109 results about "Parallel channel" patented technology

The invention discloses a big language model reasoning optimization method and system based on dynamic video memory compression and memory isomerism, and intelligent management of video memory resources is realized by integrating a dynamic compression strategy of KV Cache and a memory parallel architecture. The method comprises the following steps: 1) analyzing the spatial-temporal characteristics of the KV Cache in real time, adaptively selecting a quantization compression algorithm, a rarefaction algorithm or a low-rank decomposition algorithm, performing hierarchical storage based on attention head importance scores, keeping high precision of a core head, and implementing low-bit quantization on a secondary head; (2) the compressed inactive data are divided into a plurality of data blocks to be stored in a system memory, a parallel data channel group is established according to the number of physical channels, the compressed blocks are concurrently read through multiple channels during loading, and parallel decompression of a sparse matrix is accelerated through a GPU tensor core; and 3) constructing a KV Cache multiplexing mechanism and a parallel channel, and parallelizing a compression / decompression process and model calculation by adopting a hardware acceleration compression and asynchronous pipeline mechanism.

The invention belongs to the field of intelligent traffic systems, and relates to a multi-mode space-time traffic flow modeling method supporting large-scale road network real-time prediction, and the method comprises the steps: firstly designing a space-time prediction framework facing a dynamic traffic network, and then carrying out the training to obtain a final prediction model; the space-time prediction framework comprises a data embedding layer, a space-time coding module and a deep modeling and output module based on an expert hybrid mechanism; the data embedding layer comprises two parallel channels of time embedding and spectral domain space embedding and a time-space data fusion module; the space-time coding module comprises a block-level sparse time attention module, a space attention-message passing module and a weighted fusion layer which are parallel; the deep modeling and output module based on the expert hybrid mechanism comprises an MoE dynamic expert modeling module, a full-connection mapping module, a jump connection layer and an output layer; according to the design, the response speed, the prediction precision and the cross-regional adaptive capacity of the model in a high-heterogeneity scene are improved.

The invention relates to the field of target detection, in particular to a UMFNet-YOLO-based joint detection algorithm under a low-light condition, which adopts a multi-scale decomposition and high-frequency-low-frequency feature collaborative enhancement strategy, separates image details from global features through high / low-pass filtering, realizes multi-resolution feature complementation in combination with dynamic weight distribution, and improves the detection accuracy. Target contour and texture information degraded in severe weather can be effectively recovered; a parallel channel compression and spatial semantic enhancement module is designed, rain and fog noise interference is suppressed through cross-dimensional adaptive feature screening, and significance expression of a key target is enhanced; an image enhancement network UMFNet is cascaded with YOLOv8, and strong correlation between feature expression and a detection task is ensured by synchronously optimizing an enhancement module and detecting network parameters through back propagation. Compared with a baseline YOLOv8 detector, the method has the advantages that the average precision (mAP50) is obviously improved, and the method is obviously superior to a traditional enhancement-detection separation scheme. The method can be applied to robust target detection scenes in complex environments such as automatic driving and security monitoring.

The invention discloses a bearing fault diagnosis method based on crown porcupine optimization variable mode decomposition. A bearing vibration signal is collected; processing the bearing vibration signal by using crown porcupine optimization variable mode decomposition to obtain an intrinsic mode function IMFs; converting the IMFs into a two-dimensional wavelet time-frequency graph through continuous wavelet transform, and performing feature fusion through a convolutional neural network to form a new two-dimensional wavelet time-frequency graph; the method comprises the following steps: constructing a feature fusion network which takes a residual network and a Swin Transform network as parallel channels; and inputting the two-dimensional wavelet time-frequency graph into the feature fusion network for learning and fault diagnosis. The bearing vibration signal is subjected to variable mode decomposition and two-dimensional wavelet time-frequency graph processing, and the convolutional neural network and the parallel channel feature fusion network are combined, so that high-accuracy diagnosis of bearing faults under the conditions of limited samples and variable working conditions is realized. The method has high practicability and accuracy, and the reliability of fault diagnosis can be effectively improved.

The provided is an anti-radiation low-latency chip for neural network inference acceleration. The following steps are included: deploying a neural network inference acceleration chip at a frontend of a detector, using a streaming architecture to correspond each pipeline stage to each major neural network layer, balancing a limited on-chip memory resource and support for a large-sized input, and constructing layer parallelism, channel parallelism, and convolution kernel parallelism. This application proposes to deploy an anti-radiation, low-latency, and efficient chip for convolutional neural network (CNN) inference acceleration is deployed at the frontend of the detector to improve an intelligence level of future detector hardware. To this end, a fine-grained streaming architecture, and fine-grained storage management, flexible compression and quantization, and anti-radiation digital chip design technologies are proposed to achieve a high throughput and a low on-chip memory consumption while achieving anti-radiation and low-latency inference.

The invention discloses a dual-mode communication method and system based on link redundancy check and dynamic hybrid routing, and relates to the technical field of electricity meter reading, and the method comprises the steps: constructing a dual-link parallel channel through HPLC and HRF, and carrying out data synchronization and consistency comparison; when the link is abnormal, dynamically electing an agent coordination node and constructing a cross-protocol mixed link; and optimizing communication path selection and interference suppression by using an intelligent routing algorithm and a noise recognition mechanism. According to the method provided by the invention, the reliability and data transmission efficiency of a communication system are remarkably improved through double-link redundancy check and a hybrid link dynamic reconstruction mechanism. And a parallel acquisition and path optimization method is adopted, so that the meter reading time is shortened, the link switching delay is reduced, and the real-time performance and the resource utilization rate are enhanced. The system has good environment adaptability and can cope with topological changes caused by a complex power grid structure and distributed energy access.

One or more embodiments of the invention provide a retrieval enhancement generation method and device, equipment and a storage medium. According to the method, a plurality of parallel channels are utilized to retrieve a query request from a plurality of dimensions; the recall content of each channel is pre-filled according to the time sequence of the recall content returned by each channel, the pre-filling result of each channel is stored in a cache device, and the pre-filling result of the channel is obtained according to the pre-filling result cached by the previous channel returning the recall content; and finally, obtaining an output result of the large language model according to a pre-filling result of the channel which finally returns the recall content. The pre-filling is performed according to the time sequence of the recall content returned by the channels, so that the waiting time in the process of one-time retrieval enhancement generation request is reduced, and the pre-filling result of the current channel is calculated according to the pre-filling result cached by the previous channel returning the recall content, so that repeated calculation in the pre-filling stage is avoided.

The present invention discloses a dual-mode communication method and system based on link redundancy check and dynamic hybrid routing, which relates to the technical field of electricity metering, including building a dual-link parallel channel through HPLC and HRF to perform data synchronization and consistency comparison; when the link is abnormal, dynamically electing an agent coordination node and building a cross-protocol hybrid link; and optimizing communication path selection and interference suppression using an intelligent routing algorithm and a noise identification mechanism. The method of the present invention significantly improves the reliability and data transmission efficiency of the communication system through a dual-link redundancy check and a hybrid link dynamic reconstruction mechanism. The parallel collection and path optimization method is adopted to shorten the meter reading time, reduce the link switching delay, and enhance the real-time performance and resource utilization. The system has good environmental adaptability and can cope with the topology changes brought about by complex power grid structures and distributed energy access.

The invention provides an instant image printing task scheduling method and system based on dynamic queue management, and relates to the technical field of task scheduling, and the method comprises the steps: deploying a three-level distributed architecture, and sensing the multi-dimensional attributes of production data and video data in real time; based on the system load threshold value and the real-time requirement, the dynamic priority of the data is calculated through a multi-factor weighting formula, and priority factors comprise data freshness, event emergency degree and transmission stability; dynamically recombining the data queue, and dividing the data queue into a real-time analysis pool, a quasi-real-time buffer pool and a background processing pool; adaptive data preprocessing is executed, wherein format standardization, redundancy compression and noise filtering are included; constructing a parallel channel and transmitting data through the optimal path; sHA-256 feature code verification and full link resource recovery are carried out; the system comprises a distributed state sensing module and the like. According to the method, the real-time performance of multi-source data fusion is improved, key events can be processed preferentially, and the factory monitoring efficiency and the resource utilization rate are improved.

The invention discloses a Chinese fraudulent language detection method and system based on two channels, and the method comprises the steps: obtaining text data, generating a word vector through a BERT model, inputting the word vector into two parallel channels for processing, and enabling the first channel to generate a first feature vector through a capsule network; and the second channel combines a bidirectional gating loop unit, a text convolutional neural network and a multi-head attention mechanism to generate a second feature vector, after the two feature vectors are fused, the two feature vectors are input into the full-connection layer for feature mapping, and a Chinese fraudulent language detection result is output. Wide application prospects are realized.

The invention provides a transaction data low-delay credible acquisition method, device and equipment, and relates to the technical field of data acquisition, and the method comprises the steps: building a plurality of first data channels for collecting data of a plurality of transaction data sources, and obtaining multi-channel data; preprocessing the multi-channel data to obtain preprocessed data; dynamically adjusting bandwidth allocation of the plurality of first data channels according to the importance of the preprocessed data to obtain a plurality of second data channels; distributing the loads of the plurality of second data channels to obtain the loads of the plurality of second data channels; and transmitting the preprocessed data according to the plurality of second data channel loads. The technical problem that data transmission in a high-frequency transaction environment has a bottleneck effect and a large amount of data cannot be processed in time due to the fact that data transmission depends on a single channel or a limited parallel channel can be solved, the technical target of efficient parallel processing of multi-source transaction data is achieved, and the data transmission efficiency is improved. The technical effects of reducing the data transmission delay and improving the system resource utilization rate are achieved.

The invention provides a digital micro-fluidic system, and relates to the technical field of micro-fluidic systems. The digital micro-fluidic system provided by the invention comprises a micro-fluidic chip, and liquid drops are moved, combined and split in the horizontal direction through voltage difference; the heating plate is positioned at the bottom of the micro-fluidic chip and is used for heating the micro-fluidic chip; the magnetic control module is positioned at the bottom of the heating plate and is used for adsorbing magnetic beads on the micro-fluidic chip so as to purify liquid drops; the micro-fluidic chip is provided with a plurality of parallel channels arranged in the first direction, a plurality of electrode units used for bearing liquid drops are arranged in the parallel channels in an array mode, and the corresponding electrode units in the parallel channels are connected in parallel and electrically connected with the control circuit. The problems that the practical application of DMF is limited due to the fact that the flux of an existing passive matrix DMF is too small, the electrode driving force of an active matrix DMF is small, cost is high and the like can be solved, accurate synchronous control over multiple groups of liquid drops is achieved, and a new solution is provided for large-scale low-cost digital micro-fluidic application.

The invention discloses an image compression method based on a JPEG-LS (Joint Photographic Experts Group-Least Squares) parallel optimization algorithm, which comprises the following steps: calculating local gradient values of to-be-coded data and then merging to obtain a context index Q value, the to-be-coded data being a prediction error of a current pixel; grouping the data to be coded according to the context index Q value; if the to-be-coded number contained in the key group affects the parallelism degree, a controllable distortion value is introduced into the pixel value of the to-be-coded data by using an equalization algorithm, and the group serial number corresponding to the data in the key group is corrected; deploying processing units of parallel channels with the number equal to that of the groups, and scheduling the grouped data to be coded to the processing units to realize pipeline processing; integrating each group of data into coded and compressed data through code stream splicing; the image compression method based on the JPEG-LS parallel optimization algorithm is used for image compression, the system data size is low, and the satellite-ground transmission bandwidth pressure is small.

The invention relates to the technical field of optical computing, in particular to a large-scale reconfigurable photon large matrix model and architecture. Wherein the model comprises a mixed light calculation sub-model; the large-scale reconfigurable photon large matrix model comprises an acquisition module, a parallel input module and an integration module, and the acquisition module is used for acquiring a first large matrix and an input vector which need to be calculated; the parallel input module is used for inputting the first large matrix and the input vector into the mixed light calculation sub-model of each parallel channel in parallel based on the determined number of the parallel channels to obtain an output vector of each parallel channel; and the integration module is used for integrating the output vectors of the parallel channels to obtain a target output vector. According to the scheme, under the condition that high calculation precision and high data flux are considered, a large-scale general optical matrix operation new normal form is achieved, and therefore a complex optical general calculation task is supported.

The invention discloses a ZYNQ neural network acceleration system for aluminum surface micro-defect detection, an LADFastDet network for aluminum surface micro-defect detection is deployed on a ZYNQ platform, a PL end of the ZYNQ platform comprises an accelerator module, the accelerator module comprises a convolution IP core and a sampling IP core, all convolution in the LADFastDet network structure is designed into the convolution IP core, and the sampling IP core is used for sampling the convolution IP core. The sampling IP core is designed through up-sampling and maximum pooling, the convolution IP core is provided with multiple read-write parallel channels and interfaces, and the sampling IP core is provided with multiple read-write interfaces; the PS end is responsible for configuration of an LADFastDet network structure, calling of an accelerator module and management of an overall control process, and the PL end calls and executes functions of the convolution IP core and the sampling IP core according to the configuration of the PS end; according to the invention, the precision and speed of aluminum surface micro-defect detection under a limited resource ZYNQ platform are improved.

The invention relates to the technical field of power load power prediction, and provides a power load power prediction method and system for a sewage plant and a medium, and the method comprises the steps: obtaining the feature data of the sewage plant, including meteorological data, date, water quality data and historical power load power data; the constructed load prediction model comprises a feature extraction layer, an attention fusion layer, a GRU layer and a full connection layer; the feature extraction layer is provided with three parallel CNN channels and an attention fusion layer, and each channel sequentially comprises a plurality of convolution modules. Wherein the first channel inputs meteorological data and date, the second channel inputs water quality data, and the third channel inputs historical power load power data. The output of the three channels enters the GRU layer after being processed by the attention fusion layer, and the output of the three channels passes through the full connection layer to obtain a predicted value of the power load power of the sewage plant. According to the method, multi-type feature information can be effectively fused, and the accuracy of power load power prediction of the sewage plant is improved.

The invention belongs to the technical field of cryptography and secure multi-party computing, and provides a homomorphic encryption multiplication triple generation method and system based on the Chinese remainder theorem.A data application party and a data provider generate private random vectors respectively, and the private random vectors are decomposed into a plurality of component matrixes by means of the Chinese remainder theorem; the data provider executes ciphertext multiplication and addition operation in the corresponding parallel channels according to the number of the component matrixes, the ciphertext of each channel is calculated based on a BFV homomorphic encryption algorithm, and the ciphertext result is sent to the data application party; and the two parties decrypt each ciphertext by using the private key to obtain a plaintext component matrix, reconstruct through the Chinese remainder theorem to obtain a plaintext vector, and calculate a local triad component based on the plaintext vector to generate a triad, and the triad meets the multi-party multiplication constraint. According to the method, the Chinese remainder theorem and the homomorphic encryption algorithm are ingeniously combined, efficient and extensible multiplication triple generation is achieved, and the efficiency and space limitation in the prior art is broken through.

The invention discloses an optical fiber expansion demodulation method for a parallel measurement optical frequency domain reflection technology. The method comprises the following steps: obtaining mixed reference signals of all sensing optical fibers; acquiring a reference spectrum of each fiber bragg grating in the mixed reference signal according to the wavelength bandwidth of each fiber bragg grating; obtaining mixed measurement signals of all sensing optical fibers; according to the wavelength bandwidth of each fiber bragg grating, obtaining the measurement spectrum of each fiber bragg grating in the mixed measurement signal; performing drift demodulation on the reference spectrum and the measurement spectrum of each fiber bragg grating to obtain a central wavelength drift distance of each fiber bragg grating; and according to the central wavelength drift distance of each fiber grating, the central wavelength drift spectrum of each sensing fiber is obtained through combination. According to the optical fiber expansion demodulation method, the number of sensing optical fibers in a parallel channel can be increased without sacrificing the spatial resolution and precision of system sensing. The invention discloses an optical fiber expansion demodulation device based on the optical fiber expansion demodulation method and a parallel measurement optical frequency domain reflection system.

The invention relates to a multi-feature extraction method based on a TBA-GRU dual-channel model, and belongs to the technical field, and the method specifically comprises two parallel channels, namely a TextCNN channel and a BiGRU channel, which are used for extracting local features and time sequence features respectively, and further enhancing the attention to important features in combination with a multi-head attention mechanism. According to the XSS attack detection method, the output features of the TextCNN and BiGRU channels are spliced and fused through the double-channel structure, multi-level depth feature expression is achieved, feature mapping and classification tasks are completed through a full connection layer, the output of the two channels is spliced and fused, the model achieves multi-level depth feature expression, and therefore the accuracy and robustness of XSS attack detection are improved.

The application provides a method for obtaining a multispectral image through single exposure, an endoscope system and electronic equipment. The method comprises the following steps: obtaining an original image within a single exposure time, wherein the original image is obtained by sequentially imaging a light signal on a main light path of the endoscope system through a lens array and a continuous variable filter; dividing the original image into a plurality of sub-images corresponding to different channels in the lens array, wherein different sub-images correspond to different wave bands and different channel response coefficients; and performing normalization correction on the corresponding sub-images according to the wave bands and the channel response coefficients, so as to eliminate the response difference between different channels and obtain a sequence image of multiple wave bands. Through multiple parallel channels, a plurality of sub-images with different spectral responses are obtained within a single exposure, and the rapid acquisition of multispectral information of an endoscope is realized.

The present disclosure relates to the technical field of graphic processor, and discloses a memory management device and method of multiple parallel channels, electronic equipment, storage medium and computer program product. The device comprises a plurality of parameter stream generators and a target cache shared by the plurality of parameter stream generators. Any parameter stream generator is configured to send a data read request to the target cache, the data read request being used to obtain reference data corresponding to an effective primitive currently required to be processed by the parameter stream generator. The target cache is configured to determine the reference data corresponding to the effective primitive currently required to be processed by the parameter stream generator in response to the data read request of any parameter stream generator, and return the reference data to the parameter stream generator. The present disclosure can perform unified cache management, improve the overall hit rate of data read requests, reduce the interaction times with external storage devices, and improve processing performance.

The invention discloses a production line management and control integrated data synchronization method and system based on a big data technology, and relates to the technical field of production line data synchronization, and the method comprises the steps: marking production orders and production process data, generating production line management and control integrated data when detecting that the data changes, and triggering a synchronization mode judgment strategy; preliminarily selecting a synchronization mode according to the data volume, calculating synchronization time by combining the network bandwidths of the serial and parallel channels, and further selecting the synchronization mode; setting a serial and parallel synchronous waiting queue and a mark matching queue; different conditions of production order data and production process data are considered in design: when only the production order data is included, sorting is performed according to priority indexes; when only the production process data is contained, sorting is performed according to the production order and the task node sequence; when the two exist simultaneously, the method is synthesized for sorting; according to the design, accurate and efficient synchronization of production data is ensured, and the integration level and flexibility of production line management and control are improved.

The present disclosure relates to the field of optical computing technologies, and particularly to a large-scale reconfigurable photonic large matrix model and architecture. Among them, the model includes: a hybrid optical computing sub-model; the large-scale reconfigurable photonic large matrix model includes an acquisition module, a parallel input module, and an integration module. Among them, the acquisition module is used to acquire a first large matrix and an input vector that need to be calculated; the parallel input module is used to parallelly input the first large matrix and the input vector into the hybrid optical computing sub-model of each parallel channel based on the determined number of parallel channels, and obtain an output vector of each parallel channel; the integration module is used to integrate the output vectors of each parallel channel to obtain a target output vector. The present disclosure adopting the above solution realizes a new paradigm of large-scale general optical matrix operations while taking into account high computing accuracy and high data throughput, thereby supporting complex optical general computing tasks.

The invention provides a hybrid private chain and alliance chain-based Internet of Things data security sharing method and system, and belongs to the technical field of block chains, Internet of Things and cloud computing, and the method comprises the steps: deploying a private chain, and taking an alliance chain as a mutual trust coordination layer to achieve registration anchoring and interoperation of multi-chain data; a parallel HTLC channel and a dynamic time lock mechanism guarantee atomic exchange and efficient transmission of cross-chain data, and meanwhile, safe and credible data sharing is realized in combination with multi-party threshold notarization and zero-knowledge proof; the intelligent gateway layer provides protocol adaptation, format conversion, machine learning caching and intelligent access control, and optimizes a data publishing and prefetching process; the light client only reserves a block head and a Merkle index, and incremental verification and multi-path synchronization are carried out; the homomorphic encryption and parallel zero-knowledge proof framework gives consideration to data privacy and on-chain auditing performance, and it is ensured that limiting value judgment and compliance auditing are carried out in a ciphertext state. According to the method, credible, efficient and auditable data cross-chain sharing is realized.

The embodiments of the present application relate to the field of data transmission and provide a data transmission method, apparatus, communication system, computing device, chip, product and storage medium. The method includes: a transmitting end sends the first to Nth sub-data packets in parallel through M channels. The first to Nth sub-data packets are obtained by splitting the first data micro-slice, and M and N are positive integers. The receiving end may not receive the correct sub-data packet and sends a first retransmission indication identifier, which instructs the transmitting end to retransmit the first data micro-slice. The transmitting end sends a first retransmission identifier based on the first retransmission indication identifier and sends the first to Nth sub-data packets in parallel again through M channels. The first retransmission identifier indicates that the first to Nth sub-data packets are retransmitted data. Therefore, the receiving end can receive the retransmitted sub-data packets after receiving the first retransmission identifier. There is no need to always receive and merge the sub-data packets received through parallel channels. This saves computing resources at the receiving end.

The present invention belongs to the field of intelligent transportation systems and relates to a multimodal spatiotemporal traffic flow modeling method that supports real-time prediction of large-scale road networks. The method first designs a spatiotemporal prediction framework for dynamic traffic networks and then performs training to obtain the final prediction model. The spatiotemporal prediction framework includes a data embedding layer, a spatiotemporal coding module, and a deep modeling and output module based on an expert mixture mechanism. The data embedding layer includes two parallel channels, namely time embedding and spectral-domain spatial embedding, and a spatiotemporal data fusion module. The spatiotemporal coding module includes a parallel block-level sparse temporal attention module and a spatial attention-message passing module and a weighted fusion layer. The deep modeling and output module based on the expert mixture mechanism includes a MoE dynamic expert modeling module, a fully connected mapping module, a skip connection layer, and an output layer. This design improves the model's response speed, prediction accuracy, and cross-regional adaptability in highly heterogeneous scenarios.

One or more embodiments of the specification provide a retrieval enhancement generation method, device, equipment and storage medium. The method uses multiple parallel channels to retrieve query requests from multiple dimensions respectively; then, according to the time sequence of the return recall content of each channel, the recall content of each channel is pre-filled, wherein the pre-filled result of each channel is stored in the cache device, and the pre-filled result of the channel is obtained according to the pre-filled result cached by the previous channel returning the recall content; finally, according to the pre-filled result of the channel returning the recall content last, the output result of the large language model is obtained. Since the pre-filling is performed according to the time sequence of the return recall content of the channel, the waiting time in the retrieval enhancement generation request process is reduced, and the pre-filled result of the current channel is calculated according to the pre-filled result cached by the previous channel returning the recall content, avoiding repeated calculation in the pre-filling stage.

The invention relates to the field of semiconductor manufacturing, and provides a substrate clamping state detection method and system, and the method comprises the steps: a master station transmits a set value to a slave station; after the main station confirms that the substrate reaches the bearing table, the clamping assembly is controlled to clamp the outer edge of the substrate; in each acquisition period, the slave station acquires an acquisition value output by an analog quantity sensing module located at the outer edge of the substrate through N parallel channels; the slave station transmits the acquired value to the master station; the slave station compares the collected value with a set value based on a specified judgment method to obtain a judgment result; when the judgment result is not the over-clamping state and the fragment state, the slave station continues to execute the next collection period; when the judgment result is the over-clamping state or the fragment state, the clamping assembly is controlled to reset after a preset time period; the master station obtains the judgment result and calculates the difference value between the collection value and the set value; controlling the rotating speed of the bearing platform or updating the set value according to the difference value. The method is used for improving the data acquisition and processing speed when the clamping state of the substrate is detected.

A memory device comprises a plurality of backend interfaces that are each configured to connect to a respective media device (e.g., DDR, LPDDR DRAM), a plurality of reliability, availability and serviceability (RAS) channel datapaths with each RAS channel datapath connected to one or more backend interfaces, one or more caches that are each connected to at least one RAS channel datapath, and control circuitry. The control circuitry is configured to: in response to receiving a memory access request for a memory location, search a cache of the one or more caches for the requested memory location; and when, in response to the search, the requested memory location is not found in the cache, obtain the requested memory location from a media device connected to the at least one RAS channel datapath. Corresponding methods are also described.

The invention discloses a microfluidic large-scale manufacturing method and microfluidic large-scale manufacturing equipment for monodisperse double emulsions, and relates to the technical field of microfluidics. Flow parameters are decoupled through a multi-interface collaborative pinch-off mechanism in a stepped emulsification microfluidic unit; therefore, the influence of non-uniform distribution of fluid in a parallel stepped emulsification microfluidic network on the monodispersity of liquid drops is avoided, and large-scale manufacturing of the monodisperse double emulsion is realized through parallel connection and modular integration of a plurality of stepped emulsification microfluidic units. According to the micro-fluidic large-scale manufacturing method and equipment for the monodisperse double emulsion, the problem that the monodispersity of liquid drops is reduced due to non-uniform flow resistance distribution in parallel channels can be avoided, parallel mass production of the double emulsion with 1000 channels can be realized through parallelism of 100 stepped emulsification micro-fluidic units and integration of 10 modules, and the production cost is reduced. And the manufacturing flux can reach 1.8 L / h.