3results about How to "Guaranteed distillation effect" patented technology

This utility model discloses a purification device and an electrolyte preparation system for 2-methyl-adipic acid. The purification device first purifies the crude electrolyte with high viscosity through a thin-film evaporator, separates the light components, and then performs distillation. This combines the advantages of high evaporation efficiency of the thin-film evaporator with the distillation effect of the distillation column, resulting in a high-purity purified product, which promotes the industrial production of high-quality aluminum electrolytic capacitor electrolyte.

The utility model relates to a rectifying device for biodiesel, including esterificator, rectifying tower and condenser, rectifying tower includes reboiler and the rectifying column of solid connection in reboiler top, first heating pipe is solidly established in reboiler, the rectifying column includes the first rectifying section, secondary heating section and second rectifying section of solid connection from below to above in proper order, second heating pipe is solidly established in secondary heating section, the outlet of esterificator communicates with the import of reboiler through first pipeline, and the top outlet of second rectifying section communicates with the import of condenser through second pipeline, and the outlet of condenser communicates with the recovery mouth of esterificator, the outlet of first heating pipe communicates with the import of second heating pipe through third pipeline, the utility model improves the rectification effect and utilization ratio to methyl alcohol, and then improves methyl alcohol purity, and the methyl alcohol purity is conveniently detected, and it is more convenient to use.

The present application relates to the technical field of artificial intelligence and distributed privacy computing, and particularly relates to a global model correction method for federated learning based on adaptive screening of generated data. The method is applied to a server, the server and a plurality of clients constitute a federated learning system, and the method comprises: performing federated learning iteration with the plurality of clients to obtain a target classification model; wherein the tth iteration in the federated learning iteration is: broadcasting a (t-1)th round master classification model and a target generator to the clients, freezing the generator by the clients, reporting after training the classification model and the discriminator, aggregating the classification model parameters by the server, training the generator with the multi-discriminator parameters reported by the clients, outputting a pseudo sample set by the generator, performing weighted knowledge distillation with the aggregated model as a teacher and the master model of the last round as a student to obtain a master classification model of the present round. The present application can improve the model application effect of the global model finally output by the federated learning system.