Fog computing environment personalized deep learning method
Technical Field
The invention relates to cloud computing, Internet of things, artificial intelligence and deep learning technologies, in particular to a personalized deep learning method in a fog computing environment.
Background
In recent years, the development of artificial intelligence technology is rapid, the commercialization speed of the technology is beyond expectations, and artificial intelligence brings subversive changes to the whole society and becomes an important development strategy for countries in the future. Particularly, the algorithm evolution taking deep learning as a core has the super-strong evolution capability, and under the support of big data, a large-scale convolutional neural network similar to a human brain structure is obtained through training and construction, so that various problems can be solved.
Deep learning needs a large amount of data and computing resources for training, cloud services can meet requirements to a certain extent, a large amount of physical hardware resources are aggregated in a cloud center, unified distribution, scheduling and management of heterogeneous network computing resources are achieved by adopting a virtualization technology, and computing and storage costs are greatly reduced by building a data center in a centralized mode. However, with the increasing volume of data, the transmission rate is decreasing, even sometimes there is a large network delay, and the computation and storage cannot be all placed in the remote cloud. At the moment, fog calculation greatly improves the situation, particularly meets various requirements of the edge side such as real-time service, data optimization, bandwidth limitation, application intelligence, safety, privacy and the like, accelerates the development of fog calculation, and brings new possibility for cloud calculation and fog calculation.
With the development of deep learning technology, the cloud training learning can generate a universal data model such as object detection, however, in a specific application scenario, the universal model cannot meet the requirement of the industry personalization, the training can be completed at the cloud, and the fog computing node penetrates through the cloud and the equipment end to form a bridge between the cloud and the equipment end, so that near-end training reasoning computing service can be provided nearby. Under the circumstances, how to effectively utilize the cloud computing and fog computing capabilities to provide personalized deep learning computing capabilities becomes an urgent problem to be solved.
Disclosure of Invention
The invention provides a fog computing environment personalized deep learning method.
The technical task of the invention is realized by the following modes:
a fog computing environment personalized deep learning method is characterized in that cloud nodes carry out training on a universal model through massive training data, the universal model obtained through training is distributed to various fog computing nodes, and then the deep learning model meeting the requirements of the edge side industry is trained by utilizing the computing and storing capacity of the fog computing nodes; the data are collected from the intelligent sensing equipment, real-time reasoning is carried out on the fog computing node, and the result is output in real time; and errors occurring in reasoning are recognized, an optimization model is continuously trained, the industry personalized model can be selectively transmitted to the cloud, and meanwhile, a general model of the cloud is received and optimization is continuously improved.
The cloud end node is responsible for continuously training and optimizing the general deep learning model, distributing the general model obtained through training to each fog computing node, and meanwhile, is responsible for collecting and storing the deep learning model related to the specific industry from the fog computing node.
The cloud computing node is a bridge between a cloud end and an equipment end and is responsible for receiving the general deep learning model from the cloud end node, adding the edge side industry personalized data for training and generating the industry personalized deep learning model.
The cloud computing node provides a reasoning function, processes and feeds back input data from the edge side intelligent sensing equipment in real time, trains according to errors occurring in reasoning, compares the errors with a general model of a cloud, continuously optimizes an industry personalized deep learning model, can interact with the cloud, and uploads the industry personalized model according to the requirements of customers.
The intelligent sensing equipment collects environmental data in real time, and utilizes the fog computing node to perform real-time deep learning computation, and the obtained result is fed back to a user in time or action is taken.
The method comprises the following operation steps:
carrying out cloud training on the cloud nodes in the step 1) through a large number of universal deep learning training sets to obtain a basic model with universal cognitive recognition capability through training;
the fog computing node in the step 2) requests a deep learning model from the cloud node according to specific industry requirements close to the edge side and the credibility of a local personalized deep learning model;
the cloud node in the step 3) distributes a universal deep learning model to the fog computing node;
step 4), the cloud end node is inquired according to the specific industry requirement of the fog computing node; if the personalized deep learning model shared by the user exists, returning the model to the fog computing node;
step 5) if the fog computing node receives the personalized deep learning model shared by the cloud end user, comparing whether the model exists locally or not, and if so, turning to step 7); otherwise, go to step 6);
the fog computing node in the step 6) is trained by utilizing industry application data and locally stored training data based on a model shared by a user according to specific industry requirements close to the edge side to generate an industry personalized deep learning model, and the industry personalized deep learning model is stored locally;
the fog computing node in the step 7) receives the universal deep learning model from the cloud, compares whether the model exists locally, and if so, goes to the step 9); otherwise, go to step 8);
the fog computing node in the step 8) is trained by utilizing industry application data and locally stored training data based on a general model according to specific industry requirements close to the edge side to generate an industry personalized deep learning model, and the industry personalized deep learning model is stored locally;
step 9), the fog computing node compares the credibility of each local model, and selects the optimal model for reasoning;
step 10), the intelligent sensing equipment acquires data according to industry requirements and sends the data to the fog computing node for reasoning;
the fog computing node utilizes an industry personalized deep learning model to carry out reasoning on the collected data, completes cognitive service, realizes industry requirements, and outputs a result in real time to the intelligent sensing equipment;
the intelligent sensing equipment in the step 12) feeds back the result to an end user and executes a corresponding working instruction;
step 13), the fog computing node forms a training sample from the collected error data and stores the training sample in a local storage;
the fog computing node in the step 14) selects idle time, regularly trains training samples stored in a local storage based on the existing personalized deep learning model to form a new deep learning training model, and stores the new deep learning training model in a local cache;
the fog computing node uploads and shares the personalized deep learning model meeting the specific industry requirements to the cloud according to user permission;
and step 16) circularly executing the step 1) to the step 15), continuously performing model optimization, improving deep learning calculation reasoning capability and meeting industry personalized requirements.
The cognitive ability of the basic model in the step 1) is suitable for a general scene.
And (3) outputting the result in real time in the step 11) and sending the result to the intelligent sensing equipment, wherein the step comprises the step of forming a training sample by using the acquired data and the reasoning result and storing the training sample in a local cache storage of the fog computing node.
And 12) executing a corresponding working instruction, wherein if an error occurs, the error and a correction result are uploaded to the fog computing node.
Said step 13) is saved in the local storage, including, and calculating the credibility of the plurality of deep learning models in the local storage.
Compared with the prior art, the fog computing environment personalized deep learning method fully utilizes the characteristic that the fog computing nodes are used as a cloud and equipment end bridge, distributes deep learning computation on the cloud and the fog computing nodes, enables the cloud to be responsible for training of the general model, distributes the general model to each fog computing node, and utilizes the computing and storing capacity of the fog computing nodes and combines the specific industry application requirements of the fog computing edge side. According to the method, the individual requirements of specific industries are fully considered, high-efficiency cognitive computing capacity meeting the application of the specific industries is provided, deep learning computation is completed at the fog computing node and is close to the equipment demand side, real-time service execution efficiency is improved, meanwhile, the individual deep learning model is stored in the fog computing node, the safety of the model is guaranteed, on the other hand, the model can be shared to the cloud according to user permission, and the model is open. In addition, the fog computing node selects the idle time continuous optimization model through error reasoning feedback of practical application, computing resources are effectively utilized, the cloud general model is continuously updated, and the model with higher credibility is selected and utilized, so that the continuous optimization industry personalized deep learning computing capability is provided.
Drawings
FIG. 1 is a schematic diagram of deep learning computing node composition of a fog computing environment personalized deep learning method;
fig. 2 is a flow chart of a fog computing environment personalized deep learning method.
Detailed Description
Example 1:
the invention is further described with reference to the following figures and specific examples, which are not intended to be limiting.
As shown in fig. 1, the cloud gathers a large amount of computing resources, performs training of a general model through mass training data, distributes the general model obtained by training to each fog computing node, and trains a deep learning model meeting the requirements of the edge side industry by using the computing and storage capabilities of the fog computing nodes; the data are collected from the intelligent sensing equipment, real-time reasoning is carried out on the fog computing node, and the result is output in real time; and errors occurring in reasoning are recognized, an optimization model is continuously trained, the industry personalized model can be selectively transmitted to the cloud, and meanwhile, a general model of the cloud is received and optimization is continuously improved. Wherein,
the cloud node is responsible for continuously training and optimizing the general deep learning model, distributing the general model obtained by training to each fog computing node, and meanwhile, collecting and storing the deep learning model related to the specific industry from the fog computing node; the cloud computing node is a bridge between a cloud end and an equipment end, is responsible for receiving a general deep learning model from the cloud end, adding edge side industry personalized data for training, generating an industry personalized deep learning model, simultaneously providing a reasoning function, processing and feeding back input data from edge side intelligent sensing equipment in real time, training the cloud computing node according to errors occurring in reasoning, comparing the error with the general model of the cloud end, continuously optimizing the industry personalized deep learning model, interacting with the cloud end, and uploading the industry personalized model according to the requirements of customers; the intelligent sensing equipment collects environmental data in real time, and utilizes the fog computing node to perform real-time deep learning computation, and the obtained result is fed back to a user in time or action is taken.
Referring to fig. 2, the personalized deep learning calculation includes the steps of:
the cloud node in the step 1) carries out cloud training through a large number of universal deep learning training sets to obtain a basic model with universal cognitive recognition capability through training, and the cognitive capability of the model is suitable for a universal scene;
the fog computing node in the step 2) requests a deep learning model from the cloud node according to specific industry requirements close to the edge side and the credibility of a local personalized deep learning model;
the cloud node in the step 3) distributes a universal deep learning model to the fog computing node;
the cloud node in the step 4) queries according to the specific industry requirements of the fog computing node, and if an individualized deep learning model shared by a user exists, the model is returned to the fog computing node;
step 5) if the fog computing node receives the personalized deep learning model shared by the cloud end user, comparing whether the model exists locally or not, and if so, turning to step 7); otherwise, go to step 6);
the fog computing node in the step 6) is trained by utilizing industry application data and locally stored training data based on a model shared by a user according to specific industry requirements close to the edge side to generate an industry personalized deep learning model, and the industry personalized deep learning model is stored locally;
the fog computing node in the step 7) receives the universal deep learning model from the cloud, compares whether the model exists locally, and if so, goes to the step 9); otherwise, go to step 8);
the fog computing node in the step 8) is trained by utilizing industry application data and locally stored training data based on a general model according to specific industry requirements close to the edge side to generate an industry personalized deep learning model, and the industry personalized deep learning model is stored locally;
step 9), the fog computing node compares the credibility of each local model, and selects the optimal model for reasoning;
step 10), the intelligent sensing equipment acquires data according to industry requirements and sends the data to the fog computing node for reasoning;
step 11) the fog computing node utilizes an industry personalized deep learning model to reason the collected data, completes cognitive service, realizes industry requirements, outputs a result in real time to the intelligent sensing equipment, and simultaneously forms a training sample by the collected data and the inference result to be stored in a local cache of the fog computing node;
step 12) the intelligent sensing equipment feeds back the result to an end user, executes a corresponding work instruction, and uploads an error and a correction result to the fog computing node if the error occurs;
step 13), the fog computing node forms a training sample from the collected error data, stores the training sample in a local storage, and computes the credibility of a plurality of deep learning models in the local storage;
the fog computing node in the step 14) selects idle time, regularly trains training samples stored in a local storage based on the existing personalized deep learning model to form a new deep learning training model, and stores the new deep learning training model in a local cache;
the fog computing node uploads and shares the personalized deep learning model meeting the specific industry requirements to the cloud according to user permission;
and step 16) circularly executing the step 1) to the step 15), continuously performing model optimization, improving deep learning calculation reasoning capability and meeting industry personalized requirements.
The present invention can be easily implemented by those skilled in the art from the above detailed description. It should be understood, however, that the intention is not to limit the invention to the particular embodiments described. On the basis of the disclosed embodiments, a person skilled in the art can combine different technical features at will, thereby implementing different technical solutions.