Cloud computing power dynamic distribution system and method based on cloud server

Abstract

The invention relates to a cloud computing power dynamic distribution system and method based on a cloud server. The method comprises the following steps: establishing a mapping table between task data and computing power; creating a task prediction model and training; creating a computing power prediction model and training; predicting the task data of each terminal device through the task prediction model, and outputting a task prediction result of each intelligent terminal; predicting computing power data of the cloud server through the computing power prediction model, and outputting a computing power prediction result; dividing calculation power prediction results into different calculation power resource pools, and virtually segmenting calculation power into calculation power blocks with different sizes; extracting a task prediction result of each intelligent terminal, calling the mapping table to match a corresponding resource pool for the task of each intelligent terminal, and pre-allocating a specified computing power block to the task of each intelligent terminal; and collecting real-time task data of each intelligent terminal, and carrying out computing power redistribution on the real-time task data of the intelligent terminals which are not in the preset error range. The dynamic allocation of the computing power is realized, and the computing power allocation efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of cloud computing power allocation, and in particular to a cloud computing power dynamic allocation system and method based on a cloud server. Background Art

[0002] With the unprecedented development of artificial intelligence, a large number of intelligent devices and applications are greatly influencing and promoting the progress of economic society and human civilization. Computing power is not only the foundation of artificial intelligence, but also one of the most important driving forces for the rapid development of artificial intelligence. As one of the three basic elements of artificial intelligence development, efficient computing power plays a catalytic role in data processing, algorithm optimization, high-precision and fast interaction. The rapid and vigorous development of computing and connection technologies has made the interconnection of all things in the intelligent world an irreversible trend.

[0003] With the development of artificial intelligence, a large number of artificial intelligence device terminals and artificial intelligence applications will generate a large amount of data every moment, and it is necessary to allocate corresponding computing power to process the data generated at these moments, which will bring about the problem of computing power allocation.

[0004] In the prior art, there are currently two ways to distribute computing power: random distribution and round-robin distribution. The random distribution method has a simple distribution process, but the scale of distributed systems to which the random distribution method can be applied is relatively small. Currently, the increasingly large-scale distributed systems cannot guarantee the fairness and efficiency of task or data distribution using random distribution methods, which can easily lead to some computing nodes being overloaded and some computing nodes being underloaded. The round-robin distribution method distributes task data to different computing nodes in a fixed order, which can achieve a more even distribution of computing power and avoid node overload. However, the round-robin distribution method cannot dynamically allocate according to the actual load of the node, resulting in the node's computing power not being fully utilized.

[0005] Therefore, with the development of artificial intelligence, the scale of distributed systems is getting larger and larger. How to improve the existing computing power allocation method to avoid the situation where the random allocation method cannot guarantee the fairness and efficiency of task or data allocation, which may easily lead to some computing nodes being overloaded and some computing nodes being underloaded, or the polling allocation method cannot be dynamically allocated according to the actual load of the node, resulting in the node's computing power not being fully utilized. Summary of the invention

[0006] The purpose of the present invention is to provide a cloud computing power dynamic allocation system and method based on a cloud server, so as to improve the existing computing power allocation method and avoid the situation where the random allocation method cannot guarantee the fairness and efficiency of task or data allocation, which may easily lead to some computing nodes being overloaded while some computing nodes are underloaded or the polling allocation method cannot be dynamically allocated according to the actual load of the node, resulting in the computing power of the node not being fully utilized.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows: In a first aspect, a method for dynamically allocating cloud computing power based on a cloud server is provided, comprising the following steps: S1: Establishing a mapping table between task data and computing power, wherein the mapping table establishes a matching relationship between the type of task data and the type of computing power, and a matching relationship between the amount of task data and the size of computing power; S2: Create a task prediction model, obtain historical task data within a specified period of each terminal device, construct a historical task data set after preprocessing, and input it into the task prediction model for training, and determine whether the prediction ability of the task prediction model meets the preset indicators. If so, execute step S3; if not, adjust the parameters of the task prediction model until the task prediction ability meets the preset indicators; S3: Create a computing power prediction model, obtain the computing power data of the specified period of the cloud server, construct a historical computing power data set after preprocessing, and input it into the computing power prediction model for training, and determine whether the prediction ability of the computing power prediction model meets the preset indicators. If so, execute step S4; if not, adjust the parameters of the computing power prediction model until the computing power prediction ability meets the preset indicators; S4: predicting the task data of each terminal device at a specified time through the task prediction model, and outputting the task prediction result of each smart terminal; predicting the computing power data of the cloud server through the computing power prediction model, and outputting the computing power prediction result; S5: Based on the computing power prediction result, the computing power prediction result is divided into different computing power resource pools by using containerization technology, and the computing power is virtually divided into computing power blocks of different sizes in different computing power resource pools; S6: extracting the task prediction results of each smart terminal, calling the mapping table, matching the task of each smart terminal with a corresponding resource pool based on the mapping table, and pre-allocating the specified computing power blocks in the corresponding resource pool to the tasks of each smart terminal; S7: Collect the real-time task data of each smart terminal, and compare the real-time task data with the corresponding task prediction results, extract the real-time task data of the smart terminal whose difference between the task prediction result and the real-time task data is not within a preset error range, and redistribute based on the extracted real-time task data of the smart terminal.

[0008] Preferably, the specific process of step S1 is as follows: S11: Divide the task data of each intelligent terminal into different task types according to the processing time complexity; S12: Divide the computing power into computing power types that match different task types; S13: Match the corresponding computing power for the specified task volume of each task type; S14: construct a general mapping table of task types and computing power types, and establish a sub-mapping table of the specified task amount and the corresponding computing power of each task type under the general mapping table.

[0009] Preferably, in step S11, different task types are divided according to processing time complexity into constant-order tasks, logarithmic-order tasks and exponential-order tasks, and in step S12, computing power types matching different task types include constant-order computing power, logarithmic-order computing power and exponential-order computing power.

[0010] Preferably, the task prediction model in step S2 and the computing power prediction model in step S3 are both long short-term memory network models. The task prediction model and the computing power prediction model both include an input gate, a forget gate, a memory cell and an output gate; The input gate is used to input a historical task data set, and extract the specified task data from the input historical task data to be updated into the memory cell state; The forget gate is used to determine whether the specified task data in the input historical task data is forgotten; The memory cells are used to record time series information of task data; The output gate is used to control the output of the specified task data in the memory cell state to the hidden state.

[0011] Preferably, the input gate uses a nonlinear action function of the first neuron to determine the task information to be updated to the memory cell state, and the first nonlinear action function generates a value from 0 to 1, indicating the proportion of the task data updated to the cell state, and the specific formula is: I t = s 1 ( W i *[ h t−1, x t ]+ b i ); in, I t represents the input gate, s 1 is the first nonlinear action function, W i is the weight matrix of the input gate, h t−1 is the hidden state at the previous time step, x t is the input of the current time step, b i is the bias term of the input gate; The forget gate uses the nonlinear action function of the second neuron to determine the forgotten task information. The second nonlinear action function generates a value from 0 to 1, indicating the proportion of forgotten task data. The specific formula is: F t = s 2 ( ( W f *[ h t−1 , x t ]+ b f ); in, F t represents the forget gate, s 2 is the second nonlinear action function, W f is the weight matrix of the forget gate, h t−1 is the hidden state at the previous time step, x t is the input of the current time step, b f is the bias term of the forget gate; The output gate uses the nonlinear action function of the third neuron to determine the output task information. The third nonlinear action function generates a value from 0 to 1, indicating the proportion of the task data to be output. The specific formula is: O t = s 3 ( ( W o *[ h t−1 ,x t ]+ b o ); in, O t represents the forget gate, s 3 is the second nonlinear action function, W o is the weight matrix of the forget gate, h t−1 is the hidden state at the previous time step, x t is the input of the current time step, b o is the bias term of the output gate.

[0012] Preferably, the specific process in step S5 is as follows: S51: Identify the predicted computing power and add a computing power type identifier, and allocate it to different computing power resource pools according to the computing power type identifier based on containerization technology. The computing power resource pools match the computing power types, including a constant-order computing power resource pool, a logarithmic-order computing power resource pool, and an exponential-order computing power resource pool. S52: Virtually divide the computing power in the constant-order computing power resource pool, the logarithmic-order computing power resource pool, and the exponential-order computing power resource pool into computing power blocks of different sizes.

[0013] Preferably, the specific process in step S6 is as follows: S61: extracting task prediction results of each intelligent terminal, and identifying the type and size of the predicted task data; S62: calling the general mapping table, and extracting a matching computing resource pool type according to the general mapping table based on the task type of each smart terminal; S63: calling the sub-mapping table, and allocating computing power blocks of the corresponding computing power resource pool according to the sub-mapping table according to the task data size of each smart terminal; When the corresponding computing power resource pool in step S63 cannot match a computing power block of the corresponding size, the computing power blocks in the corresponding computing power resource pool are combined according to the size of the task data, and then the combined computing power blocks are allocated.

[0014] Preferably, the process of reallocating the real-time task data of the smart terminal whose difference between the task prediction result and the real-time task data is not within the preset error range based on the extracted real-time task data of the smart terminal in step S7 is the process of steps S62-S63.

[0015] In a second aspect, a cloud computing power dynamic allocation system based on a cloud server is provided, which is used to implement any one of the cloud computing power dynamic allocation methods based on a cloud server, including a mapping table generation module, a model creation module, a task prediction model, a computing power prediction model, a computing power resource pool, a computing power virtual segmentation module, a data acquisition module, a data comparison module, and a computing power allocation module; the model creation module is connected to the task prediction model and the computing power prediction model, the computing power resource pool is connected to the computing power virtual segmentation module, the computing power prediction model is connected to the computing power resource pool, the task prediction model is connected to the data acquisition module, and the computing power resource pool is connected to the computing power allocation module; The mapping table generation module is used to generate a mapping table between task data and computing power, wherein the mapping table establishes a matching relationship between the type of task data and the type of computing power, and a matching relationship between the amount of task data and the size of computing power; The model creation module is used to create a task prediction model and a computing power prediction model; The task prediction model is used to predict the task data of each terminal device at a specified time and output the task prediction result of each intelligent terminal; The computing power prediction model is used to predict the computing power data of the cloud server through the computing power prediction model and output the computing power prediction result; The computing power resource pool is used to accommodate different types of computing power; The computing power virtual division module is used to virtually divide the computing power of the computing power resource pool; The data acquisition module is used to collect real-time task data of each intelligent terminal; The data comparison module is used to compare the real-time task data with the corresponding task prediction results, and extract the real-time task data of the intelligent terminal where the difference between the task prediction result and the real-time task data is not within a preset error range; The computing power allocation module is used to call the mapping table and match the corresponding resource pool for the task of each smart terminal based on the mapping table, pre-allocate the specified computing power blocks in the corresponding resource pool to the tasks of each smart terminal, and redistribute them based on the extracted real-time task data of the smart terminal.

[0016] The beneficial effects of the present invention include: The cloud computing power dynamic allocation system and method based on the cloud server provided by the present invention establishes a mapping table between task data and computing power; creates and trains a task prediction model; creates and trains a computing power prediction model; predicts the task data of each terminal device through the task prediction model, and outputs the task prediction result of each intelligent terminal; predicts the computing power data of the cloud server through the computing power prediction model, and outputs the computing power prediction result; divides the computing power prediction result into different computing power resource pools, and virtually divides the computing power into computing power blocks of different sizes; extracts the task prediction result of each intelligent terminal, and calls the mapping table to match the corresponding resource pool for the task of each intelligent terminal, and pre-allocates the specified computing power block to the task of each intelligent terminal; collects the real-time task data of each intelligent terminal, and redistributes the computing power for the real-time task data of the intelligent terminal that is not within the preset error range, thereby realizing the dynamic allocation of computing power and improving the computing power allocation efficiency.

[0017] First, by establishing a mapping table between task data and computing power, the mapping table establishes the matching relationship between the type of task data and the type of computing power, and the matching relationship between the amount of task data and the size of computing power. This makes it unnecessary to use various algorithms to calculate which computing power is allocated to which task data when allocating computing power, but directly look up the mapping table to allocate computing power, which greatly saves computing power allocation time and improves computing power allocation efficiency.

[0018] Secondly, by creating a task prediction model and a computing power prediction model, the task data of each smart terminal at a specified time point can be predicted, as well as the computing power of each cloud server at a specified time point. Based on the predicted task data and the predicted computing power, computing power is pre-allocated. When real-time data is subsequently allocated, if the predicted tasks are the same, they are directly allocated according to the pre-allocation results. If the predicted task data of a smart terminal exceeds the error range, it can be allocated separately, which can further save computing power allocation time and improve computing power allocation efficiency.

[0019] Again, the task prediction model and the computing power prediction model set an input gate, a forget gate, a memory cell and an output gate. The input gate uses the nonlinear action function of the first neuron to determine the task information to be updated to the memory cell state, the forget gate uses the nonlinear action function of the second neuron to determine the forgotten task information, and the output gate uses the nonlinear action function of the third neuron to determine the output task information, thereby realizing accurate prediction of the task data of the smart terminal and the computing power of the cloud server, and providing an accurate data basis for the subsequent pre-allocation of computing power.

[0020] In addition, the computing power prediction results are divided into different computing power resource pools through containerization technology based on the computing power prediction results, and the computing power is virtually divided into computing power blocks of different sizes in different computing power resource pools. When the computing power blocks of the corresponding sizes cannot be matched in the corresponding computing power resource pools, the computing power blocks in the corresponding computing power resource pools are combined according to the size of the task data, and then the combined computing power blocks are allocated. It can realize the one-time allocation of computing power, reduce the number of times data is allocated again, further save computing power allocation time, and improve computing power allocation efficiency.

[0021] Finally, by identifying the type of predicted computing power and adding a computing power type identifier, and based on containerization technology, allocating it to different computing power resource pools according to the computing power type identifier, the computing power resource pool is matched with the computing power type, including constant-order computing power resource pool, logarithmic-order computing power resource pool, and exponential-order computing power resource pool, so that different task data can be allocated to the appropriate computing power for processing, ensuring that each type of task can be processed efficiently and reducing data processing delays. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic flow chart of the cloud computing power dynamic allocation method based on cloud server of the present invention.

[0023] Figure 2 It is a schematic diagram of the architecture of the cloud computing power dynamic allocation system based on cloud servers of the present invention.

[0024] Figure 3 A schematic diagram of the process of allocating computing power according to the present invention. DETAILED DESCRIPTION

[0025] The following is combined with Figure 1~Figure 3 The present invention is further described in detail: Example 1 See attached Figure 1 As shown, the cloud computing power dynamic allocation method based on the cloud server includes the following steps: S1: Establish a mapping table between task data and computing power, the mapping table establishes the matching relationship between the type of task data and the type of computing power, and the matching relationship between the amount of task data and the size of computing power. By establishing a mapping table between task data and computing power, the mapping table establishes the matching relationship between the type of task data and the type of computing power, and the matching relationship between the amount of task data and the size of computing power. When allocating computing power, there is no need to use various algorithms to calculate which computing power is allocated to which task data, but directly look up the mapping table to allocate computing power, which greatly saves computing power allocation time and improves computing power allocation efficiency.

[0026] S2: Create a task prediction model, obtain the historical task data within the specified period of each terminal device, construct a historical task data set after preprocessing, and input it into the task prediction model for training, and determine whether the prediction ability of the task prediction model meets the preset indicators. If so, execute step S3; if not, adjust the parameters of the task prediction model until the task prediction ability meets the preset indicators.

[0027] S3: Create a computing power prediction model, obtain the computing power data of the specified period of the cloud server, and construct a historical computing power data set after preprocessing, and input it into the computing power prediction model for training, and determine whether the prediction ability of the computing power prediction model meets the preset indicators. If so, execute step S4, if not, adjust the parameters of the computing power prediction model until the computing power prediction ability meets the preset indicators. By creating a task prediction model and a computing power prediction model, the task data of each smart terminal at a specified time point can be predicted, and the computing power of each cloud server at a specified time point can be predicted, and the computing power can be pre-allocated based on the predicted task data and the predicted computing power. When real-time data is subsequently allocated, if the predicted tasks are the same, they are directly allocated according to the pre-allocation results. If the predicted task data of a smart terminal exceeds the error range, it is allocated separately, which can further save computing power allocation time and improve computing power allocation efficiency.

[0028] S4: predicting the task data of each terminal device at a specified time through the task prediction model, and outputting the task prediction result of each smart terminal; predicting the computing power data of the cloud server through the computing power prediction model, and outputting the computing power prediction result; S5: Based on the computing power prediction result, the computing power prediction result is divided into different computing power resource pools by using containerization technology, and the computing power is virtually divided into computing power blocks of different sizes in different computing power resource pools; S6: extracting the task prediction results of each smart terminal, calling the mapping table, matching the task of each smart terminal with a corresponding resource pool based on the mapping table, and pre-allocating the specified computing power blocks in the corresponding resource pool to the tasks of each smart terminal; S7: Collect the real-time task data of each smart terminal, and compare the real-time task data with the corresponding task prediction results, extract the real-time task data of the smart terminal whose difference between the task prediction result and the real-time task data is not within a preset error range, and redistribute based on the extracted real-time task data of the smart terminal.

[0029] In this embodiment, the specific process of step S1 is as follows: S11: Divide the task data of each intelligent terminal into different task types according to the processing time complexity; S12: Divide the computing power into computing power types that match different task types; S13: Match the corresponding computing power for the specified task volume of each task type; S14: construct a general mapping table of task types and computing power types, and establish a sub-mapping table of the specified task amount and the corresponding computing power of each task type under the general mapping table.

[0030] In step S11, different task types are divided according to the processing time complexity into constant-order tasks, logarithmic-order tasks and exponential-order tasks. In step S12, the computing power types matching the different task types include constant-order computing power, logarithmic-order computing power and exponential-order computing power.

[0031] Example 2 On the basis of Example 1, the task prediction model in step S2 and the computing power prediction model in step S3 are both long short-term memory network models. The task prediction model and the computing power prediction model both include an input gate, a forget gate, a memory cell and an output gate; the input gate is used to input a historical task data set, and extract the specified task data in the input historical task data to be updated to the memory cell state; the forget gate is used to determine that the specified task data in the input historical task data is forgotten; the memory cell is used to record the time series information of the task data; the output gate is used to control the output of the specified task data in the memory cell state to the hidden state. The memory cell state is the core of the long short-term memory network model. It records the sequence information and passes it to the next moment. The update of the cell state depends on the results of the input gate and the forget gate. It can selectively remember or forget past information, thereby solving the gradient vanishing or gradient exploding problems encountered by traditional RNN when processing long sequences, so that the long short-term memory network model can better capture long-term dependencies when processing sequence data, and realize accurate prediction of task data and computing power data based on time series.

[0032] In this embodiment, the input gate uses the nonlinear action function of the first neuron to determine the task information updated to the memory cell state. The first nonlinear action function generates a value from 0 to 1, indicating the proportion of the task data updated to the cell state. The specific formula is: I t = s 1 ( W i *[ h t−1 , x t ]+ b i ); in, It represents the input gate, s 1 is the first nonlinear action function, W i is the weight matrix of the input gate, h t−1 is the hidden state at the previous time step, x t is the input of the current time step, b i is the bias term of the input gate; The forget gate uses the nonlinear action function of the second neuron to determine the forgotten task information. The second nonlinear action function generates a value from 0 to 1, indicating the proportion of forgotten task data. The specific formula is: F t = s 2 ( ( W f *[ h t−1 , x t ]+ b f ); in, F t represents the forget gate, s 2 is the second nonlinear action function, W f is the weight matrix of the forget gate, h t−1 is the hidden state at the previous time step, x t is the input of the current time step, b f is the bias term of the forget gate; The output gate uses the nonlinear action function of the third neuron to determine the output task information. The third nonlinear action function generates a value from 0 to 1, indicating the proportion of the task data to be output. The specific formula is: O t = s 3 ( ( W o *[ h t−1 , x t ]+ b o ); in, Ot represents the forget gate, s 3 is the second nonlinear action function, W o is the weight matrix of the forget gate, h t−1 is the hidden state at the previous time step, x t is the input of the current time step, b o is the bias term of the output gate.

[0033] The input gate, forget gate, memory cell and output gate are set through the task prediction model and the computing power prediction model. The input gate uses the nonlinear action function of the first neuron to determine the task information to be updated to the memory cell state, the forget gate uses the nonlinear action function of the second neuron to determine the forgotten task information, and the output gate uses the nonlinear action function of the third neuron to determine the output task information. This achieves accurate prediction of the task data of the smart terminal and the computing power of the cloud server, providing an accurate data basis for the subsequent pre-allocation of computing power.

[0034] Example 3 Based on Example 1 or Example 2, the specific process in step S5 is as follows: S51: Identify the predicted computing power and add a computing power type identifier, and allocate it to different computing power resource pools according to the computing power type identifier based on containerization technology. The computing power resource pools match the computing power types, including a constant-order computing power resource pool, a logarithmic-order computing power resource pool, and an exponential-order computing power resource pool. S52: Virtually divide the computing power in the constant-order computing power resource pool, the logarithmic-order computing power resource pool, and the exponential-order computing power resource pool into computing power blocks of different sizes.

[0035] By using containerization technology to divide the computing power prediction results into different computing power resource pools based on the computing power prediction results, and virtually dividing the computing power into computing power blocks of different sizes in different computing power resource pools, and when the computing power blocks of the corresponding size cannot be matched in the corresponding computing power resource pool, the computing power blocks in the corresponding computing power resource pool are combined according to the size of the task data, and then the combined computing power blocks are allocated. It can achieve a one-time allocation of computing power, reduce the number of times data is allocated again, further save computing power allocation time, and improve computing power allocation efficiency.

[0036] In this embodiment, see Figure 3 , the specific process in step S6 is as follows: S61: extracting task prediction results of each intelligent terminal, and identifying the type and size of the predicted task data; S62: calling the general mapping table, and extracting a matching computing resource pool type according to the general mapping table based on the task type of each smart terminal; S63: calling the sub-mapping table, and allocating computing power blocks of the corresponding computing power resource pool according to the sub-mapping table according to the task data size of each smart terminal; When the corresponding computing power resource pool in step S63 cannot match the computing power block of the corresponding size, the computing power blocks in the corresponding computing power resource pool are combined according to the size of the task data, and then the combined computing power blocks are allocated. Through the mapping table between task data and computing power, the mapping table establishes the matching relationship between the type of task data and the type of computing power, and the matching relationship between the amount of task data and the size of computing power. When allocating computing power, there is no need to use various algorithms to calculate which computing power is allocated to which task data, but directly look up the mapping table for computing power allocation, which greatly saves computing power allocation time and improves computing power allocation efficiency.

[0037] The process of reallocating the real-time task data of the smart terminal whose difference between the task prediction result and the real-time task data is not within the preset error range based on the extracted real-time task data of the smart terminal in step S7 is a process of steps S62-S63.

[0038] By identifying the type of predicted computing power and adding a computing power type identifier, and allocating it to different computing power resource pools according to the computing power type identifier based on containerization technology, the computing power resource pool is matched with the computing power type, including constant-order computing power resource pool, logarithmic-order computing power resource pool and exponential-order computing power resource pool, so that different task data can be allocated to appropriate computing power for processing, ensuring that each type of task can be processed efficiently and reducing data processing delays.

[0039] See also Figure 2 , a cloud computing power dynamic allocation system based on a cloud server, used to implement any one of the cloud computing power dynamic allocation methods based on a cloud server, including a mapping table generation module, a model creation module, a task prediction model, a computing power prediction model, a computing power resource pool, a computing power virtual segmentation module, a data acquisition module, a data comparison module, and a computing power allocation module; the model creation module is connected to the task prediction model and the computing power prediction model, the computing power resource pool is connected to the computing power virtual segmentation module, the computing power prediction model is connected to the computing power resource pool, the task prediction model is connected to the data acquisition module, and the computing power resource pool is connected to the computing power allocation module.

[0040] The mapping table generation module is used to generate a mapping table between task data and computing power, and the mapping table establishes a matching relationship between the type of task data and the type of computing power, and a matching relationship between the amount of task data and the size of computing power; the model creation module is used to create a task prediction model and a computing power prediction model; the task prediction model is used to predict the task data of each terminal device at a specified time, and output the task prediction result of each smart terminal; the computing power prediction model is used to predict the computing power data of the cloud server through the computing power prediction model, and output the computing power prediction result; the computing power resource pool is used to accommodate different types of computing power; the computing power virtual The quasi-segmentation module is used to virtually divide the computing power of the computing power resource pool; the data collection module is used to collect the real-time task data of each intelligent terminal; the data comparison module is used to compare the real-time task data with the corresponding task prediction results, and extract the real-time task data of the intelligent terminal whose difference between the task prediction result and the real-time task data is not within a preset error range; the computing power allocation module is used to call the mapping table, and match the corresponding resource pool for the task of each intelligent terminal based on the mapping table, pre-allocate the specified computing power blocks in the corresponding resource pool to the tasks of each intelligent terminal, and redistribute them based on the extracted real-time task data of the intelligent terminal.

[0041] In summary, the cloud computing power dynamic allocation system and method based on the cloud server provided by the present invention establishes a mapping table between task data and computing power, and the mapping table establishes the matching relationship between the type of task data and the type of computing power and the matching relationship between the amount of task data and the size of computing power. When allocating computing power, there is no need to use various algorithms to calculate which computing power is allocated to which task data, but directly look up the mapping table to allocate computing power, which greatly saves computing power allocation time and improves computing power allocation efficiency. By creating a task prediction model and a computing power prediction model, the task data of each smart terminal at a specified time point is predicted, and the computing power of each cloud server at a specified time point is predicted, and the computing power is pre-allocated based on the predicted task data and the predicted computing power. When real-time data is subsequently allocated, if the predicted tasks are the same, they are directly allocated according to the pre-allocation results. If the predicted task data of the smart terminal exceeds the error range, it is allocated separately, which can further save computing power allocation time and improve computing power allocation efficiency.

[0042] The task prediction model and the computing power prediction model are provided with input gates, forget gates, memory cells and output gates. The input gate uses the nonlinear action function of the first neuron to determine the task information updated to the memory cell state, the forget gate uses the nonlinear action function of the second neuron to determine the forgotten task information, and the output gate uses the nonlinear action function of the third neuron to determine the output task information, which realizes the accurate prediction of the task data of the intelligent terminal and the computing power of the cloud server, and provides an accurate data basis for the subsequent pre-allocation of computing power. The computing power prediction results are divided into different computing power resource pools based on the computing power prediction results through containerization technology, and the computing power is virtually divided into computing power blocks of different sizes in different computing power resource pools. When the computing power blocks of the corresponding size cannot be matched in the corresponding computing power resource pool, the computing power blocks in the corresponding computing power resource pool are combined according to the size of the task data, and then the combined computing power blocks are allocated. It can realize the one-time allocation of computing power, reduce the number of times data is allocated again, further save computing power allocation time, and improve computing power allocation efficiency. By identifying the type of predicted computing power and adding a computing power type identifier, and allocating it to different computing power resource pools according to the computing power type identifier based on containerization technology, the computing power resource pool is matched with the computing power type, including constant-order computing power resource pool, logarithmic-order computing power resource pool and exponential-order computing power resource pool, so that different task data can be allocated to appropriate computing power for processing, ensuring that each type of task can be processed efficiently and reducing data processing delays.

Claims

1. A cloud computing power dynamic allocation method based on a cloud server, characterized in that: The following steps are involved: S1: Establishing a mapping table between task data and computing power, wherein the mapping table establishes a matching relationship between the type of task data and the type of computing power, and a matching relationship between the amount of task data and the size of computing power; S2: Create a task prediction model, obtain historical task data within a specified period of each terminal device, construct a historical task data set after preprocessing, and input it into the task prediction model for training, and determine whether the prediction ability of the task prediction model meets the preset indicators. If so, execute step S3; if not, adjust the parameters of the task prediction model until the task prediction ability meets the preset indicators; S3: Create a computing power prediction model, obtain the computing power data of the specified period of the cloud server, construct a historical computing power data set after preprocessing, and input it into the computing power prediction model for training, and determine whether the prediction ability of the computing power prediction model meets the preset indicators. If so, execute step S4; if not, adjust the parameters of the computing power prediction model until the computing power prediction ability meets the preset indicators; S4: predicting the task data of each terminal device at a specified time through the task prediction model, and outputting the task prediction result of each smart terminal; predicting the computing power data of the cloud server through the computing power prediction model, and outputting the computing power prediction result; S5: Based on the computing power prediction result, the computing power prediction result is divided into different computing power resource pools by using containerization technology, and the computing power is virtually divided into computing power blocks of different sizes in different computing power resource pools; S6: extracting the task prediction results of each smart terminal, calling the mapping table, matching the task of each smart terminal with a corresponding resource pool based on the mapping table, and pre-allocating the specified computing power blocks in the corresponding resource pool to the tasks of each smart terminal; S7: Collect the real-time task data of each smart terminal, and compare the real-time task data with the corresponding task prediction results, extract the real-time task data of the smart terminal whose difference between the task prediction result and the real-time task data is not within a preset error range, and redistribute based on the extracted real-time task data of the smart terminal.

2. The cloud computing power dynamic allocation method based on cloud servers according to claim 1 is characterized in that: The specific process of step S1 is as follows: S11: Divide the task data of each intelligent terminal into different task types according to the processing time complexity; S12: Divide the computing power into computing power types that match different task types; S13: Match the corresponding computing power for the specified task volume of each task type; S14: construct a general mapping table of task types and computing power types, and establish a sub-mapping table of the specified task amount and the corresponding computing power of each task type under the general mapping table.

3. The cloud computing power dynamic allocation method based on cloud servers according to claim 2 is characterized in that: In step S11, different task types are divided according to the processing time complexity into constant-order tasks, logarithmic-order tasks and exponential-order tasks. In step S12, the computing power types matching the different task types include constant-order computing power, logarithmic-order computing power and exponential-order computing power.

4. The cloud computing power dynamic allocation method based on cloud servers according to claim 1 is characterized in that: The task prediction model in step S2 and the computing power prediction model in step S3 are both long short-term memory network models. The task prediction model and the computing power prediction model both include an input gate, a forget gate, a memory cell and an output gate; The input gate is used to input a historical task data set, and extract the specified task data from the input historical task data to be updated into the memory cell state; The forget gate is used to determine whether the specified task data in the input historical task data is forgotten; The memory cells are used to record time series information of task data; The output gate is used to control the output of the specified task data in the memory cell state to the hidden state.

5. The cloud computing power dynamic allocation method based on cloud servers according to claim 4 is characterized in that: The input gate uses the nonlinear action function of the first neuron to determine the task information updated to the memory cell state. The nonlinear action function of the first neuron generates a value from 0 to 1, indicating the proportion of the task data updated to the cell state. The specific formula is: I t = σ 1 ( W i *[ h t−1 , x t ]+ b i )； in, I t represents the input gate, σ 1 is the nonlinear action function of the first neuron, W i is the weight matrix of the input gate, h t−1 is the hidden state at the previous time step, x t is the input of the current time step, b i is the bias term of the input gate; The forget gate uses the nonlinear action function of the second neuron to determine the forgotten task information. The nonlinear action function of the second neuron generates a value from 0 to 1, indicating the proportion of forgotten task data. The specific formula is: F t = σ 2 ( ( W f *[ h t−1 , x t ]+ b f )； in, F t represents the forget gate, σ 2 is the nonlinear action function of the second neuron, W f is the weight matrix of the forget gate, h t−1 is the hidden state at the previous time step, x t is the input of the current time step, b f is the bias term of the forget gate; The output gate uses the nonlinear action function of the third neuron to determine the output task information. The nonlinear action function of the third neuron generates a value from 0 to 1, indicating the proportion of the task data to be output. The specific formula is: O t = σ 3 ( ( W o *[ h t−1 , x t ]+ b o )； in, O t represents the forget gate, σ 3 is the nonlinear action function of the third neuron, W o is the weight matrix of the forget gate, h t−1 is the hidden state at the previous time step, x t is the input of the current time step, b o is the bias term of the output gate.

6. The cloud computing power dynamic allocation method based on cloud servers according to claim 1 is characterized in that: The specific process in step S5 is as follows: S51: Identify the predicted computing power and add a computing power type identifier, and allocate it to different computing power resource pools according to the computing power type identifier based on containerization technology. The computing power resource pools match the computing power types, including a constant-order computing power resource pool, a logarithmic-order computing power resource pool, and an exponential-order computing power resource pool. S52: Virtually divide the computing power in the constant-order computing power resource pool, the logarithmic-order computing power resource pool, and the exponential-order computing power resource pool into computing power blocks of different sizes.

7. The cloud computing power dynamic allocation method based on cloud servers according to claim 2 is characterized in that: The specific process in step S6 is as follows: S61: extracting task prediction results of each intelligent terminal, and identifying the type and size of the predicted task data; S62: calling the general mapping table, and extracting a matching computing resource pool type according to the general mapping table based on the task type of each smart terminal; S63: calling the sub-mapping table, and allocating computing power blocks of the corresponding computing power resource pool according to the sub-mapping table according to the task data size of each smart terminal; When the corresponding computing power resource pool in step S63 cannot match a computing power block of the corresponding size, the computing power blocks in the corresponding computing power resource pool are combined according to the size of the task data, and then the combined computing power blocks are allocated.

8. The cloud computing power dynamic allocation method based on cloud servers according to claim 1 is characterized in that: The process of reallocating the real-time task data of the smart terminal whose difference between the task prediction result and the real-time task data is not within the preset error range based on the extracted real-time task data of the smart terminal in step S7 is a process of steps S62-S63.

9. A cloud computing power dynamic allocation system based on a cloud server, used to implement the cloud computing power dynamic allocation method based on a cloud server according to any one of claims 1 to 8, characterized in that: It includes a mapping table generation module, a model creation module, a task prediction model, a computing power prediction model, a computing power resource pool, a computing power virtual segmentation module, a data acquisition module, a data comparison module, and a computing power allocation module; the model creation module is connected to the task prediction model and the computing power prediction model, the computing power resource pool is connected to the computing power virtual segmentation module, the computing power prediction model is connected to the computing power resource pool, the task prediction model is connected to the data acquisition module, and the computing power resource pool is connected to the computing power allocation module; The mapping table generation module is used to generate a mapping table between task data and computing power, wherein the mapping table establishes a matching relationship between the type of task data and the type of computing power and a matching relationship between the amount of task data and the size of computing power; The model creation module is used to create a task prediction model and a computing power prediction model; The task prediction model is used to predict the task data of each terminal device at a specified time and output the task prediction result of each intelligent terminal; The computing power prediction model is used to predict the computing power data of the cloud server through the computing power prediction model and output the computing power prediction result; The computing power resource pool is used to accommodate different types of computing power; The computing power virtual division module is used to virtually divide the computing power of the computing power resource pool; The data acquisition module is used to collect real-time task data of each intelligent terminal; The data comparison module is used to compare the real-time task data with the corresponding task prediction results, and extract the real-time task data of the intelligent terminal where the difference between the task prediction result and the real-time task data is not within a preset error range; The computing power allocation module is used to call the mapping table and match the corresponding resource pool for the task of each smart terminal based on the mapping table, pre-allocate the specified computing power blocks in the corresponding resource pool to the tasks of each smart terminal, and redistribute them based on the extracted real-time task data of the smart terminal.

Images