Implementation method of virtual power plant multi-source data fusion processing hardware

Abstract

The invention provides an implementation method of virtual power plant multi-source data fusion processing hardware, and relates to the technical field of virtual power plant data processing. In order to solve the technical defects of poor real-time performance, high power consumption, insufficient stability, weak expansibility and the like due to the fact that multi-source data fusion of an existing virtual power plant depends on a general processor and software processing, a heterogeneous integrated architecture is adopted, and multi-module collaborative special processing hardware is built; multi-source data collection, hardware preprocessing, FPGA + MCU heterogeneous parallel fusion calculation, closed-loop control and dual-mode communication, hierarchical storage and hardware integrated debugging are sequentially achieved, and meanwhile power consumption management, redundant backup and standardized modular design are integrated. The core data processing flow is subjected to hardware processing, the data processing delay and power consumption are greatly reduced, the data fusion precision and the system stability are improved, multi-source heterogeneous data sources of different scales of virtual power plants can be flexibly adapted, seamless connection with an existing regulation and control platform is achieved, and the deployment and maintenance cost is reduced.

Description

Technical Field

[0001] This invention belongs to the field of virtual power plant data processing technology, and more specifically, it relates to a method for implementing hardware for multi-source data fusion processing in a virtual power plant. Background Technology

[0002] With the rapid development of new energy technologies, virtual power plants, as the core carrier integrating distributed resources such as distributed power sources, energy storage devices, and controllable loads, are playing an increasingly important role in peak shaving and valley filling of the power grid, improving energy utilization efficiency, and ensuring the stable operation of the power grid. The efficient operation of virtual power plants relies on the accurate fusion and processing of multi-source heterogeneous data generated by various distributed resources, grid-side data, and environmental monitoring equipment. This data covers multiple dimensions such as voltage, current, power, SOC, environmental parameters, and dispatch commands, and is characterized by dispersed data sources, diverse interface types, large differences in data transmission rates, and high real-time requirements.

[0003] Currently, the multi-source data fusion processing in virtual power plants mainly adopts a "general-purpose processor + software algorithm" approach. This involves using a general-purpose computer or industrial control unit as the processing core, relying on software programs to complete data acquisition, preprocessing, and fusion calculations. However, this traditional approach has many technical shortcomings and struggles to meet the actual operational needs of virtual power plants.

[0004] First, there is insufficient real-time performance. General-purpose processors use a serial computing architecture, which makes it difficult to handle the parallel processing requirements of multi-source heterogeneous data. Furthermore, the execution of software preprocessing and software fusion algorithms has high latency, with the overall data processing latency typically exceeding 1000ms. This cannot meet the requirements of real-time control of virtual power plants, easily leading to delayed control commands and affecting the stability of power grid operation.

[0005] Secondly, the power consumption is relatively high. The hardware architecture of general-purpose processors is not customized for virtual power plant data fusion scenarios, and has many redundant functions. Even under low load conditions, it maintains high power consumption, usually exceeding 150W. This does not conform to the development trend of low power consumption and energy saving in industrial equipment, and increases the operating cost of virtual power plants.

[0006] Secondly, stability and reliability are poor. Traditional solutions rely on software algorithms to perform data preprocessing and fusion calculations, which are susceptible to software vulnerabilities, system crashes, and other factors. Furthermore, they lack hardware-level redundancy and fault-tolerant design. When the processing core fails, the entire data processing system will be paralyzed, making it impossible to guarantee the 24-hour uninterrupted operation of the virtual power plant. At the same time, software processing methods are susceptible to electromagnetic interference, leading to data loss or processing abnormalities.

[0007] Furthermore, scalability and adaptability are weak. Traditional hardware architectures are monolithic designs, lacking standardized modular structures. When the scale of the virtual power plant expands or new data sources are added, the original hardware architecture needs to be completely redesigned, requiring redesigned circuits and software programs, resulting in high upgrade and maintenance costs. Simultaneously, the interface adaptability of general-purpose processors is limited, making it difficult to accommodate the interface requirements of different types of data sources, necessitating the addition of additional interface adapter hardware, increasing deployment complexity. Additionally, existing solutions often rely on software encryption to ensure data transmission security, but this encryption is inefficient and easily cracked, posing a risk of data leakage.

[0008] Therefore, developing a dedicated hardware implementation method that can solve the above-mentioned technical defects, achieve low-latency, high-precision, high-reliability, and low-power fusion processing of multi-source data in virtual power plants, and has good scalability and adaptability has become a key issue that urgently needs to be addressed in the current field of virtual power plant technology. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention provides a hardware implementation method for multi-source data fusion processing in virtual power plants. This method solves the technical problems of existing multi-source data fusion processing in virtual power plants, which relies on "general-purpose processors + software algorithms" and suffers from poor real-time performance, high power consumption, insufficient stability and reliability, weak scalability and adaptability, low data accuracy and security, and high deployment and maintenance costs.

[0010] A method for implementing hardware for multi-source data fusion processing in a virtual power plant is disclosed. The hardware adopts a heterogeneous integrated architecture, building a dedicated processing hardware with multi-module collaboration to realize real-time acquisition, preprocessing, fusion calculation, and output control of multi-source heterogeneous data in the virtual power plant. The specific implementation steps include:

[0011] S1: Construct a multi-source data acquisition module, integrate various dedicated interface circuits, adapt to different data interface types of distributed power sources, energy storage devices, controllable loads, grid side and environmental monitoring equipment in the virtual power plant, and synchronously acquire multi-dimensional heterogeneous data such as voltage, current, power, SOC, environmental parameters and dispatch instructions;

[0012] S2: A hardware-based data preprocessing module is built, using dedicated filtering circuits and logic operation units to perform noise reduction, outlier removal, format standardization, and synchronization alignment on the collected multi-source data. This eliminates the need for software algorithm intervention, reducing data transmission latency and main control unit computing power consumption.

[0013] S3: Design a dedicated fusion processing module, adopting an FPGA+MCU heterogeneous computing architecture, embedding multi-source data fusion algorithms into the FPGA chip to achieve parallel fusion computing. The MCU is responsible for the dynamic control of the fusion strategy and the verification of the calculation results, solving the problems of poor real-time performance and high power consumption of traditional general-purpose processors in fusion processing.

[0014] S4: Integrates the main control module and the two-way communication module. The main control module, together with the preprocessing module, the fusion processing module, and the storage module, forms a closed-loop control. It receives the fusion processing results and outputs control commands. The communication module adopts a wired + wireless dual-mode redundancy design to realize data interaction with the virtual power plant control platform and terminal equipment.

[0015] S5: Build a hierarchical storage module and adopt a hardware architecture of cache + non-volatile storage. The cache module is used to temporarily store real-time acquisition and fusion intermediate data, and the non-volatile storage module is used to persistently store fusion results and key operation data to ensure that data is not lost.

[0016] S6: Complete hardware integration and debugging, integrate each module on the same hardware substrate through a high-speed bus, optimize the wiring design between modules, reduce signal interference, and ensure the stability and reliability of multi-module collaborative operation through hardware-level timing calibration and fault-tolerant design, ultimately achieving low-latency and high-precision fusion processing of multi-source data in the virtual power plant.

[0017] Preferably, in step S1, the dedicated interface circuit integrated by the multi-source data acquisition module includes an RS485 interface, a CAN bus interface, an Ethernet interface, and an ADC analog signal acquisition interface. The ADC interface uses a 16-bit high-precision analog-to-digital converter chip, and the sampling frequency can be adjusted by a hardware DIP switch with an adjustment range of 10Hz-1kHz to adapt to the data transmission rate requirements of different types of data sources. At the same time, it integrates a data acquisition synchronization trigger to ensure the timing consistency of multi-channel data acquisition, with a synchronization error ≤10μs.

[0018] Preferably, in step S2, the hardware-based data preprocessing module includes an RC low-pass filter circuit, an outlier detection logic unit, and a format conversion circuit. The outlier detection logic unit is implemented using a hardware comparator and a threshold register. By setting a hardware threshold (which can be dynamically configured by the main control module), it can determine in real time whether the collected data exceeds a reasonable range, and perform hardware-level masking and replacement processing on the outlier data. The replacement value is the average of the first three valid data. The preprocessing delay is ≤50μs, which is more than 60% lower than the software preprocessing delay.

[0019] Preferably, in step S3, the FPGA chip adopts a high-performance industrial-grade chip, integrating at least 8 parallel computing units, and solidifies the spatiotemporal fusion network architecture (combining CNN convolutional neural network to extract spatial features and LSTM long short-term memory network to capture temporal dependencies) into hardware logic to realize parallel fusion computing of multi-source data, with a fusion computing latency of ≤200ms; the MCU adopts a low-power industrial-grade microcontroller, which communicates with the FPGA chip through a high-speed SPI interface, receives the intermediate results of the fusion computing of the FPGA in real time, and dynamically adjusts the fusion weight parameters according to the virtual power plant operation status (grid load, distributed power output fluctuation), with a weight adjustment response time of ≤10ms.

[0020] Preferably, in step S4, the main control module adopts an ARM architecture industrial-grade chip and integrates a hardware encryption unit to perform AES-128 hardware encryption processing on the transmitted fusion results and control commands to ensure data transmission security. In the bidirectional communication module, wired communication adopts a gigabit Ethernet interface, and wireless communication integrates a 5G / NB-IoT dual-mode module, which can automatically switch communication modes according to communication distance and environmental complexity. The 5G mode communication rate is ≥1Gbps, and the NB-IoT mode communication power consumption is ≤50mW, which is suitable for the communication requirements of distributed deployment of virtual power plants. At the same time, it supports hardware adaptation of multiple communication protocols such as Modbus, MQTT and RESTfulAPI without the need for software protocol conversion.

[0021] Preferably, in step S5, the cache module uses a DDR4 high-speed cache chip with a cache capacity of ≥4GB and a read / write speed of ≥2400Mbps, used for temporary storage of real-time acquired raw data and intermediate data of fusion processing, reducing data read / write latency; the non-volatile storage module uses an industrial-grade SSD with a storage capacity of ≥128GB, supports power loss protection, and ensures persistent storage of fusion results and key operating data (at least 30 days) through a hardware-level bad block management mechanism, with a data read speed of ≥500Mbps.

[0022] Preferably, in step S6, the high-speed bus adopts a PCIe 4.0 interface, the data transmission rate between modules is ≥8Gbps, the wiring design adopts differential wiring to reduce electromagnetic interference, and at the same time integrates a hardware-level redundant backup unit to provide dual backup for the core circuits of the acquisition module and the fusion processing module. When the main module fails, the backup module can automatically switch to operation within 5ms to ensure the continuous operation of the hardware system, with an average fault-free working time of ≥20,000 hours.

[0023] Preferably, the hardware also integrates a hardware-based power consumption management module, which adopts an intelligent power supply regulation circuit. It can dynamically adjust the power supply voltage and current of each module according to the load demand of virtual power plant data acquisition and fusion processing. When the load rate is lower than 30%, it automatically switches to a low-power mode, reducing power consumption to less than 40% of the normal mode. At the same time, it integrates a power consumption monitoring circuit to collect power consumption data of each module in real time and feed it back to the main control module to realize visualized control and optimization of power consumption.

[0024] Preferably, in step S3, the fusion processing module also integrates a hardware data calibration unit, which performs real-time calibration of the fusion calculation results through a preset calibration coefficient (which can be dynamically updated by the main control module), with a calibration error ≤0.5%; at the same time, it supports hardware switching of multiple fusion algorithms, and can select the corresponding fusion algorithm (weighted average fusion algorithm, Bayesian estimation fusion algorithm, neural network fusion algorithm) through a hardware DIP switch according to the data source type (such as renewable energy data, load data), without the need to re-burn the program;

[0025] The hardware adopts a standardized modular design, and each functional module (acquisition module, preprocessing module, and fusion processing module) is pluggable. It is connected to the base plate through a standardized interface. The number and type of modules can be flexibly increased or decreased according to the scale of the virtual power plant and the type of data source. No modification to the original hardware architecture is required during the expansion process, reducing hardware upgrade and maintenance costs. At the same time, it supports seamless integration with the existing virtual power plant control platform without the need for additional adapter hardware.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] This invention employs an FPGA+MCU heterogeneous computing architecture, embedding the multi-source data fusion algorithm within the FPGA chip. Leveraging the parallel computing advantages of the FPGA, it achieves parallel fusion computation of multi-source data. Simultaneously, a hardware-based data preprocessing module performs noise reduction and outlier removal at the hardware level, eliminating the need for software algorithm intervention and significantly reducing data processing latency. Actual measurements show that this invention achieves a data acquisition synchronization error ≤10μs, preprocessing latency ≤50μs, fusion computation latency ≤200ms, and overall data processing latency ≤300ms. Compared to traditional solutions, this represents a latency reduction of over 70%, enabling rapid response to the control requirements of virtual power plants and ensuring grid operational stability.

[0028] This invention employs low-power industrial-grade chips (FPGA, MCU, main control chip) and integrates a dedicated hardware power management module. It can dynamically adjust the power supply voltage and current of each module according to the load requirements of data acquisition and fusion processing. Under low load conditions, it automatically switches to low-power mode, reducing power consumption to less than 40% of normal mode (normal mode power consumption ≤50W, low-power mode power consumption ≤20W). Compared with traditional general-purpose processor solutions, power consumption is reduced by more than 60%, significantly reducing the operating cost of the virtual power plant and conforming to the trend of energy-saving development in industrial equipment.

[0029] The system's stability and reliability are significantly improved, ensuring uninterrupted operation. This invention employs multiple fault-tolerant designs at the hardware level: modules are connected via a high-speed PCIe 4.0 bus, and differential cabling is used to reduce electromagnetic interference; the core circuits of the acquisition module and the fusion processing module are doubly redundantly backed up, and the backup module can automatically switch to operation within 5ms in the event of a main module failure; simultaneously, the hardware-based processing avoids the impact of software vulnerabilities and system crashes. Actual testing shows that the hardware system of this invention has a mean time between failures (MTBF) of ≥20,000 hours, enabling 24-hour uninterrupted and stable operation of the virtual power plant, thus improving the reliability of data processing.

[0030] It boasts strong scalability and adaptability, and low upgrade and maintenance costs. This invention adopts a standardized modular design, with each functional module (acquisition module, preprocessing module, fusion processing module, etc.) being pluggable. They connect to the hardware substrate via standardized interfaces. When the virtual power plant expands in scale or new data sources are added, the number and type of modules can be flexibly increased or decreased without modifying the original hardware architecture, significantly reducing upgrade and maintenance costs. Simultaneously, the multi-source data acquisition module integrates various dedicated interface circuits, directly adapting to the interface requirements of various data sources within the virtual power plant without the need for additional adapter hardware, thus reducing deployment complexity.

[0031] The data processing module of this invention boasts high accuracy and security, ensuring reliable and usable data. It integrates a hardware-based data calibration unit, which uses preset, dynamically updatable calibration coefficients to calibrate the fusion calculation results in real time, achieving a calibration error of ≤0.5%. Simultaneously, the hardware-based preprocessing module accurately identifies and removes abnormal data, achieving a data acquisition accuracy of ≥99.9%, ensuring high precision of the fused data. Furthermore, the main control module integrates a hardware encryption unit, employing AES-128 hardware encryption to encrypt the fusion results and control commands. The encryption key is unreadable and unmodifiable, offering higher encryption efficiency and stronger security compared to traditional software encryption, effectively avoiding the risk of data leakage.

[0032] With strong adaptability and practicality, this invention is easy to promote and apply. The hardware system supports hardware adaptation of multiple communication protocols, including Modbus, MQTT, and RESTful API, without the need for software protocol conversion. It can seamlessly integrate with existing virtual power plant control platforms without requiring modifications to existing platforms, thus reducing deployment costs. Furthermore, the hardware architecture design is tailored to the actual scenarios of distributed deployment and multi-source heterogeneous data sources in virtual power plants, making it widely applicable to virtual power plants of different sizes (small, medium, and large). Its strong adaptability and high practical value demonstrate its promising prospects for widespread adoption.

[0033] High hardware integration and convenient deployment. This invention integrates various functional modules onto the same hardware substrate via a high-speed bus, resulting in a compact overall structure and small size. Compared to traditional multi-device combinations, it offers higher integration, occupies less space, and features standardized installation and commissioning processes, eliminating the need for complex debugging by professional personnel. This further reduces the deployment difficulty and labor costs of virtual power plants. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the present invention. Detailed Implementation

[0035] The following detailed description, with reference to specific embodiments, further illustrates the implementation method of a virtual power plant multi-source data fusion processing hardware according to the present invention. This embodiment is merely for explaining the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0036] The core objective of this invention is to address the technical shortcomings of existing virtual power plant multi-source data fusion processing, which relies on general-purpose processors and software preprocessing, resulting in poor real-time performance, high power consumption, insufficient stability, and weak scalability. By implementing core processes such as data acquisition, preprocessing, and fusion computing in hardware through a heterogeneous integrated dedicated hardware architecture, this invention balances low latency, high precision, and high reliability, and adapts to the actual needs of distributed deployment and multi-source heterogeneous data source access in virtual power plants. The specific implementation method is as follows.

[0037] I. Overall Hardware Architecture Setup:

[0038] The hardware of this invention adopts a heterogeneous integrated architecture, which is divided into seven functional modules: a multi-source data acquisition module, a hardware-based data preprocessing module, a dedicated fusion processing module (FPGA+MCU heterogeneous computing architecture), a main control module, a bidirectional communication module, a hierarchical storage module, and a hardware-based power management module. Each module is integrated on the same industrial-grade hardware substrate (made of FR-4 material, with a size of 300mm×200mm) via a PCIe 4.0 high-speed bus. The substrate surface adopts a differential line design to reduce electromagnetic interference between modules, and at the same time, it reserves standardized pluggable interfaces for the installation, replacement and expansion of each functional module.

[0039] The hardware board integrates a power interface (supporting 220V AC input and 24V DC output), a grounding interface, and a debugging interface. The power interface connects to an external industrial-grade regulated power supply to provide stable power to each module. The grounding interface adopts a single-point grounding design to reduce signal interference. The debugging interface uses a USB-Type-C interface for hardware debugging, parameter configuration, and program burning (only used when the fusion algorithm is first solidified and the calibration coefficient is updated).

[0040] II. Functional Modules:

[0041] Multi-source data acquisition module:

[0042] The multi-source data acquisition module, serving as the data input terminal of the hardware system, plays a crucial role in adapting to the interface differences of various data sources within the virtual power plant, enabling the synchronous acquisition of multi-dimensional heterogeneous data. The specific implementation is as follows:

[0043] (1) Interface circuit selection and integration: The system integrates four types of dedicated interface circuits: RS485 interface, CAN bus interface, Ethernet interface and ADC analog signal acquisition interface. Each type of interface circuit is designed independently and does not interfere with each other. The specific selection is as follows:

[0044] ①RS485 interface: The MAX485 chip is used as the transceiver, which supports half-duplex communication. The communication rate can be adjusted by hardware DIP switch (1200bps-115200bps). It is compatible with the RS485 communication interface of energy storage devices and controllable loads in virtual power plants. It is used to collect data such as SOC (State of Charge, remaining power), charging and discharging current, charging and discharging power of energy storage devices, and operating status and load power of controllable loads.

[0045] ② CAN bus interface: The TJA1050 chip is used as the CAN transceiver, which supports the CAN2.0A / B protocol and has a communication rate of 500kbps. It is compatible with the CAN bus interface of distributed power sources (photovoltaic inverters, wind turbine controllers) and is used to collect data such as output power, voltage, current and operating status of distributed power sources.

[0046] ③ Ethernet interface: The DP83848 chip is used as the Ethernet physical layer chip, and the STM32F407 chip is used as the interface controller. It supports Gigabit Ethernet communication and is compatible with the Ethernet interface of the power grid side control equipment. It is used to collect data such as voltage, frequency and dispatching instructions from the power grid side.

[0047] ④ ADC Analog Input Interface: It adopts the ADS1115 16-bit high-precision analog-to-digital converter chip. The sampling frequency can be adjusted by hardware DIP switch, with an adjustment range of 10Hz-1kHz. It is compatible with the analog output interface of environmental monitoring equipment (temperature sensor, humidity sensor, light sensor) and is used to collect environmental parameters such as ambient temperature (-40℃~85℃), ambient humidity (0%~100%RH), and light intensity (0~10000lux).

[0048] (2) Synchronous acquisition implementation: A CD4060 counter is integrated in the acquisition module as a data acquisition synchronization trigger. The synchronization trigger is synchronized with the clock signal (100MHz) of the main control module and sends synchronous acquisition commands to various interface circuits to ensure the timing consistency of multi-channel data acquisition. According to actual measurement, the synchronization error of multi-channel data acquisition is ≤10μs, which meets the requirements of multi-source data synchronous acquisition in virtual power plants.

[0049] (3) Interface protection design: All types of interface circuits integrate TVS transient suppression diodes and self-resetting fuses. TVS diodes are used to suppress external surge voltages (maximum withstand voltage 60V), and self-resetting fuses are used to limit the interface loop current (maximum limit current 1A) to avoid damage to the acquisition module by abnormal external voltage and current, and improve the reliability of the acquisition module.

[0050] Hardware-based data preprocessing module:

[0051] The core function of the hardware-based data preprocessing module is to perform data denoising, outlier removal, format standardization, and synchronization alignment at the hardware level, without relying on software algorithms. This reduces data transmission latency and the computing power consumption of the main control unit. The specific implementation is as follows:

[0052] (1) Module composition: The preprocessing module consists of three parts: RC low-pass filter circuit, outlier detection logic unit and format conversion circuit. Each part of the circuit is integrated on the same PCB board and connected to the acquisition module and fusion processing module through high-speed signal lines. The preprocessing delay is ≤50μs, which is more than 60% lower than the traditional software preprocessing delay.

[0053] (2) RC low-pass filter circuit: A second-order RC low-pass filter circuit is adopted. The resistor is a 1kΩ metal film resistor and the capacitor is a 0.1μF ceramic capacitor. The cutoff frequency is set to 50Hz to filter out high-frequency noise (such as noise signals generated by electromagnetic interference) in the acquired data and ensure the smoothness of the acquired data. The output of the filter circuit is connected to an operational amplifier (LM324) for signal amplification and buffering to avoid signal attenuation.

[0054] (3) Outlier Detection Logic Unit: An LM339 quad voltage comparator is used as the hardware comparator, paired with a 74HC573 latch as the threshold register. The threshold register presets reasonable ranges for various types of data (which can be dynamically configured through the main control module, and the configuration command is transmitted through the SPI interface). For example, the reasonable range for the SOC of the energy storage device is 0%~100%, and the reasonable range for the output power of the distributed power source is 0~500kW. The hardware comparator compares the preprocessed collected data with the preset thresholds in the threshold register in real time. If the data exceeds the reasonable range, it is determined to be abnormal data, and the masking and replacement mechanism is immediately triggered. The replacement value is the average of the first 3 valid data to ensure the validity of the data.

[0055] (4) Format standardization and synchronization alignment: The 74HC165 shift register is used as the format conversion circuit to convert the different format data (RS485 differential signal, CAN bus CAN signal, ADC analog signal) collected by various interface circuits into a unified 8-bit parallel digital signal. At the same time, the converted digital signal is time-aligned according to the clock signal of the synchronous flip-flop to ensure that the data input to the fusion processing module is uniform in format and synchronized in timing.

[0056] Dedicated fusion processing module:

[0057] The dedicated fusion processing module adopts an FPGA+MCU heterogeneous computing architecture. Its core function is to realize parallel fusion computing of multi-source data, while also performing dynamic adjustment of the fusion strategy and verification and calibration of the calculation results. This solves the problems of poor real-time performance and high power consumption in traditional general-purpose processor fusion processing. The specific implementation is as follows:

[0058] (1) Chip selection:

[0059] ① The FPGA chip selected is the Xilinx Artix-7 series industrial-grade chip (model XC7A35T). This chip integrates 8 parallel computing units, with a maximum clock frequency of 150MHz. It supports hardware logic solidification and is used to solidify multi-source data fusion algorithms into hardware logic to achieve parallel fusion computing.

[0060] ② The MCU selected is the STM32L431 low-power industrial-grade microcontroller. This microcontroller has low power consumption (power consumption ≤100mW in normal working mode and ≤10mW in sleep mode), and integrates an SPI interface for communication with the FPGA chip to realize dynamic control of the fusion strategy and verification of calculation results.

[0061] (2) Hardware implementation of fusion algorithm: The spatiotemporal fusion network architecture (combining CNN convolutional neural network to extract spatial features and LSTM long short-term memory network to capture temporal dependencies) is solidified in the FPGA chip. The specific implementation process is as follows: the hardware logic code of CNN and LSTM is written in VHDL hardware description language, and the code is burned into the Flash memory of the FPGA chip. After the FPGA chip is powered on, the hardware logic is automatically loaded and the preprocessed multi-source data is fused in parallel. The fusion calculation delay is ≤200ms.

[0062] (3) Dynamic adjustment of fusion strategy: The MCU receives the intermediate fusion calculation results from the FPGA chip in real time through a high-speed SPI interface (communication rate ≥ 1Mbps), and simultaneously receives the virtual power plant operation status data (grid load, distributed power generation output fluctuations) transmitted by the main control module. Based on the operation status data, the fusion weight parameters (weight parameters range from 0 to 1) are dynamically adjusted. The adjustment command is transmitted to the FPGA chip through the SPI interface. The FPGA chip updates the fusion calculation logic according to the adjusted weight parameters. The weight adjustment response time is ≤ 10ms. For example: when the grid load fluctuates greatly, the fusion weight of the grid-side data is increased; when the distributed power generation output fluctuates greatly, the fusion weight of the distributed power generation data is increased.

[0063] (4) Hardware-based data calibration and algorithm switching:

[0064] ① The fusion processing module integrates an AD8421 instrumentation amplifier as a hardware data calibration unit. The calibration coefficients are preset in the MCU's Flash memory (which can be dynamically updated by the main control module, and the update command is transmitted through the UART interface). The calibration unit calibrates the fusion calculation results of the FPGA in real time. According to actual measurements, the calibration error is ≤0.5%.

[0065] ② Supports hardware switching of multiple fusion algorithms. A 3-bit hardware DIP switch is set on the PCB board of the fusion processing module. The corresponding fusion algorithm is selected by the DIP switch: DIP switch 001 corresponds to the weighted average fusion algorithm, DIP switch 010 corresponds to the Bayesian estimation fusion algorithm, and DIP switch 100 corresponds to the neural network fusion algorithm. The switching process does not require re-programming, and the switching response time is ≤5ms.

[0066] Main control module and bidirectional communication module:

[0067] The main control module, as the core control unit of the hardware system, is responsible for the coordinated control, data interaction, and command output of various modules; the two-way communication module is responsible for the data interaction between the hardware system and the virtual power plant control platform and terminal equipment, and is implemented as follows:

[0068] (1) Implementation of main control module: The main control module adopts an ARM architecture industrial-grade chip (model STM32H743). This chip integrates a Cortex-M7 core with a maximum clock frequency of 480MHz. It integrates a hardware encryption unit, SPI interface, UART interface and Ethernet interface for communication with various functional modules to form a closed-loop control.

[0069] ① Closed-loop control implementation: The main control module communicates with the preprocessing module and the fusion processing module through the SPI interface to receive preprocessed data and fusion calculation results; it communicates with the storage module through the UART interface to control data reading and writing; and it communicates with the communication module through the Ethernet interface to transmit fusion results and control commands, thereby realizing coordinated control of each module.

[0070] ② Hardware encryption implementation: The hardware encryption unit of the main control module supports the AES-128 encryption algorithm, and performs hardware encryption processing on the transmitted fusion results and control commands. The encryption key is stored in the encrypted storage area of ​​the chip (unreadable and unmodifiable) to ensure data transmission security.

[0071] (2) Implementation of bidirectional communication module: The communication module adopts a wired + wireless dual-mode redundancy design. The two communication modes work independently and can automatically switch according to the communication distance and environmental complexity. The specific implementation is as follows:

[0072] ① Wired communication: It adopts a gigabit Ethernet interface (sharing the DP83848 chip with the Ethernet interface of the acquisition module), supports TCP / IP protocol, and has a communication rate of ≥1Gbps. It is used for wired connection with the virtual power plant control platform to transmit a large amount of fusion results and historical data.

[0073] ② Wireless Communication: Integrates the SIM8200 5G module and the BC95 NB-IoT module. Both modules are connected to the main control module via an SPI interface. The main control module automatically switches the communication mode based on the communication distance (5G module communication distance ≥ 1km, NB-IoT module communication distance ≥ 10km) and environmental complexity (such as obstruction): When there is no obstruction and short-distance communication, it switches to 5G mode (communication rate ≥ 1Gbps); when there is obstruction and long-distance communication, it switches to NB-IoT mode (communication power consumption ≤ 50mW).

[0074] ③ Hardware adaptation of communication protocols: The hardware logic for protocol conversion is integrated into the communication module (implemented through 74HC373 latch), which supports hardware adaptation of multiple communication protocols such as Modbus, MQTT and RESTfulAPI. No software protocol conversion is required, and protocol compatibility with different terminal devices and control platforms can be directly achieved.

[0075] Tiered storage module:

[0076] The tiered storage module adopts a hardware architecture of cache + non-volatile storage. Its core function is to achieve classified storage of real-time data, intermediate data, and result data, ensuring that data is not lost. The specific implementation is as follows:

[0077] (1) Cache module: It adopts a DDR4 high-speed cache chip (model MT41K256M16TW-107), with a cache capacity of ≥4GB and a read / write rate of ≥2400Mbps. It is connected to the main control module and the fusion processing module through the DDR4 interface. It is used to temporarily store the raw data collected in real time and the intermediate data of fusion processing, reduce data read / write latency, and ensure the real-time performance of fusion computing. The cache module supports a refresh mechanism (refresh frequency of 15.625ms) to avoid data loss.

[0078] (2) Non-volatile storage module: adopts industrial-grade SSD (model Samsung 870EVO), storage capacity ≥128GB, connected to the main control module through SATAIII interface, used for persistent storage of fusion results and key operation data (at least 30 days); SSD integrates power loss protection circuit (using supercapacitor, capacity ≥5F), when a sudden power failure occurs, the supercapacitor provides temporary power to the SSD (power supply time ≥10s) to ensure complete data writing; at the same time, through the hardware-level bad block management mechanism (implemented by the SSD controller), bad blocks are automatically detected and shielded to ensure the reliability of data storage, and the data read rate is ≥500Mbps.

[0079] (3) Data read and write control: The main control module controls the data read and write of the storage module through the SATA controller and GPIO interface. It adopts the DMA direct memory access method to reduce the computing power consumption of the main control module and the data read and write latency is ≤10ms. At the same time, the data storage strategy is set: the real-time collected data is stored in the cache module for ≤1 hour and is automatically overwritten after the timeout; the fusion results and key operation data are stored in the SSD, classified by timestamp, and the oldest data is automatically deleted after 30 days to release storage space.

[0080] Hardware-based power management module:

[0081] The core function of the hardware-based power management module is to dynamically adjust the power supply voltage and current of each module, reduce the power consumption of the hardware system, and simultaneously achieve real-time monitoring and control of power consumption. The specific implementation is as follows:

[0082] (1) Module composition: The power management module consists of an intelligent power supply regulation circuit and a power consumption monitoring circuit, which are integrated at the power interface of the hardware substrate and connected to the power interface of each functional module.

[0083] (2) Intelligent power supply regulation: The TPS63070 power management chip is used as the core of the intelligent power supply regulation circuit. This chip supports wide voltage input (2.7V~20V) and can dynamically adjust the power supply voltage and current of each module according to the load demand signal transmitted by the main control module.

[0084] ① When the load rate (actual power consumption of each module / rated power consumption) is ≥30%, the power supply regulation circuit outputs the rated voltage (3.3V for FPGA, 3.3V for MCU, and 5V for main control module), and each module works normally;

[0085] ② When the load rate is less than 30%, the power supply regulation circuit automatically reduces the output voltage (2.5V for FPGA, 2.5V for MCU, and 3.3V for main control module) and switches to low power mode. At this time, the power consumption of the hardware system is reduced to less than 40% of the normal mode (normal mode power consumption ≤50W, low power mode power consumption ≤20W).

[0086] (3) Real-time power consumption monitoring: The INA219 current sensor is used as the core of the power consumption monitoring circuit. The sensor can collect the power supply current and voltage of each module in real time, calculate the actual power consumption of each module, and transmit the monitoring data to the main control module through the I2C interface. The main control module stores the power consumption data in the storage module and transmits it to the virtual power plant control platform through the communication module to realize the visualization control and optimization of power consumption.

[0087] Standardized modular design and redundant backup design:

[0088] (1) Standardized and modular design: Each functional module (acquisition module, preprocessing module, fusion processing module, communication module, and power management module) adopts a pluggable structure. The PCB board size of the modules is uniformly 100mm×80mm. A standardized 20-pin interface is set at the bottom of the module (the pin definitions are uniform: pins 1-5 are power pins, pins 6-10 are signal input pins, pins 11-15 are signal output pins, and pins 16-20 are ground pins), which are connected to the hardware base plate through the standardized interface. The number and type of modules can be flexibly increased or decreased according to the scale of the virtual power plant and the type of data source: for example, a small virtual power plant (≤10 data sources) can be configured with only 1 acquisition module and 1 preprocessing module; a large virtual power plant (≥20 data sources) can be configured with 2-3 acquisition modules. During the expansion process, there is no need to modify the original hardware architecture, reducing the cost of hardware upgrades and maintenance. At the same time, the module supports seamless docking with the existing virtual power plant control platform without the need to add additional adapter hardware. The docking process only requires configuring the communication protocol parameters.

[0089] (2) Redundancy Backup Design: A hardware-level redundancy backup unit is integrated on the hardware substrate to provide dual backup for the core circuits of the acquisition module and the fusion processing module. The hardware configurations of the main module and the backup module are completely identical. Fault detection and automatic switching are achieved through a 74HC138 decoder. When the main module fails (such as no data output from the acquisition module or abnormal calculation in the fusion processing module), the decoder detects the fault signal and immediately triggers a switching command. The backup module automatically switches to operation within 5ms to ensure the continuous operation of the hardware system. According to actual measurements, the mean time between failures (MTBF) of the hardware system is ≥20,000 hours.

[0090] Hardware integration and debugging steps:

[0091] The hardware integration and debugging steps of this invention are strictly performed according to the following process to ensure the stability and reliability of the coordinated operation of each module. The specific steps are as follows:

[0092] Step 1: Hardware assembly. Install each functional module to its corresponding position on the hardware substrate through standardized interfaces. Secure the modules with screws to ensure good contact between the modules and the substrate. Connect the power pins, signal pins, and ground pins of each module to the PCIe 4.0 high-speed bus of the substrate through high-speed signal lines. Use differential routing with a spacing of ≥2mm to avoid signal interference. Connect the power interface, ground interface, and debugging interface of the hardware substrate to complete the hardware assembly.

[0093] Step 2: Hardware power-on test. Connect the 220V AC power supply to the power interface of the hardware board. The voltage is converted to 24V DC output by the regulated power supply to power each module. Use a multimeter to test the power supply voltage and current of each module to ensure that the voltage and current meet the design requirements (3.3V for FPGA, 3.3V for MCU, and 5V for main control module). Check the grounding of each module to ensure good grounding (grounding resistance ≤1Ω) to avoid electromagnetic interference.

[0094] Step 3: Individual module debugging. Connect the computer and hardware system through the debugging interface to debug each functional module individually:

[0095] ① Data Acquisition Module Debugging: Connect to various data sources (energy storage devices, distributed power sources, environmental monitoring devices), adjust the sampling frequency, collect various types of data, and check the accuracy and synchronization of the collected data through computer software to ensure that the collected data is without loss or noise and the synchronization error is ≤10μs;

[0096] ② Preprocessing module debugging: Input simulated data containing noise and outliers, check the output data after preprocessing, and ensure that noise is effectively filtered out and outliers are correctly removed. The preprocessing delay should be ≤50μs.

[0097] ③ Debugging the fusion processing module: Input the preprocessed multi-source data, switch between different fusion algorithms, and check the accuracy and real-time performance of the fusion calculation results. Ensure that the fusion calculation delay is ≤200ms and the calibration error is ≤0.5%.

[0098] ④ Communication module debugging: Test the communication rate and stability of wired and wireless communication modes respectively to ensure that the communication rate of 5G mode is ≥1Gbps, the power consumption of NB-IoT mode is ≤50mW, and there is no packet loss (packet loss rate ≤0.1%).

[0099] ⑤ Storage module debugging: Test the read and write speeds of the cache module and SSD to ensure that the cache read and write speed is ≥2400Mbps and the SSD read speed is ≥500Mbps. Test the power loss protection function to ensure that data is not lost after power failure.

[0100] ⑥ Power management module debugging: Adjust the load rate of each module, check the power mode switching status, and ensure that the module automatically switches to low power mode when the load rate is <30%, and the power consumption is reduced to less than 40% of the normal mode.

[0101] Step 4: System integration and debugging. After completing the individual debugging of each module, the system is integrated and debugged. The hardware system is started, and each module works together to collect multi-source data from the virtual power plant, complete preprocessing, fusion calculation, storage and transmission, and run continuously for 72 hours. The operating status of each module, data processing latency, power consumption and communication stability are monitored to ensure that the system is fault-free, data processing is accurate and communication is smooth.

[0102] Step 5: Timing calibration and fault tolerance testing. Input timing calibration commands to the main control module through the debugging interface to calibrate the clock signals of each module and ensure timing synchronization between modules; simulate a main module failure (such as disconnecting the power supply of the acquisition module) to test the automatic switching function of the backup module and ensure that the switching time is ≤5ms; simulate external electromagnetic interference and voltage fluctuations to test the fault tolerance capability of the hardware system and ensure that the system operates normally without data loss or calculation abnormalities.

[0103] Step 6: Debugging and optimization. Based on the results of system integration and fault tolerance testing, optimize the wiring design between modules, adjust the filtering parameters of the preprocessing module and the weight parameters of the fusion processing module to reduce data processing latency and power consumption; repair the faults found during debugging to ensure that the hardware system meets the requirements of low latency and high precision fusion processing of multi-source data in the virtual power plant, and complete hardware integration and debugging.

[0104] IV. Verification of Implementation Results:

[0105] The virtual power plant multi-source data fusion processing hardware implemented in this invention was applied to a virtual power plant in an industrial park (including 10 photovoltaic inverters, 5 energy storage devices, 20 controllable loads, 3 sets of environmental monitoring equipment, and 1 set of grid-side control equipment) for a 30-day actual operation test. The test results are as follows:

[0106] 1. Real-time data processing: Multi-source data acquisition synchronization error ≤10μs, preprocessing delay ≤50μs, fusion calculation delay ≤200ms, overall data processing delay ≤300ms. Compared with the traditional software fusion processing solution based on general-purpose processors (overall delay ≥1000ms), the delay is reduced by more than 70%, meeting the real-time control requirements of virtual power plants.

[0107] 2. Data processing accuracy: After hardware calibration, the fusion calculation results have a calibration error of ≤0.5%, a data acquisition accuracy of ≥99.9%, and an abnormal data identification accuracy of ≥99.5%, ensuring the reliability of the fusion data and providing accurate data support for virtual power plant control decisions;

[0108] 3. System stability: The hardware system can run continuously for 30 days with a fault-free operating time of ≥720 hours and an average fault-free operating time of ≥20,000 hours. When the main module fails, the backup module switching time is ≤5ms, ensuring continuous system operation and meeting the requirements of 24-hour uninterrupted operation of the virtual power plant.

[0109] 4. Power consumption performance: Under normal operating conditions, the hardware system power consumption is ≤50W; under low power mode, the power consumption is ≤20W. Compared with traditional general-purpose processor solutions (power consumption ≥150W), the power consumption is reduced by more than 60%, which meets the development needs of low power consumption in industrial equipment.

[0110] 5. Scalability and compatibility: By adding a data acquisition module, it can easily adapt to the 15 new data sources without modifying the original hardware architecture during the expansion process; the hardware system can seamlessly connect with the existing virtual power plant control platform, with no packet loss in communication and smooth protocol adaptation, without the need to add additional adapter hardware, thus reducing deployment costs.

[0111] Test results show that the virtual power plant multi-source data fusion processing hardware implemented in this invention can effectively solve the defects of the existing technology, realize low-latency, high-precision, and high-reliability fusion processing of multi-source heterogeneous data in virtual power plants, and has good practicality, stability, and scalability, fully meeting the actual operation requirements of virtual power plants.

[0112] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A method for implementing hardware for multi-source data fusion processing in a virtual power plant, characterized in that: The hardware adopts a heterogeneous integrated architecture, building a dedicated processing hardware with multi-module collaboration to realize real-time acquisition, preprocessing, fusion calculation, and output control of multi-source heterogeneous data in the virtual power plant. The specific implementation steps include: S1: Construct a multi-source data acquisition module, integrate various dedicated interface circuits, adapt to different data interface types of distributed power sources, energy storage devices, controllable loads, grid side and environmental monitoring equipment in the virtual power plant, and synchronously acquire multi-dimensional heterogeneous data such as voltage, current, power, SOC, environmental parameters and dispatch instructions; S2: A hardware-based data preprocessing module is built, using dedicated filtering circuits and logic operation units to perform noise reduction, outlier removal, format standardization, and synchronization alignment on the collected multi-source data. This eliminates the need for software algorithm intervention, reducing data transmission latency and main control unit computing power consumption. S3: Design a dedicated fusion processing module, adopting an FPGA+MCU heterogeneous computing architecture, embedding multi-source data fusion algorithms into the FPGA chip to achieve parallel fusion computing. The MCU is responsible for the dynamic control of the fusion strategy and the verification of the calculation results, solving the problems of poor real-time performance and high power consumption of traditional general-purpose processors in fusion processing. S4: Integrates the main control module and the two-way communication module. The main control module, together with the preprocessing module, the fusion processing module, and the storage module, forms a closed-loop control. It receives the fusion processing results and outputs control commands. The communication module adopts a wired + wireless dual-mode redundancy design to realize data interaction with the virtual power plant control platform and terminal equipment. S5: Build a hierarchical storage module and adopt a hardware architecture of cache + non-volatile storage. The cache module is used to temporarily store real-time acquisition and fusion intermediate data, and the non-volatile storage module is used to persistently store fusion results and key operation data to ensure that data is not lost. S6: Complete hardware integration and debugging, integrate each module on the same hardware substrate through a high-speed bus, optimize the wiring design between modules, reduce signal interference, and ensure the stability and reliability of multi-module collaborative operation through hardware-level timing calibration and fault-tolerant design, ultimately achieving low-latency and high-precision fusion processing of multi-source data in the virtual power plant.

2. The implementation method according to claim 1, characterized in that, In step S1, the dedicated interface circuit integrated by the multi-source data acquisition module includes an RS485 interface, a CAN bus interface, an Ethernet interface, and an ADC analog signal acquisition interface. The ADC interface uses a 16-bit high-precision analog-to-digital converter chip, and the sampling frequency can be adjusted by a hardware DIP switch. The adjustment range is 10Hz-1kHz, which can adapt to the data transmission rate requirements of different types of data sources. At the same time, a data acquisition synchronization trigger is integrated to ensure the timing consistency of multi-channel data acquisition, with a synchronization error ≤10μs.

3. The implementation method according to claim 1, characterized in that, In step S2, the hardware-based data preprocessing module includes an RC low-pass filter circuit, an outlier detection logic unit, and a format conversion circuit. The outlier detection logic unit is implemented using a hardware comparator and a threshold register. By setting a hardware threshold, it can determine in real time whether the collected data exceeds a reasonable range and perform hardware-level masking and replacement processing on the outlier data. The replacement value is the average of the first three valid data. The preprocessing delay is ≤50μs, which is more than 60% lower than the software preprocessing delay.

4. The implementation method according to claim 1, characterized in that, In step S3, the FPGA chip adopts a high-performance industrial-grade chip, integrates at least 8 parallel computing units, solidifies the spatiotemporal fusion network architecture into hardware logic, realizes parallel fusion computing of multi-source data, and the fusion computing latency is ≤200ms. The MCU adopts a low-power industrial-grade microcontroller and communicates with the FPGA chip through a high-speed SPI interface. It receives the intermediate results of the fusion calculation of the FPGA in real time and dynamically adjusts the fusion weight parameters according to the virtual power plant's operating status. The weight adjustment response time is ≤10ms.

5. The implementation method according to claim 1, characterized in that, In step S4, the main control module adopts an ARM architecture industrial-grade chip and integrates a hardware encryption unit to perform AES-128 hardware encryption processing on the transmitted fusion results and control instructions to ensure data transmission security. In the bidirectional communication module, wired communication uses a gigabit Ethernet interface, while wireless communication integrates a 5G / NB-IoT dual-mode module. It can automatically switch communication modes according to communication distance and environmental complexity. The 5G mode communication rate is ≥1Gbps, and the NB-IoT mode communication power consumption is ≤50mW, which is suitable for the communication requirements of distributed deployment of virtual power plants. It also supports hardware adaptation of multiple communication protocols such as Modbus, MQTT and RESTfulAPI, without the need for software protocol conversion.

6. The implementation method according to claim 1, characterized in that, In step S5, the cache module uses a DDR4 high-speed cache chip with a cache capacity of ≥4GB and a read / write rate of ≥2400Mbps. It is used to temporarily store the raw data collected in real time and the intermediate data of the fusion processing, thereby reducing data read / write latency. The non-volatile storage module uses industrial-grade SSDs with a storage capacity of ≥128GB and supports power loss protection. Through a hardware-level bad block management mechanism, it ensures the persistent storage of fusion results and critical operational data, with a data read rate of ≥500Mbps.

7. The implementation method according to claim 1, characterized in that, In step S6, the high-speed bus adopts a PCIe 4.0 interface, the data transmission rate between modules is ≥8Gbps, the wiring design adopts differential wiring to reduce electromagnetic interference, and at the same time integrates a hardware-level redundant backup unit to provide dual backup for the core circuits of the acquisition module and the fusion processing module. When the main module fails, the backup module can automatically switch to operation within 5ms to ensure the continuous operation of the hardware system, with an average fault-free working time of ≥20,000 hours.

8. The implementation method according to claim 1, characterized in that, The hardware also integrates a hardware-based power management module, which adopts an intelligent power supply regulation circuit. It can dynamically adjust the power supply voltage and current of each module according to the load demand of virtual power plant data acquisition and fusion processing. When the load rate is lower than 30%, it automatically switches to a low-power mode, reducing power consumption to less than 40% of the normal mode. At the same time, it integrates a power consumption monitoring circuit to collect power consumption data of each module in real time and feed it back to the main control module to realize visualized control and optimization of power consumption.

9. The implementation method according to claim 1, characterized in that, In step S3, the fusion processing module also integrates a hardware data calibration unit, which performs real-time calibration on the fusion calculation results using preset calibration coefficients, with a calibration error ≤0.5%. It also supports hardware switching of multiple fusion algorithms. Depending on the data source type, the corresponding fusion algorithm can be selected through a hardware DIP switch without reprogramming.

10. The implementation method according to claim 1, characterized in that, The hardware adopts a standardized modular design, with each functional module being a pluggable structure. It connects to the baseboard through a standardized interface, allowing for flexible addition or removal of modules based on the scale of the virtual power plant and the type of data source. During expansion, no modifications to the original hardware architecture are required, reducing hardware upgrade and maintenance costs. It also supports seamless integration with existing virtual power plant control platforms without the need for additional adapter hardware.

Images