270results about How to "Reduce running time" patented technology

The invention discloses a twin network performance prediction method and device based on a graph neural network and a storage medium, and belongs to the technical field of communication networks, and the method comprises the steps: carrying out the preprocessing of the obtained historical information of a physical communication network, and obtaining a dependency graph data set; initializing the state of each node in the dependency graph data set to obtain the initial hidden state of each node; the initial hidden state is used to train a pre-constructed network performance prediction model to obtain a trained network performance prediction model, and the network performance prediction model obtains iteratively updated hidden state information and adjusts model parameters through cooperative work of an attention mechanism, a message passing neural network and a gating circulation unit; and inputting the obtained real-time information of the physical communication network into the trained network performance prediction model to obtain a performance prediction result, thereby improving the stability, accuracy and efficiency of performance prediction of the digital twin communication network.

The invention discloses an electric power system climbing demand evaluation and optimization method considering load side resources, and the method comprises the steps: providing a calculation method for maximum upward and downward climbing demands in different time periods, including a calculation method for deterministic and uncertain climbing demands, in order to effectively cope with the fluctuation and uncertainty of loads or new energy; a climbing demand optimization model which takes minimization of the uncertainty climbing demand as an objective function and takes constraint conditions such as load adjusting quantity and demand response into consideration is provided; a gradeability optimization model which takes minimization of comprehensive consumables of the power system as an optimization target and considers constraints such as node balance and line power flow is provided; modeling is carried out on AMPL software, a solver GUROBI is called for solving, and an electric power system climbing demand evaluation and optimization scheme is obtained. The comprehensive operation consumables of the power system can be effectively reduced, and the net load peak load shifting effect on the power system is achieved, so that the operation reliability of the power system is improved, and the dispatching pressure of the power system is relieved.

The invention relates to an attenuation self-adaptive intelligent fuel cell automobile optimal energy management system, an energy management strategy of the system is realized by adopting a multi-core controller, and the multi-core controller comprises a CPU0 (central processing unit 0) for estimating the SOH (state of health) of a fuel cell in real time and obtaining a fuel cell efficiency curve under the current SOH to update a fuel cell automobile model; the CPU1 is used for performing prediction based on machine learning to predict and obtain an optimal co-state variable; and the CPU 2 performs a real-time optimal energy management strategy. The invention ensures that the energy management strategy can be adjusted along with the change of the state of the fuel cell, ensures that the strategy is always consistent with the actual state of the fuel cell, improves the accuracy and adaptability of energy management, and is beneficial to prolonging the service life of the fuel cell and improving the overall performance of the vehicle. According to the invention, the execution efficiency of the energy management strategy is optimized, the real-time performance of the system is improved, and the optimal energy management of the fuel cell vehicle under the actual driving condition is ensured.

The invention discloses a GPGPU (General Purpose Graphics Processing Unit)-based singular value decomposition acceleration method and equipment, belongs to the technical field of computer technologies, and is used for solving the problems that when an SVD (Singular Value Decomposition) technology is applied to hardware, a relatively large calculation amount and a large amount of memory occupation can be caused, and different index strategies influence the parallel operation efficiency of the hardware, so that the hardware performance is poor. And the overall performance of hardware is reduced. The method comprises the steps of obtaining an index sequence of a to-be-decomposed matrix; generating a plurality of groups of index sequence pairs based on the index sequence; sending a kernel function kernel to the GPGPU, and traversing each group of index sequence pairs through each block in the kernel function kernel; decomposing matrix data corresponding to the first group of index sequences in the to-be-decomposed matrix through the queue in each block to obtain a first decomposed matrix; and decomposing matrix data corresponding to the second group of index sequences in the first decomposition matrix to obtain a second decomposition matrix, and so on until the last group of decomposition matrix is obtained, thereby obtaining a singular value decomposition result.

The invention discloses a false news detection method and system based on a partition filtering network, and the method comprises the steps: carrying out the fusion of news text features and comment text features, and obtaining event features; inputting the news text and the comment text into a dual-emotion acquisition module to obtain dual-emotion features; inputting the word vector of the comment text into a partition filtering network to obtain an initial vertical feature and an initial rumor feature; fusing the event feature with the initial vertical feature to obtain a corrected vertical feature; classifying the corrected vertical features to obtain a first classification result; fusing the initial rumor feature, the dual emotion feature and the corrected standing feature to obtain a corrected rumor feature; classifying the corrected rumor features to obtain a second classification result; and fusing the first classification result and the second classification result to obtain a final classification result.

The invention discloses a hierarchical power electronic circuit layout method based on reinforcement learning, which belongs to the field of integrated circuits and mainly comprises three levels of element sequence generation, layout optimization and wiring optimization. The method comprises the following steps: firstly, analyzing the power flow direction of a power electronic circuit, and reasonably arranging an element layout sequence, so as to realize basic optimization from a top layer; then, on the basis of the element sequence, a convolutional neural network is used for extracting layout features, a graph convolutional neural network is used for extracting graph features of a circuit connection relation, specific design rules of the power electronic circuit are embedded into a reinforcement learning reward system, and a deep Q network (DQN) is used for achieving layout of elements; through reinforcement learning, the intelligent agent can summarize experience in trial and error without human intervention, and finally a reward maximized layout result is generated.

The invention relates to an MMC (Modular Multilevel Converter) optimal zero-sequence signal calculation method for an overload operation demand, which belongs to the field of power electronic converter signal modulation, and comprises the following steps: constructing an optimal zero-sequence signal calculation operation domain; establishing an optimal zero sequence signal solving mathematical model; continuously iterating the active power and the reactive power, inputting the optimal zero-sequence signal to solve the mathematical model, solving to obtain a plurality of groups of maximum modulation signal margin values Mar, amplitudes and phase angles of corresponding zero-sequence signals, namely amplitudes and phase angles of optimal zero-sequence signals, outputting and recording, and respectively forming an OPA3 (P, Q) table and an OPAlf3 (P, Q) table; and fitting the table by adopting a function fitting mode to obtain a final fitting functional expression which is used for subsequently and quickly calculating the amplitude and the phase angle of the optimal zero-sequence signal. According to the invention, the operation time of modulation signal overmodulation caused by overload operation of the MMC can be greatly reduced, and the problem that the service life of a switching device is shortened due to modulation signal overmodulation is reduced.

The invention relates to the field of semiconductor lasers and deep learning, and discloses a DFB laser spectral parameter reverse design method based on a DNN model, which comprises the following steps: giving spectral parameters of a DFB laser, and obtaining a threshold spectrum of the DFB laser through a transmission matrix method; extracting characteristic parameters of the threshold spectrum; recording values of the characteristic parameters and spectral parameter values of the corresponding DFB laser to generate a data set; initializing the weight and the back propagation error of the DNN model by adopting a particle swarm algorithm; inputting the data set into the initialized model for training; giving a target threshold spectrum, and extracting characteristic parameters; and inputting the characteristic parameters into the trained model, outputting spectral parameters of the DFB laser, and realizing reverse design of the spectral parameters. The method provided by the invention can reduce the tedious operation of acquiring the threshold spectrum by using a spectrograph for many times, improves the accuracy, and reduces the used time and memory proportion, thereby optimizing the spectral characteristics more quickly and more accurately.

The invention discloses a high-voltage cascade energy storage system hardware-in-the-loop simulation model design method and simulation system, and the method comprises the steps: constructing a plurality of component simulation models needed in the hardware-in-the-loop simulation test process according to the hardware-in-the-loop simulation test content, and obtaining the time change characteristics of the parameters of the component simulation models; embedding the component simulation model into a large-step simulation model or a small-step simulation model based on the hardware-in-the-loop simulation test instruction and the time change characteristics of the parameters of the component simulation model; acquiring to-be-exchanged data between the large-step simulation model and the small-step simulation model; dividing the to-be-exchanged data into a plurality of groups according to time change characteristics of the to-be-exchanged data, and establishing a communication mechanism between the large-step simulation model and the small-step simulation model according to a grouping result; and performing hardware-in-the-loop simulation test, running the large-step simulation model and the small-step simulation model, and updating the step length of the large-step simulation model.

The invention relates to a point cloud real-time semantic segmentation automatic driving roadblock detection method and system, and the method comprises the following steps: obtaining the three-dimensional point cloud data of a scene through a laser radar, marking a corresponding label for each frame of the collected point cloud data, and obtaining a data set; the labels comprise fence objects, bicycles, buses, cars, trailers, trucks, motorcycles, sidewalks, grassland and walkers; a point cloud segmentation model is constructed, and the point cloud segmentation model adopts a deep learning architecture PCformer and comprises a point cloud embedding stage, a sharing stage and a segmentation head stage; a deep learning architecture PCform is realized by using Python, C + + and CUDA are used for acceleration, and after data set training is used, a trained point cloud segmentation model is obtained and is used for real-time detection and classification of roadblock information in an automatic driving scene. According to the invention, the operation time of the system can be obviously reduced, and real-time detection of the roadblock is realized.

The invention relates to the technical field of probe card testing, and particularly discloses a probe card full-automatic testing system and a testing method.The probe card full-automatic testing system comprises a testing box, a conveying frame is fixedly arranged in the testing box, the left side and the right side of the conveying frame penetrate through the left side and the right side of the testing box, and a discharging frame is fixedly arranged on the rear portion in the testing box; and one side of the discharging frame extends to the rear part of the testing box. By synchronously recording the test parameters of each probe card, detailed data support is provided for subsequent quality analysis, the surface quality of the probe cards is checked before testing, the defective probe cards can be removed in time, invalid test processing on the defective products is avoided, and for the probe cards produced in batches, the test efficiency is greatly improved. The pre-screening mode can greatly improve the overall test efficiency, not only saves the test time, but also improves the utilization rate of test resources, reduces the operation time and energy consumption of equipment, and reduces the operation cost.

The invention belongs to the technical field of retired battery sorting, and particularly relates to a retired battery secondary sorting method, system and equipment based on random charging fragments and a medium. The method comprises the steps of generating a single battery identification code, constructing a battery code database, and performing charging and discharging control on a single battery based on a preset charging and discharging strategy. And obtaining a time sequence sample data set of temperature, voltage and capacity change, converting the time sequence sample data set into two-dimensional image data and storing the two-dimensional image data, and training a prediction model through a two-dimensional convolutional neural network to estimate the maximum available capacity of the battery so as to realize secondary sorting of the battery. The system comprises a code scanning module, a charging and discharging control module, a data acquisition module, a data processing module and a sorting module. The method can reduce test time and energy consumption, improve battery sorting efficiency and precision, reduce memory occupation and rear-end model operation time, and is suitable for efficient and accurate sorting of retired batteries.

The utility model discloses a holding furnace with a holding furnace cover, which comprises a bottom plate, a holding furnace body is arranged at the top of the bottom plate, a furnace cover body is in threaded connection with the top of the holding furnace body, two handles are fixedly connected with the top of the furnace cover body, and moving wheels are fixedly connected with the bottom of the bottom plate at equal intervals. A rotating rod is rotationally connected between the bottom of the inner wall of the heat preservation furnace body and the bottom of the furnace cover body, a motor is arranged at the top of the furnace cover body, the bottom of an output shaft of the motor is fixedly connected with the top of the rotating rod, and a stirring assembly is arranged on the outer side of the rotating rod. According to the heat preservation furnace with the heat preservation furnace cover, the novel stirring assembly is arranged, liquid can be more fully mixed in a revolution and rotation stirring mode, mixing is more uniform and thorough, the mixing efficiency is improved, the stirring time is shortened, and the uniformity and stability of alloy are improved.

The invention discloses a heterogeneous platform calculation task scheduling method based on a machine learning algorithm. The scheduling method comprises the following steps of 1, analyzing and collecting static characteristics and dynamic characteristics of a data set program and hardware characteristics of an operation platform; 2, selecting a proper machine learning scheduling model and optimizing hyper-parameters, and evaluating the performance of the scheduling model through a plurality of indexes; and 3, selecting optimal hardware equipment during operation according to a model prediction result to realize efficient task scheduling. The invention also discloses a scheduling system for realizing the scheduling method and a corresponding application, and has a wide application scene.

The invention discloses a method for evaluating the service life of lubricating grease for a sliding friction pair in a high and low temperature environment, which comprises the following steps: a first-stage test: adjusting the lubricating grease at a set temperature, operating the lubricating grease under a rated load to obtain first-stage test data, and judging whether the lubricating grease passes the first-stage test according to the first-stage test data; and if the lubricating grease passes the first-stage test, further performing a second-stage test on the lubricating grease on the basis of the first-stage test: operating under a load higher than a rated load to obtain second-stage test data, and judging the service life grade of the lubricating grease according to the second-stage test data. According to the method for evaluating the service life of the lubricating grease for the sliding friction pair in the high-low temperature environment, the service life of the lubricating grease for the sliding friction pair is tested, the service life testing speed of the lubricating grease for the sliding friction pair is high, evaluation is accurate, and the method can be used in the high-temperature environment and the low-temperature environment; the application range of the lubricating grease life evaluation method is expanded.

According to the elevator automatic permission opening method and system, after one-time registration of a mobile device APP, automatic permission opening of an elevator can be achieved, extra physical media do not need to be carried, the use convenience is greatly improved, a cloud server and a property background management system jointly check user information, authenticity and accuracy are ensured, and the user experience is improved. Meanwhile, specific secret key identity verification is adopted, the system safety is enhanced, a data encryption and verification mechanism in the Bluetooth communication process is adopted, safe transmission of user information is further guaranteed, a property background management system can manage the user information and authority in real time, the traditional process is simplified, and an elevator main control system automatically records and updates the use condition of a user; according to the method, the mobile device APP, the cloud server and the elevator master control system are integrated, intelligence and automation of elevator control are achieved, invalid operation and waiting time are shortened, energy consumption and emission are reduced, and intelligent development of buildings is promoted.

The invention relates to the technical field of container processing, and provides a container processing method based on a host machine, the host machine, a system, equipment and a medium so as to improve the virtual system creation efficiency, the invention provides the container processing method based on the host machine, the host machine is deployed with a virtualization service container, and the virtual system creation efficiency is improved. The host machine is deployed with the virtualization service container, the virtualization service container runs a virtualization service starting monitoring mode so as to monitor a creation request triggered by a client of each operating system container, and the creation request is used for requesting to create a virtual operating system; when a plurality of creation requests are monitored, loading an operating system mirror image corresponding to each creation request based on a mirror image storage path of the operating system container; and correspondingly creating each virtual operating system in the virtualization service container in parallel based on the loading resources on the host machine and the operating system mirror image corresponding to each creation request.

The invention relates to a coarse-grained reconfigurable processor operator memory data multiplexing design method, which comprises the following steps of: acquiring operator related information of all operators in a to-be-processed model, then determining an operator fusion strategy according to the operator related information of all the operators, and finally fusing all the operators according to the operator fusion strategy, so that the memory data multiplexing design of the coarse-grained reconfigurable processor operator is realized. And the fused operator is processed. According to the method, the fused operators are uniformly processed on the on-chip memory, or the on-chip memory is reused to more quickly process the intermediate data in the memory, so that the repeated carrying of the intermediate data can be reduced, the data carrying power consumption of the whole network operation can be reduced, the utilization rate of the computing unit can be improved, and the whole operation time can be shortened; and the data processing efficiency is improved.

The invention relates to the technical field of plastic injection molding, in particular to a plastic injection molding machine which comprises an injection molding machine, and through an energy-saving type base anti-cracking mechanism, when a mold mounting base A and a mold mounting base B of which the outer surfaces are provided with injection molds are attached to each other, the mold mounting base A and the mold mounting base B can be automatically fixed; during cooling, the injection molding opening A can be automatically sealed through the injection molding rod-free injection molding mechanism, so that plastic rods are prevented from appearing on a finished product, and through the technical scheme, the production steps are reduced, and meanwhile, a hydraulic air cylinder does not need to be started all the time in the injection molding process; when a plastic injection molding machine is used, in order to prevent mutually attached molds from being separated due to internal force, a hydraulic air cylinder and hot melt plastic located in a mold cavity need to be started all the time for cooling; and hot melt plastic which is located in the injection molding opening and connected with the injection molding opening is cooled along with the injection molding opening, so that a plastic rod is formed on the outer surface of a finished product.

The embodiment of the invention provides a yaw angle rapid optimization method and system for active yaw control, and the method comprises the steps: setting an initial yaw matrix formed by all fan yaw angles as a zero matrix, calculating the corresponding power of each fan traversing a preset fan yaw angle candidate set, and calculating the corresponding power of the fan yaw angle candidate set; selecting a first fan yaw angle corresponding to the maximum power of each fan to form a first yaw matrix of all fans; for each fan, after different offsets are added on the basis of the first fan yaw angle, the corresponding power is calculated, the offset value is in a preset offset angle candidate set, and a second fan yaw angle corresponding to the maximum power of each fan is selected to form a second yaw matrix of all the fans; and for each fan, different correction disturbance values are added on the basis of the second fan yaw angle, the power of the wind field is calculated, and a third fan yaw angle corresponding to the maximum power of each fan is selected to form a third yaw matrix of all the fans.

The invention provides a power line impedance matching method and system based on space division, and belongs to the technical field of power line impedance matching based on deep learning. Acquiring a power line terminal load value; based on a database query function, querying capacitance and inductance element values and an optimal target function value during optimal matching corresponding to the obtained power line terminal load value in a database; wherein the database stores the optimal values of capacitance and inductance devices under the optimal matching condition corresponding to different power line terminal load values obtained by using a deep reinforcement learning algorithm model for optimization. According to the method, the program running time is effectively shortened, the problem of unsatisfactory agent optimization caused by too large search space is improved, and the problem of poor optimization result caused by insufficient training of the agent is avoided; the search space is reduced, and the search accuracy is improved; the space is subjected to non-uniform segmentation, and compared with multi-agent distributed optimization based on uniform segmentation, the search effect of the agents is obviously better.

The invention discloses a thermal management system for a hybrid power vehicle and a use method of the thermal management system. The system comprises a refrigerant system, a first cooling liquid system and a second cooling liquid system, the refrigerant system is used for absorbing heat through a first heat exchanger by adopting a heat pump technology so as to heat a passenger compartment and / or a battery, or dissipating heat through the first heat exchanger so as to refrigerate the passenger compartment and / or the power battery; the first cooling liquid system is used for dissipating heat generated when a heat source component of the electric drive system works into the atmosphere or through the first heat exchanger, absorbing heat through the first heat exchanger and dissipating the absorbed heat into the atmosphere, and absorbing heat through the second heat exchanger and dissipating the absorbed heat into the atmosphere. The absorbed heat is dissipated through the first heat exchanger; and the second cooling liquid system is used for dissipating heat generated by working of the engine into the atmosphere or dissipating the heat through the second heat exchanger.

The invention discloses a viewpoint planning method based on deep reinforcement learning. The method comprises the following steps: acquiring a sheltered image and a camera attitude and obtaining state information; inputting the state information into a trained DQN network to obtain a camera motion action at the next moment; and controlling the mechanical arm to move to adjust the camera position. In order to improve the training efficiency of the DQN network, a training strategy of global trend guide learning is creatively provided. The global trend guided learning strategy refers to a method for forming a global trend by recording viewpoint areas of successfully detected fruit stems in the same scene in a learning process, and then guiding a network learning process by using the global trend. According to the invention, viewpoint planning is realized by a brand-new technical means, so that the problem of shielding of the grape picking robot in the picking operation process is solved. According to the method, the viewpoint planning time is greatly shortened under the condition that the high picking success rate is kept, and a new technical scheme is provided for the field of active vision technology application.

The utility model relates to the technical field of liquefied natural gas (LNG) receiving stations, in particular to a system for mutually utilizing liquid nitrogen cold energy and boil-off gas (BOG) heat energy, which comprises a BOG direct output system, a BOG recondensation system and a liquid nitrogen system, the BOG direct output system comprises an LNG storage tank, the LNG storage tank is sequentially connected with a first BOG compressor and a first heat exchanger for providing a cold source by the liquid nitrogen system through a BOG pipeline, and the first heat exchanger is connected with a second BOG compressor through a BOG pipeline. The BOG outlet end of the first heat exchanger is communicated with a natural gas pipe network; the BOG recondensation system comprises a second BOG compressor, a second heat exchanger, a recondenser, a high-pressure pump and a first gasifier which are sequentially connected with the LNG storage tank through a BOG pipeline, the liquid nitrogen system provides a cold source through the second heat exchanger, the recondenser, the high-pressure pump and the first gasifier, and the outlet end of the first gasifier is communicated with a natural gas pipe network; and the in-tank low-pressure pump is connected with the re-condenser through a low-pressure output pipeline. The system is safe and reliable, energy stored by media of the system is mutually utilized, electricity charge is saved, noise is reduced, efficiency is improved, and compared with the prior art, the system has obvious advantages.

The invention relates to the technical field of database management, and provides a database external sorting method, system and equipment, and the method comprises the steps: segmenting original to-be-sorted data into locally ordered data blocks in an external sorting segmentation stage; performing data format configuration on the locally ordered data blocks obtained by segmentation, and establishing a pre-cache space for the data blocks subjected to data format configuration; and in a merging stage of external sorting, locally ordered data blocks are merged in a pre-cache space through an optimized septicem tree algorithm in each round of merging process. According to the database external sorting method, system and equipment, the overall use efficiency of the cache and the disk can be remarkably improved, repeated I / O is reduced, the performance is improved, and the running time is shortened.

The utility model relates to the technical field of medical product production, and discloses a powder-liquid double-chamber bag sterilizing and drying system which comprises a sterilizing and drying chamber, a fan, a circulating air pipe and an air-water separator, the sterilizing and drying chamber, the gas-water separator and the fan are communicated through circulating air pipes and form a circulating loop, the gas-water separator is arranged on the air outlet side of the sterilizing and drying chamber in the circulating loop, and the gas-water separator is used for separating and discharging moisture in clean air in the circulating loop. According to the utility model, the circulation loop of the clean air is constructed, the dehumidification equipment capable of removing moisture in the clean air is arranged in the loop, and the moisture in the drying chamber is continuously brought out by utilizing the continuously circulated clean air, so that the drying device has the advantages of high drying degree, short drying period, reduced production cost and the like.

A reliability prediction method for a power field effect transistor belongs to the technical field of power MOSFET reliability modeling. The present invention is to solve the problem of low work efficiency caused by manually adjusting model hyperparameters in the existing deep learning model for reliability prediction of power devices. It includes: S1, collecting sensor data as initial data and performing data processing on the initial data; S2, optimizing the hyperparameters of the Transformer model using a genetic algorithm to construct a GA-Transformer model; S3, transmitting the data obtained by S1 to the GA-Transformer model obtained by S2, evaluating the accuracy of the prediction based on performance indicators, predicting the remaining service life RUL, and outputting the reliability prediction result. The present invention is used for power MOSFET reliability prediction.

The invention discloses a thermal management method, a unit, a computer readable storage medium and an energy storage system.The thermal management method is used for the unit, the unit is in communication connection with a battery management system, and the thermal management method comprises the steps that battery cell temperature information, sent by the battery management system, of battery cells in a target battery pack is obtained; determining cold and hot requirements of the battery according to the battery cell temperature information; and controlling a target refrigerator in the unit to operate in a target mode according to the battery cold and hot demand. The thermal management method can accurately respond to temperature changes, more accurate temperature control is achieved, unnecessary operation time is shortened, system energy consumption is reduced, the service life of devices is prolonged, and maintenance cost is reduced. Moreover, the method is used for a unit, is clear in division of labor with a battery management system, reduces the workload of the battery management system, and can make a more timely response to heat management.

The invention provides an architecture method and a control device for a control function of a synchronous servo controller, and solves the technical problems of poor efficiency, universality and expandability of a control process of an existing synchronous servo controller. The method comprises the following steps: packaging a control process for a motor body to form a motor control module, and packaging a control process for a peripheral circuit to form a service task; the execution of the control module is triggered through the EPWM interrupt service, and counting statistics of the interrupt service is carried out in the execution process; task scheduling cycles are formed according to counting statistics of the interrupt services, all the task scheduling cycles are coordinated to form a task scheduling process, and repeated execution of the task scheduling process is formed through a main control process. The running time of the interrupt service program is shortened, and the risk that the running time of the interrupt program exceeds the interrupt period is reduced. A scheduling architecture for executing the control process in a time slice polling manner is formed, and the operation efficiency, universality and expandability of control software of the synchronous servo controller are improved.

The present invention discloses a heat dissipation device and control method for a photovoltaic inverter, wherein the heat dissipation device comprises a heat dissipation chamber housing, a control unit, a fan group, a radiator and a temperature collection unit, the control unit is electrically connected to the fan group, the radiator and the temperature collection unit in a one-to-one correspondence; the radiator and the fan group are located on the heat dissipation chamber housing, the temperature collection unit and the fan group are both arranged at any position close to the radiator, and the radiator is arranged adjacent to the power device of the photovoltaic inverter; wherein at least two fans are arranged in the fan group. The present invention collects the temperature of the radiator through the temperature collection unit, thereby reducing the loss of the fan and prolonging the life of the fan; and by providing multiple fans for air cooling and heat dissipation, when one of the fans fails, because there are other fans that can perform air cooling and heat dissipation, there is no need to directly reduce the output power of the photovoltaic inverter, thereby improving the power generation efficiency and reliability of the photovoltaic inverter.