Variable computing power edge intelligent gateway and dynamic resource adjustment method thereof

Abstract

The application discloses a variable computing power edge intelligent gateway and a dynamic resource adjustment method thereof, wherein the intelligent gateway comprises a computing power calling and distributing module, a multi-protocol communication module and a computing power strategy generating module; the computing power calling and distributing module is used for calling the computing power resources of CPUs and GPUs according to a computing power calling strategy, and then distributing the called computing power resources to the running processing of to-be-processed data; the multi-protocol communication module is used for accessing Internet of Things devices to obtain various to-be-processed data; the computing power strategy generating module is used for constructing a resource state matrix according to resource state information, constructing a demand state matrix according to various to-be-processed data, and then generating a real-time computing power calling strategy, which is used for setting the computing power resources called by CPUs and GPUs respectively. The gateway and the adjustment method thereof can make each to-be-processed data quickly obtain sufficient and appropriate computing power resources for operation processing, and ensure the system data flow speed and response capability.

Description

Technical Field

[0001] This invention relates to an intelligent gateway and its resource adjustment method, and more particularly to a variable computing power edge intelligent gateway and its dynamic resource adjustment method. Background Technology

[0002] Edge gateways are devices that connect terminal devices and cloud servers, playing a crucial role in IoT systems. They act as an intermediary layer between terminal devices and cloud servers, responsible for collecting data from terminal devices, processing and storing it, and sending the processed and packaged data to the cloud server. By connecting to IoT devices, edge gateways can achieve real-time control and monitoring of devices, and analyze and make decisions based on the data. Simultaneously, edge gateways can coordinate communication between terminal devices, thereby enabling effective management and control of the entire IoT environment.

[0003] However, existing edge gateways still have some technical pain points, namely fixed computing power. Traditional edge gateways use fixed computing power configurations, which cannot be dynamically adjusted according to business load, resulting in resource waste or performance bottlenecks, affecting data flow speed and system responsiveness. Summary of the Invention

[0004] The purpose of this invention is to provide a variable computing power edge smart gateway, including a computing power allocation module, a multi-protocol communication module, and a computing power strategy generation module;

[0005] The computing power allocation module is used to provide computing power resources combining CPU and GPU, and to call up the computing power resources of CPU and GPU according to the computing power call strategy of the computing power strategy generation module, and then allocate the called computing power resources for the processing of data to be processed by various IoT devices.

[0006] The multi-protocol communication module provides conversion interfaces between industrial protocols and IoT protocols, and connects to IoT devices through the conversion interfaces to obtain the data to be processed from each IoT device;

[0007] The computing power strategy generation module is used to collect the resource status information of the gateway, construct a resource status matrix based on the resource status information, construct a demand status matrix based on each piece of data to be processed, and then generate a real-time computing power call strategy based on the resource status matrix and the demand status matrix. The computing power call strategy is used to set the computing power resources to be called by the CPU and GPU respectively.

[0008] Furthermore, the computing power allocation module includes a computing power allocation unit and a computing power allocation unit;

[0009] The computing power allocation unit is used to obtain the computing power allocation strategy from the computing power strategy generation module, and set the computing power allocation values ​​for CPU and GPU respectively according to the computing power allocation strategy;

[0010] The computing power allocation unit is used to generate various data processing processes for each data to be processed, and then set the processing priority of each data processing process according to the processing priority of each data to be processed. Then, each data processing process loads its corresponding data to be processed to form a process to be run. Finally, the computing power call values ​​of the CPU and GPU are allocated to each process to be run according to the processing priority.

[0011] Furthermore, the multi-protocol communication module includes a data conversion unit and a data receiving unit;

[0012] The data conversion unit is used to identify the device model of each connected IoT device, and provide the corresponding industrial protocol and IoT protocol conversion interface according to the identified device model. Then, it establishes a data transmission channel with the corresponding IoT device through the industrial protocol or IoT protocol to realize the access of IoT devices.

[0013] The data receiving unit is used to obtain the collected data from various IoT devices through the conversion interface, preprocess the collected data from various IoT devices to obtain data to be processed in a unified data packet format, set the processing priority of each data to be processed, and store each data to be processed in a hierarchical manner.

[0014] Furthermore, the specific steps for the data receiving unit to preprocess the collected data from each IoT device are as follows:

[0015] First, obtain the key data from each collected data set. The key data includes data information and valid data. The data information includes the device model, data type, data size, and collection time.

[0016] Then, based on the device model, the corresponding anomaly verification template is obtained from the anomaly data verification database. The anomaly verification template is used to evaluate the data anomalies of each collected data. If the collected data matches the anomaly verification template, the matching collected data is deleted; otherwise, the corresponding collected data is retained.

[0017] Then, the packet size of each data to be processed is preset according to the data type, and then a portion of the data is copied from the valid data of the corresponding data type and packaged into each prediction data packet according to the preset packet size;

[0018] Then, the CPU's prediction unit computing power is set and a prediction running process is generated. The prediction running process is then used to run and process the prediction data packets of each data type with the prediction unit computing power to obtain the processing time of the prediction data packets of each data type.

[0019] Then, an intermediate value is selected from the processing times of each data type as the determination time. The size of the corresponding predicted data packet is then adjusted according to the difference between the determination time and the other processing times, so that the error ratio between the final processing time of each data type and the determination time is less than 5%.

[0020] Then, the packet size of the predicted data packets of each data type is recorded, and the packet size of each data to be processed is set according to the recorded packet size of each data type, so that the data to be processed of each data type has similar processing time under the same computing power conditions.

[0021] Finally, based on the packet size set for each data type, the valid data of the corresponding data type is divided into packets. Each data packet obtained from the packet division is then packaged separately with the corresponding data information to obtain the data to be processed for each data type. The data to be processed is then sent into the processing queue in a unified data packet format according to the order of collection time.

[0022] Furthermore, the specific steps for the data receiving unit to set the processing priority of each piece of data to be processed and to store each piece of data to be processed in different levels are as follows:

[0023] First, a corresponding cache queue is established according to the preset processing priority. The processing priority includes first-level priority, second-level priority and third-level priority, and the processing priority of first-level priority, second-level priority and third-level priority decreases in sequence. The established cache queue includes first-level cache queue, second-level cache queue and third-level cache queue.

[0024] Then, each piece of data to be processed is taken out sequentially from the tail of the queue. The data type of each piece of data to be processed is obtained based on the data information in the data to be processed. The emergency assessment of the effective data in the data to be processed is then performed based on the data type to obtain the emergency assessment result of each piece of data to be processed.

[0025] Then, based on the emergency assessment results, set the processing priority for each piece of data to be processed, setting Level 1 emergency as Level 1 priority, Level 2 emergency as Level 2 priority, and Level 3 emergency as Level 3 priority.

[0026] Finally, each piece of data to be processed is stored in a first-level cache queue, a second-level cache queue, and a third-level cache queue according to its corresponding processing priority, thus realizing hierarchical storage of each piece of data to be processed.

[0027] Furthermore, the specific steps for emergency assessment of valid data within the data to be processed, based on data type, are as follows:

[0028] First, obtain the level classification rules corresponding to each data type. The level classification rules set the feature data classification intervals and feature data extraction rules in the effective data of each data type. Each feature data classification interval is set with three classification sub-intervals, which correspond to the first-level emergency, the second-level emergency, and the third-level emergency, respectively.

[0029] Then, feature data is extracted from the effective data of each data type according to the feature data extraction rules. Feature data is the feature value that reflects the urgency of the target. The feature data extraction rules set the extraction method for feature data of each data type, so as to quickly find the corresponding feature data in the effective data.

[0030] Finally, the obtained feature data is matched with the three hierarchical sub-intervals of the corresponding data type. The emergency situation corresponding to the feature data is determined based on the hierarchical sub-interval in which the feature data falls. Then, the emergency situation of the corresponding valid data is determined as a first-level emergency situation, a second-level emergency situation, or a third-level emergency situation based on the emergency situation of the feature data.

[0031] Furthermore, the computing power strategy generation module includes a state acquisition unit, a state matrix construction unit, a demand matrix construction unit, a computing power prediction unit, and a computing power allocation unit;

[0032] The status acquisition unit is used to set the acquisition time interval and the number of acquisition points for resource status information, and then collects and stores the gateway's resource status information according to the acquisition time interval and the number of acquisition points.

[0033] The state matrix construction unit is used to construct a resource state matrix based on resource state information;

[0034] The demand matrix construction unit is used to construct a demand state matrix based on each piece of data to be processed;

[0035] The computing power prediction unit is used to calculate the computing power resource demand at the current moment based on the resource status matrix and the demand status matrix.

[0036] The computing power allocation unit is used to generate real-time computing power allocation strategies based on computing power resource requirements.

[0037] Furthermore, the specific steps for the status acquisition unit to collect and store the gateway's resource status information based on the acquisition time interval and the number of acquisition points are as follows: First, collect the gateway's resource status information according to the acquisition time interval and store the collected resource status information in the corresponding status information queues, which include a memory utilization queue and a bandwidth utilization queue; then, set the amount of collected data stored in each status information queue according to the number of acquisition points, and when the amount of collected data reaches the number of acquisition points, after collecting each piece of resource status information, first delete one piece of resource status information from the tail of the corresponding status information queue, and then store the latest collected resource status information into the head of the corresponding status information queue, thus realizing the dynamic updating of each status information queue; finally, output the dynamically updated memory utilization queue and bandwidth utilization queue.

[0038] The specific steps for the state matrix construction unit to construct the resource state matrix based on resource state information are as follows: First, a temporary state matrix is ​​constructed based on the number of queues in the state information queue and the number of acquisition points. The number of rows in the temporary state matrix is ​​equal to the number of queues in the state information queue, and the number of columns in the temporary state matrix is ​​equal to the number of acquisition points. Then, the resource state information in the dynamically updated memory utilization queue and bandwidth utilization queue is acquired, and after each dynamic update, the resource state information in the memory utilization queue and bandwidth utilization queue is copied to the temporary state matrix in an overwrite manner, so that each state information queue occupies a row of the temporary state matrix, and each resource state information in the state information queue occupies a matrix element in its row. Finally, the temporary state matrix is ​​used as the resource state matrix, and the resource state matrix is ​​dynamically updated after each state information queue is dynamically updated.

[0039] Furthermore, the specific steps for the demand matrix construction unit to construct the demand state matrix based on each piece of data to be processed are as follows: First, a temporary demand matrix is ​​constructed based on the number of IoT devices and the number of cache queues. The number of rows in the temporary demand matrix equals the number of cache queues, and the number of columns in the temporary demand matrix equals the number of IoT devices. Then, the computing power requirements for the data to be processed at the first-level priority, second-level priority, and third-level priority are set to K1, K2, and K3, respectively, with K1 > K2 > K3. Next, the number of data items loaded at one time in the first-level cache queue, second-level cache queue, and third-level cache queue after each allocation of computing power resources is set to N1, N2, and N3, respectively, with N1 > N2 > N3. Then, K1, K2 and K3 are used as the three rows of matrix elements in the temporary demand matrix. The number of cached data to be processed in the first-level cache queue, the second-level cache queue, and the third-level cache queue is obtained in real time. Then, it is determined whether the number of cached data in the cache queue is less than the number of columns in the temporary demand matrix. If the number of cached data in the current cache queue is less than the number of columns in the matrix, the difference between the number of columns in the matrix and the number of cached data in the current cache queue is calculated as M, and the M matrix elements at the end of the matrix row corresponding to the current cache queue in the temporary demand matrix are set to 0. Finally, the temporary demand matrix is ​​used as the demand status matrix, and the corresponding matrix elements in the demand status matrix are updated after the data to be processed in the first-level cache queue, the second-level cache queue, and / or the third-level cache queue is updated.

[0040] The specific steps by which the computing power prediction unit calculates the current computing power resource demand based on the resource state matrix and the demand state matrix are as follows: First, obtain the gateway's processing hardware information, including CPU hardware configuration information and GPU hardware configuration information. Then, calculate the theoretical computing power value of the CPU and the theoretical computing power value of the GPU based on the CPU hardware configuration information and the GPU hardware configuration information, respectively. Then, calculate the remaining computing power value of the CPU and the remaining computing power value of the GPU based on the current CPU utilization and GPU utilization, respectively, C1 and G1. Next, average the elements of each row of the resource state matrix to obtain the average memory usage and average bandwidth usage, respectively, P and Q. Then, calculate the computing power impact factor as U, U=w1×(1-P)+w2×(1-Q), where w1 w1 is the preset memory impact weight, w2 is the preset bandwidth impact weight, and w1+w2=1. Then, the computing power impact factor is multiplied by the CPU remaining computing power value and the GPU remaining computing power value respectively to obtain the CPU available computing power value and GPU available computing power value, which are C0 and G0 respectively, C0=U×C1, G0=U×G1. Then, the first N1 matrix elements of the first row of the demand state matrix are added to obtain the first-level demand computing power value R1. Then, the first N2 matrix elements of the second row of the demand state matrix are added to obtain the second-level demand computing power value R2. Then, the first N3 matrix elements of the third row of the demand state matrix are added to obtain the third-level demand computing power value R3. The first-level demand computing power value, the second-level demand computing power value, and the third-level demand computing power value are added to obtain the computing power resource demand R, R=R1+R2+R3.

[0041] The specific steps for the computing power allocation unit to generate a real-time computing power allocation strategy based on computing power resource requirements are as follows: First, calculate the computing power trigger threshold L for triggering GPU acceleration, where L = C0 × (1-E), and E is the proportion of reserved computing power resources; then, establish the computing power allocation strategy as follows: when R < L / 2, first calculate the ratio S1 of L / 2 to R, then multiply the current computing power demand values ​​K1, K2, and K3 by S1 and assign them to K1, K2, and K3, and then return to the demand matrix construction unit to construct the demand state matrix based on each piece of data to be processed; when R ≥ C0 + G0, first calculate the ratio S1 of R to (C0 + G0)... The ratio of (0+G0)×(1-E) is S2. Then, the current computing power demand values ​​K1, K2 and K3 are all divided by S2 and then assigned to K1, K2 and K3 respectively. Then, the step of constructing the demand state matrix based on each data to be processed is returned to the demand matrix construction unit. When L / 2≤R<L, the CPU provides computing power resources alone and sets the CPU's computing power call value to be equal to R. When L≤R<C0+G0, the CPU and GPU provide computing power resources together and set the CPU's computing power call value to be Y1 and the GPU's computing power call value to be Y2, and Y1<L, Y1+Y2=R.

[0042] This invention also provides a dynamic resource adjustment method for a variable computing power edge smart gateway, comprising the following steps:

[0043] Step 1: The data conversion unit of the multi-protocol communication module identifies the device model of each connected IoT device, and provides the corresponding industrial protocol and IoT protocol conversion interface according to the identified device model. Then, a data transmission channel is established with the corresponding IoT device through the industrial protocol or IoT protocol to realize the access of IoT devices.

[0044] Step 2: The data receiving unit of the multi-protocol communication module obtains the collected data from each IoT device through the conversion interface, preprocesses the collected data from each IoT device to obtain data to be processed in a unified data packet format, sets the processing priority of each data to be processed, and stores each data to be processed in different levels.

[0045] Step 3: The status acquisition unit of the computing power strategy generation module sets the acquisition time interval and the number of acquisition points for resource status information, and then collects and stores the resource status information of the gateway according to the acquisition time interval and the number of acquisition points.

[0046] Step 4: The state matrix construction unit of the computing power strategy generation module constructs a resource state matrix based on the resource state information, and then the demand matrix construction unit constructs a demand state matrix based on each piece of data to be processed.

[0047] Step 5: The computing power prediction unit of the computing power strategy generation module calculates the current computing power resource demand based on the resource status matrix and the demand status matrix, and then the computing power allocation unit generates a real-time computing power call strategy based on the computing power resource demand.

[0048] Step 6: The computing power allocation module obtains the computing power strategy from the computing power strategy generation module, and sets the computing power allocation values ​​for CPU and GPU respectively according to the computing power allocation strategy.

[0049] Step 7: The computing power allocation unit of the computing power allocation module generates each data processing process for each data to be processed. Then, according to the processing priority of each data to be processed, the processing priority of each data processing process is set. Then, each data processing process loads its corresponding data to be processed to form a process to be run. Then, the computing power call value of CPU and GPU is allocated to each process to be run according to the processing priority, so that the process to be run with higher processing priority gets more computing power value.

[0050] Compared with existing technologies, the advantages of this invention are as follows: The multi-protocol communication module enables IoT devices to access both industrial and IoT protocols, thereby stably obtaining the corresponding data to be processed; the computing power strategy generation module generates real-time computing power allocation strategies based on resource and demand state matrices, enabling dynamic and variable settings of CPU and GPU computing power resources, ensuring that each piece of data to be processed can quickly receive sufficient and appropriate computing power resources for computation, guaranteeing system data flow speed and responsiveness; the computing power allocation module allocates CPU and GPU computing power resources according to the computing power allocation strategy generated by the computing power strategy generation module, accurately allocating the allocated computing power resources to the data to be processed of each IoT device for processing, achieving reliable and efficient allocation of computing power resources. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the variable computing power system architecture of the present invention;

[0052] Figure 2 This is a flowchart illustrating the execution of the variable computing power method of the present invention. Detailed Implementation

[0053] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the embodiments described.

[0054] like Figure 1 As shown, the variable computing power edge smart gateway provided by the present invention includes: a computing power allocation module, a multi-protocol communication module, and a computing power strategy generation module;

[0055] The computing power allocation module is used to provide computing power resources combining CPU and GPU, and to call up the computing power resources of CPU and GPU according to the computing power call strategy of the computing power strategy generation module, and then allocate the called computing power resources for the processing of data to be processed by various IoT devices.

[0056] The multi-protocol communication module provides conversion interfaces between industrial protocols and IoT protocols, and connects to IoT devices through the conversion interfaces to obtain the data to be processed from each IoT device;

[0057] The computing power strategy generation module is used to collect the resource status information of the gateway, including memory occupancy and memory bandwidth occupancy. Based on the resource status information, a resource status matrix is ​​constructed, and based on each piece of data to be processed, a demand status matrix is ​​constructed. Then, based on the resource status matrix and the demand status matrix, a real-time computing power call strategy is generated. The computing power call strategy is used to set the computing power resources to be called by the CPU and GPU respectively.

[0058] The multi-protocol communication module enables IoT devices to access industrial and IoT protocols, thereby stably obtaining the corresponding data to be processed. The computing power strategy generation module generates real-time computing power allocation strategies based on resource and demand state matrices, enabling dynamic and variable settings of CPU and GPU computing power resources. This ensures that each piece of data to be processed can quickly receive sufficient and appropriate computing power resources for computation, guaranteeing the system's data flow speed and responsiveness. The computing power allocation module allocates CPU and GPU computing power resources according to the computing power allocation strategies generated by the computing power strategy generation module, thereby accurately allocating the allocated computing power resources to the data to be processed by each IoT device for processing, achieving reliable and efficient allocation of computing power resources.

[0059] Furthermore, it also includes a dynamic power management module. This module acquires the current computing power demand and dynamically adjusts the operating frequency and voltage of the CPU and GPU based on this demand using voltage and frequency regulation technology. By utilizing the dynamic power management module, the operating frequency and voltage of the CPU and GPU can be dynamically adjusted, thereby ensuring stable computing power output from the CPU and GPU.

[0060] Furthermore, the dynamic power management module includes a demand acquisition module and a power distribution unit;

[0061] The demand acquisition module is used to obtain the current computing power call values ​​of the CPU and GPU based on the current computing power call strategy.

[0062] The power distribution unit is used to dynamically adjust the operating frequency and operating voltage of the CPU and GPU based on their respective computing power allocation values ​​using voltage and frequency regulation technology.

[0063] Furthermore, the computing power allocation module includes a computing power allocation unit and a computing power allocation unit;

[0064] The computing power allocation unit is used to obtain the computing power allocation strategy from the computing power strategy generation module, and set the computing power allocation values ​​for CPU and GPU respectively according to the computing power allocation strategy;

[0065] The computing power allocation unit is used to generate various data processing processes for each data to be processed, and then set the processing priority of each data processing process according to the processing priority of each data to be processed. Then, each data processing process loads its corresponding data to be processed to form a process to be run. Finally, the computing power call values ​​of CPU and GPU are allocated to each process to be run according to the processing priority, so that the process to be run with a higher processing priority can obtain more computing power.

[0066] The computing power allocation unit can allocate the computing power call values ​​of CPU and GPU to each running process according to the processing priority, so that the running process with higher processing priority can obtain more computing power value, thereby realizing the on-demand dynamic and variable allocation of the computing power call values ​​of CPU and GPU.

[0067] Furthermore, the specific steps for allocating CPU and GPU computing power allocation values ​​to each process to be run according to processing priority are as follows: first, determine the priority of the process to be run, and then set the computing power allocated to the first-priority process to be run to K1, the computing power allocated to the second-priority process to K2, and the computing power allocated to the third-priority process to K3.

[0068] Furthermore, the multi-protocol communication module includes a data conversion unit and a data receiving unit;

[0069] The data conversion unit is used to identify the device models of each connected IoT device, and provide corresponding industrial protocol and IoT protocol conversion interfaces based on the identified device models. Then, it establishes a data transmission channel with the corresponding IoT device through the industrial protocol or IoT protocol to realize the access of multiple IoT devices and thus successfully obtain the collected data from each IoT device.

[0070] The data receiving unit is used to obtain the collected data from various IoT devices through the conversion interface, preprocess the collected data from various IoT devices to obtain data to be processed in a unified data packet format, set the processing priority of each data to be processed, and store each data to be processed in different levels, namely, storing the data to be processed with first-level priority, second-level priority, and third-level priority separately.

[0071] Furthermore, the specific steps for the data receiving unit to preprocess the collected data from each IoT device are as follows:

[0072] First, obtain the key data from each collected data set. The key data includes data information and valid data. The data information includes the device model, data type, data size, and collection time.

[0073] Then, based on the device model, the corresponding anomaly verification template is obtained from the anomaly data verification library. The anomaly verification template is used to evaluate the data anomalies of each collected data. If the collected data matches the anomaly verification template, the matching collected data is deleted; otherwise, the corresponding collected data is retained. By comparing the anomaly verification template, the identification of various types of data anomalies in the collected data can be achieved quickly.

[0074] Then, the packet size of each data to be processed is preset according to the data type, and then a portion of the data is copied from the valid data of the corresponding data type and packaged into each prediction data packet according to the preset packet size;

[0075] Then, the CPU's prediction unit computing power is set and a prediction running process is generated. The prediction running process is then used to run and process the prediction data packets of each data type using a preset processing algorithm with the prediction unit computing power to obtain the processing time of the prediction data packets of each data type. The prediction data packets and the prediction running process can be used to predict the processing time of each data type, which makes it easier to adjust the packet size of the data to be processed of each data type in the future.

[0076] Then, an intermediate value is selected from the processing times of each data type as the determination time. The size of the corresponding predicted data packet is then adjusted according to the difference between the determination time and the other processing times, so that the error ratio between the final processing time of each data type and the determination time is less than 5%. By adjusting the processing time of each data type to be processed to the set error ratio range, it is easier to ensure the accuracy of the allocation of computing resources for each data to be processed in the subsequent processing, so that the processing time ratio between data of different priority levels has a relatively stable ratio during actual processing.

[0077] Then, the packet size of the predicted data packets of each data type is recorded, and the packet size of each data to be processed is set according to the recorded packet size of each data type, so that the data to be processed of each data type has similar processing time under the same computing power conditions.

[0078] Finally, based on the packet size set for each data type, the valid data of the corresponding data type is divided into packets. Each data packet obtained from the packet division is then packaged separately with the corresponding data information to obtain the data to be processed for each data type. The data to be processed is then sent to the processing queue in a unified data packet format according to the order of collection time. The processing queue is used to store the data to be processed, which facilitates the subsequent extraction of the data to be processed for emergency assessment.

[0079] Furthermore, the specific steps for the data receiving unit to set the processing priority of each piece of data to be processed and to store each piece of data to be processed in different levels are as follows:

[0080] First, a corresponding cache queue is established according to the preset processing priority. The processing priority includes first-level priority, second-level priority and third-level priority, and the processing priority of first-level priority, second-level priority and third-level priority decreases in sequence. The established cache queue includes first-level cache queue, second-level cache queue and third-level cache queue.

[0081] Then, each piece of data to be processed is taken out sequentially from the tail of the queue. The data type of each piece of data to be processed is obtained based on the data information in the data to be processed. The emergency assessment of the effective data in the data to be processed is then performed based on the data type to obtain the emergency assessment result of each piece of data to be processed.

[0082] Then, based on the emergency assessment results, set the processing priority for each piece of data to be processed, setting Level 1 emergency as Level 1 priority, Level 2 emergency as Level 2 priority, and Level 3 emergency as Level 3 priority.

[0083] Finally, each piece of data to be processed is stored in a first-level cache queue, a second-level cache queue, and a third-level cache queue according to its corresponding processing priority, thus realizing hierarchical storage of each piece of data to be processed.

[0084] Furthermore, the specific steps for emergency assessment of valid data within the data to be processed, based on data type, are as follows:

[0085] First, obtain the level classification rules corresponding to each data type. The level classification rules set the feature data classification intervals and feature data extraction rules in the effective data of each data type. Each feature data classification interval is set with three sub-intervals. The ranges of the three sub-intervals are sequentially continuous. For example, the ranges of the three sub-intervals are set as follows: T < 60℃ for Level 3 emergency, 60℃ ≤ T < 140℃ for Level 2 emergency, and T ≥ 140℃ for Level 1 emergency. The three sub-intervals correspond to Level 1 emergency, Level 2 emergency, and Level 3 emergency, respectively.

[0086] Then, feature data is extracted from the effective data of each data type according to the feature data extraction rules. The feature data is the feature value reflecting the emergency situation of the target (such as the local temperature extreme value). The feature data extraction rules set the extraction method of feature data for each data type (for example, judging whether there is a fire from the infrared image frame, determining whether there is a fire by judging the pixel value of each pixel in the infrared image frame, and determining the processing priority of the infrared image frame by the size of such feature value, setting the more urgent infrared image frame as a higher emergency situation, so as to quickly perform image recognition and find the location of the burning equipment as soon as possible). Thus, the corresponding feature data can be quickly found in the effective data. The feature data extraction methods used in this invention are all conventional extraction methods in the prior art, and are set according to the collection object during implementation.

[0087] Finally, the obtained feature data is matched with the three hierarchical sub-intervals of the corresponding data type. The emergency situation corresponding to the feature data is determined based on the hierarchical sub-interval in which the feature data falls. Then, the emergency situation of the corresponding valid data is determined to be either Level 1 emergency, Level 2 emergency, or Level 3 emergency based on the emergency situation of the feature data.

[0088] Furthermore, the computing power strategy generation module includes a state acquisition unit, a state matrix construction unit, a demand matrix construction unit, a computing power prediction unit, and a computing power allocation unit;

[0089] The status acquisition unit is used to set the acquisition time interval and the number of acquisition points for resource status information, and then collects and stores the gateway's resource status information according to the acquisition time interval and the number of acquisition points.

[0090] The state matrix construction unit is used to construct a resource state matrix based on resource state information, thereby quickly integrating the recorded information of resource state information for easy subsequent analysis.

[0091] The demand matrix construction unit is used to construct a demand state matrix based on each piece of data to be processed, thereby integrating the state information of the data to be processed, which facilitates the subsequent computational power calculation and prediction.

[0092] The computing power prediction unit is used to calculate the computing power resource demand at the current moment based on the resource status matrix and the demand status matrix. The calculated computing power resource demand enables computing power matching according to the running processing requirements of the data to be processed, so that each data to be processed can be processed in the appropriate time.

[0093] The computing power allocation unit is used to generate real-time computing power call strategies based on computing power resource requirements. The generated computing power call strategies can clearly explain the computing power allocation of CPU and GPU, thereby facilitating the computing power call and allocation module to quickly execute the call and allocation of computing power according to the strategies.

[0094] Furthermore, the specific steps for the status acquisition unit to collect and store the gateway's resource status information based on the acquisition time interval and the number of acquisition points are as follows: First, collect the gateway's resource status information according to the acquisition time interval and store the collected resource status information in the corresponding status information queue. The status information queue includes a memory utilization rate queue and a bandwidth utilization rate queue. That is, the collected memory utilization rate is stored in the memory utilization rate queue in the order of acquisition time, and the collected memory bandwidth utilization rate is stored in the bandwidth utilization rate queue in the order of acquisition time. Then, the amount of collected data stored in each status information queue is set according to the number of acquisition points. When the amount of collected data reaches the number of acquisition points, after collecting each piece of resource status information, a piece of resource status information is first deleted from the tail of the corresponding status information queue, and then the latest collected resource status information is stored in the head of the corresponding status information queue, thereby realizing the dynamic update of each status information queue. Finally, the dynamically updated memory utilization rate queue and bandwidth utilization rate queue are output.

[0095] The specific steps of the state matrix construction unit in constructing the resource state matrix based on resource state information are as follows: First, a temporary state matrix is ​​constructed based on the number of queues in the state information queue and the number of acquisition points. The number of rows in the temporary state matrix is ​​equal to the number of queues in the state information queue. In this invention, there are two state information queues: a memory utilization queue and a bandwidth utilization queue, i.e., the number of queues is 2. The number of columns in the temporary state matrix is ​​equal to the number of acquisition points. In this invention, the number of acquisition points is set to 10. Then, the resource state information in the dynamically updated memory utilization queue and bandwidth utilization queue is acquired, and after each dynamic update, the resource state information in the memory utilization queue and bandwidth utilization queue is copied to the temporary state matrix in an overwrite manner, so that each state information queue occupies a row of the temporary state matrix, and each resource state information in the state information queue occupies a matrix element in its row. Finally, the temporary state matrix is ​​used as the resource state matrix, and the resource state matrix is ​​dynamically updated accordingly after each state information queue is dynamically updated.

[0096] Furthermore, the specific steps of the demand matrix construction unit in constructing the demand state matrix based on each piece of data to be processed are as follows: First, a temporary demand matrix is ​​constructed based on the number of IoT devices and the number of cache queues. The number of rows in the temporary demand matrix is ​​equal to the number of cache queues (in this invention, the number of cache queues is 3), and the number of columns in the temporary demand matrix is ​​equal to the number of IoT devices. Then, the computing power requirements for the data to be processed at the first, second, and third priorities are set to K1, K2, and K3, respectively, with K1 > K2 > K3. Next, the number of data items loaded at one time in the first, second, and third cache queues after each allocation of computing power resources is set to N1, N2, and N3, respectively, with N1 > N2 > N3. Then, K1, K2, and K3 are used as the three rows of the temporary demand matrix. The system then retrieves the cached data in the first-level, second-level, and third-level cache queues in real time. It then determines whether the cached data in each queue is less than the number of columns in the temporary demand matrix. If the cached data in the current queue is less than the number of columns, the difference M between the number of columns and the cached data in the current queue is calculated. The last M elements of the row corresponding to the current cached queue in the temporary demand matrix are set to 0. For example, if there are only two data items to be processed in the first-level cache queue, the first row of the demand state matrix will only have two K1 values ​​at the beginning, and the other positions in the first row will be 0. Finally, the temporary demand matrix is ​​used as the demand state matrix, and the corresponding matrix elements in the demand state matrix are updated after the data to be processed in the first-level, second-level, and / or third-level cache queues are updated.

[0097] Furthermore, the specific steps for the computing power prediction unit to calculate the current computing power resource demand based on the resource state matrix and the demand state matrix are as follows: First, obtain the gateway's processing hardware information, including CPU hardware configuration information and GPU hardware configuration information. CPU hardware configuration information includes CPU utilization, number of CPU cores, single-core clock speed, and single-core single-cycle floating-point calculation coefficient. GPU hardware configuration information includes GPU utilization, number of CUDA cores, single-core clock speed, and single-core single-cycle floating-point calculation coefficient. Then, calculate the theoretical computing power value of the CPU and the theoretical computing power value of the GPU based on the CPU and GPU hardware configuration information respectively. Next, calculate the remaining computing power value of the CPU and the remaining computing power value of the GPU based on the current CPU utilization and GPU utilization, respectively, as C1 and G1. When calculating the remaining computing power value of the CPU, first calculate the current computing power value of the CPU using the CPU utilization, and then subtract the current computing power value of the CPU from the theoretical computing power value to obtain the remaining computing power value of the CPU. When calculating the remaining computing power value of the GPU, first calculate the current computing power value of the GPU using the GPU utilization, and then subtract the current computing power value of the CPU from the theoretical computing power value of the GPU. Subtract the current GPU computing power to obtain the remaining GPU computing power. Then, average the elements of each row of the resource status matrix to obtain the average memory usage and average bandwidth usage, P and Q, respectively. Calculate the computing power impact factor U, U = w1 × (1 - P) + w2 × (1 - Q), where w1 is the preset memory impact weight, w2 is the preset bandwidth impact weight, and w1 + w2 = 1. Then, multiply the computing power impact factor by the remaining CPU computing power and GPU computing power, respectively, to obtain the available CPU computing power and available GPU computing power. Set the available CPU computing power... The computing power value and the available GPU computing power value are C0 and G0, respectively, where C0 = U × C1 and G0 = U × G1. Then, the first N1 matrix elements of the first row of the demand state matrix are added together to obtain the first-level demand computing power value R1. Then, the first N2 matrix elements of the second row of the demand state matrix are added together to obtain the second-level demand computing power value R2. Then, the first N3 matrix elements of the third row of the demand state matrix are added together to obtain the third-level demand computing power value R3. The computing power resource requirement is obtained by adding the first-level demand computing power value, the second-level demand computing power value, and the third-level demand computing power value, R = R1 + R2 + R3.

[0098] The specific steps of the computing power allocation unit in generating a real-time computing power allocation strategy based on computing power resource requirements are as follows: First, calculate the computing power trigger threshold L for triggering GPU acceleration, where L = C0 × (1 - E), and E is the proportion of reserved computing power resources. This ensures that the CPU's computing power resources are not fully utilized during operation, preventing system crashes. Then, establish the computing power allocation strategy: when R < L / 2, first calculate the ratio S1 of L / 2 to R, then multiply the current computing power requirement values ​​K1, K2, and K3 by S1 before assigning them to K1, K2, and K3. Finally, return to the requirement matrix construction unit to construct the requirement state matrix based on each piece of data to be processed. This allows for the appropriate increase of the computing power requirement value for each piece of data to be processed when the current computing power resource requirement is low, thereby improving the processing efficiency of each piece of data. When R ≥ C0 + G0, the ratio of R to (C0 + G0) × (1 - E) is first calculated as S2. Then, the current computing power demand values ​​K1, K2, and K3 are all divided by S2 and assigned to K1, K2, and K3 respectively. Then, the process returns to the demand matrix construction unit to construct the demand state matrix based on each piece of data to be processed. This allows for timely reduction of the computing power demand values ​​of each piece of data to be processed when the current computing power resource demand is too high, thus ensuring the normal operation of the system. When L / 2 ≤ R < L, the CPU provides computing power resources alone, and the CPU's computing power call value is set to equal R. When L ≤ R < C0 + G0, the CPU and GPU provide computing power resources together, and the CPU's computing power call value is set to Y1 and the GPU's computing power call value is set to Y2, where Y1 < L and Y1 + Y2 = R.

[0099] like Figure 2 As shown, the dynamic resource adjustment method for a variable computing power edge smart gateway provided by the present invention includes the following steps:

[0100] Step 1: The data conversion unit of the multi-protocol communication module identifies the device model of each connected IoT device, and provides the corresponding industrial protocol and IoT protocol conversion interface according to the identified device model. Then, a data transmission channel is established with the corresponding IoT device through the industrial protocol or IoT protocol to realize the access of IoT devices.

[0101] Step 2: The data receiving unit of the multi-protocol communication module obtains the collected data from each IoT device through the conversion interface, preprocesses the collected data from each IoT device to obtain data to be processed in a unified data packet format, sets the processing priority of each data to be processed, and stores each data to be processed in different levels.

[0102] Step 3: The status acquisition unit of the computing power strategy generation module sets the acquisition time interval and the number of acquisition points for resource status information, and then collects and stores the resource status information of the gateway according to the acquisition time interval and the number of acquisition points.

[0103] Step 4: The state matrix construction unit of the computing power strategy generation module constructs a resource state matrix based on the resource state information, and then the demand matrix construction unit constructs a demand state matrix based on each piece of data to be processed.

[0104] Step 5: The computing power prediction unit of the computing power strategy generation module calculates the current computing power resource demand based on the resource status matrix and the demand status matrix, and then the computing power allocation unit generates a real-time computing power call strategy based on the computing power resource demand.

[0105] Step 6: The computing power allocation module obtains the computing power strategy from the computing power strategy generation module, and sets the computing power allocation values ​​for CPU and GPU respectively according to the computing power allocation strategy.

[0106] Step 7: The computing power allocation unit of the computing power allocation module generates each data processing process for each data to be processed. Then, according to the processing priority of each data to be processed, the processing priority of each data processing process is set. Then, each data processing process loads its corresponding data to be processed to form a process to be run. Then, the computing power call value of CPU and GPU is allocated to each process to be run according to the processing priority, so that the process to be run with higher processing priority gets more computing power value.

[0107] As described above, although the invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. A variable computing power edge intelligent gateway, characterized in that, It includes a computing power allocation module, a multi-protocol communication module, and a computing power strategy generation module; The computing power allocation module is used to provide computing power resources combining CPU and GPU, and to call up the computing power resources of CPU and GPU according to the computing power call strategy of the computing power strategy generation module, and then allocate the called computing power resources for the processing of data to be processed by various IoT devices. The multi-protocol communication module provides conversion interfaces between industrial protocols and IoT protocols, and connects to IoT devices through the conversion interfaces to obtain the data to be processed from each IoT device; The computing power strategy generation module is used to collect the resource status information of the gateway, construct a resource status matrix based on the resource status information, construct a demand status matrix based on each piece of data to be processed, and then generate a real-time computing power call strategy based on the resource status matrix and the demand status matrix. The computing power call strategy is used to set the computing power resources to be called by the CPU and GPU respectively. The multi-protocol communication module includes a data conversion unit and a data receiving unit; The data conversion unit is used to identify the device model of each connected IoT device, and provide the corresponding industrial protocol and IoT protocol conversion interface according to the identified device model. Then, it establishes a data transmission channel with the corresponding IoT device through the industrial protocol or IoT protocol to realize the access of IoT devices. The data receiving unit is used to obtain the collected data from various IoT devices through the conversion interface, preprocess the collected data from various IoT devices to obtain data to be processed in a unified data packet format, set the processing priority of each data to be processed, and store each data to be processed in different levels. The specific steps for the data receiving unit to preprocess the data collected from various IoT devices are as follows: First, obtain the key data from each collected data set. The key data includes data information and valid data. The data information includes the device model, data type, data size, and collection time. Then, based on the device model, the corresponding anomaly verification template is obtained from the anomaly data verification database. The anomaly verification template is used to evaluate the data anomalies of each collected data. If the collected data matches the anomaly verification template, the matching collected data is deleted; otherwise, the corresponding collected data is retained. Then, the packet size of each data to be processed is preset according to the data type, and then a portion of the data is copied from the valid data of the corresponding data type and packaged into each prediction data packet according to the preset packet size; Then, the CPU's prediction unit computing power is set and a prediction running process is generated. The prediction running process is then used to run and process the prediction data packets of each data type with the prediction unit computing power to obtain the processing time of the prediction data packets of each data type. Then, an intermediate value is selected from the processing times of each data type as the determination time. The size of the corresponding predicted data packet is then adjusted according to the difference between the determination time and the other processing times, so that the error ratio between the final processing time of each data type and the determination time is less than 5%. Then, the packet size of the predicted data packets of each data type is recorded, and the packet size of each data to be processed is set according to the recorded packet size of each data type, so that the data to be processed of each data type has similar processing time under the same computing power conditions. Finally, based on the packet size set for each data type, the valid data of the corresponding data type is divided into packets. Each data packet obtained from the packet division is then packaged separately with the corresponding data information to obtain the data to be processed for each data type. The data to be processed is then sent into the processing queue in a unified data packet format according to the order of collection time.

2. The variable computing power edge smart gateway according to claim 1, characterized in that, The computing power allocation module includes a computing power allocation unit and a computing power allocation unit; The computing power allocation unit is used to obtain the computing power allocation strategy from the computing power strategy generation module, and set the computing power allocation values ​​for CPU and GPU respectively according to the computing power allocation strategy; The computing power allocation unit is used to generate various data processing processes for each data to be processed, and then set the processing priority of each data processing process according to the processing priority of each data to be processed. Then, each data processing process loads its corresponding data to be processed to form a process to be run. Finally, the computing power call values ​​of the CPU and GPU are allocated to each process to be run according to the processing priority.

3. The variable computing power edge intelligent gateway according to claim 1, characterized in that, The specific steps for the data receiving unit to set the processing priority of each piece of data to be processed and to store each piece of data in a hierarchical manner are as follows: First, a corresponding cache queue is established according to the preset processing priority. The processing priority includes first-level priority, second-level priority and third-level priority, and the processing priority of first-level priority, second-level priority and third-level priority decreases in sequence. The established cache queue includes first-level cache queue, second-level cache queue and third-level cache queue. Then, each piece of data to be processed is taken out sequentially from the tail of the queue. The data type of each piece of data to be processed is obtained based on the data information in the data to be processed. The emergency assessment of the effective data in the data to be processed is then performed based on the data type to obtain the emergency assessment result of each piece of data to be processed. Then, based on the emergency assessment results, set the processing priority for each piece of data to be processed, setting Level 1 emergency as Level 1 priority, Level 2 emergency as Level 2 priority, and Level 3 emergency as Level 3 priority. Finally, each piece of data to be processed is stored in a first-level cache queue, a second-level cache queue, and a third-level cache queue according to its corresponding processing priority, thus realizing hierarchical storage of each piece of data to be processed.

4. The variable computing power edge smart gateway according to claim 3, characterized in that, The specific steps for emergency assessment of valid data in the data to be processed, based on data type, are as follows: First, obtain the level classification rules corresponding to each data type. The level classification rules set the feature data classification intervals and feature data extraction rules in the effective data of each data type. Each feature data classification interval is set with three classification sub-intervals, which correspond to the first-level emergency, the second-level emergency, and the third-level emergency, respectively. Then, feature data is extracted from the effective data of each data type according to the feature data extraction rules. Feature data is the feature value that reflects the urgency of the target. The feature data extraction rules set the extraction method for feature data of each data type, so as to quickly find the corresponding feature data in the effective data. Finally, the obtained feature data is matched with the three hierarchical sub-intervals of the corresponding data type. The emergency situation corresponding to the feature data is determined based on the hierarchical sub-interval in which the feature data falls. Then, the emergency situation of the corresponding valid data is determined as a first-level emergency situation, a second-level emergency situation, or a third-level emergency situation based on the emergency situation of the feature data.

5. The variable computing power edge intelligent gateway according to claim 3, characterized in that, The computing power strategy generation module includes a state acquisition unit, a state matrix construction unit, a demand matrix construction unit, a computing power prediction unit, and a computing power allocation unit; The status acquisition unit is used to set the acquisition time interval and the number of acquisition points for resource status information, and then collects and stores the gateway's resource status information according to the acquisition time interval and the number of acquisition points. The state matrix construction unit is used to construct a resource state matrix based on resource state information; The demand matrix construction unit is used to construct a demand state matrix based on each piece of data to be processed; The computing power prediction unit is used to calculate the computing power resource demand at the current moment based on the resource status matrix and the demand status matrix. The computing power allocation unit is used to generate real-time computing power allocation strategies based on computing power resource requirements.

6. The variable computing power edge intelligent gateway according to claim 5, characterized in that, The specific steps for the status acquisition unit to collect and store the gateway's resource status information based on the acquisition time interval and the number of acquisition points are as follows: First, collect the gateway's resource status information according to the acquisition time interval and store the collected resource status information in the corresponding status information queues, which include a memory utilization queue and a bandwidth utilization queue; then, set the amount of collected data stored in each status information queue according to the number of acquisition points, and when the amount of collected data reaches the number of acquisition points, after collecting each piece of resource status information, first delete one piece of resource status information from the tail of the corresponding status information queue, and then store the latest collected resource status information into the head of the corresponding status information queue, thus realizing the dynamic updating of each status information queue; finally, output the dynamically updated memory utilization queue and bandwidth utilization queue. The specific steps for the state matrix construction unit to construct the resource state matrix based on resource state information are as follows: First, a temporary state matrix is ​​constructed based on the number of queues in the state information queue and the number of acquisition points. The number of rows in the temporary state matrix is ​​equal to the number of queues in the state information queue, and the number of columns in the temporary state matrix is ​​equal to the number of acquisition points. Then, the resource state information in the dynamically updated memory utilization queue and bandwidth utilization queue is acquired, and after each dynamic update, the resource state information in the memory utilization queue and bandwidth utilization queue is copied to the temporary state matrix in an overwrite manner, so that each state information queue occupies a row of the temporary state matrix, and each resource state information in the state information queue occupies a matrix element in its row. Finally, the temporary state matrix is ​​used as the resource state matrix, and the resource state matrix is ​​dynamically updated after each state information queue is dynamically updated.

7. The variable computing power edge intelligent gateway according to claim 5, characterized in that, The specific steps for the demand matrix construction unit to construct the demand state matrix based on each piece of data to be processed are as follows: First, a temporary demand matrix is ​​constructed based on the number of IoT devices and the number of cache queues. The number of rows in the temporary demand matrix is ​​equal to the number of cache queues, and the number of columns in the temporary demand matrix is ​​equal to the number of IoT devices. Then, the computing power requirements for the data to be processed at priority levels 1, 2, and 3 are set to K1, K2, and K3, respectively, with K1 > K2 > K3. Next, the number of data items loaded at one time in each cache queue (level 1, 2, and 3) after each allocation of computing power resources is set to N1, N2, and N3, respectively, with N1 > N2 > N3. First, K1, K2, and K3 are used as the three rows of elements in a temporary demand matrix. Then, the cached data quantity in the level 1, 2, and 3 cache queues is obtained in real time. Next, it is determined whether the cached data quantity in the cache queues is less than the number of columns in the temporary demand matrix. If the cached data quantity in the current cache queue is less than the number of columns, the difference between the number of columns and the cached data quantity in the current cache queue is calculated as M, and the last M matrix elements at the end of the row corresponding to the current cache queue in the temporary demand matrix are set to 0. Finally, the temporary demand matrix is ​​used as the demand state matrix, and the corresponding matrix elements in the demand state matrix are updated after the data to be processed in the level 1, 2, and / or 3 cache queues are updated. The specific steps by which the computing power prediction unit calculates the current computing power resource demand based on the resource state matrix and the demand state matrix are as follows: First, obtain the gateway's processing hardware information, including CPU hardware configuration information and GPU hardware configuration information. Then, calculate the theoretical computing power value of the CPU and the theoretical computing power value of the GPU based on the CPU hardware configuration information and the GPU hardware configuration information, respectively. Then, calculate the remaining computing power value of the CPU and the remaining computing power value of the GPU based on the current CPU utilization and GPU utilization, respectively, C1 and G1. Next, average the elements of each row of the resource state matrix to obtain the average memory usage and average bandwidth usage, respectively, P and Q. Then, calculate the computing power impact factor as U, U=w1×(1-P)+w2×(1-Q), where w1 w1 is the preset memory impact weight, w2 is the preset bandwidth impact weight, and w1+w2=1. Then, the computing power impact factor is multiplied by the CPU remaining computing power value and the GPU remaining computing power value respectively to obtain the CPU available computing power value and GPU available computing power value, which are C0 and G0 respectively, C0=U×C1, G0=U×G1. Then, the first N1 matrix elements of the first row of the demand state matrix are added to obtain the first-level demand computing power value R1. Then, the first N2 matrix elements of the second row of the demand state matrix are added to obtain the second-level demand computing power value R2. Then, the first N3 matrix elements of the third row of the demand state matrix are added to obtain the third-level demand computing power value R3. The first-level demand computing power value, the second-level demand computing power value, and the third-level demand computing power value are added to obtain the computing power resource demand R, R=R1+R2+R3. The specific steps for the computing power allocation unit to generate a real-time computing power call strategy based on computing power resource requirements are as follows: First, calculate the computing power trigger threshold L that triggers GPU acceleration, where L = C0 × (1 - E), and E is the proportion of reserved computing power resources; then, establish the computing power call strategy as follows: when R < L / 2, first calculate the ratio S1 of L / 2 to R, then multiply the current computing power demand values ​​K1, K2, and K3 by S1 and assign them to K1, K2, and K3, and then return to the demand matrix construction unit to construct the demand state matrix based on each piece of data to be processed. When R≥C0+G0, the ratio of R to (C0+G0)×(1-E) is first calculated as S2. Then, the current computing power demand values ​​K1, K2, and K3 are all divided by S2 and then assigned to K1, K2, and K3 respectively. Then, the process returns to the demand matrix construction unit to construct the demand state matrix based on each piece of data to be processed. When L / 2≤R<L, the CPU provides computing power resources alone, and the CPU's computing power call value is set to R. When L≤R<C0+G0, the CPU and GPU provide computing power resources together, and the CPU's computing power call value is set to Y1 and the GPU's computing power call value is set to Y2, where Y1<L and Y1+Y2=R.

8. A dynamic resource adjustment method for a variable computing power edge intelligent gateway according to claim 5, characterized in that, Includes the following steps: Step 1: The data conversion unit of the multi-protocol communication module identifies the device model of each connected IoT device, and provides the corresponding industrial protocol and IoT protocol conversion interface according to the identified device model. Then, a data transmission channel is established with the corresponding IoT device through the industrial protocol or IoT protocol to realize the access of IoT devices. Step 2: The data receiving unit of the multi-protocol communication module obtains the collected data from each IoT device through the conversion interface, preprocesses the collected data from each IoT device to obtain data to be processed in a unified data packet format, sets the processing priority of each data to be processed, and stores each data to be processed in different levels. Step 3: The status acquisition unit of the computing power strategy generation module sets the acquisition time interval and the number of acquisition points for resource status information, and then collects and stores the resource status information of the gateway according to the acquisition time interval and the number of acquisition points. Step 4: The state matrix construction unit of the computing power strategy generation module constructs a resource state matrix based on the resource state information, and then the demand matrix construction unit constructs a demand state matrix based on each piece of data to be processed. Step 5: The computing power prediction unit of the computing power strategy generation module calculates the computing power resource demand at the current moment based on the resource status matrix and the demand status matrix, and then the computing power allocation unit generates a real-time computing power call strategy based on the computing power resource demand. Step 6: The computing power allocation module obtains the computing power strategy from the computing power strategy generation module, and sets the computing power allocation values ​​for CPU and GPU respectively according to the computing power allocation strategy. Step 7: The computing power allocation unit of the computing power allocation module generates each data processing process for each data to be processed. Then, the processing priority of each data processing process is set according to the processing priority of each data to be processed. Each data processing process loads its corresponding data to be processed to form a process to be run. Then, the computing power call value of CPU and GPU is allocated to each process to be run according to the processing priority, so that the process to be run with higher processing priority gets more computing power value.

Images