence of the booster station environment, such as insufficient lighting, the shooting angle of inspection robots, overexposure and other problems, the quality of the captured photos is poor, which affects the recognition accuracy of meter readings.

[0004] In addition, the meter readings of different booster station equipment are diverse. Usually, a specific model needs to be designed for a specific type of meter (digital display screen), and the model is iteratively updated offline. The degree of popularization and application is limited.

[0005] Existing implementation technologies: Chinese patent applications with application numbers CN202011364877.0 and CN202111223903.2 can only realize the recognition of meter readings in a single scenario. For the recognition of meter readings in a new scenario, a sufficient amount of labeled pictures need to be obtained to reconstruct the model. This reconstruction method is time-consuming. It is impossible to obtain a sufficient amount of labeled pictures in the early stage, and the newly constructed model in the new scenario has problems of insufficient generalization ability and poor accuracy. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the prior art and propose a method and system for identifying meter readings of a photovoltaic booster station based on enhanced transfer learning. In the scenario of identifying meter readings of a photovoltaic booster station, the transfer learning method is used to realize the efficient recognition of meter readings in a new scenario lacking labeled pictures.

[0007] In view of this, the present invention proposes a method for identifying meter readings of a photovoltaic booster station based on enhanced transfer learning, including:

[0008] Step 1) Based on the established meter recognition model in the source domain, using the transfer learning method, a new meter recognition model in the target domain is trained from a small number of labeled pictures in the target domain;

[0009] Step 2) Input the pictures in the target domain into the new meter recognition model to obtain new recognition labeled pictures, and then perform enhanced training on the new meter recognition model to obtain an enhanced new meter recognition model;

[0010] Step 3) Using the transfer learning method, a fused scenario meter reading recognition model is trained from the scenario labeled pictures in the source domain and the target domain;

[0011] Step 4) Based on the fusion scenario meter reading recognition model, realize the recognition of meter readings including the source domain and the target domain.

[0012] Preferably, the method includes, before step 1), training the established meter recognition model in the source domain with a large number of labeled pictures in the source domain to obtain a trained established meter recognition model.

[0013] Preferably, the large number of labeled pictures in the source domain is at least two orders of magnitude more than the small number of labeled pictures in the target domain.

[0014] Preferably, the method further includes:

[0015] Based on the fusion scenario meter reading recognition model, using the transfer learning method, training a new scenario meter recognition model from a small number of new scenario labeled pictures.

[0016] In a second aspect, the present invention provides a photovoltaic booster station meter reading recognition system enhanced by transfer learning, including:

[0017] A new meter recognition model training module, configured to train a new meter recognition model in the target domain from a small number of labeled pictures in the target domain based on the established meter recognition model in the source domain using the transfer learning method;

[0018] A new meter recognition model enhanced training module, configured to input pictures in the target domain into the new meter recognition model to obtain new recognition labeled pictures, and then perform enhanced training on the new meter recognition model to obtain an enhanced new meter recognition model;

[0019] A fusion scenario meter reading recognition model training module, configured to train a fusion scenario meter reading recognition model from the scenario labeled pictures in the source domain and the target domain using the transfer learning method; and

[0020] A result display module, configured to realize the recognition of meter readings including the source domain and the target domain based on the fusion scenario meter reading recognition model.

[0021] In a third aspect, the present invention provides a computer program product, including a computer program, which when executed by a processor implements the steps of any of the above methods.

[0022] Compared with the prior art, the advantages of the present invention are:

[0023] The present invention can realize the transfer and expansion ability of the source domain model, and avoid the problems that the model cannot be constructed due to the lack of labeled data in the target domain or the poor accuracy of the newly built model.

[0024] The present invention can complete incremental training based on pictures of target field applications, realize enhanced learning of the transfer learning model in the target field, and improve the accuracy and generalization ability of the newly built model. Description of the Drawings

[0025] Figure 1 It is a flow chart of the method for identifying meter readings in a photovoltaic step-up substation enhanced by transfer learning according to the present invention. Detailed Embodiments

[0026] The ability to enhance meter reading recognition by the transfer learning technology of the present invention is a method that uses the mature meter recognition ability and knowledge of the "source field" (established meter reading recognition scenarios with a large number of labeled pictures) to help the "target field" (newly added meter reading recognition scenarios with a small number of labeled pictures) learn internal scene features, as well as the similarity and correlation features between the source field and the target field, and constructs a meter reading recognition model for the new scene to achieve accurate recognition of meter readings in the new scene.

[0027] The method of the present invention includes:

[0028] Step 0) Training the established meter recognition model in the source field with a large number of labeled pictures in the source field to obtain a trained established meter recognition model.

[0029] Step 1) Based on the established meter recognition model in the source field, using the transfer learning method, training the newly built meter recognition model in the target field with a small number of labeled pictures in the target field;

[0030] Step 2) Inputting pictures in the target field into the newly built meter recognition model to obtain newly recognized labeled pictures, and then performing enhanced training on the newly built meter recognition model to obtain an enhanced newly built meter recognition model;

[0031] Step 3) Using the transfer learning method, training the integrated scene meter reading recognition model with the scene labeled pictures in the source field and the target field;

[0032] Step 4) Based on the integrated scene meter reading recognition model, realizing the recognition of meter readings including those in the source field and the target field.

[0033] The large number of labeled pictures in the source field mentioned above includes at least more than 1000 in quantity. The small number of labeled pictures in the target field refers to a scenario with less than 10 or almost none, and the two differ by at least more than 2 orders of magnitude.

[0034] The system consists of an image receiving module, a meter reading recognition module, and a result display module. The specific structure is as Figure 1 shown.

[0035] (1) The image receiving module mainly realizes the reception of meter reading pictures.

[0036] (2) The meter reading recognition module mainly recognizes the training and recognition of the meter reading model.

[0037] The training process includes:

[0038] A. Implement transfer learning to train and construct a new meter reading recognition model in the target domain

[0039] B. Train and construct a meter reading recognition model for the fusion scenario

[0040] C. Train and construct a meter reading recognition model for the new scenario.

[0041] Referring to the above figure, the specific process is described as follows:

[0042] ·a. In the source domain, based on a large number of labeled images in different scenarios, use the established meter reading recognition model to implement the transfer learning method.

[0043] ·b. In the target domain, select 1 new meter (digital display screen) reading recognition scenario. Based on a small number of labeled images, use the transfer learning method to expand the training of the established model to achieve the training and construction of the new meter reading recognition model. It mainly includes: based on the limited labeled meter reading images in the new scenario, use the transfer learning method to extract the source domain features and the transfer features of the target domain, and measure the similarity relationship between the source scenario and the new scenario, and fine-tune the training to obtain a new meter reading recognition model.

[0044] ·c. Based on the scenario images in the source domain and the target domain, use the transfer learning method to achieve the training and construction of the meter reading recognition model for the fusion scenario. The new transfer learning model supports the recognition of the meter reading model in the new scenario.

[0045] ·d. In the target domain scenario, the target domain images can be input, and the meter reading of the target domain images can be recognized using the newly established meter reading recognition model.

[0046] ·e. During the application process in the target domain scenario, new images can be recognized and labeled through the new transfer learning model. The new images can be used to enhance the training of the newly established meter reading recognition model, improve the recognition accuracy of the transfer learning model for the meter reading images in the target domain, achieve the scenario transfer of the source model, and weaken the problem of insufficient labeled samples in the new scenario.

[0047] ·f. In the new scenario transfer learning, based on the established meter reading recognition model for the fusion scenario, referring to steps a, b, and c, use the transfer learning method to achieve the training and construction of the meter reading recognition model for the new scenario.

[0048] (3) The result shows that the model realizes the display of the labeled images for meter reading recognition.

[0049] The technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

[0050] Embodiment 1

[0051] Embodiment 1 of the present invention provides a method for identifying the meter readings of a photovoltaic booster station enhanced by transfer learning, including the following steps:

[0052] Collect the source domain image set Set_A and construct the source domain meter reading recognition model Model_A;

[0053] Collect the target domain image set Set_B;

[0054] Use transfer learning technology to implement transfer learning of the source domain meter reading recognition model Model_A, and based on the target domain image set Set_B, learn to construct the target domain meter reading recognition model Model_B;

[0055] Use the target domain meter reading recognition model Model_B to identify and label the target domain meter reading images;

[0056] Collect the newly recognized and labeled target domain meter reading image set Set_B_Add;

[0057] Based on the newly recognized and labeled target domain meter reading image set Set_B_Add, enhance the training of the target domain meter reading recognition model Model_B, and obtain the enhanced target domain meter reading recognition model Model_B_Enhance.

[0058] Using the enhanced target domain meter reading recognition model Model_B_Enhance can realize the recognition of the target domain meter reading images and improve the recognition accuracy relative to the source model.

[0059] Embodiment 2

[0060] Embodiment 2 of the present invention provides a system for identifying the meter readings of a photovoltaic booster station enhanced by transfer learning, which is implemented based on the method of Embodiment 1. The system includes:

[0061] A newly built meter recognition model training module, which is used to obtain a newly built meter recognition model for the target domain by training a small number of target domain labeled images using transfer learning methods based on the existing meter recognition model in the source domain;

[0062] A newly built meter recognition model enhancement training module, which is used to input target domain images into the newly built meter recognition model to obtain newly recognized and labeled images, and then enhance the training of the newly built meter recognition model to obtain an enhanced newly built meter recognition model;

[0063] The integrated scenario meter reading recognition model training module is used to train an integrated scenario meter reading recognition model by using transfer learning method with the scenario annotation pictures of the source domain and the target domain;

[0064] The result display module is used to realize the recognition of the meter readings including those in the source domain and the target domain based on the integrated scenario meter reading recognition model.

[0065] It should be noted that in the embodiments of the above recognition system, the included modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional modules are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0066] Embodiment 3

[0067] Embodiment 3 of the present invention provides a computer program product including a computer program. When the computer program is executed by a processor, each step in the above method embodiments can be realized.

[0068] Innovation points:

[0069] The present invention can integrate the meter reading recognition models of the source domain and the target domain, use transfer learning technology to extract the associated features between the source domain and the target domain, use the source domain model to extract the source domain features, and based on the above features, use deep learning technology to construct a new model, and at the same time realize the recognition of the meter readings in the scenarios of the source domain and the target domain, enhance the generalization ability of the model, and have more transfer learning and transfer application abilities for new scenarios.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the embodiments, those of ordinary skill in the art should understand that any modification or equivalent replacement of the technical solutions of the present invention does not depart from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying meter readings of a photovoltaic booster station enhanced by transfer learning, comprising: Step 1) Based on the established meter identification model in the source domain, using the transfer learning method, a newly established meter identification model in the target domain is trained from a small number of labeled pictures in the target domain; Step 2) Input pictures in the target domain into the newly established meter identification model to obtain newly identified labeled pictures, and then perform enhanced training on the newly established meter identification model to obtain an enhanced newly established meter identification model; Step 3) Using the transfer learning method, a fused-scene meter reading identification model is trained from the scene-labeled pictures in the source domain and the target domain; Step 4) Based on the fused-scene meter reading identification model, the identification of meter readings including those in the source domain and the target domain is realized.

2. The method for identifying the meter reading of a photovoltaic booster station enhanced based on transfer learning according to claim 1, wherein The method includes, before step 1), training the established meter identification model in the source domain with a large number of labeled pictures in the source domain to obtain a trained established meter identification model.

3. The method for identifying meter readings of a photovoltaic booster station enhanced based on transfer learning according to claim 1, characterized in that, The large number of labeled pictures in the source domain is at least two orders of magnitude more than the small number of labeled pictures in the target domain.

4. The method for identifying the meter reading of a photovoltaic booster station enhanced based on transfer learning according to claim 1, wherein The method further includes: Based on the fused-scene meter reading identification model, using the transfer learning method, a new-scene meter identification model is trained from a small number of newly scene-labeled pictures.

5. A photovoltaic booster station meter reading recognition system enhanced based on transfer learning, characterized in that Including: A newly established meter identification model training module, which is used to train a newly established meter identification model in the target domain from a small number of labeled pictures in the target domain based on the established meter identification model in the source domain by using the transfer learning method; A newly established meter identification model enhanced training module, which is used to input pictures in the target domain into the newly established meter identification model to obtain newly identified labeled pictures, and then perform enhanced training on the newly established meter identification model to obtain an enhanced newly established meter identification model; A fused-scene meter reading identification model training module, which is used to train a fused-scene meter reading identification model from the scene-labeled pictures in the source domain and the target domain by using the transfer learning method; and A result display module, which is used to realize the identification of meter readings including those in the source domain and the target domain based on the fused-scene meter reading identification model.

6. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it realizes the steps of the method according to any one of claims 1-4.

Images