Abnormity recognition early warning method and device based on deep transfer learning

Abstract

The invention discloses an abnormity identification early warning method and device based on deep transfer learning, and the method comprises the steps: obtaining the power utilization data and state data of a first equipment set, carrying out the preprocessing operation of the power utilization data and state data, and obtaining a preprocessed first data set, inputting the first data set into a preset fault recognition model for training to obtain a trained first training model, obtaining equipment state information of the second equipment set, performing preprocessing operation on the equipment state information to obtain a second data set, inputting the second data set into the first training model for re-training to obtain a fault recognition result; and based on a deep transfer learning algorithm, carrying out tuning operation on the middle layer and the weight of the first training model to obtain a target fault recognition model. According to the invention, the method can utilize the deep transfer learning algorithm to optimize the preset fault recognition model and carry out recognition and early warning on the fault, and facilitates the improvement of the efficiency and accuracy of monitoring and early warning of the operation state of the equipment.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an abnormality recognition and early warning method and device based on deep transfer learning. Background Art

[0002] As the scale and complexity of energy equipment increases, it is urgent and important to monitor and diagnose abnormalities of energy equipment in production in a timely and effective manner through reliable condition monitoring technology in order to ensure the safety and stability of the production process. However, the traditional regular maintenance method of energy equipment requires regular manual condition monitoring and data recording, which can no longer meet the needs of timely detection of potential problems.

[0003] Therefore, a method and device for abnormality identification and early warning based on deep transfer learning is currently provided, which can introduce advanced monitoring technology and intelligent algorithms, timely detect equipment abnormalities, prevent potential failures and accidents, ensure the safety and stability of energy supply, reduce maintenance costs, and improve the efficiency and accuracy of equipment operation status monitoring and early warning. Summary of the invention

[0004] The present invention provides a method and device for abnormality identification and early warning based on deep transfer learning, which can accurately and effectively prevent potential failures and accidents, and is conducive to improving the efficiency and accuracy of equipment operation status monitoring and early warning.

[0005] In order to solve the above technical problems, the first aspect of the present invention discloses an abnormality recognition and early warning method based on deep transfer learning, the method comprising:

[0006] Acquire power usage data and status data of a first set of devices, and perform preprocessing operations on the power usage data and the status data to obtain a preprocessed first data set;

[0007] Inputting the first data set into a preset fault identification model for training to obtain a trained first training model;

[0008] Acquire device status information of a second device set, and perform a preprocessing operation on the device status information to obtain a second data set;

[0009] The second data set is input into the first training model for retraining, and the intermediate layers and weights of the first training model are tuned based on the deep transfer learning algorithm to obtain a target fault identification model, which is used to identify and warn faults in the operating data of the second equipment set.

[0010] As an optional embodiment, in the first aspect of the present invention, the power consumption data of the first device set includes: real-time power consumption data and historical power consumption data, and the status data of the first device set includes: at least one of the matching binding information, usage record information, regional location information, operation status information, and auxiliary equipment association information of each first device in the first device set; the real-time power consumption data includes: at least one of the power consumption, voltage value, current value, power value, load rate, and abnormal power consumption status information of each first device in the first device set collected in real time; and the historical power consumption data includes: at least one of the power consumption, voltage value, current value, power value, load rate, and abnormal power consumption status information of the first device set in a historical period.

[0011] As an optional implementation manner, in the first aspect of the present invention, the preprocessing operation is performed on the power consumption data and the state data to obtain a preprocessed first data set, including:

[0012] Performing data cleaning and data normalization on the power consumption data and the status data to obtain processed power consumption data and processed status data;

[0013] Performing a fusion operation on the processed power consumption data and the processed status data to obtain a preprocessed first data set;

[0014] Among them, the operations of the data cleaning processing include: deleting or marking duplicate data in the power usage data and the status data, deleting or replacing invalid abnormal data in the power usage data and the status data, and deleting or filling missing data in the power usage data and the status data.

[0015] As an optional implementation, in the first aspect of the present invention, at least one of the matching binding information, usage record information, regional location information, operating status information, and accessory device association information of each second device in the second device set, the preset fault identification model is a fault identification model based on a deep neural network algorithm, and the first training model includes at least a plurality of intermediate layers;

[0016] The deep transfer learning algorithm is used to tune the intermediate layers and weights of the first training model to obtain a target fault recognition model, including:

[0017] The last intermediate layer of the first training model is deleted, and a fully connected layer is added to the position after the last intermediate layer is deleted to obtain a second training model;

[0018] Randomizing the weights of the fully connected layer and freezing all weights of all intermediate layers in the second training model;

[0019] The second training model with frozen weights is retrained to obtain a target fault recognition model.

[0020] As an optional implementation, in the first aspect of the present invention, after the randomization operation is performed on the weights of the fully connected layer, the method further includes:

[0021] Classify and analyze the data types of the first data set and the data types of the second data set to obtain data structure information of the first data set and data structure information of the second data set;

[0022] Calculate the data structure difference between the first data set and the second data set according to the data structure information of the first data set, the data structure information of the second data set and a preset analysis algorithm;

[0023] Determine the target weights that need to be frozen for all intermediate layers in the second training model according to the data structure difference, and freeze the target weights for all intermediate layers in the second training model;

[0024] The second training model with frozen weights is retrained to obtain a target fault recognition model.

[0025] As an optional embodiment, in the first aspect of the present invention, the fault type corresponding to the fault anomaly of the second device set identified by the target fault identification model includes: at least one of: control system failure, communication failure, circuit failure, software failure and mechanical failure.

[0026] As an optional embodiment, in the first aspect of the present invention, the method further comprises:

[0027] Inputting the operation data of the second equipment set within a preset historical period into the target fault identification model for identification, and obtaining fault abnormality data of the second equipment set;

[0028] Classify and summarize the identified fault anomaly data of the second device set to obtain the fault anomaly occurrence frequency of each fault type corresponding to the second device set within a preset historical period;

[0029] The fault anomaly occurrence frequencies of all fault types are arranged in descending order to obtain the target fault type corresponding to the highest fault anomaly occurrence frequency in a preset historical period, and the fault anomaly data related to the target fault type is used as a subsequent training data set to retrain and update the target fault recognition model.

[0030] The second aspect of the present invention discloses an abnormality recognition and early warning device based on deep transfer learning, the device comprising:

[0031] An acquisition module, used to acquire power consumption data and status data of the first device set;

[0032] A preprocessing module, configured to perform a preprocessing operation on the power consumption data and the state data acquired by the acquisition module to obtain a preprocessed first data set;

[0033] A training module, used for inputting the first data set obtained by the preprocessing module into a preset fault identification model for training to obtain a first training model after training;

[0034] The acquisition module is further used to acquire device status information of the second device set;

[0035] The preprocessing module is further used to perform a preprocessing operation on the device status information acquired by the acquisition module to obtain a second data set;

[0036] The training module is further used to input the second data set processed by the preprocessing module into the first training model for retraining;

[0037] A tuning module is used to tune the intermediate layers and weights of the first training model obtained by training the training module based on a deep transfer learning algorithm to obtain a target fault identification model, and the target fault identification model is used to identify and warn of fault anomalies of the second device set.

[0038] As an optional embodiment, in the second aspect of the present invention, the power consumption data of the first device set includes: real-time power consumption data and historical power consumption data, the status data of the first device set includes: at least one of the matching binding information, usage record information, regional location information, operation status information, and auxiliary equipment association information of each first device in the first device set, the real-time power consumption data includes: at least one of the power consumption, voltage value, current value, power value, load rate, and abnormal power consumption status information of each first device in the first device set collected in real time, and the historical power consumption data includes: at least one of the power consumption, voltage value, current value, power value, load rate, and abnormal power consumption status information of the first device set in a historical period.

[0039] As an optional implementation, in the second aspect of the present invention, the preprocessing module performs a preprocessing operation on the power consumption data and the state data to obtain the preprocessed first data set in a specific manner as follows:

[0040] Performing data cleaning and data normalization on the power consumption data and the status data to obtain processed power consumption data and processed status data;

[0041] Performing a fusion operation on the processed power consumption data and the processed status data to obtain a preprocessed first data set;

[0042] Among them, the operations of the data cleaning processing include: deleting or marking duplicate data in the power usage data and the status data, deleting or replacing invalid abnormal data in the power usage data and the status data, and deleting or filling missing data in the power usage data and the status data.

[0043] As an optional implementation, in the second aspect of the present invention, at least one of the matching binding information, usage record information, regional location information, operating status information, and accessory device association information of each second device in the second device set, the preset fault identification model is a fault identification model based on a deep neural network algorithm, and the first training model includes at least a plurality of intermediate layers;

[0044] The tuning module performs tuning operations on the intermediate layers and weights of the first training model based on the deep transfer learning algorithm to obtain the target fault recognition model in the following specific manner:

[0045] The last intermediate layer of the first training model is deleted, and a fully connected layer is added to the position after the last intermediate layer is deleted to obtain a second training model;

[0046] Randomizing the weights of the fully connected layer and freezing all weights of all intermediate layers in the second training model;

[0047] The second training model with frozen weights is retrained to obtain a target fault recognition model.

[0048] As an optional implementation, in the second aspect of the present invention, the device further includes:

[0049] an analysis module, configured to analyze the data type of the first data set and the data type of the second data set obtained by the preprocessing module after the tuning module performs a randomization operation on the weight of the fully connected layer, so as to obtain data structure information of the first data set and data structure information of the second data set;

[0050] a calculation module, configured to calculate the data structure difference between the first data set and the second data set according to the data structure information of the first data set and the data structure information of the second data set obtained by the analysis module and a preset analysis algorithm;

[0051] A determination module, used to determine the target weights required to be frozen for all intermediate layers in the second training model according to the size of the data structure difference calculated by the calculation module;

[0052] The tuning module is further used to freeze the target weights of all intermediate layers in the second training model;

[0053] The training module is also used to retrain the second training model after the tuning module freezes the weights to obtain a target fault recognition model.

[0054] As an optional implementation, in the second aspect of the present invention, the fault type corresponding to the fault anomaly of the second device set identified by the target fault identification model includes: at least one of: control system failure, communication failure, circuit failure, software failure and mechanical failure.

[0055] As an optional implementation, in the second aspect of the present invention, the device further includes:

[0056] an identification module, configured to input the operation data of the second equipment set in a preset historical period into the target fault identification model tuned by the tuning module for identification, and obtain fault abnormality data of the second equipment set;

[0057] a classification module, configured to classify and summarize the fault anomaly data of the second device set identified by the identification module, and obtain the fault anomaly occurrence frequency of each fault type corresponding to the second device set within a preset historical period;

[0058] A sorting module, used to arrange the fault anomaly occurrence frequencies of all fault types classified by the classification module in descending order, and obtain the target fault type corresponding to the highest fault anomaly occurrence frequency in a preset historical period;

[0059] The training module is further configured to use the fault anomaly data related to the target fault type obtained by sorting the sorting module as a subsequent training data set to retrain and update the target fault identification model.

[0060] The third aspect of the present invention discloses another abnormality recognition and early warning device based on deep transfer learning, the device comprising:

[0061] A memory storing executable program code;

[0062] a processor coupled to the memory;

[0063] The processor calls the executable program code stored in the memory to execute the abnormality identification and early warning method based on deep transfer learning disclosed in the first aspect of the present invention.

[0064] The fourth aspect of the present invention discloses a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute the abnormality identification and early warning method based on deep transfer learning disclosed in the first aspect of the present invention.

[0065] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0066] The present invention provides an abnormality identification and early warning method and device based on deep transfer learning, the method comprising: obtaining power consumption data and status data of a first device set, preprocessing the power consumption data and status data to obtain a preprocessed first data set, inputting the first data set into a preset fault identification model for training to obtain a trained first training model, obtaining device status information of a second device set, preprocessing the device status information to obtain a second data set, inputting the second data set into the first training model for retraining, and tuning the intermediate layer and weights of the first training model based on a deep transfer learning algorithm to obtain a target fault identification model, the target fault identification model being used to identify and warn of fault anomalies of the second device set. It can be seen that the implementation of the present invention can obtain the power consumption data and status data of the first device set, perform preprocessing operations on the power consumption data and status data, and obtain the preprocessed first data set, which is beneficial to improving the recognition accuracy of the model obtained by subsequent training; obtain the device status information of the second device set, perform preprocessing operations on the device status information, and obtain the second data set, and input the second data set into the first training model for retraining, which is beneficial to improving the generalization ability and universality of the fault identification model, so that the final model obtained by training can identify various fault anomalies occurring in the equipment, and based on the deep transfer learning algorithm, the intermediate layer and weights of the first training model are tuned to obtain the target fault identification model, which is further beneficial to improving the accuracy and efficiency of the fault identification model's recognition warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0068] Figure 1 It is a flow chart of an abnormality recognition and early warning method based on deep transfer learning disclosed in an embodiment of the present invention;

[0069] Figure 2 It is a schematic diagram of a construction process of another fault identification model based on deep transfer learning disclosed in an embodiment of the present invention;

[0070] Figure 3 It is a flowchart of another abnormality recognition and early warning method based on deep transfer learning disclosed in an embodiment of the present invention;

[0071] Figure 4 It is a structural schematic diagram of an abnormality recognition and early warning device based on deep transfer learning disclosed in an embodiment of the present invention;

[0072] Figure 5 It is a structural schematic diagram of another abnormality recognition and early warning device based on deep transfer learning disclosed in an embodiment of the present invention;

[0073] Figure 6 It is a structural schematic diagram of another abnormality recognition and early warning device based on deep transfer learning disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not 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.

[0075] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or end including a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or ends.

[0076] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0077] The present invention discloses an abnormality recognition and early warning method and device based on deep transfer learning, which can obtain the power consumption data and status data of a first device set, perform preprocessing operations on the power consumption data and status data, and obtain a preprocessed first data set, which is conducive to improving the recognition accuracy of the model obtained by subsequent training; obtain the device status information of the second device set, perform preprocessing operations on the device status information, obtain a second data set, and input the second data set into the first training model for retraining, which is conducive to improving the generalization ability and universality of the fault recognition model, so that the final model obtained by training can recognize multiple fault anomalies occurring in the equipment, and based on the deep transfer learning algorithm, the middle layer and weight of the first training model are tuned to obtain a target fault recognition model, which is further conducive to improving the accuracy and efficiency of the fault recognition model recognition and early warning. The following are detailed descriptions.

[0078] Embodiment 1

[0079] See also Figure 1 , Figure 1 : is a flow chart of an abnormality recognition and early warning method based on deep transfer learning disclosed in an embodiment of the present invention. Figure 1 The described anomaly recognition and early warning method based on deep transfer learning can be applied to an anomaly recognition and early warning device based on deep transfer learning, and the device can be applied to an intelligent management system of energy equipment, which is not limited in the embodiments of the present invention. Optionally, the system can be applied to a local terminal device (e.g., a PC terminal, a local server, etc.) or a mobile terminal device (e.g., a smart phone, a tablet computer, a PDA, a mobile Internet device, etc.), which is not limited in the embodiments of the present invention. Figure 1 As shown, the abnormality recognition and early warning method based on deep transfer learning may include the following operations:

[0080] 101. Obtain power usage data and status data of a first device set, and perform preprocessing operations on the power usage data and status data to obtain a preprocessed first data set.

[0081] In the embodiment of the present invention, optionally, Figure 2 As shown, Figure 2This is a schematic diagram of the construction process of a fault identification model based on deep transfer learning. The power consumption data of the first device set may include: real-time power consumption data and historical power consumption data. The status data of the first device set may include: at least one of the matching binding information, usage record information, regional location information, operating status information, and auxiliary equipment association information of each first device in the first device set. The real-time power consumption data may include: at least one of the power consumption, voltage value, current value, power value, load rate, and abnormal power consumption status information of each first device in the first device set collected in real time. The historical power consumption data includes: at least one of the power consumption, voltage value, current value, power value, load rate, and abnormal power consumption status information of the first device set in a historical period. The embodiment of the present invention does not limit this.

[0082] In the embodiment of the present invention, preprocessing the power consumption data and the state data to obtain a preprocessed first data set may include:

[0083] The power consumption data and the status data are cleansed and normalized to obtain processed power consumption data and processed status data.

[0084] The processed power consumption data and the processed status data are fused to obtain a preprocessed first data set.

[0085] Among them, the data cleaning processing operations may include: deleting or marking duplicate data in the power usage data and status data, deleting or replacing invalid abnormal data in the power usage data and status data, deleting or filling missing data in the power usage data and status data, etc.

[0086] In the embodiment of the present invention, optionally, performing a fusion operation on the processed power consumption data and the processed status data to obtain a pre-processed first data set may include:

[0087] The processed power consumption data corresponding to each first device and the processed status data corresponding to each first device are merged to obtain a preprocessed first data set.

[0088] The richer the power consumption data and status data of the first device set obtained in this way, the more conducive it is to improving the efficiency of subsequent model training and the recognition accuracy of the recognition model obtained by training; and the operations of data cleaning and data normalization on the power consumption data and status data can unify the data standards to be trained, which is beneficial to the accuracy and efficiency of the subsequent fault recognition model recognition warning.

[0089] 102. Input the first data set into a preset fault identification model for training to obtain a trained first training model.

[0090] In an embodiment of the present invention, optionally, the preset fault identification model may be a fault identification model based on a deep neural network algorithm, and the first training model may include at least a plurality of intermediate layers (hidden layers).

[0091] In the embodiment of the present invention, optionally, a loss function may be used to converge the trained fault identification model, and finally a converged first training model is obtained.

[0092] In this way, inputting the first data set into the preset fault identification model for training can utilize the large amount of data and learned knowledge in the source field to perform effective migration in the target field, thereby improving the generalization ability and predictive performance of the fault identification model obtained through subsequent training in identifying other types of abnormal faults, which is beneficial to improving the accuracy and efficiency of the fault identification model's recognition and warning.

[0093] 103. Obtain device status information of the second device set, and perform a preprocessing operation on the device status information to obtain a second data set.

[0094] In the embodiment of the present invention, optionally, Figure 2 As shown, the second device set may include: at least one of: photovoltaic power generation equipment, energy storage equipment, lighting equipment, low-voltage equipment, high-voltage equipment, etc., and the device status information of the second device set may include: at least one of: matching binding information, usage record information, regional location information, operating status information, and auxiliary equipment association information of each second device in the second device set. The device status information of the second device set can be obtained through the Internet of Things platform connected to it, which is not limited in the embodiments of the present invention.

[0095] In the embodiment of the present invention, performing a preprocessing operation on the device status information to obtain a second data set may include:

[0096] The device status information is subjected to data cleaning and data normalization operations to obtain a processed second data set.

[0097] Among them, the data cleaning processing operation may include: deleting or marking duplicate data in the device status information, deleting or replacing invalid abnormal data in the device status information, deleting or filling missing data in the device status information, etc.

[0098] The richer the device status information of the second device set obtained in this way, the more conducive it is to improving the efficiency of subsequent model training and the recognition accuracy of the recognition model obtained by training; the operations of data cleaning and data normalization on the device status information of the second device set can unify the data standards to be trained, which is beneficial to the accuracy and efficiency of the subsequent fault recognition model recognition warning.

[0099] 104. Input the second data set into the first training model for retraining, and perform tuning operations on the intermediate layers and weights of the first training model based on the deep transfer learning algorithm to obtain a target fault recognition model.

[0100] In an embodiment of the present invention, tuning the intermediate layers and weights of the first training model based on the deep transfer learning algorithm to obtain the target fault recognition model may include:

[0101] The last intermediate layer of the first training model is deleted, and a fully connected layer is added to the position after the last intermediate layer is deleted to obtain the second training model.

[0102] Randomize the weights of the fully connected layers and freeze all weights of all intermediate layers in the second trained model.

[0103] The second training model with frozen weights is retrained to obtain a target fault recognition model.

[0104] In an embodiment of the present invention, optionally, the target fault identification model is used to identify and warn the fault anomalies of the second device set, and the fault types corresponding to the fault anomalies of the second device set identified by the target fault identification model may include: control system failure, communication failure, circuit failure, software failure and mechanical failure, etc. At least one of the following.

[0105] In this way, the second data set is input into the first training model and the intermediate layers and weights of the first training model are tuned based on deep transfer learning. It is possible to utilize the large amount of data and learned knowledge in the source field, use the new second data set for parameter training of the new connection layer, and perform effective migration in the target field, thereby improving the model's generalization ability and predictive performance for identifying other types of abnormal faults. Since energy equipment data often have the problem of insufficient labeling, deep transfer learning can also accelerate the model's training process by migrating the labeling information of the source field, and better adapt to the data characteristics of the target field, thereby achieving more accurate fault identification and real-time abnormal warning, thereby ensuring the safe operation and continuous supply of energy equipment.

[0106] In an optional embodiment, after executing step 104, the method may further include the following operations:

[0107] The operation data of the second equipment set in a preset historical period is input into the target fault identification model for identification, so as to obtain the fault abnormality data of the second equipment set.

[0108] The identified fault anomaly data of the second device set are classified and summarized to obtain the fault anomaly occurrence frequency of each fault type corresponding to the second device set within a preset historical period.

[0109] The fault anomaly occurrence frequencies of all fault types are arranged in descending order to obtain the target fault type corresponding to the highest fault anomaly occurrence frequency in the preset historical period, and the fault anomaly data related to the target fault type is used as the subsequent training data set to retrain and update the target fault recognition model.

[0110] It can be seen that this optional embodiment can subsequently adjust the recognition ability of the recognition model according to the fault type and recognition accuracy in the recognition results of the fault recognition model trained last time. When the model recognition shows that the frequency of a certain fault type in a specific time period or equipment location is high, the equipment operation data corresponding to the fault type is used as a reference value to adjust the training data samples of the fault model for subsequent training, which is beneficial to improve the differentiation ability of the fault recognition model and the accuracy and efficiency of recognition and early warning.

[0111] Embodiment 2

[0112] See also Figure 3 , Figure 3 : is a flow chart of an abnormality recognition and early warning method based on deep transfer learning disclosed in an embodiment of the present invention. Figure 3 The described anomaly recognition and early warning method based on deep transfer learning can be applied to an anomaly recognition and early warning device based on deep transfer learning, and the device can be applied to an intelligent management system of energy equipment, which is not limited in the embodiments of the present invention. Optionally, the system can be applied to a local terminal device (e.g., a PC terminal, a local server, etc.) or a mobile terminal device (e.g., a smart phone, a tablet computer, a PDA, a mobile Internet device, etc.), which is not limited in the embodiments of the present invention. Figure 3 As shown, the abnormality recognition and early warning method based on deep transfer learning may include the following operations:

[0113] 201. Obtain power usage data and status data of a first device set, and perform preprocessing operations on the power usage data and status data to obtain a preprocessed first data set.

[0114] 202. Input the first data set into a preset fault identification model for training to obtain a trained first training model.

[0115] 203. Obtain device status information of the second device set, and perform a preprocessing operation on the device status information to obtain a second data set.

[0116] 204. Input the second data set into the first training model for retraining, delete the last intermediate layer in the intermediate layers of the first training model, and add a fully connected layer at the position after the last intermediate layer is deleted to obtain a second training model.

[0117] 205. Randomize the weights of the fully connected layer, and perform structural analysis on the first data set and the second data set according to the data type of the first data set and the data type of the second data set to obtain data structure information of the first data set and data structure information of the second data set.

[0118] In an embodiment of the present invention,

[0119] 206. Calculate the data structure difference between the first data set and the second data set according to the data structure information of the first data set, the data structure information of the second data set, and a preset comparison algorithm.

[0120] In an embodiment of the present invention, optionally, the preset comparison algorithm may adopt a comparison analysis algorithm such as a correlation analysis algorithm and a regression analysis algorithm to analyze and calculate the data structure information of the first data set and the data structure information of the second data set to obtain the data structure difference between the first data set and the second data set.

[0121] 207. Determine the target weights that need to be frozen for all the intermediate layers in the second training model according to the data structure difference, and freeze the target weights for all the intermediate layers in the second training model.

[0122] In the embodiment of the present invention, optionally, the smaller the data structure difference is, the fewer the target weights required to be frozen in all the intermediate layers in the second training model are, and vice versa.

[0123] 208. Retrain the second training model after freezing the weights to obtain a target fault recognition model.

[0124] In the embodiment of the present invention, for other descriptions of step 201 to step 204, please refer to the detailed description of step 101 to step 104 in the first embodiment, and the embodiment of the present invention will not be repeated.

[0125] It can be seen that the implementation of the embodiment of the present invention can freeze part of the weights of the training network in a targeted manner according to the differences between different initial training data samples (the first data set corresponding to the first device set after processing and the second data set corresponding to the second device set after processing), and focus more computing resources on specific parts of the training network, thereby improving the final model's recognition efficiency and accuracy for a specific type of fault.

[0126] It can be seen that implementation Figure 3 The described abnormality identification and early warning method based on deep transfer learning can obtain the power consumption data and status data of the first device set, preprocess the power consumption data and status data, and obtain the preprocessed first data set, which is conducive to improving the recognition accuracy of the model obtained by subsequent training; obtain the device status information of the second device set, preprocess the device status information, obtain the second data set, input the second data set into the first training model for retraining, so that the final model obtained by training can identify various fault anomalies occurring in the equipment, and tune the intermediate layer and weights of the first training model based on the deep transfer learning algorithm to obtain the target fault identification model, which is conducive to improving the generalization ability and universality of the fault identification model; and according to the difference between different initial training data samples (the first data set corresponding to the first device set after processing and the second data set corresponding to the second device set after processing), freeze part of the weights of the training network in a targeted manner, and focus more computing resources on specific parts of the training network, so as to improve the recognition efficiency and accuracy of the final model for a specific type of fault.

[0127] Embodiment 3

[0128] See also Figure 4 , Figure 4 : is a schematic diagram of the structure of an abnormality recognition and early warning device based on deep transfer learning disclosed in an embodiment of the present invention. Figure 4 The described anomaly recognition and early warning device based on deep transfer learning can execute the above-mentioned anomaly recognition and early warning method based on deep transfer learning, and the device can be applied to the intelligent management system of energy equipment, which is not limited by the embodiments of the present invention. Optionally, the system can be applied to local terminal devices (for example: PC terminals, local servers, etc.) or mobile terminal devices (for example: smart phones, tablet computers, PDAs, mobile Internet devices, etc.), which is not limited by the embodiments of the present invention. Figure 4 As shown, the abnormality recognition and early warning device based on deep transfer learning may include: an acquisition module 301, a preprocessing module 302, a training module 303, and a tuning module 304, wherein:

[0129] The acquisition module 301 is used to acquire power usage data and status data of a first device set.

[0130] The preprocessing module 302 is used to perform a preprocessing operation on the power consumption data and the state data acquired by the acquisition module 301 to obtain a preprocessed first data set.

[0131] The training module 303 is used to input the first data set obtained by the preprocessing module 302 into a preset fault identification model for training, so as to obtain a first training model after training.

[0132] The acquisition module 301 is further configured to acquire device status information of the second device set.

[0133] The preprocessing module 302 is further used to perform a preprocessing operation on the device status information acquired by the acquisition module 301 to obtain a second data set.

[0134] The training module 303 is further used to input the second data set processed by the preprocessing module 302 into the first training model for retraining.

[0135] The tuning module 304 is used to tune the intermediate layer and weights of the first training model trained by the training module 303 based on the deep transfer learning algorithm to obtain a target fault identification model, which is used to identify and warn fault anomalies of the second device set.

[0136] It can be seen that implementation Figure 4 The described abnormality identification and early warning device based on deep transfer learning can obtain the power consumption data and status data of the first equipment set, preprocess the power consumption data and status data, and obtain the preprocessed first data set, which is conducive to improving the recognition accuracy of the model obtained by subsequent training; obtain the equipment status information of the second equipment set, preprocess the equipment status information, and obtain the second data set, and input the second data set into the first training model for retraining, which is conducive to improving the generalization ability and universality of the fault identification model, so that the final model obtained by training can identify various fault anomalies occurring in the equipment, and based on the deep transfer learning algorithm, the intermediate layer and weights of the first training model are tuned to obtain the target fault identification model, which is further conducive to improving the accuracy and efficiency of the fault identification model's identification and early warning.

[0137] In an optional embodiment, the electricity usage data of the first device set includes: real-time electricity usage data and historical electricity usage data, the status data of the first device set includes: at least one of the matching binding information, usage record information, regional location information, operation status information, and auxiliary equipment association information of each first device in the first device set, the real-time electricity usage data includes: at least one of the electricity usage, voltage value, current value, power value, load rate, and abnormal electricity usage status information of each first device in the first device set collected in real time, and the historical electricity usage data includes: at least one of the electricity usage, voltage value, current value, power value, load rate, and abnormal electricity usage status information of the first device set in a historical period.

[0138] It can be seen that implementation Figure 4 The more abundant the power consumption data and status data of the first device set that the described abnormality recognition and early warning device based on deep transfer learning can obtain, the more conducive it is to improving the efficiency of subsequent model training and the recognition accuracy of the recognition model obtained by training.

[0139] In another optional embodiment, Figure 4 As shown, the preprocessing module 302 performs preprocessing operations on the power consumption data and the state data to obtain the preprocessed first data set in the following specific manner:

[0140] The power consumption data and the status data are cleansed and normalized to obtain processed power consumption data and processed status data.

[0141] The processed power consumption data and the processed status data are fused to obtain a preprocessed first data set.

[0142] Among them, the operations of data cleaning processing include: deleting or marking duplicate data in power usage data and status data, deleting or replacing invalid abnormal data in power usage data and status data, and deleting or filling missing data in power usage data and status data.

[0143] It can be seen that implementation Figure 4 The described abnormality recognition and early warning device based on deep transfer learning can perform data cleaning and data normalization processing on power consumption data and status data, and can unify the data standards to be trained, which is beneficial to the accuracy and efficiency of subsequent fault recognition model recognition and early warning.

[0144] In another optional embodiment, at least one of the matching binding information, usage record information, regional location information, operating status information, and auxiliary device association information of each second device in the second device set, the preset fault identification model is a fault identification model based on a deep neural network algorithm, and the first training model includes at least multiple intermediate layers.

[0145] As well as Figure 4 As shown, the tuning module 304 performs tuning operations on the intermediate layers and weights of the first training model based on the deep transfer learning algorithm, and the specific manner of obtaining the target fault identification model is as follows:

[0146] The last intermediate layer of the first training model is deleted, and a fully connected layer is added to the position after the last intermediate layer is deleted to obtain the second training model.

[0147] Randomize the weights of the fully connected layers and freeze all weights of all intermediate layers in the second trained model.

[0148] The second training model with frozen weights is retrained to obtain a target fault recognition model.

[0149] It can be seen that implementation Figure 4 The described abnormality identification and early warning device based on deep transfer learning can input the second data set into the first training model and perform tuning operations on the intermediate layer and weights of the first training model based on deep transfer learning. It can use a large amount of data and learned knowledge in the source field to use the new second data set for parameter training of the new connection layer, and perform effective migration in the target field, thereby improving the generalization ability and prediction performance of the model for other types of abnormal fault recognition. Since energy equipment data often has the problem of insufficient labeling, deep transfer learning can also accelerate the training process of the model by migrating the labeling information of the source field, better adapt to the data characteristics of the target field, thereby achieving more accurate fault identification and real-time abnormal warning, thereby ensuring the safe operation and continuous supply of energy equipment.

[0150] In yet another optional embodiment, Figure 5 As shown, the device also includes:

[0151] The analysis module 305 is used to analyze the data type of the first data set and the data type of the second data set obtained by the preprocessing module after the tuning module 304 performs a randomization operation on the weights of the fully connected layer, so as to obtain data structure information of the first data set and data structure information of the second data set.

[0152] The calculation module 306 is used to calculate the data structure difference between the first data set and the second data set according to the data structure information of the first data set and the data structure information of the second data set analyzed by the analysis module 305 and a preset analysis algorithm.

[0153] The determination module 307 is used to determine the target weights required to be frozen for all intermediate layers in the second training model according to the size of the data structure difference calculated by the calculation module 306.

[0154] The tuning module 304 is also used to freeze the target weights of all intermediate layers in the second training model determined by the determination module 307.

[0155] The training module 303 is also used to retrain the second training model after the weights are frozen by the tuning module 304 to obtain a target fault recognition model.

[0156] It can be seen that implementation Figure 5 The described abnormality identification and early warning device based on deep transfer learning can freeze part of the weights of the training network in a targeted manner according to the differences between different initial training data samples (the first data set corresponding to the first device set after processing and the second data set corresponding to the second device set after processing), and focus more computing resources on specific parts of the training network, thereby improving the final model's recognition efficiency and accuracy for a specific type of fault.

[0157] In yet another optional embodiment, the fault type corresponding to the fault anomaly of the second device set identified by the target fault identification model includes at least one of: a control system fault, a communication fault, a circuit fault, a software fault and a mechanical fault.

[0158] It can be seen that implementation Figure 5 The described abnormality recognition and early warning device based on deep transfer learning can identify a variety of abnormal faults that occur during the operation of different types of equipment, thereby improving the universality of the fault recognition model.

[0159] In yet another optional embodiment, Figure 5 As shown, the device also includes:

[0160] The identification module 308 is used to input the operation data of the second equipment set in a preset historical period into the target fault identification model optimized by the optimization module 304 for identification, so as to obtain the fault abnormality data of the second equipment set.

[0161] The classification module 309 is used to classify and summarize the fault anomaly data of the second device set identified by the identification module 308 to obtain the fault anomaly occurrence frequency of each fault type corresponding to the second device set within a preset historical period.

[0162] The sorting module 310 is used to arrange the fault anomaly occurrence frequencies of all fault types classified by the classification module 309 in descending order to obtain the target fault type corresponding to the highest fault anomaly occurrence frequency in a preset historical period.

[0163] The training module 303 is further configured to use the fault anomaly data related to the target fault type sorted by the sorting module 310 as a subsequent training data set to retrain and update the target fault recognition model.

[0164] It can be seen that the implementation Figure 5 The described abnormality identification and early warning device based on deep transfer learning can subsequently adjust the recognition ability of the recognition model according to the fault type and recognition accuracy in the recognition results of the fault recognition model trained last time. When the model recognition shows that the frequency of a certain fault type in a specific time period or a certain equipment location is high, the equipment operation data corresponding to the fault type is used as a reference value to adjust the training data samples of the fault model for subsequent training, which is beneficial to improve the differentiation ability of the fault recognition model and the accuracy and efficiency of the recognition and early warning.

[0165] Embodiment 4

[0166] See also Figure 6 , Figure 6 is a structural diagram of another abnormality recognition and early warning device based on deep transfer learning disclosed in an embodiment of the present invention. Figure 6 The described anomaly recognition and early warning device based on deep transfer learning can execute the above-mentioned anomaly recognition and early warning method based on deep transfer learning, and the device can be applied to the intelligent management system of energy equipment, which is not limited by the embodiments of the present invention. Optionally, the system can be applied to local terminal devices (for example: PC terminals, local servers, etc.) or mobile terminal devices (for example: smart phones, tablet computers, PDAs, mobile Internet devices, etc.), which is not limited by the embodiments of the present invention. Figure 6 As shown, the abnormality recognition and early warning device based on deep transfer learning may include:

[0167] A memory 401 storing executable program codes;

[0168] a processor 402 coupled to the memory 401;

[0169] The processor 402 calls the executable program code stored in the memory 401 to execute the steps in the abnormality identification and early warning method based on deep transfer learning described in Embodiment 1 of the present invention or Embodiment 2 of the present invention.

[0170] Embodiment 5

[0171] An embodiment of the present invention discloses a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute the steps in the abnormality identification and early warning method based on deep transfer learning described in Embodiment 1 of the present invention or Embodiment 2 of the present invention.

[0172] Embodiment 6

[0173] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute the steps in the abnormality identification and early warning method based on deep transfer learning described in Example 1 or Example 2.

[0174] The device embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, i.e., they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0175] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution can be essentially or partly contributed to the prior art in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0176] Finally, it should be noted that the abnormality identification and warning method and device based on deep transfer learning disclosed in the embodiment of the present invention discloses only the preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, ordinary technicians in this field should understand that it is still possible to modify the technical solutions recorded in the aforementioned embodiments, or to replace some of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An abnormality recognition and early warning method based on deep transfer learning, characterized in that: The method comprises: Acquire power usage data and status data of a first set of devices, and perform preprocessing operations on the power usage data and the status data to obtain a preprocessed first data set; Inputting the first data set into a preset fault identification model for training to obtain a trained first training model; Acquire device status information of a second device set, and perform a preprocessing operation on the device status information to obtain a second data set; The second data set is input into the first training model for retraining, and the intermediate layers and weights of the first training model are tuned based on the deep transfer learning algorithm to obtain a target fault identification model, which is used to identify and warn fault anomalies of the second device set.

2. The abnormality recognition and early warning method based on deep transfer learning according to claim 1 is characterized in that: The power consumption data of the first device set includes: real-time power consumption data and historical power consumption data. The status data of the first device set includes: at least one of the matching binding information, usage record information, regional location information, operation status information, and auxiliary equipment association information of each first device in the first device set. The real-time power consumption data includes: at least one of the power consumption, voltage value, current value, power value, load rate, and abnormal power consumption status information of each first device in the first device set collected in real time. The historical power consumption data includes: at least one of the power consumption, voltage value, current value, power value, load rate, and abnormal power consumption status information of the first device set in a historical period.

3. The abnormality recognition and early warning method based on deep transfer learning according to claim 2 is characterized in that: The preprocessing operation is performed on the power consumption data and the state data to obtain a preprocessed first data set, including: Performing data cleaning and data normalization on the power consumption data and the status data to obtain processed power consumption data and processed status data; Performing a fusion operation on the processed power consumption data and the processed status data to obtain a preprocessed first data set; Among them, the operations of the data cleaning processing include: deleting or marking duplicate data in the power usage data and the status data, deleting or replacing invalid abnormal data in the power usage data and the status data, and deleting or filling missing data in the power usage data and the status data.

4. The abnormality recognition and early warning method based on deep transfer learning according to claim 1 is characterized in that: at least one of the matching binding information, usage record information, regional location information, operating status information, and accessory device association information of each second device in the second device set, the preset fault identification model is a fault identification model based on a deep neural network algorithm, and the first training model includes at least a plurality of intermediate layers; The deep transfer learning algorithm is used to tune the intermediate layers and weights of the first training model to obtain a target fault recognition model, including: The last intermediate layer in the intermediate layers of the first training model is deleted, and a fully connected layer is added to the position after the last intermediate layer is deleted to obtain a second training model; Randomizing the weights of the fully connected layer and freezing all weights of all intermediate layers in the second training model; The second training model with frozen weights is retrained to obtain a target fault recognition model.

5. The abnormality recognition and early warning method based on deep transfer learning according to claim 4 is characterized in that: After the randomization operation is performed on the weights of the fully connected layer, the method further includes: According to the data type of the first data set and the data type of the second data set, performing structural analysis on the first data set and the second data set to obtain data structure information of the first data set and data structure information of the second data set; Calculate the data structure difference between the first data set and the second data set according to the data structure information of the first data set, the data structure information of the second data set and a preset comparison algorithm; Determine the target weights that need to be frozen for all intermediate layers in the second training model according to the data structure difference, and freeze the target weights of all intermediate layers in the second training model; The second training model with frozen weights is retrained to obtain a target fault recognition model.

6. The abnormality recognition and early warning method based on deep transfer learning according to claim 1 or 5, characterized in that: The fault type corresponding to the fault anomaly of the second device set identified by the target fault identification model includes: at least one of a control system fault, a communication fault, a circuit fault, a software fault and a mechanical fault.

7. The abnormality recognition and early warning method based on deep transfer learning according to claim 6 is characterized in that: The method further comprises: Inputting the operation data of the second equipment set within a preset historical period into the target fault identification model for identification, and obtaining fault abnormality data of the second equipment set; Classify and summarize the identified fault anomaly data of the second device set to obtain the fault anomaly occurrence frequency of each fault type corresponding to the second device set within a preset historical period; The fault anomaly occurrence frequencies of all fault types are arranged in descending order to obtain the target fault type corresponding to the highest fault anomaly occurrence frequency in a preset historical period, and the fault anomaly data related to the target fault type is used as a subsequent training data set to retrain and update the target fault recognition model.

8. An abnormality recognition and early warning device based on deep transfer learning, characterized in that: The device comprises: An acquisition module, used to acquire power consumption data and status data of the first device set; A preprocessing module, used for performing a preprocessing operation on the power consumption data and the state data acquired by the acquisition module to obtain a preprocessed first data set; A training module, used for inputting the first data set obtained by the preprocessing module into a preset fault recognition model for training to obtain a first training model after training; The acquisition module is further used to acquire device status information of the second device set; The preprocessing module is further used to perform a preprocessing operation on the device status information acquired by the acquisition module to obtain a second data set; The training module is further used to input the second data set processed by the preprocessing module into the first training model for retraining; A tuning module is used to tune the intermediate layers and weights of the first training model obtained by training the training module based on a deep transfer learning algorithm to obtain a target fault identification model, and the target fault identification model is used to identify and warn of fault anomalies of the second device set.

9. An abnormality recognition and early warning device based on deep transfer learning, characterized in that: The device comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the abnormality identification and early warning method based on deep transfer learning as described in any one of claims 1-7.

10. A computer storage medium, characterized in that: The computer storage medium stores computer instructions, which, when called, are used to execute the abnormality identification and early warning method based on deep transfer learning as described in any one of claims 1 to 7.

Images