Research On Super-Frequency Reconstruction Method For Electrical Data And Its Applications

The digital power grid is the development goal of the future power grid.The current power Internet of Things and information network establish a data foundation for the digitization of the power grid through the interconnection of everything and massive information.Through the benefits brought by initial development of 5G technology,it can be foreseen that this technology will greatly improve data communication capabilities and become an important auxiliary means for the data transmission process in the future.Obviously,the quantity and quality of electrical data are the foundation and focus of the above-mentioned emerging concepts and technologies.Electrical data with large quantity,wide range,high feature dimension,and rich high-frequency details are of great significance to the business scenarios of power grids such as improvement of situational awareness accuracy,monitoring level,and auxiliary service quality.At present,considering the cost issue,most data sets of the power grid are low in sampling frequency,small in electrical quantity,and large in quantity,which limits the promotion of power Internet of Things and information networks.Due to the limitations of communication and storage capabilities,although smart meters can collect high frequency electrical data,they are usually compressed to low frequency before transmission.Traditional reconstruction algorithms are difficult to achieve high-precision super frequency reconstruction,which will cause a lot of information to be lost.Therefore,this paper uses an improved generative adversarial network to reconstruct low-frequency electrical data into high-frequency.The construction method of converting time series data into electrical image is proposed,and the neural network method can efficiently extract electrical image features.The generator based on deep residual network and the improved residual block structure are used to improve the feature learning ability of the generator.In addition,the generator loss function considers the difference in low-dimensional or high-dimensional features of real samples and generated samples.Take the standard data set I-BLEND as an example for algorithm verification.The results show that compared with traditional reconstruction methods,the proposed method has higher reconstruction accuracy and high-frequency detail reduction,and can be generalized in different datasets.In the future,the rapid development of 5G technology will greatly improve electrical data communication capabilities,but at the same time it will lead to a mismatch between weak data storage capabilities and strong data communication capabilities.Therefore,this paper proposes a feature-driven based integrated architecture between compression and reconstruction to improve the storage capacity of data center.In terms of compression,Gaussian smoothing is introduced to make low-frequency electrical images rich in original feature information.In terms of reconstruction,a multi-scale feature extraction network is built,including an improved deep residual network,a shallow feature extraction module,and a deep feature extraction module.Further,deep feature extraction module is composed of deep feature extraction units,which include residual blocks based on external channel attention mechanism and residual blocks based on internal channel attention mechanism.Taking the standard data set I-BLEND as an example for algorithm verification,the simulation results show that the proposed method can effectively compress electrical data and reduce data storage pressure.At the same time,it has a more accurate reconstruction effect than other advanced super-resolution reconstruction methods.In the end,this paper uses an example of load forecasting in the field of energy and power to verify the actual engineering application effect of the proposed super-frequency reconstruction method for electrical data.A load forecasting method based on super-frequency reconstruction is proposed.First,this paper verifies the prediction accuracy of two classic load forecasting methods and a new load forecasting method on low-frequency electrical data input,and then verifies the prediction accuracy of the proposed method on the same input.Comparing the simulation results of the above-mentioned methods,the proposed super reconstruction method can effectively improve load forecast accuracy.