Edge computing method and system for edge server

Abstract

The present invention relates to an edge computing method and system for an edge server, which sends the data to be processed generated by a terminal device to the nearest edge server for data complexity calculation; when the complexity is not greater than a preset complexity threshold, the data to be processed is sent to a matching edge server for processing, otherwise the data is decomposed to obtain data blocks of different complexities, and different data blocks are transmitted to a matching edge server for processing according to the data type and the corresponding complexity, and sent to the nearest edge server for fusion processing and returned to the terminal device; data prediction is performed through a data prediction model, and the prediction result is transmitted to a data decomposition module for pre-decomposition, which is decomposed according to a pre-decomposition scheme and then sent to a matching edge server for data processing, and the matching edge server transmits the data processing result to the nearest edge server for data fusion and then returns it to the terminal device, thereby realizing the processing of large-scale data and complex computing tasks.

Description

Technical Field

[0001] The present invention belongs to the field of edge computing technology, and in particular relates to an edge computing method and system for an edge server. Background Art

[0002] ‌Edge computing‌ is a distributed computing framework that transfers data processing and analysis tasks from central cloud servers to the edge of the network, that is, close to the data source of the terminal device, and processes the data through the edge server. This computing mode allows IoT devices to quickly process data and take actions locally, thereby reducing the need to transmit data to remote data centers, reducing latency and increasing processing speed.

[0003] The edge computing process in the prior art is as follows: the terminal device generates a large amount of data to be processed in real time during operation, and transmits the data to be processed to the nearest edge server through the communication channel between the terminal device and the edge server. After receiving the data to be processed, the edge server performs data identification on it, identifies the data that it can process and the data that it cannot process, that is, the data that needs to be processed by the cloud server, and transmits the results processed by the edge server and the data that needs to be processed by the cloud server to the cloud server through the data transmission channel between the edge server and the cloud server. The cloud server processes the data that needs to be processed by the cloud server and returns the data processing results to the edge server. The edge server sends the data results processed by itself and the data results processed by the cloud server to the terminal device.

[0004] The above-mentioned edge computing process processes part of the data locally. However, due to the development of terminal devices, large-scale data and complex computing tasks are generated. As a result, the computing power of a single edge server can no longer meet the processing of large-scale data and complex computing tasks, which makes the development of edge computing reach a bottleneck.

[0005] Therefore, how to improve the edge computing process of existing edge servers to realize the processing of large-scale data and complex computing tasks and enhance edge computing capabilities is a technical problem that needs to be solved urgently. Summary of the invention

[0006] The purpose of the present invention is to provide an edge computing method and system for an edge server, so as to improve the edge computing process of an existing edge server, so as to realize the processing of large-scale data and complex computing tasks and enhance the edge computing capability.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0008] In a first aspect, an edge computing method of an edge server is provided, comprising the following steps:

[0009] S1: The terminal device generates data to be processed, and sends the data to be processed to the nearest edge server, and the nearest edge server performs data complexity calculation on the data to be processed;

[0010] S2: Determine whether the complexity calculation result of the data to be processed is greater than a preset complexity threshold. If not, identify the category of the data to be processed, and directly process the data by the nearest edge server or send the data to be processed to a matching edge server for data processing according to the identification result. If yes, execute step S3;

[0011] S3: The data decomposition module decomposes the data to be processed to obtain data blocks of different complexity, adds marks to the decomposed data blocks according to the positional relationship before decomposition, and performs data identification on the decomposed data blocks to obtain corresponding data types;

[0012] S4: transmitting different data blocks to the matching edge server according to the data type and the corresponding complexity, and the edge server processes the received data blocks to be processed and sends the data processing results to the nearest edge server in step S1;

[0013] S5: the nearest edge server performs data fusion processing on the data processing results received from different edge servers according to the tags, and transmits the data fusion results to the terminal device;

[0014] S6: After a specified time period, annotate the data to be processed of the acquired terminal device, create a data prediction model, input the annotated data to be processed into the data prediction model to train the model, evaluate the performance of the data prediction model, and determine whether the performance of the data prediction model meets the preset requirements. If yes, execute step S7; if not, retrain the model until it meets the preset requirements.

[0015] S7: predicting the data to be processed at a specified time point by using the data prediction model, transmitting the prediction result of the data to be processed to the data decomposition module for pre-decomposition, and matching the corresponding edge server according to the pre-decomposition result;

[0016] S8: collect the data to be processed of the terminal device in real time, and compare the predicted data to be processed with the real-time collected data to be processed, and determine whether the predicted data to be processed is within the error range. If so, execute step S9; if not, redistribute the data other than the predicted data to be processed by differential calculation;

[0017] S9: After decomposing according to the pre-decomposition scheme, the data is sent to the matching edge server for data processing. The matching edge server transmits the data processing result to the nearest edge server for data fusion and then returns it to the terminal device.

[0018] Preferably, the specific process of calculating the data complexity of the data to be processed in step S1 is as follows:

[0019] S11: dividing the data to be processed into virtual data blocks, obtaining the time complexity of each virtual data block, and adding the time complexity of each virtual data block to obtain a time complexity polynomial;

[0020] S12: Take the highest-order term in the time complexity polynomial, change its coefficient to 1, and use it as the time complexity of the data to be processed;

[0021] S13: calculating the size of the memory space required to be occupied when the virtual data blocks are executed using the corresponding algorithm, and summing the sizes of the memory space required to be occupied by all the virtual data blocks to obtain the space complexity of the data to be processed;

[0022] The formula of the time complexity polynomial in step S11 is:

[0023] T( n )=T1( n )+T2( n )+···+T m ( n )

[0024] =O( f 1( n ))+O( f 2( n ))+···+ O( f m ( n ))

[0025] =O( max ( f 1( n ), f 2( n ),··· f m ( n )));

[0026] Among them, T( n ) is the total time complexity of the data to be processed, T1( n ) is the time complexity of the first virtual data block, T2( n ) is the time complexity of the second virtual data block, T m (n ) is the m The time complexity of virtual data blocks is O( f 1( n )) is the order of the time complexity of the first virtual data block, O( f 2( n )) is the order of the time complexity of the second virtual data block, O( f m ( n )) is the m The order of the time complexity of the virtual data block, max (·) is the function that takes the maximum value among them.

[0027] Preferably, the specific process of determining whether the complexity calculation result of the data to be processed is greater than a preset complexity threshold in step S2 is as follows:

[0028] S21: Determine whether the time complexity of the data to be processed is greater than a preset time complexity threshold, if so, execute step S3, if not, execute step S22;

[0029] S22: Determine whether the space complexity of the data to be processed is greater than a preset space complexity threshold. If so, execute step S3; if not, execute the subsequent data identification and data sending actions of step S2.

[0030] Preferably, the specific process of the data decomposition module in step S3 performing data decomposition on the data to be processed is as follows:

[0031] S31: Perform preliminary data decomposition according to the virtual data blocks in step S2, obtain the time complexity polynomial of the data to be processed, and extract the items greater than the specified order;

[0032] S32: Decompose the data blocks corresponding to the items with a value greater than the specified order again, so that the time complexity of the decomposed data blocks is less than a preset time complexity threshold.

[0033] Preferably, the specific processing process of data fusion in step S5 is as follows:

[0034] S51: extracting features from the data processing results received from different edge servers respectively;

[0035] S52: constructing different feature vectors by extracting features from the data processing results received from different edge servers, and adding the tags to the different feature vectors;

[0036] S53: merging different feature vectors based on the tags to obtain a final feature vector, and converting the final feature vector to obtain a data processing result.

[0037] Preferably, the specific process of merging different feature vectors based on the tags in step S53 is as follows:

[0038] S531: assigning corresponding preset weights to the different feature vectors;

[0039] S532: Merge the feature vectors by weighted summation based on preset weights to obtain a final feature vector.

[0040] Preferably, the data prediction model in step S6 is a long short-term memory network model, which includes multiple LSTM units, each of which includes a memory unit, an input gate, a forget gate and an output gate, the input gate is connected to the memory unit and the forget gate, and the output gate is connected to the memory unit and the forget gate.

[0041] Preferably, the training process of the long short-term memory network model is as follows:

[0042] S61: The labeled data to be processed are sequentially fed into the first LSTM unit, the forget gate of the first LSTM unit determines the forgotten data therein, and the input data to be processed are mapped to the interval [0, 1] through the ‌Sigmoid function;

[0043] S62: Determine through the input gate that the unforgotten data to be processed is added to the cell state, generate candidate values ​​through the tanh function, determine the updated candidate values ​​through the ‌Sigmoid function, determine the output cell state through the output gate, and determine the output value through the Sigmoid function;

[0044] S63: Calculate the loss according to the output value and data annotation through the cross entropy loss function, calculate the gradient of the loss function with respect to each weight, and iteratively update the model weights and hyperparameters.

[0045] In a second aspect, an edge computing system of an edge server is provided, which is used to implement the edge computing method of an edge server, characterized in that it includes a data acquisition module, a complexity calculation module, a data decomposition module, a data identification module, a data fusion processing module, a data prediction model and a data matching module, wherein the data acquisition module is connected to the complexity calculation module, the complexity calculation module is connected to the data decomposition module, the data decomposition module is connected to the data identification module and the edge server, the edge server is connected to the data fusion processing module, the data fusion processing module is connected to a terminal device, the data prediction model is connected to the data decomposition module, and the data matching module is connected to the data identification module;

[0046] The data acquisition module is used to collect the data to be processed of the terminal device in real time;

[0047] The complexity calculation module is used to calculate the data complexity of the data to be processed;

[0048] The data identification module is used to identify the category of the data to be processed;

[0049] The data matching module is used to match the corresponding edge server for the data block to be processed according to the data identification result;

[0050] The data decomposition module is used to decompose the data to be processed to obtain data blocks of different complexities, and add marks to the decomposed data blocks according to the positional relationship before decomposition;

[0051] The data fusion processing module is used to perform data fusion processing on the data processing results received from different edge servers according to the tags, and transmit the data fusion results to the terminal device;

[0052] The data prediction model is used to predict the data to be processed at a specified time point.

[0053] The beneficial effects of the present invention include:

[0054] The edge computing method and system of the edge server provided by the present invention send the data to be processed generated by the terminal device to the nearest edge server for data complexity calculation; when the complexity is not greater than the preset complexity threshold, the data to be processed is sent to the matching edge server for processing, otherwise the data is decomposed to obtain data blocks of different complexities, and different data blocks are transmitted to the matching edge server for processing according to the data type and the corresponding complexity, and sent to the nearest edge server for data fusion processing, and returned to the terminal device; after creating a data prediction model for parallel training, the data to be processed is predicted, and the prediction result is transmitted to the data decomposition module for pre-decomposition and matching the edge server; after decomposition according to the pre-decomposition scheme, it is sent to the matching edge server for data processing, and the matching edge server transmits the data processing result to the nearest edge server for data fusion and returns it to the terminal device.

[0055] First, the data to be processed is divided into virtual data blocks, and its time complexity is obtained and added to obtain the time complexity polynomial; the highest-order term in the time complexity polynomial is taken, and its coefficient is changed to 1 as the time complexity of the data to be processed, and the size of the memory space required for the processing of the virtual data blocks is calculated respectively, and the space complexity of the data to be processed is obtained by summing up. The precise time complexity and space complexity of data processing are obtained, which is convenient for the subsequent selection of more suitable edge servers for data processing. When the complexity is low, it can be processed by a single server. When it exceeds the preset threshold, it needs to be decomposed and collaboratively processed by multiple edge services, which effectively solves the problem that the existing edge computing cannot realize the processing of large-scale data and complex computing tasks, and improves data processing efficiency and reduces data processing latency.

[0056] Secondly, when the time complexity of the data to be processed is not greater than the preset time complexity threshold, it is further determined whether the space complexity of the data to be processed is greater than the preset space complexity threshold, and the processing methods of data to be processed with different complexities are further refined to match them with better processing methods to ensure the processing efficiency of the data to be processed and the accuracy of the processing results.

[0057] Thirdly, by performing preliminary data decomposition on the virtual data blocks, the time complexity polynomial of the data to be processed is obtained, and the items greater than the specified order are extracted. The data blocks corresponding to the items greater than the specified order are decomposed again, so that the time complexity of the decomposed data blocks is less than the preset time complexity threshold. The high-order complexity data blocks are decomposed, so that the time complexity order is effectively reduced, and through collaborative processing by different edge servers, the data processing time is greatly reduced, and the data processing efficiency is improved.

[0058] Thirdly, by extracting features from the data processing results received from different edge servers respectively, different feature vectors are constructed by the features extracted from the data processing results received from different edge servers, and tags are added to different feature vectors; different feature vectors are merged based on the tags to obtain the final feature vector, and the final feature vector is converted to obtain the data processing result, so as to achieve accurate fusion of the processing results of different edge servers, obtain accurate processing results of the data to be processed, and achieve collaborative processing by different edge servers for large-scale and complex tasks to obtain ideal data processing results, further expand the processing capacity of edge computing, reduce the data processing volume of cloud servers, and improve data security.

[0059] Finally, by setting the long short-term memory network model to include multiple LSTM units, each LSTM unit includes a memory unit, an input gate, a forget gate, and an output gate structure, as well as a subsequent model training process, it is possible to accurately capture the relationship between time and data sequences, achieve accurate prediction of data, and then perform subsequent data decomposition and edge server matching in advance, thereby improving the processing efficiency of edge computing, reducing the latency of edge computing, and further improving edge computing performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of the flow of the edge computing method of the edge server of the present invention.

[0061] Figure 2 It is a flowchart of data complexity calculation of the present invention.

[0062] Figure 3 A schematic diagram of the flow of an edge computing system of an edge server of the present invention. DETAILED DESCRIPTION

[0063] The following is combined with Figure 1~Figure 3 The present invention is further described in detail:

[0064] Example 1

[0065] See attached Figure 1 As shown, an edge computing method for an edge server includes the following steps:

[0066] S1: The terminal device generates data to be processed and sends the data to be processed to the nearest edge server, which performs data complexity calculation on the data to be processed. The terminal device generates data to be processed in real time and then sends it to the nearest edge server for processing. However, the nearest edge server may not be able to independently process all the data generated by the terminal device. Even if it can process, it will take a lot of time. However, due to the latency requirements of the terminal device, the nearest edge server may not be able to complete it within the latency requirements. Therefore, after the edge server receives the data to be processed, the present invention will first perform complexity calculation and then further determine the data processing method.

[0067] S2: Determine whether the complexity calculation result of the data to be processed is greater than the preset complexity threshold. If not, identify the category of the data to be processed, and directly process it by the nearest edge server according to the identification result or send the data to be processed to the matching edge server for data processing. If yes, execute step S3. A complexity threshold is set in advance, and the complexity threshold is set according to the ability of the nearest edge server to independently process the data to be processed. If it is greater than the threshold, it means that the data to be processed cannot be processed independently by the edge server, and needs to be processed collaboratively by multiple edge servers.

[0068] S3: The data decomposition module decomposes the data to be processed to obtain data blocks of different complexity, and adds tags to the decomposed data blocks according to the positional relationship before decomposition. The tags are added to facilitate the fusion of the results after the subsequent processing by multiple edge servers. The decomposed data blocks are respectively identified to obtain the corresponding data type. Since multiple edge servers are required to collaboratively process the data to be processed, the data to be processed needs to be decomposed. By decomposing it into multiple data blocks and then sending them to the matching edge servers for simultaneous processing, the edge computing capability can be improved.

[0069] S4: Different data blocks are transmitted to the matching edge server according to the data type and corresponding complexity. The edge server processes the received data blocks to be processed and sends the data processing results to the nearest edge server in step S1. The above process can match the decomposed data blocks with more suitable edge servers for processing, thereby improving the edge computing efficiency and the accuracy of the calculation results.

[0070] S5: The nearest edge server performs data fusion processing on the data processing results received from different edge servers according to the tags, and transmits the data fusion results to the terminal device. The data fusion process makes the collaborative processing of multiple edge servers possible.

[0071] S6: After a specified time period, the acquired data to be processed of the terminal device is labeled, a data prediction model is created, and the labeled data to be processed is input into the data prediction model to train the model, and the performance of the data prediction model is evaluated to determine whether the performance of the data prediction model meets the preset requirements. If yes, step S7 is executed, if not, the model is retrained until the preset requirements are met;

[0072] S7: predicting the data to be processed at a specified time point through the data prediction model, transmitting the prediction result of the data to be processed to the data decomposition module for pre-decomposition, and matching the corresponding edge server according to the pre-decomposition result;

[0073] S8: collect the data to be processed of the terminal device in real time, and compare the predicted data to be processed with the real-time collected data to be processed, and determine whether the predicted data to be processed is within the error range. If so, execute step S9; if not, redistribute the data other than the predicted data to be processed by differential calculation;

[0074] S9: After decomposition according to the pre-decomposition scheme, the data is sent to the matching edge server for data processing. The matching edge server transmits the data processing result to the nearest edge server for data fusion and then returns it to the terminal device.

[0075] In the existing edge computing, the terminal device generates a large amount of data to be processed in real time during operation, and transmits the data to be processed to the nearest edge server through the communication channel between the terminal device and the edge server. After receiving the data to be processed, the edge server performs data identification on it, identifies the data that it can process and the data that it cannot process, that is, the data that needs to be processed by the cloud server, and transmits the results processed by the edge server and the data that needs to be processed by the cloud server to the cloud server through the data transmission channel between the edge server and the cloud server. The cloud server processes the data that needs to be processed by the cloud server and returns the data processing results to the edge server. The edge server sends the data results processed by itself and the data results processed by the cloud server to the terminal device. Since this data processing method of a single edge server greatly limits the edge computing capability, the present invention hopes to expand the edge computing model by improving the edge computing model.

[0076] In the present invention, the data to be processed generated by the terminal device is sent to the nearest edge server for data complexity calculation; when the complexity is not greater than the preset complexity threshold, the data to be processed is sent to the matching edge server for processing, otherwise the data is decomposed to obtain data blocks of different complexities, and different data blocks are transmitted to the matching edge server for processing according to the data type and the corresponding complexity, and sent to the nearest edge service for data fusion processing, and returned to the terminal device; after creating a data prediction model for parallel training, the data to be processed is predicted, and the prediction result is transmitted to the data decomposition module for pre-decomposition and matched to the edge server; after decomposition according to the pre-decomposition scheme, it is sent to the matching edge server for data processing, and the matching edge server transmits the data processing result to the nearest edge server for data fusion and returns it to the terminal device. This edge computing method realizes the processing of large-scale data and complex computing tasks through collaborative computing between multiple edge servers, expands the edge computing mode, greatly improves the edge computing capability, and enables some data that originally had to be sent to the cloud server to be realized through collaborative computing of multiple edge servers.

[0077] Example 2

[0078] Based on Example 1, the specific process of calculating the data complexity of the data to be processed in step S1 is as follows:

[0079] S11: dividing the data to be processed into virtual data blocks, obtaining the time complexity of each virtual data block, and adding the time complexity of each virtual data block to obtain a time complexity polynomial;

[0080] S12: Take the highest-order term in the time complexity polynomial, change its coefficient to 1, and use it as the time complexity of the data to be processed;

[0081] S13: calculating the size of the memory space required to be occupied when the virtual data blocks are executed using the corresponding algorithm, and summing the sizes of the memory space required to be occupied by all the virtual data blocks to obtain the space complexity of the data to be processed;

[0082] The formula of the time complexity polynomial in step S11 is:

[0083] T( n )=T1( n )+T2( n )+···+T m ( n )

[0084] =O( f 1( n ))+O( f 2( n ))+···+ O( f m ( n ))

[0085] =O( max ( f 1( n ), f 2( n ),··· f m ( n )));

[0086] Among them, T( n ) is the total time complexity of the data to be processed, T1( n ) is the time complexity of the first virtual data block, T2( n ) is the time complexity of the second virtual data block, T m ( n ) is the m The time complexity of virtual data blocks is O( f 1(n )) is the order of the time complexity of the first virtual data block, O( f 2( n )) is the order of the time complexity of the second virtual data block, O( f m ( n )) is the m The order of the time complexity of the virtual data block, max (·) is the function that takes the maximum value among them.

[0087] The data to be processed is divided into virtual data blocks, and their time complexities are obtained and added to obtain the time complexity polynomial; the highest-order term in the time complexity polynomial is taken, and its coefficient is changed to 1 as the time complexity of the data to be processed, and the size of the memory space required for the processing of the virtual data blocks is calculated respectively, and the space complexity of the data to be processed is obtained by summing up. The precise time complexity and space complexity of data processing are obtained, which is convenient for the subsequent selection of more suitable edge servers for data processing. When the complexity is low, it can be processed by a single server. When it exceeds the preset threshold, it needs to be decomposed and collaboratively processed by multiple edge services, which effectively solves the problem that the existing edge computing cannot realize the processing of large-scale data and complex computing tasks, and improves data processing efficiency and reduces data processing latency.

[0088] In this embodiment, the specific process of determining whether the complexity calculation result of the data to be processed is greater than the preset complexity threshold in step S2 is as follows:

[0089] S21: Determine whether the time complexity of the data to be processed is greater than a preset time complexity threshold, if so, execute step S3, if not, execute step S22;

[0090] S22: Determine whether the space complexity of the data to be processed is greater than a preset space complexity threshold. If so, execute step S3; if not, execute the subsequent data identification and data sending actions of step S2.

[0091] Since data complexity is generally measured by time complexity and space complexity, in the present invention, the time complexity needs to be considered. The latency of edge computing is a very important aspect. When the time complexity of the data to be processed is not greater than the preset time complexity threshold, it is further judged whether the space complexity of the data to be processed is greater than the preset space complexity threshold, and the processing methods of the data to be processed with different complexities are further refined to match them with better processing methods to ensure the processing efficiency of the data to be processed and the accuracy of the processing results.

[0092] Example 3

[0093] On the basis of Example 1 or Example 2, the specific process of the data decomposition module in step S3 performing data decomposition on the data to be processed is as follows:

[0094] S31: Perform preliminary data decomposition according to the virtual data blocks in step S2, obtain the time complexity polynomial of the data to be processed, and extract the items greater than the specified order;

[0095] S32: Decompose the data blocks corresponding to the items with a value greater than the specified order again, so that the time complexity of the decomposed data blocks is less than a preset time complexity threshold.

[0096] By performing preliminary data decomposition by dividing virtual data into blocks, obtaining the time complexity polynomial of the data to be processed, extracting the items greater than the specified order, and decomposing the data blocks corresponding to the items greater than the specified order again, the time complexity of the decomposed data blocks is less than the preset time complexity threshold. The high-order complexity data blocks are decomposed, so that the time complexity order is effectively reduced, and through collaborative processing by different edge servers, the data processing time is greatly reduced and the data processing efficiency is improved.

[0097] The specific processing process of data fusion in step S5 is as follows:

[0098] S51: Extract features from the data processing results received from different edge servers, and obtain features representing data information in the data through feature extraction. By preprocessing the data processing results, the terminal device generates real-time data to be processed. In the process of edge computing, the processing results of the data to be processed are ultimately required to inform the terminal device of the execution action. Therefore, the subsequent data fusion is achieved by extracting the features of the processing results of multiple edge servers.

[0099] S52: construct different feature vectors by extracting features from the data processing results received from different edge servers, and add the tags to the different feature vectors. Construct feature vectors from the extracted features to achieve mathematical expression of each processing result, which is convenient for subsequent feature merging to obtain the final processing result.

[0100] S53: merging different feature vectors based on the tags to obtain a final feature vector, and converting the final feature vector to obtain a data processing result.

[0101] The specific process of merging different feature vectors based on the tags in step S53 is as follows:

[0102] S531: assigning corresponding preset weights to the different feature vectors;

[0103] S532: Merge the feature vectors by weighted summation based on preset weights to obtain a final feature vector.

[0104] In this embodiment, features are extracted from data processing results received from different edge servers respectively, different feature vectors are constructed by features extracted from data processing results received from different edge servers, and labels are added to different feature vectors; different feature vectors are merged based on the labels to obtain a final feature vector, and the final feature vector is converted to obtain the data processing result, so as to achieve accurate fusion of the processing results of different edge servers, obtain accurate processing results of the data to be processed, and achieve collaborative processing by different edge servers for large-scale and complex tasks to obtain ideal data processing results, further expand the processing capacity of edge computing, reduce the data processing volume of cloud servers, and improve data security.

[0105] Example 4

[0106] Based on Example 1 or Example 2 or Example 3, the data prediction model in step S6 is a long short-term memory network model, and the long short-term memory network model includes multiple LSTM units, each of which includes a memory unit, an input gate, a forget gate and an output gate, the input gate is connected to the memory unit and the forget gate, and the output gate is connected to the memory unit and the forget gate.

[0107] In this embodiment, the training process of the long short-term memory network model is as follows:

[0108] S61: The labeled data to be processed are sequentially fed into the first LSTM unit, the forget gate of the first LSTM unit determines the forgotten data therein, and the input data to be processed are mapped to the interval [0, 1] through the ‌Sigmoid function;

[0109] S62: Determine through the input gate that the unforgotten data to be processed is added to the cell state, generate candidate values ​​through the tanh function, determine the updated candidate values ​​through the ‌Sigmoid function, determine the output cell state through the output gate, and determine the output value through the Sigmoid function;

[0110] S63: Calculate the loss according to the output value and data annotation through the cross entropy loss function, calculate the gradient of the loss function with respect to each weight, and iteratively update the model weights and hyperparameters.

[0111] An edge computing system of an edge server, used to implement an edge computing method of an edge server, characterized in that it includes a data acquisition module, a complexity calculation module, a data decomposition module, a data identification module, a data fusion processing module, a data prediction model and a data matching module, the data acquisition module is connected to the complexity calculation module, the complexity calculation module is connected to the data decomposition module, the data decomposition module is connected to the data identification module and the edge server, the edge server is connected to the data fusion processing module, the data fusion processing module is connected to a terminal device, the data prediction model is connected to the data decomposition module, and the data matching module is connected to the data identification module.

[0112] The data acquisition module is used to collect the data to be processed of the terminal device in real time. The complexity calculation module is used to calculate the data complexity of the data to be processed; the data identification module is used to identify the category of the data to be processed; the data matching module is used to match the corresponding edge server for the data block to be processed according to the data identification result; the data decomposition module is used to decompose the data to be processed to obtain data blocks of different complexities, and add tags to the decomposed data blocks according to the position relationship before decomposition; the data fusion processing module is used to perform data fusion processing on the data processing results received from different edge servers according to the tags, and transmit the data fusion results to the terminal device; the data prediction model is used to predict the data to be processed at a specified time point.

[0113] In summary, the edge computing method and system of the edge server provided by the present invention sends the data to be processed generated by the terminal device to the nearest edge server for data complexity calculation; when the complexity is not greater than the preset complexity threshold, the data to be processed is sent to the matching edge server for processing, otherwise the data is decomposed to obtain data blocks of different complexities, and different data blocks are transmitted to the matching edge server for processing according to the data type and the corresponding complexity, and sent to the nearest edge service for data fusion processing, and returned to the terminal device; after creating a data prediction model for parallel training, the data to be processed is predicted, and the prediction result is transmitted to the data decomposition module for pre-decomposition and matching the edge server; after decomposition according to the pre-decomposition scheme, it is sent to the matching edge server for data processing, and the matching edge server transmits the data processing result to the nearest edge server for data fusion and returns it to the terminal device.

[0114] The data to be processed is divided into virtual data blocks, and its time complexity is obtained and added to obtain the time complexity polynomial calculation as the time complexity of the data to be processed, and the size of the memory space required for the virtual data block processing is calculated respectively, and the space complexity of the data to be processed is obtained by summing up to obtain the accurate time complexity and space complexity of the data processing. It is convenient to select a more suitable edge server for data processing in the future. When the complexity is low, it can be processed by a single server. When it exceeds the preset threshold, it needs to be decomposed and processed collaboratively by multiple edge services, which effectively solves the problem that the existing edge computing cannot realize the processing of large-scale data and complex computing tasks, and improves data processing efficiency and reduces data processing delay. When the time complexity of the data to be processed is not greater than the preset time complexity threshold, it is further judged whether the space complexity of the data to be processed is greater than the preset space complexity threshold, and the processing methods of the data to be processed with different complexities are further refined to match them with a better processing method to ensure the processing efficiency of the data to be processed and the accuracy of the processing results.

[0115] By performing preliminary data decomposition by dividing virtual data into blocks, obtaining the time complexity polynomial of the data to be processed, extracting the items greater than the specified order, and decomposing the data blocks corresponding to the items greater than the specified order again, so that the time complexity of the decomposed data blocks is less than the preset time complexity threshold, the decomposition of the data blocks with high-order complexity effectively reduces the time complexity order, and through the collaborative processing of different edge servers, the data processing time is greatly reduced and the data processing efficiency is improved. By extracting features from the data processing results received from different edge servers respectively, constructing different feature vectors through the features extracted from the data processing results received from different edge servers, and adding tags to different feature vectors; merging different feature vectors based on the tags to obtain the final feature vector, and converting the final feature vector to obtain the data processing result, the processing results of different edge servers are accurately integrated to obtain the accurate processing results of the data to be processed, and different edge servers are collaboratively processed for large-scale and complex tasks to obtain ideal data processing results, further expanding the processing capacity of edge computing, reducing the data processing volume of cloud servers, and improving data security.

[0116] By setting up multiple LSTM units of the long short-term memory network model, each LSTM unit includes the structure of a memory unit, an input gate, a forget gate and an output gate, as well as a subsequent model training process, it is possible to accurately capture the relationship between time and data sequences, achieve accurate prediction of data, and then perform subsequent data decomposition and edge server matching in advance, thereby improving the processing efficiency of edge computing, reducing the latency of edge computing, and further improving edge computing performance.

Claims

1. An edge computing method for an edge server, characterized in that: The following steps are involved: S1: The terminal device generates data to be processed, and sends the data to be processed to the nearest edge server, and the nearest edge server performs data complexity calculation on the data to be processed; S2: Determine whether the complexity calculation result of the data to be processed is greater than a preset complexity threshold. If not, identify the category of the data to be processed, and directly process the data by the nearest edge server or send the data to be processed to a matching edge server for data processing according to the identification result. If yes, execute step S3; S3: The data decomposition module decomposes the data to be processed to obtain data blocks of different complexity, adds marks to the decomposed data blocks according to the positional relationship before decomposition, and performs data identification on the decomposed data blocks to obtain corresponding data types; S4: transmitting different data blocks to the matching edge server according to the data type and the corresponding complexity, and the edge server processes the received data blocks to be processed and sends the data processing results to the nearest edge server in step S1; S5: the nearest edge server performs data fusion processing on the data processing results received from different edge servers according to the tags, and transmits the data fusion results to the terminal device; S6: After a specified time period, annotate the data to be processed of the acquired terminal device, create a data prediction model, input the annotated data to be processed into the data prediction model to train the model, evaluate the performance of the data prediction model, and determine whether the performance of the data prediction model meets the preset requirements. If yes, execute step S7; if not, retrain the model until it meets the preset requirements. S7: predicting the data to be processed at a specified time point by using the data prediction model, transmitting the prediction result of the data to be processed to the data decomposition module for pre-decomposition, and matching the corresponding edge server according to the pre-decomposition result; S8: collect the data to be processed of the terminal device in real time, and compare the predicted data to be processed with the real-time collected data to be processed, and determine whether the predicted data to be processed is within the error range. If so, execute step S9; if not, redistribute the data other than the predicted data to be processed by differential calculation; S9: Decomposing the data according to the pre-decomposition scheme and sending the data to the matching edge server for data processing. The matching edge server transmits the data processing result to the nearest edge server for data fusion and then returns it to the terminal device; The specific process of calculating the data complexity of the data to be processed in step S1 is as follows: S11: dividing the data to be processed into virtual data blocks, obtaining the time complexity of each virtual data block, and adding the time complexity of each virtual data block to obtain a time complexity polynomial; S12: Take the highest-order term in the time complexity polynomial, change its coefficient to 1, and use it as the time complexity of the data to be processed; S13: calculating the size of the memory space required to be occupied when the virtual data blocks are executed using the corresponding algorithm, and summing the sizes of the memory space required to be occupied by all the virtual data blocks to obtain the space complexity of the data to be processed; The formula of the time complexity polynomial in step S11 is: T( n )=T1( n )+T2( n )+···+T m ( n ); =O( f 1( n ))+O( f 2( n ))+···+ O( f m ( n )) =O( max ( f 1( n ), f 2( n ),··· f m ( n )))； Among them, T( n ) is the total time complexity of the data to be processed, T1( n ) is the time complexity of the first virtual data block, T2( n ) is the time complexity of the second virtual data block, T m ( n ) is the m The time complexity of virtual data blocks is O( f 1( n )) is the order of the time complexity of the first virtual data block, O( f 2( n )) is the order of the time complexity of the second virtual data block, O( f m ( n )) is the m The order of the time complexity of the virtual data block, max (·) is the function that takes the maximum value; The specific process of determining whether the complexity calculation result of the data to be processed is greater than the preset complexity threshold in step S2 is as follows: S21: Determine whether the time complexity of the data to be processed is greater than a preset time complexity threshold, if so, execute step S3, if not, execute step S22; S22: Determine whether the space complexity of the data to be processed is greater than a preset space complexity threshold. If so, execute step S3; if not, execute the subsequent data identification and data sending actions of step S2; The specific process of the data decomposition module in step S3 performing data decomposition on the data to be processed is as follows: S31: Perform preliminary data decomposition according to the virtual data blocks in step S2, obtain the time complexity polynomial of the data to be processed, and extract the items greater than the specified order; S32: Decomposing the data blocks corresponding to the items with a value greater than the specified order again, so that the time complexity of the decomposed data blocks is less than a preset time complexity threshold; The specific processing process of data fusion in step S5 is as follows: S51: extracting features from the data processing results received from different edge servers respectively; S52: constructing different feature vectors by extracting features from the data processing results received from different edge servers, and adding the tags to the different feature vectors; S53: merging different feature vectors based on the mark to obtain a final feature vector, and converting the final feature vector to obtain a data processing result; The specific process of merging different feature vectors based on the tags in step S53 is as follows: S531: assigning corresponding preset weights to the different feature vectors; S532: merging feature vectors by weighted summation calculation based on preset weights to obtain a final feature vector; The data prediction model in step S6 is a long short-term memory network model, and the long short-term memory network model includes a plurality of LSTM units, each of which includes a memory unit, an input gate, a forget gate and an output gate, the input gate is connected to the memory unit and the forget gate, and the output gate is connected to the memory unit and the forget gate; The training process of the long short-term memory network model is as follows: S61: The labeled data to be processed are sequentially fed into the first LSTM unit, the forget gate of the first LSTM unit determines the forgotten data therein, and the input data to be processed are mapped to the interval [0, 1] through the ‌Sigmoid function; S62: Determine through the input gate that the unforgotten data to be processed is added to the cell state, generate candidate values ​​through the tanh function, determine the updated candidate values ​​through the ‌Sigmoid function, determine the output cell state through the output gate, and determine the output value through the Sigmoid function; S63: Calculate the loss according to the output value and data annotation through the cross entropy loss function, calculate the gradient of the loss function with respect to each weight, and iteratively update the model weights and hyperparameters.

2. An edge computing system of an edge server, used to implement the edge computing method of an edge server according to claim 1, characterized in that: It includes a data acquisition module, a complexity calculation module, a data decomposition module, a data identification module, a data fusion processing module, a data prediction model and a data matching module, wherein the data acquisition module is connected to the complexity calculation module, the complexity calculation module is connected to the data decomposition module, the data decomposition module is connected to the data identification module and the edge server, the edge server is connected to the data fusion processing module, the data fusion processing module is connected to the terminal device, the data prediction model is connected to the data decomposition module, and the data matching module is connected to the data identification module; The data acquisition module is used to collect the data to be processed of the terminal device in real time; The complexity calculation module is used to calculate the data complexity of the data to be processed; The data identification module is used to identify the category of the data to be processed; The data matching module is used to match the corresponding edge server for the data block to be processed according to the data identification result; The data decomposition module is used to decompose the data to be processed to obtain data blocks of different complexities, and add marks to the decomposed data blocks according to the positional relationship before decomposition; The data fusion processing module is used to perform data fusion processing on the data processing results received from different edge servers according to the tags, and transmit the data fusion results to the terminal device; The data prediction model is used to predict the data to be processed at a specified time point.

Images