Edge computing processing system and task optimization method based on regional integrated energy station

Abstract

The present invention discloses an edge computing processing system and a task optimization method based on a regional integrated energy station. The edge computing processing system includes: a data input module, a data processing module, a storage module, a communication module and a power conversion module; the data input module converts the analog stream output from the regional integrated energy station into data information; the data processing module assigns tasks to the data information in a timing pipeline manner; the storage module stores and calls the task assignments; the communication module transmits the task assignments to a server; the power conversion module is used to supply power to the edge computing processing system; the data processing module assigns tasks to the data information in a timing pipeline manner; thereby, multiple tasks are queued for processing according to computing overhead, cost, required resources and maximum tolerable delay, and are sent one by one to the neural network processing unit for processing, thereby realizing rapid processing of data tasks of the regional integrated energy station.

Description

Technical Field

[0001] The present invention belongs to the technical field of regional integrated energy system data processing, and in particular relates to an edge computing processing system and a task optimization method based on a regional integrated energy station. Background Art

[0002] With the large-scale rollout of new digitally driven power systems and the increasing penetration of renewable energy, regional integrated energy systems, centered around integrated energy stations, are becoming a key approach to localizing and regulating renewable energy, as a grid architecture designed to optimize the coordinated operation of multiple energy sources within the periphery of large power grids. However, a prominent challenge facing regional integrated energy stations is that their operation on the distribution network side generates massive amounts of multi-source, heterogeneous energy data, exponentially increasing the volume of time series data. The current centralized data processing approach based on the distribution network middleware requires that this massive amount of data be directly reported to the middleware over the network for unified classification and processing. This places significant strain on network bandwidth and the distribution network data middleware, and also creates significant time delays, making it unacceptable for time-sensitive energy dispatch. This has become one of the core issues hindering the development of new power systems. The recent adoption of edge computing technology is highly aligned with the technical requirements of current regional integrated energy stations, enabling local, rational control of system equipment and efficient utilization of system data. Edge computing, a product of fifth-generation communications technology, is designed to process data from terminals at the edge of the network. With the advent of the Internet of Everything (IoE), the massive number of connected devices and explosive data growth are solidifying the position of edge computing technology in IoT applications. Compared to traditional cloud computing, edge computing offers significant advantages in areas such as conserving network bandwidth, reducing communication latency, and protecting user privacy. Edge computing can leverage historical data and other information for day-ahead scheduling, while also utilizing its own 24 / 7 computing power for real-time control. This seamless integration of the day-ahead scheduling and intraday control functions of traditional integrated energy stations improves the efficiency of regional integrated energy stations.

[0003] Therefore, to address the massive data processing pressure faced by distribution network mid-stations, a distributed, decentralized edge computing data processing approach, with microgrids as the core processing unit, will be the primary solution for future digital traction power systems. On the microgrid side, due to their proximity to user demand and the complex multi-source and multi-load situation, the current traditional grid-based radial energy supply and demand response technology architecture is no longer sufficient. There is an urgent need to develop terminal hardware architectures for massive energy data at the distribution network edge, particularly around regional energy stations, along with algorithms and protocols for data processing, storage, and real-time communication. This will reduce the processing and storage pressure and minimize data latency. Furthermore, at the network edge, near individual data collection terminals near energy stations, deploying a new architecture of edge computing equipment and leveraging data processing technologies to enable local load forecasting will significantly reduce the load on the distribution network mid-station and improve communication bandwidth efficiency. Furthermore, it will effectively enhance the distribution network's ability to optimize the coordinated generation, distribution, load, and storage of energy, laying a critical technical foundation for subsequent multi-energy complementary energy scheduling. Summary of the Invention

[0004] In response to the technical problems existing in the prior art, the present invention provides an edge computing processing system and a data mining method based on a regional integrated energy station; the edge computing processing system in the present invention reduces the hardware overhead cost and improves the stability of the operation of the edge computing system of the regional integrated energy station; the task optimization method in the present invention adopts a pipeline task allocation optimization mechanism. The task optimization method establishes a queue for processing for multiple tasks according to the computing overhead, cost, required resources and maximum tolerable delay, and sends them one by one to the neural network processing unit for processing, thereby realizing the processing method under single-core hardware resources and improving the data processing speed.

[0005] In order to solve the problems existing in the prior art, the present invention adopts the following technical solutions to implement:

[0006] An edge computing processing system based on a regional integrated energy station, the edge computing processing system comprising: a data input module, a data processing module, a storage module, a communication module, and a power conversion module;

[0007] The data input module converts the analog flow from the regional integrated energy station into data information;

[0008] The data processing module distributes tasks to the data information in a time-series pipeline manner;

[0009] The storage module stores and calls the task assignments;

[0010] The communication module transmits the task assignment to the server;

[0011] The power conversion module is used to supply power to the edge computing processing system; wherein:

[0012] The data processing module includes a processor, a neural network operation module, an AXI bus, an internal cache unit and a 3-DMA controller; the neural network operation module includes an input buffer unit, an intermediate calculation buffer unit, an average pooling unit, a control unit and a convolution operation unit; wherein:

[0013] The control unit is mainly responsible for the configuration before data calculation, the interaction between various modules and the control during the calculation process; the calculation results are transmitted from the convolution operation unit to the average pooling unit through the control module. The average pooling module contains 8 pooling units, each of which is composed of a three-layer addition tree and a data buffer; after pooling, 1024 classifications are obtained, which are stored in the output buffer and then transmitted to the external memory via the AXI bus. Finally, the final result is obtained using the MaxSoft classifier.

[0014] Furthermore, the convolution operation unit is composed of a data buffer, a weight register, an offset register, a convolution layer, a feature buffer and a logic controller; the convolution layer includes 4 PE units, forming 2 groups of 8×8 parallel convolution acceleration arrays; the feature buffer is composed of a first feature buffer and a second feature buffer; the first feature buffer will store the operation results of the previous layer network, and the second feature buffer will store the operation results of the current layer network; the logic controller is a time series processor.

[0015] Furthermore, the data input module is composed of an analog quantity acquisition unit, a data quantity acquisition unit, an A / D conversion unit and an electromagnetic isolation unit.

[0016] Furthermore, the neural network computing unit is a NEON GX680 architecture GPU programmable soft core.

[0017] Furthermore, the data processing module is composed of an FPGA integrated circuit.

[0018] Furthermore, the power conversion module is composed of AC / DC conversion, voltage reduction and backup capacitor circuits.

[0019] Furthermore, the neural network computing module is a programmable soft core of the GPU based on the NEON GX680 architecture.

[0020] In order to solve the problems existing in the prior art, the present invention can also be implemented by adopting the following technical solutions:

[0021] A method for optimizing data tasks of a regional integrated energy station based on an edge computing processing system includes the following steps:

[0022] Step 1: Construct the preset conditions for regional integrated energy station data tasks;

[0023] Step 2: Divide different types of data tasks according to the data representation of the regional energy station and then build a processing task timing logic queue under multi-cycle coupling. The different data types include time request tasks TSDP and resource demand tasks NTSDP;

[0024] Step 3: Construct an optimized resource model for the data processing module based on the following formula according to the task types represented by different representations:

[0025]

[0026] Where: C(t) is the total cost of the task processing by the computing unit at time t; e is the number of the PE inside the computing unit; C 1 e (t), C 2 e (t), C 3 e (t) are the computation, delay, and communication costs incurred when the PE inside the computing unit processes tasks;

[0027] Step 4: Construct the optimized load model of the data processing module according to the following formula based on the task types with different representations:

[0028] θ a =[p a ,w a (t),T a ],a∈[1,m]

[0029] w a (t) = q a u a (t)

[0030] W=[w1(t),w2(t),…,w m (t)]

[0031] Where: θ a is the feature model of task a; p a is the type of task a; wa(t) is the computational load of task a at time t; T a is the delay constraint of task a; q a To process p a The computational load per unit data volume for class tasks; u a (t) is the amount of data for task a at time t; W is a vector consisting of m tasks; m tasks are divided into m1 TSDPs and m2 NTSDPs, then m1 + m2 = m; after the computing unit first satisfies the processing requirements of the TSDP, the remaining resources are called residual resources. The amount of residual resources affects the computational delay when the computing unit processes the TSDP.

[0032] Step 5: Construct the optimized cost model of the data processing module according to the following formula based on the task types with different representations:

[0033]

[0034]

[0035]

[0036] Where: d ae (t) is the time required for the computing unit to process task a; f ae (t) is the computing resource allocated by the operation unit to the buffer of task a, C 1 e (t) is the cost of unit computing resources occupied by the computing unit; F e is the total amount of computing resources of the operation unit; ∝ is the proportional symbol; Formula (1) is the computing resource constraint that should be met when the operation unit is assigned task a; further, the process of constructing the processing task timing logic queue under multi-cycle coupling includes the following steps:

[0037] For time request tasks TSDP, they are arranged in priority according to the time request;

[0038] For resource demand tasks NTSDP, they are arranged by size according to the calculated task load requirements;

[0039] According to the time request task TSDP, a delay constraint model for processing time request tasks is established according to the following formula:

[0040] 2d ie +d ae χ ie χ ae ≤T m1

[0041] Where: d ie represents the time required for task i of the computing processing unit; d ae represents the time required for the processing unit task a; ie represents the chi-square distribution coefficient of task i; ae represents the chi-square distribution coefficient of task a; T mi Represents the delay constraint of the request task TSDP;

[0042] According to the resource demand task NTSDP, a delay constraint model for processing resource demand tasks is established according to the following formula;

[0043]

[0044] Where: d ee’represents the time required for the task e' of the computing processing unit; ee’ represents the chi-square distribution coefficient of task e'; T m2 Represents the delay constraint of the request task TSDP;

[0045]

[0046] In the above formula: T max Delay tolerance for data processing services; Indicates that when the value of m1 is equal to 1 or 2, at T m1 Take the maximum value; It means that when m1 is equal to 3, T m1 the limit;

[0047] According to the periodicity of data processing tasks, the period of a data processing module triggering is from T0 to T5, where T0 is the start time of the data processing task and T5 is the end time of the task; at time t, there is a single task request or multiple parallel task requests in the task timing logic queue in the processing state;

[0048] There will be multiple task cycles at time t. The data processing module forms task queues under different resource load spaces according to the timing logic queue; by establishing the characteristics of various data processing tasks under different time requirements, the processing timing of different types of tasks is characterized.

[0049] The present invention provides the following beneficial effects:

[0050] 1. The edge computing system involved in the present invention is a lightweight, modular, low-cost, plug-and-play integrated circuit. The system utilizes the discrete and coupled characteristics of energy data in time series and the parallel computing characteristics within the integrated circuit, divides the received data into modules according to data processing tasks, and designs it in a pipeline manner. At the same time, it fully considers the plug-and-play characteristics between software and hardware. The present invention adopts high-speed parallel communication between modules and asynchronous soft switches to control the modules, so that the addition and removal of new tasks will not affect the execution of other tasks.

[0051] 2. The present invention adopts a lightweight, pipelined, modular design edge computing system based on FPGA to reduce hardware overhead costs while improving the stability of the edge computing system operation in regional integrated energy stations; it solves the technical problem that most edge computing terminals in the prior art are computer architectures with multiple GPU cores, with complex hardware structures, which cannot be plug-and-play, and the hardware cost is very high, which is not conducive to large-scale batch deployment. At the same time, the present invention solves the technical problem that under the condition of limited existing hardware resources, traditional short-term electric load forecasting algorithms can no longer be directly applied. The present invention can realize the technical problem that the edge computing multi-task allocation mechanism based on multi-core GPU processors has high requirements for hardware resources and is completely directly deployed on edge computing terminals with limited performance and resources.

[0052] 3. Based on the discrete coupling characteristics of the energy data of the regional integrated energy station in the time series, the present invention proposes a pipeline-based task allocation optimization mechanism, which establishes a queue for processing for multiple tasks according to the computing overhead, cost, required resources and maximum tolerable delay, and sends them one by one to the neural network processing unit for processing, thereby realizing the rapid processing of data tasks of the regional integrated energy station.

[0053] 4. The neural network computing unit in the present invention fully utilizes the programmable characteristics of the GPU soft core to maximize the use of internal resources. At the same time, it fully considers the discrete time characteristics and coupling of power data, arranges the single-core GPU operation in a pipeline manner, maximizes the release of the internal computing power of the GPU, and improves work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a schematic diagram of the structure of the edge computing processing system based on the regional integrated energy station of the present invention;

[0055] Figure 2 This is a schematic diagram of the FPGA core processing module in the edge computing processing system based on the regional integrated energy station of the present invention;

[0056] Figure 3 This is a schematic diagram of the neural network operation module in the edge computing processing system based on the regional integrated energy station of the present invention;

[0057] Figure 4 Schematic diagram of the task optimization module in the edge computing processing system based on the regional integrated energy station of the present invention;

[0058] Figure 5 This is a flowchart of the task optimization method in the edge computing processing system based on the regional integrated energy station of the present invention.

[0059] Figure 6This is a flow chart of the task optimization method in the edge computing processing system based on the regional integrated energy station of the present invention. DETAILED DESCRIPTION

[0060] The following is combined with Figure 1-6 The present invention is described in detail:

[0061] 1. Power conversion module: Consists of AC / DC conversion, voltage reduction, and backup capacitor circuits. The ADDC converts 110-380V AC power into a 48V DC output. The DCDC step-down circuit steps down the 48V DC power to eight voltages: 15V, 10V, 5V, 3.3V, 1.8V, 1.5V, 1.2V, and 1V, for use by various circuits within the terminal. The backup capacitor is a 0.5F supercapacitor, charged by a 5V power supply. In the event of an external power failure, the backup capacitor supplies current, allowing the terminal to continue operating for approximately 3 seconds. Simultaneously, an interrupt signal is sent to the FPGA processor, alerting it to a fault state. Upon receiving the interrupt signal, the processor immediately enters low-power mode, saving all program data and status.

[0062] 2 Data Input Module: 2-1 is divided into four parts: the analog acquisition unit, 2-2 the digital acquisition unit, 2-3 the A / D conversion unit, and 2-4 the electromagnetic isolation unit. The analog acquisition unit can simultaneously acquire eight channels of analog voltage and current with 1-14-bit accuracy. The digital acquisition unit provides six common industrial digital interfaces: RS485, RS232, USB, CAN, and I / O. The A / D conversion unit converts analog signals into digital signals, while the electromagnetic isolation unit provides electrical isolation between the external circuitry and the internal FPGA core circuitry, ensuring that surges, pulses, and other factors do not interfere with the normal operation of the internal core unit. The data input module utilizes a plug-and-play modular design, and the data protocol uses a serial asynchronous transparent transmission mode, ensuring that the data input module can be added or removed without affecting the FPGA core processor.

[0063] FPGA core processing module: It is the core of the entire architecture and plays the role of control, calculation, processing and other functions. The internal architecture of the FPGA core module is as follows Figure 2As shown. The FPGA core module is divided into five parts: 3-1 processor, 3-2 neural network operation module, 3-3 AXI bus, 3-4 internal cache unit and 3-5 DMA controller. The processor module is an ARM Cortex-A72 core, which mainly performs logical control on all units inside the FPGA, including data input, preprocessing, interruption, task calling and allocation, etc. The program running inside the Cortex-A72 core is a tailored RT-Linux system. The neural network operation unit is a NEON GX680 architecture GPU programmable soft core. The highest clock of the soft core is 1.5G FPGA processor clock, that is, the time series processor is the most core operation unit for realizing data edge side processing. Taking into full consideration the plug-and-play, modular and pipeline design requirements, the design architecture of the soft core of the present invention is as follows. Figure 3 shown.

[0064] Figure 3 In the FPGA, the convolution operation unit is the core of the neural network operation module, which consists of N×M PE units and eigenvalue buffers. Since there are a total of 1,800 PEs inside the FPGA, and each PE is responsible for a 3×3 convolution operation, 128 PEs were finally designed to form two groups of 8×8 parallel convolution acceleration arrays. The feature buffer is a double buffer structure, one is the working buffer responsible for storing the operation results of the previous layer of the network, and the other is the result buffer responsible for storing the operation results of the current layer of the network. Due to the limited storage resources on the FPGA chip, the present invention includes 2 buffers shared by every 4 PEs, for a total of 32 eigenvalue buffers. The double buffer structure makes it unnecessary to transfer the intermediate operation results of the network to the external memory, realizes inter-layer data reuse, and greatly improves the operation speed of the entire neural network. The control unit is mainly responsible for the configuration before the data operation, the interaction between the units and the control during the operation process. The control unit transfers the computation results from the convolution module to the average pooling unit, which contains eight pooling units, each consisting of a three-layer addition tree and a data buffer. Pooling yields 1024 classifications, which are stored in the output buffer and then transferred to external memory via the AXI bus. The final result is obtained using the MaxSoft classifier. The entire neural network computation module fully utilizes the programmable nature of the GPU soft core, maximizing internal resources. While fully considering the discrete time characteristics and coupling of power data, the single-core GPU computation is orchestrated in a pipelined design, maximizing the GPU's internal computing power.

[0065] The 4 storage module consists of an 8GB DDR4 memory and a 128GB mobile solid-state drive.

[0066] 5. Communication module: It consists of a Gigabit Ethernet port, a 5G mobile communication module and a fiber optic communication port to realize communication with the upper server and the lower computer.

[0067] The present invention adopts a task allocation optimization model process based on a sequential pipeline:

[0068] Step 1: Create pre-conditions

[0069] Condition 1: To ensure the reliability of data processing, the changes in the amount of data processed by each PE within the neural network operation module can be measured in real time by register reading. In addition, the fluctuations in the PE data volume satisfy the normal distribution and are independent of each other, and the relative positions of the PEs are known.

[0070] Condition 2: The computational load per unit data volume of each data processing task is known, and the processing timing logic of the business is known.

[0071] Condition 3: The controller's time allocating computing resources for data buffering within the arithmetic unit is negligible, and buffer creation and destruction can be completed in real time.

[0072] Step 2: Characterize the types of different data processing tasks

[0073] Based on the differences in data sources of regional energy stations, tasks are abstracted into two categories: time request tasks and resource demand requests, namely TSDP (Time-Sensitive Data Protocol) and NTSDP (Non-Time-Sensitive Data Protocol). All task requests are sent to the time series processor, which processes them as follows: for TSDP, it is assigned the highest priority and sorted according to the time queue, while calculating the respective maximum delay tolerance value and resource load requirements. For time-insensitive requests, after calculating the task load requirements, they are sorted according to the load requirements. The time queue processing block diagram for TSDP is as follows Figure 5 shown.

[0074] When the convolutional operation unit of the neural network is running in a steady state, the data processing task of the front-side input is periodic. Figure 5 The trigger period of the energy data management service is from T0 to T5, where T0 is the start time of the data processing task and T5 is the end time of the task. Due to the succession relationship during the TSDP processing of power data, at time t, a single task request or multiple parallel task requests may be in the processing state in the sequential logic chain of the processing task at the same time.

[0075] When the data service request of the neural network convolution operation unit reaches a steady state, there will be multiple task cycles at time t, such as Figure 5The processing timing logic chain under multi-cycle coupling in the [1] is shown. The computing unit allocates independent resource space to each task, while decoupling and making different tasks independent from each other. Task queues with different resource load spaces are formed according to the timing logic chain. By establishing characteristics for various data processing tasks under different time requirements, different types of task processing are characterized while also meeting certain computing resources, delay tolerance, and objective functions.

[0076] Step 3: Establish the objective function of the optimization model

[0077] Data processing tasks require the computational and communication resources of the neural network's convolutional processing units, generating energy consumption. Because the computational load of different types of tasks varies with the amount of data processed, and these changes vary significantly across time and space, the computational and communication resources used during data processing and transmission also vary accordingly.

[0078] Computational delay constraints

[0079] Because much power energy data is time-sensitive, processing tasks must have a maximum latency tolerance. This means that the sum of the latency of each connecting component in the sequential logic chain and the maximum latency of each parallel component must not exceed the latency tolerance. Task service latency is primarily comprised of the transmission latency of data uplinks and instructions downlinks executed by the arithmetic unit, as well as the computational latency incurred during data processing. For priority processing tasks in a pipeline architecture, TSDP, the latency constraint model is shown below.

[0080] 2d ie +d ae χ ie χ ae ≤T m1

[0081] For non-priority processing tasks in the pipeline architecture, such as NTSDP, the service delay is different from TSDP in the processing timing logic chain, resulting in different data transmission directions within the operation unit. The processing delay constraint model of NTSDP tasks is shown as follows:

[0082]

[0083]

[0084] In the above formula: T max Delay tolerance for data processing services.

[0085] Step 3: Construct the objective function of the optimization model, as shown below:

[0086]

[0087] Where: C(t) is the total cost of the computing unit for processing the task at time t; e is the number of the PE inside the computing unit; They are the computation, delay, and communication costs generated when PE processes tasks inside the computing unit.

[0088] Step 4: Calculate the load

[0089] The demand for computing resources of the computing unit is represented by the computing load of the task. The size of the computing load is related to the task type and the amount of data to be processed. The calculation formula is as follows:

[0090] θ a =[p a ,w a (t),T a ],a∈[1,m]

[0091] w a (t) = q a u a (t)

[0092] W=[w1)t),w2(t),…,w m (t)]

[0093] Where: θ a is the feature model of task a; p a is the type of task a; wa(t) is the computational load of task a at time t; T a is the delay constraint of task a; q a To process p a The computational load per unit data volume for class tasks; u a (t) is the data volume of task a at time t; W is a vector consisting of m tasks. If m tasks are divided into m1 TSDPs and m2 NTSDPs, then m1 + m2 = m. After the computing unit first satisfies the processing requirements of the TSDP, the remaining resources are called residual resources. The amount of residual resources affects the computational latency of the computing unit when processing the TSDP.

[0094] Step 5: Calculate the cost

[0095] In the neural network convolution operation unit, different types of TSDP and NTSDP tasks are placed in different computation buffers, and a certain amount of computation and communication resources are allocated to each buffer according to the task attributes, so that more residual resources can be provided for NTSDP while giving priority to meeting the delay constraints of TSDP tasks. In practical applications, since the computation load varies with the data type, the computation cost expression is

[0096]

[0097]

[0098]

[0099] Where: d ae (t) is the time required for the computing unit to process task a; f ae (t) is the computing resource allocated by the operation unit to the buffer of task a, C 1 e (t) is the cost of unit computing resources occupied by the computing unit; F e is the total amount of computing resources of the operation unit; ∝ is the proportional symbol; Formula (1) is the computing resource constraint that should be met when the operation unit is assigned task a.

[0100] The above embodiments are intended only to illustrate the technical concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. They are not intended to limit the scope of protection of the present invention. Any equivalent changes or modifications made in accordance with the spirit of the present invention are intended to be covered by the scope of protection of the present invention.

Claims

1. A method for optimizing regional integrated energy tasks based on an edge computing processing system, characterized by: The data processing module in the edge computing processing system distributes the data information to tasks in a time-series pipeline manner; the steps include: Step 1: Construct the preset conditions for regional integrated energy station data tasks; Step 2: Divide different types of data tasks according to the data representation of the regional energy station and then build a processing task timing logic queue under multi-cycle coupling. The different types of data include time request tasks and resource demand tasks ; Step 3: Construct an optimized resource model for the data processing module based on the following formula according to the task types represented by different representations: ； Where: For the moment The total cost of the computing unit for task processing; Internal to the arithmetic unit Number; 、 They are the computation, delay, and communication costs incurred when PE processes tasks within the computing unit; Step 4: Construct the optimized load model of the data processing module according to the following formula based on the task types with different representations: ； ； ； Where: For the task The characteristic model of For the task Type; For the task At the moment The computational load; For the task Delay constraints; For processing The computational load per unit data volume for class tasks; For the task At the moment The amount of data; for A vector of tasks; The tasks are divided into indivual and indivual ,but ; The operation unit first satisfies After the processing requirements are met, the remaining resources are called residual resources. The amount of residual resources affects the processing of the operation unit. Calculation delay of 1 second; Step 5: Construct the optimized cost model of the data processing module according to the following formula based on the task types with different representations: ； ； （1）； Where: Processing tasks for the computing unit the time required; Assign tasks to computing units The computing resources of the buffer to which it belongs, The cost of computing resources for each unit that occupies the computing unit; is the total amount of computing resources of the computing unit; is the proportional symbol; Formula (1) is the task allocation of the operation unit The computing resource constraints that should be met when 2. The method for optimizing regional comprehensive energy tasks based on an edge computing processing system according to claim 1, characterized in that: The process of constructing a sequential logic queue for processing tasks under multi-cycle coupling includes the following steps: For time request tasks , requests are prioritized according to time; For resource-demanding tasks , according to the calculated task load requirements, they are arranged in order of size; Request tasks based on time The delay constraint model for processing time request tasks is established according to the following formula: ； in: Indicates the task of the processing unit the time required; Indicates the task of the processing unit the time required; Indicates a task The chi-square distribution coefficient of Indicates a task The chi-square distribution coefficient of Indicates a request task Delay constraints; Tasks based on resource requirements Establish a delay constraint model for processing resource demand tasks according to the following formula; ； in: Indicates the task of the processing unit the time required; Indicates a task The chi-square distribution coefficient of Indicates a request task Delay constraints; ； In the above formula: Delay tolerance for data processing services; Indicates when When the value is equal to 1 or 2, Take the maximum value; Indicates when When it is equal to 3, the limit; According to the periodicity of data processing tasks, the cycle of a data processing module trigger is to ,in, is the start time of the data processing task, The end time of the task; at time Under this condition, there is a single task request or multiple parallel task requests in the task timing logic queue in the processing state; time There will be multiple task cycles. The data processing module forms task queues under different resource load spaces according to the timing logic queue; the processing timing of different types of tasks is characterized by establishing the characteristics of various data processing tasks under different time requirements.

3. An edge computing processing system optimized for tasks in a regional integrated energy station, comprising: a data input module, a data processing module, a storage module, a communication module, and a power conversion module; characterized in that: The data input module converts the analog flow from the regional integrated energy station into data information; The data processing module distributes tasks to the data information in a time-series pipeline manner; The storage module stores and calls the task assignments; The communication module transmits the task assignment to the server; The power conversion module is used to supply power to the edge computing processing system; wherein: The data processing module includes a processor, a neural network operation module, Bus, internal cache unit based on edge computing processing system and Controller; the neural network operation module includes an input buffer unit, an intermediate calculation buffer unit, an average pooling unit, a control unit and a convolution operation unit; wherein: The control unit is mainly responsible for the configuration before data operation, the interaction between each module and the control during the operation; the operation result is transferred from the convolution operation unit to the average pooling unit through the control unit. The average pooling unit contains 8 pooling units, each pooling unit consists of a three-layer addition tree and a data buffer; after pooling, 1024 categories are obtained, which are stored in the output buffer and then transferred to the external memory via the AXI bus. Finally, The classifier produces the final result.

4. The edge computing processing system for optimizing tasks in a regional integrated energy station according to claim 3, characterized in that: The convolution operation unit is composed of a data buffer, a weight register, an offset register, a convolution layer, a feature buffer and a logic controller; the convolution layer includes 4 PE units, forming 2 groups of 8×8 parallel convolution acceleration arrays; the feature buffer is composed of a first feature buffer and a second feature buffer; the first feature buffer will store the operation results of the previous layer of the network, and the second feature buffer will store the operation results of the current layer of the network; the logic controller is a 1.5G processor clock.

5. The edge computing processing system for optimizing tasks in a regional integrated energy station according to claim 3, characterized in that: The data input module is composed of an analog quantity acquisition unit, a data quantity acquisition unit, an A / D conversion unit and an electromagnetic isolation unit.

6. The edge computing processing system for optimizing tasks in a regional integrated energy station according to claim 3, characterized in that: The neural network operation unit is a of Architectural Programmable soft core.

7. The edge computing processing system for optimizing tasks in a regional integrated energy station according to claim 3, characterized in that: The data processing module is Integrated circuit composition.

8. The edge computing processing system for optimizing tasks in a regional integrated energy station according to claim 3, characterized in that: The power conversion module consists of It consists of conversion, step-down and backup capacitor circuits.

9. The edge computing processing system for optimizing tasks in a regional integrated energy station according to claim 3, characterized in that: The neural network operation module is of Architectural Programmable soft core.

Images