45results about How to "Preserve diversity" patented technology

The invention provides an adaptive state early warning system and method of working conditions of an oil transmission pump. The method comprises steps of carrying out online acquisition on characteristic parameters of operation working stations and deterioration states of the oil transmission pump, wherein the characteristic parameters comprise working condition characteristic parameters and deterioration state representation parameters; carrying out related analysis on the deterioration state representation parameters to acquire related coefficients between the deterioration state representation parameters; according to the related coefficients and the working condition characteristic parameters, dividing the working conditions of the oil transmission pump into multiple working condition sub-spaces, determining deterioration state representation signals of the working condition sub-spaces from history operation data of the oil transmission pump, and carrying out characteristic extraction and fusion on the deterioration state representation signals to acquire a deterioration evaluation index in each working condition sub-space; and according to the deterioration evaluation index in each working condition sub-space, predicating a deterioration state value in each working condition sub-space of the oil transmission pump in real time, comparing the deterioration state values with a corresponding alarming threshold and if the deterioration state values exceed the alarming threshold, outputting corresponding alarming information. According to the invention, deterioration states of the oil transmission pump can be precisely predicated under variable operation working conditions.

The invention provides a particle filter weight processing and resampling method based on an FPGA. The particle filter weight processing and resampling method has the advantages that on one hand, a weight normalization process is abandoned, parallelization sorting processing is carried out on weights, all particles can be compared and exchanged by M / 2 comparators and controllers within one period, and the comparison and exchanging efficiency of a sort algorithm is improved; on the other hand, a threshold value of a resampling part is set according to permutation distribution of the weights, the large-weight particles, the middle-weight particles and the small-weight particles are determined through Nh and N1, and the threshold value is not a preset threshold value. The large-weight particles and the middle-weight particles are reserved in the resampling process, the small-weight particles are eliminated, the weights and the position information of the large-weight particles are input into an FIFO, one circuit of the weights and the position information are output through an FIFOOUT to replace a weight of a small-weight particle register, the other circuit of the weights and the position information are input into the FIFO, and the function of copying the large-weight particles is achieved. By means of the particle filter weight processing and resampling method, the accuracy of target position calculation, the diversity of particle samples and the high particle replacement efficiency in the hardware environment are guaranteed.

The invention relates to a biochar-mediated solid biofilm anaerobic digestion promoter which comprises the following components: ferrous sulfate, cobalt chloride, nickel chloride, composite vitaminsand an anaerobic digestion solid biofilm. The anaerobic digestion solid biological membrane is prepared by freeze-drying biochar subjected to anaerobic digestion domestication and a biological membrane on the biochar. The invention also discloses a preparation method of the biochar-mediated solid biological membrane anaerobic digestion promoter. The anaerobic digestion promoter disclosed by the invention has the advantages of long effective service life, convenience in storage and transportation and simplicity and quickness in operation. When the anaerobic digestion promoter is added into an anaerobic digestion reactor, anaerobic digestion can be rapidly started, microorganisms can be rapidly enriched and stabilized, the sufficiency of anaerobic digestion organic matter degradation can beeffectively improved, the hydraulic retention period can be shortened, and the methane yield can be increased.

The invention is suitable for the field of machine learning and medical image processing technology, and provides a thresholding method based on weight distribution. The method comprises the followingsteps of S1, dividing a training set sample to two sample sets based on a class label, namely a patient set and a normal person set; S2, for aiming at all samples in the two sample sets, using a Pearson correlation coefficient between time sequences of brain regions as a connecting weight, and constructing a brain network adjacent matrix; and S3, based on the function connecting distribution difference of the two sample sets, constructing a threshold for aiming at each brain region. Through each function connection in the network, the weight distribution is used for adaptively constructing anoptimal threshold, thereby constructing different thresholds for the function connection between different brain regions, thereby constructing a threshold matrix for the whole brain network, thresholding the brain network, keeping function connection diversity between the different brain regions, and settling defects in a single threshold or percentage method.

The invention discloses a short-term household electrical load prediction method and system, a storage medium and equipment. The method comprises the following steps: determining the architecture of a BP neural network, and selecting a training set sample, a test set sample and related parameters; then initializing the particle swarm; determining a particle individual historical extreme value and a group historical extreme value according to the initialized parameters; updating the position and the speed of the particle according to the fitness value F obtained by calculation, and completing iteration when a preset error range is reached or the maximum number of iterations is completed, at the moment, the gbest position being the optimal weight and threshold; performing BP neural network training on the obtained optimal weight and threshold; and comparing the actual load value with the predicted load value, calculating an average relative error of the prediction day, stopping iteration and ending training when the average relative error meets a preset error allowable range, and outputting a prediction result through error judgment to complete load prediction. The structure is simple, and the reliability of the prediction result is enhanced.

The invention relates to the technical field of molecular biology and in particular relates to a novel DNA library construction method and an application thereof. According to the DNA library construction method disclosed by the invention, an adhesive end joint can be directly mediated by a common ligase to be connected to a single-stranded DNA denatured from a sample genome DNA; the novel DNA library construction method can be applied to a sample with low initial mass; and compared with previous methods, the novel DNA library construction method has the advantages that the connection efficiency of samples is greatly increased, the diversity of the samples are retained to the greater extent, the stability is good, and library bias caused by artificial factors is shortened.

The invention relates to a teaching content configuration and interaction scheme implementation method suitable for AR education application. Firstly, a system developer designs an interaction scheme at a client in advance, then a teacher configures and standardizes teaching contents at a server through a Web page, and finally sends teaching data to the client, the client adopts the corresponding interaction scheme to implement, and a virtual scene generated by rendering is submitted to students to use. According to the invention, the architecture of the client / server and the idea of augmented reality education application are adopted, the teaching content is configured through the webpage of the server, and the interaction scheme is implemented at the client, so that a set of augmented reality scene for students to learn is presented; finally, the augmented reality courses learned by the students are autonomously designed and managed according to the willingness of the teacher.

The invention provides an in-situ water body remediation material adopting natural minerals. The in-situ water body remediation material adopting natural minerals is prepared from raw materials as follows: volcanic ash, dolomite, clinoptilolite, rock-forming minerals and clay minerals with high ion substitution capacity. The prepared in-situ water body remediation material adopting the natural minerals can be completely applied to pollution treatment of rivers, lakes and other water bodies and can decompose pollutants in various water bodies and notably reduce the turbidity. Green algae, red tide and eutrophication of the water bodies are effectively prevented, the water quality and the habitat environment of aquatic organisms are improved, the biodiversity is kept to the greatest extent,and the in-situ water body remediation material has the characteristics of being energy-saving, environmentally friendly, remarkable in effects, free from secondary pollution, capable of taking effectrapidly, wide in application range, convenient to use and free from restriction by environmental conditions.

The invention relates to an upper limb posture recognition algorithm based on a genetic algorithm for super-parameter optimization. At that move end, the upper limb posture movement is collected, andthrough the analysis of the collected data, six kinds of motion information are identified. The upper limb posture data are time series, time window length, data overlap rate and the number of hiddenneurons, which will affect the accuracy of recognition. Therefore, a genetic algorithm is used to optimize the super-parameters and find the most appropriate set of solutions. The traditional geneticalgorithm is slow and easy to fall into the local solution, the invention sorts the population according to the fitness function, the sorted populations are divided into four parts, namely, Top, Best,Normal and Worse. All the good populations are selected and the best, Normal and Worse populations are selected proportionally. After selection, the inadequate populations are randomly generated to ensure that the number of population remains unchanged in each round. The invention accelerates the speed of parameter optimization, and simultaneously finds the global optimal solution.

The invention relates to the internet of things technology, and discloses an image segmentation method comprising the following steps: using a pre-constructed image segmentation model comprising multi-scale convolution kernels and multi-voidage void convolution to carry out feature extraction on a to-be-segmented image to acquire a feature map; splicing the feature map and executing up-sampling operation to obtain a restored image; and performing pixel segmentation on the restored image by using an image segmentation algorithm to obtain a segmented image. The invention further provides an image segmentation method and device, electronic equipment and a computer readable storage medium. According to the invention, the problem of low accuracy of the segmentation result of the traditional segmentation network can be solved.

The invention provides a free-collapsed warp-knitted jacquard fabric and a manufacturing method. The free-collapsed warp-knitted jacquard fabric comprises a mesh fabric body formed by integral warp knitting, wherein the mesh fabric body is composed of a transparent mesh jacquard surface layer, a collapsed layer, a pattern multi-color jacquard layer, a pile height layer and a bottom layer from top to bottom; the transparent mesh jacquard surface layer is composed of two semi-number jacquard guide bars JB2 and JB3; the collapsed layer is composed of four semi-number jacquard guide bars JB2 and JB3 andjacquard guide bars JB4 and JB5 which are matched with each other; and the pattern multi-color jacquard layer is composed of two semi-number jacquard guide bars JB6 and JB7 which are used for jacquard on the transparent mesh jacquard surface layer and the pile height layer. According to the free-collapsed warp-knitted jacquard fabric and the manufacturing method provided by the invention,the jacquard mesh fabric with high performance and unique appearance patterns can be produced by combining a two-needle-bar three-jacquard machine with the guide bar empty threading change and matching the pattern design with the chemical fiber yarns.

The invention relates to a processing method and an electronic device for attribute data applied to a knowledge map. In the data import stage: receiving user's data import configuration in multiple formats for data with multiple formats, and completing the data import Finally, record the mapping relationship between attributes, data formats, and data types; the data import configuration is performed by the user through the data compatibility module configured on the front-end page; map operation stage: receive the user's operation on the attribute data in the map, and According to the mapping relationship recorded in the data import stage, the data filled in the operation is automatically stored in a format corresponding to the attribute selected by the user in the operation. By configuring the data compatibility module on the front-end page to provide multiple storage formats, users can complete the storage of attribute data without calling the mapping configuration table, which is convenient for operation, and does not need to convert data in multiple formats into the same format for storage. Therefore, no additional resources will be occupied and resource consumption will be reduced.

The invention discloses a salient target detection method based on a refined spatial consistency two-stage graph, which mainly solves the problem that the prior art cannot detect salient targets completely and consistently in complex scenes. The implementation plan is as follows: 1. Over-segment the input image to obtain several superpixels; 2. Construct the first-stage graph; 3. Construct a weighted joint robust sparse representation model; 4. Find the optimal weighted robust sparse representation model Optimal solution; 5. Calculate the saliency value of each node in the first stage graph; 6. Refine the spatial consistency and construct the second stage graph; 7. Use the popular ranking model to calculate the saliency value of each node in the second stage graph ; 8. Calculate the saliency value of each superpixel of the input image; 9. Output the pixel-level saliency map of the input image. The invention has good effects of foreground segmentation and background suppression, can completely and consistently detect image salient objects in complex scenes, and can be used for image preprocessing in computer vision.