85 results about "Time series processing" patented technology

The invention discloses an underwater sonar target recognition system based on multi-domain feature fusion and lightweight modeling, and the system comprises a Trifusion block, a novel lightweight attention residual network, a long and short time attention LSTM and a Mamba module which are connected in sequence, and achieves target recognition through multi-domain parallel extraction of fusion features, lightweight convolution and attention optimization, long and short time dependence capture and long sequence modeling. The method has the advantages that complex noise is comprehensively represented, the model efficiency and stability are improved, the bottleneck of time sequence processing is broken through, and the method has high performance, high robustness and wide adaptability.

The invention provides a crop planting suitability dynamic evaluation method and system and a storage medium, and the method comprises the steps: S1, carrying out the time sequence processing of collected NDVI, meteorological data and static data; s2, dividing the phenological period of the target crop into a plurality of stages; s3, constructing a seasonal suitability function for the NDVI and the meteorological data to calculate scores of the data in each stage of a phenological period; s4, determining the score of the static data in each stage according to a preset membership score and grade division; s5, according to the historical NDVI, the meteorological data and the static data, the correlation coefficient of each data and the historical target crop yield and the entropy weight of each data are calculated respectively, and the proportion of the data and the historical target crop yield is adjusted through empirical parameters to obtain the dynamic weight corresponding to each data; and S6, performing weighted summation on the data based on the scores obtained in the steps S3 and S4 in combination with corresponding dynamic weights to obtain a crop planting suitability dynamic index. Therefore, the phenological accuracy of the planting suitability evaluation result of the target crop is improved.

The invention provides a loop current prediction method based on a CNN and a BiLSTM neural network. The loop current prediction method is used for improving the prediction precision of proton flux data in space weather. The method comprises the following implementation steps of: acquiring and preprocessing data; a CNN neural network model and a BiLSTM neural network model are constructed; training the model; and evaluating and visualizing the model. According to the method, the advantages of the CNN in the aspect of spatial feature extraction and the powerful time sequence processing capability of the BiLSTM neural network are utilized, and finally, the outputs of the CNN and the BiLSTM neural network are fused, so that the spatial-temporal dynamic distribution of the ring current proton flux can be more effectively captured. Experimental results show that the method has excellent performance in a loop current prediction task, is suitable for real-time detection and early warning of space weather, and has relatively high prediction precision and stability.

The invention provides a structure health early warning method and system based on space-time correlation characteristics and digital twinning, and relates to the technical field of data processing. The method comprises the following steps: acquiring multi-source heterogeneous data of a target structure; performing dynamic sampling alignment and wavelet packet decomposition on the multi-source heterogeneous data to extract energy features to obtain synchronous data, and performing abnormal data filtering on the synchronous data to obtain cleaned fusion data; performing wavelet decomposition on a high-frequency vibration signal in the fused data to obtain a damage impact feature, performing time sequence processing on low-frequency temperature data in the fused data to obtain a temperature time feature, and performing dynamic graph convolutional network processing on strain data in the fused data to obtain a spatial correlation feature; constructing an input vector; calculating a damage degree index; and according to the damage degree indexes, early warning grades are divided, and corresponding control instructions are triggered for different early warning grades. By implementing the technical scheme provided by the invention, the accuracy of structural health early warning can be improved.

The invention discloses a method and system for constructing a city information model, and belongs to the field of city intelligent management, and the method comprises the steps: carrying out the spatial registration and semantic alignment of future multi-source heterogeneous data through a unified coordinate reference, and constructing a multi-dimensional incidence relation between a spatial entity and the attribute of the spatial entity through a graph database; performing time serialization processing on the city information model based on the real-time sensing data, and performing real-time correction on the geometric state and the attribute of the model by adopting an incremental modeling algorithm; performing automatic calibration on the model data in combination with rule reasoning and probability correction methods; predictive calculation is carried out on multiple scenes, and parameterized optimization is carried out on the city information model according to a calculation result; and carrying out adaptive hierarchical abstraction on the city information model, and automatically generating model subsets with different precision levels. According to the method, spatial registration and semantic integration are performed on the data, so that the global consistency of the city information model is realized, and the problems of data inconsistency, repetition and redundancy in a traditional method are solved.

The invention relates to a teaching quality dynamic evaluation system based on big data, and the system comprises a data collection module which is used for obtaining an original teaching time series data flow containing student scores, classroom performance and homework quality in real time; the time sequence processing module is used for adding a time stamp to the original teaching time sequence data stream and verifying the time stamp to generate a teaching time mark data set, performing time sequence analysis on the data set to obtain a teaching fluctuation feature set, and performing multi-dimensional decomposition on the feature set to extract a teaching quality feature set; and the real-time monitoring module is used for constructing a teaching monitoring model based on the teaching target parameters, calculating a multi-dimensional deviation between the current teaching quality feature set and the historical teaching quality feature set by using the model, and generating a multi-dimensional teaching early warning signal when the deviation exceeds a preset multi-dimensional joint early warning threshold. According to the invention, real-time dynamic monitoring and risk early warning of teaching quality are realized, and evaluation efficiency and accuracy are improved.

The invention discloses a power transmission tower deformation monitoring method and device and a computer program product, and the method comprises the steps: obtaining a multi-angle collection image of a random speckle pattern on the surface of a power transmission tower; constructing an image sequence based on the multi-angle acquired images, reconstructing a three-dimensional displacement field on the surface of the power transmission tower by using a stereoscopic vision reconstruction algorithm, and exporting a strain tensor field; performing time sequence processing on the strain tensor field under a plurality of time nodes through a strain evolution analysis model of a continuous time sequence, and extracting a strain change trend of each preset structure part; carrying out space consistency mapping and weighted interpolation processing according to the strain change trends of the plurality of preset structure parts, and calculating to obtain the overall structural dependent variable of the power transmission tower; and S5, in combination with historical strain data and the current overall structural dependent variable, identifying an abnormal deformation area of the power transmission tower by comparing the spatio-temporal evolution mode of the strain tensor field. The method has the effect of improving the accuracy of deformation monitoring of the power transmission tower.

The invention relates to the technical field of laboratory environment control, and discloses an environment parameter control method and system for a complex laboratory, and the method comprises the steps: obtaining environment index data, and carrying out the time sequence processing, and obtaining a real-time change trend; according to the real-time change trend, predicting a future offset by using a prediction model, and performing data fusion when the offset exceeds a threshold value to obtain an accurate environment state estimated value; according to the accurate environmental state estimation value, quantifying the mutual influence among the environmental indexes to obtain the mutual influence relationship of the environmental indexes; determining an environment index priority sequence according to the mutual influence relationship; generating and executing a control instruction according to the priority sequence; and when continuous deviation exists, the prediction model and the mutual influence relation are updated through feedback circulation. According to the method, the dynamic mutual influence among the environmental indexes can be quantified, and accurate and stable cooperative control of multiple parameters in a complex dynamic environment is realized.

The invention relates to the technical field of laboratory instrument management and control, in particular to a laboratory instrument management system based on artificial intelligence, which comprises the following steps: acquiring instrument operation data and artificial use data in a target laboratory, constructing a use behavior sequence in combination with time sequence processing, and generating an abnormal residual vector through a residual generation network; according to the method, tiny deviation characteristics in operation of an instrument are depicted, abnormal judgment is carried out in combination with a contrast learning neural network, and an abnormal event probability value is output, so that early recognition of a potential fault trend is realized, and the problem that the fault is detected after being serialized is avoided. And the alarm generation module dynamically generates alarm information according to the abnormal event probability value, and performs alarm processing according to a preset strategy. The problems that in an existing laboratory instrument management system, fault response lags behind, the monitoring dimension is single, and manual operation behaviors are ignored are effectively solved.

The invention discloses a concrete dam crack opening degree prediction method based on feature optimization recognition. Comprising the following steps: S1, data preprocessing; S2, saliency feature extraction of a two-channel point riding attention mechanism module; S3, convolutional neural network feature extraction and fusion; S4, time sequence feature learning of a long and short term memory network; the beneficial effects of the invention are that deep learning and cooperative training are combined; a two-channel point riding attention mechanism is introduced to enhance the representation of key features, and the feature extraction capability of the convolutional neural network and the time sequence processing capability of the long and short term memory network are fused; self-adaptive weighted learning of multivariate environment characteristic information in crack opening monitoring data is realized, and the opening response of the crack under the influence of a complex environment is accurately predicted; the prediction capability of the concrete dam crack opening effect can be improved, and an innovative calculation scheme can be provided for concrete dam crack monitoring, early warning, prevention and control.

The invention relates to the technical field of IT support, and provides a time sequence processing method, device and equipment adopting a quantum pulse neural network, and the method comprises the steps: obtaining time sequence data which comprises network alarm data, network equipment performance index data and network operation and maintenance work order data; encoding the time sequence data into a first quantum state by using a quantum preprocessing layer, inputting the first quantum state into a pulse neural network layer, converting the first quantum state into a time sequence pulse sequence, and processing the time sequence pulse sequence to obtain an output result; by utilizing a quantum attention enhancement mechanism, calculating attention weight of an output result in a quantum state space, and weighting to obtain a second quantum state; and decoding the second quantum state by using the hybrid decoding layer to obtain a final prediction result. The final prediction result is used for realizing fault root cause positioning, abnormal work order identification or network service quality prediction. According to the method, the parallelism of quantum calculation and the superposition characteristic of the quantum state are utilized, and the calculation efficiency can be improved when large-scale time sequence data are processed.

The invention discloses a clothing demand dynamic prediction method and system based on multi-source data fusion and machine learning, and the method comprises the steps: collecting and preprocessing multi-source heterogeneous data, and obtaining basic time sequence features, static attribute features and external situation features containing fashion trend quantification features; inputting the basic time sequence and the static attribute characteristics into a time sequence processing network to obtain a reference trend prediction value; inputting the external situation features into a situation feature processing network to obtain a situation influence vector; generating a dynamic adjustment coefficient by the vector through a gating unit; and finally, performing fusion calculation according to the reference trend prediction value and the dynamic adjustment coefficient to obtain a final demand prediction quantity. According to the invention, through a double-flow network structure and a gating fusion mechanism, effective modeling is carried out on an internal sales law and external situation impact, the prediction accuracy and the response speed to market changes are significantly improved, and accurate and dynamic decision support is provided for a clothing supply chain.

The invention discloses a distributed power supply cooperative control method based on transient dynamic characteristic adaptive driving, which adopts a multi-dimensional transient performance coupling evaluation function, unifies multiple targets including frequency, voltage, power angle and operation economy into a global optimization target, and fundamentally solves the defect of single target in the prior art. A collaborative strategy network based on dynamic context awareness is adopted, and a time sequence processing technology based on an attention mechanism is utilized, so that an intelligent agent can extract key transient dynamic characteristics from a local observation sequence, and the limitation of no memory and non-self-adaption in the prior art is broken through; according to the method, a cooperative control flow of a cooperative control strategy is provided, a high-performance learning type strategy is combined with a high-reliability safety monitor, the problem that safety and performance are difficult to be compatible is solved, fundamental transformation of distributed power supply control from passive, local and static rule-based modes is achieved, and the transient stability margin of a novel power system is remarkably improved.

The embodiment of the invention discloses a database dynamic query optimization and resource scheduling method and device and a medium, belongs to the technical field of databases, and solves the problems that the mode of manually adjusting PostgreSQL resource parameters is low in efficiency and causes resource waste. Obtaining query feature data and resource state data corresponding to the database, and performing time sequence processing on the resource state data to obtain a resource prediction result; inputting the query feature data and the resource state data into a preset AI strategy optimization model to output an optimization execution plan and optimization resource allocation parameters; inputting the optimization execution plan into a dynamic injection module; dynamically adjusting the operation parameters of the PostgreSQL based on the optimized resource allocation parameters and the resource prediction result; and in response to the plan injection instruction, selecting a required execution plan in the dynamic injection module so as to execute the optimized query task based on the adjusted operation parameters and the required execution plan.

The invention discloses a triple scene graph video understanding method fused with a time sequence perception moving window, which comprises the following five core steps of: inputting a video clip, acquiring a frame sequence through an encoder, carrying out space division on each frame and adding time embedding; executing multi-head self-attention calculation of space-time offset enhancement in each sliding window, and integrating space structure modeling and time sequence processing; constructing a cross-space-time sliding window based on continuous frames to extract a token subset; compressing the window output into a fixed-length video token (triple scene graph) through a sliding query converter; and inputting the triple scene graph of the video content and a user question into the multi-modal large model to realize semantic analysis and content understanding. According to the method, the video understanding precision is remarkably improved while the space-time relation modeling capability is enhanced, the average recall rate of behavior recognition is improved by 10%, a high-precision video understanding benchmark covering 25 object classes and 26 relation classes is constructed, the labeling error rate is lower than 3%, and the shielding problem and the long tail deviation problem are effectively solved.

The invention discloses an MEMS inertial sensor reliability analysis method based on long and short term memory deep learning, and belongs to the technical field of microelectronic reliability evaluation, and the method comprises the steps: collecting an original output signal and environmental parameter data of an MEMS inertial sensor in a target scene, and recording a collection timestamp; then preprocessing is carried out to obtain a time sequence processing data set; performing feature extraction on the time sequence processing data set to form a multi-dimensional feature vector; carrying out correlation analysis on the multi-dimensional feature vectors, and screening out a key feature set; constructing an LSTM model based on the key feature set, and performing training and parameter optimization on the model to enable the model to capture a time sequence dependency relationship of sensor performance degradation; inputting the real-time key feature set subjected to preprocessing and feature extraction into the trained LSTM model, and outputting probability distribution of the current health state of the sensor; and then reliability evaluation is carried out and an early warning signal is output through the system terminal to prompt maintenance personnel to carry out timely intervention.

The invention discloses a power equipment inspection device fault detection and diagnosis method based on deep learning, and the method comprises the following steps: collecting multi-modal data, and generating an original data set; performing time calibration normalization on the original data set to generate a preprocessed data packet; performing anomaly detection on the preprocessed data packet, outputting an anomaly candidate region set and generating a feature slice set; executing group convolution operation on the feature slice set, and outputting structure preserving representation; performing classification and calibration on the structure preserving representation, and outputting a fault type and a confidence coefficient vector; performing risk assessment on the fault type, and outputting a risk level code; performing time sequence processing on the structure retention representation, and outputting a fault trend curve and a degradation rate; and sorting the abnormal candidate region set, archiving the risk level code, marking the fault trend curve and the degradation rate, and generating diagnosis report data. According to the invention, fault detection and diagnosis of the power equipment inspection device are realized.

The invention discloses a multi-source collaborative dynamic time sequence TCP prediction method and system, a medium and a program product, and the method comprises the steps: collecting multi-source heterogeneous data, including DVH data, CT data, CPP data and longitudinal follow-up data of a plurality of time nodes, of a patient in a whole radiotherapy period; performing hierarchical preprocessing and data format unification on the multi-source heterogeneous data; constructing a dynamic time sequence feature fusion matrix containing a time dependency relationship; constructing a deep learning model integrating a time sequence processing unit and a cross-modal fusion unit, wherein the model supports an incremental learning iteration updating mechanism; using the model to output TCP prediction results of a new patient at different time nodes; and if the new patient is a special case, calling the clinical rule adaptation model, and correcting the TCP prediction result by adopting a mixed correction strategy combining rule matching and doctor experience weight. According to the method, multi-source heterogeneous data can be deeply fused, time dynamic association is captured, modal feature contribution degree is quantified, continuous iterative updating is supported, and individual differences are adapted.

The invention provides a corrosion life prediction method, system and equipment for a high-cleanliness chemical supply system pipeline and a storage medium. The corrosion life prediction method comprises the following steps: obtaining target parameters related to corrosion in a high-cleanliness chemical supply system pipeline, wherein the target parameters comprise a pipeline corrosion state parameter, a medium characteristic parameter, an operation environment parameter and historical failure data; inputting the target parameter into a pre-constructed corrosion life prediction model to enable the corrosion life prediction model to extract a potential corrosion characteristic sequence based on the target parameter, and performing time sequence processing on the potential corrosion characteristic sequence to generate a corrosion life prediction value; and grading the corrosion risk of the high-cleanliness chemical supply system pipeline according to the corrosion life prediction value. According to the technical scheme, quantitative evaluation and real-time prediction of the corrosion state of the pipeline of the high-cleanliness chemical supply system can be achieved.

A system includes a machine-learning model implemented on a data processing apparatus, which features a parallel processor with a memory hierarchy. The machine-learning model is a recurrent neural network (RNN) with a multi-head architecture, comprising multiple sub- vectors that process parallel data streams. The RNN's weight matrix is structured as a block-diagonal matrix, allowing for parallel processing of the sub-vectors. A fused computational kernel executes an entire time-series processing loop for the multi-head RNN, maintaining the weight matrix blocks in on-chip memory and performing matrix multiplications and element-wise operations for each sub-vector in a single kernel execution.

The invention relates to a leaf area index time sequence processing method, which comprises the following steps of: inputting a time sequence remote sensing image and suburb forest and economic forest classification data of the same region; calculating a normalized differential vegetation index of the research area and carrying out time sequence sorting; calculating vegetation coverage data of the research area and carrying out time sequence sorting; synthesizing to obtain monthly FVC time sequence data of the suburb forest region and the economic forest region; calculating a suburb forest leaf area index LAI in the research area and performing time sequence sorting; calculating the economic forest leaf area index LAI of the research area and performing time sequence sorting; performing seasonal decomposition on the suburb forest time sequence data after adaptive filtering; performing seasonal decomposition on the economic forest time series data after adaptive filtering; merging the suburb forest time sequence data and the economic forest time sequence data; and outputting to obtain final optimized data. The invention further relates to a leaf area index time sequence processing system. According to the invention, a purification LAI time sequence product which clearly represents long-term trend, mutation and gradual change signals of respective vegetation canopy structures can be output.

Automated forecasting and replenishment planning system for pharmacy stocks, comprising a computer-aided infrastructure for recording dispensing data and stock levels, a computer-based time series processing unit for generating demand forecasts, and a replenishment control unit for automatically triggering replenishment measures based on the forecasts.

The present application relates to a kind of single-station hourly temperature prediction methods based on data fusion and mixed convolution, belong to meteorological prediction field.The present application selects important single-station meteorological observation element, obtains single-station historical observation sequence;Extract multiple-element prediction data, obtain numerical prediction space-time sequence;Data normalization;A three-dimensional convolution module is constructed, input after numerical prediction space-time sequence of normalization, finally produce time series containing spatial features;Build interactive learning model ICM, to single-station historical observation sequence and the time series extracted by three-dimensional convolution processing, generate the time characteristic information of two kinds of data;Build MCNN network model, fuse the two branch time characteristic information obtained in S3, after again interactive learning and fully connected, form prediction result;Temperature prediction is carried out using the model.The time series prediction model built by the present application improves stability and prediction accuracy, and the method is suitable for single-station 72-120 hours of hourly temperature prediction.

The present invention provides a method and system for dynamic inertia control of a virtual synchronous machine based on artificial intelligence technology, which is applied to the control of a voltage source inverter. The main steps include: data acquisition and preprocessing, and outputting dynamic parameters through a time series neural network for the control links of the virtual synchronous machine respectively; calculating a phase compensation angle by using a convolutional neural network for phase correction of pulse width modulation, and then using a deep reinforcement learning network to output a duty cycle correction amount to adjust the pulse width modulation for collaborative control of the dynamic inertia of the virtual synchronous machine. The present invention obtains dynamically adjusted parameters through data time series processing and neural network application, and can realize adaptive adjustment of core parameters; uses a convolutional neural network to participate in the calculation of modulation signals, effectively compensates for hardware phase delay, and improves the dynamic response accuracy; and then realizes collaborative optimization of inertia control and dynamic response by combining deep reinforcement learning, reduces current overshoot and suppresses harmonics, and improves power quality and system stability.

The application discloses a kind of GoldenDB database performance bottleneck evaluation method, equipment and medium, related to database data processing technical field, method includes the following steps: the general performance index of each node in cluster and GoldenDB database unique performance index are collected, and the data collected are format unified and time series processing, form standardization data;The standardization data is aggregated under different time dimensions, according to the preset aggregation index, aggregated data is converted into the aggregation data reflecting performance change trend with discrete data points;The aggregation data is respectively input into the corresponding performance evaluation model deployed on each node for analysis, obtains the performance bottleneck evaluation result and performance change trend prediction result of each node and cluster;The performance bottleneck evaluation result and performance change trend prediction result distributed in each node are summarized, and unified database performance bottleneck evaluation and prediction report is generated.

The invention discloses a rainfall prediction method based on a Transform architecture, and the method comprises the steps: constructing a Transform rainfall model, carrying out the full extraction of the features of different time scales, enabling the interior of the Transform rainfall model to be provided with a multi-scale channel module, a Transform module and a prediction head module, carrying out the time sequence processing and abnormal value processing of the rainfall data of a laser radar through the Transform rainfall model, and carrying out the prediction of the abnormal value of the rainfall data of the laser radar. According to the method, a data set with reasonable data distribution is provided, and a multi-scale channel attention mechanism is provided after the data set is provided and is used for extracting features in a channel time space, performing comprehensive experimental verification on the data set and completing rainfall prediction of a target area; the method is composed of three key parts, namely a multi-scale channel module, a Transform module and a prediction head module, the multi-scale channel module is introduced for a similar three-dimensional structure of laser radar data, the feature extraction capability of the Transform is improved, the problem of memory deterioration in a long-time sequence is solved, and feature fusion is realized by allocating different weights to two channels.

The invention relates to a ferroelectric memristor array and a sequential sequence processing method adopting the ferroelectric memristor array, and the ferroelectric memristor array comprises a bottom array gating structure which is used for providing unit-by-unit addressing and suppressing an undercurrent path of a non-gated unit in the array when the array scale is expanded; the adjustable conductive channel structure is used for generating a continuous photoconductive effect which can be accumulated and attenuated along with time under optical excitation; the in-plane polarization regulation and control structure modifies the conductive degree of the semiconductor channel by regulating and controlling the surface potential barrier of the channel so as to form a non-volatile weight state; and the multi-physical-quantity coupled electrode structure is used as a unified channel for weight regulation and control, conductivity reading and photoelectric response. Compared with the prior art, the semiconductor channel potential barrier is dynamically regulated and controlled through in-plane polarization of the ferroelectric film, non-volatile multi-stage conductance weight is achieved, synaptic-like photoresponse and short-time memory characteristics are obtained through the continuous photoconduction effect of a semiconductor material, and the continuous light input natural accumulation and attenuation process is achieved.

The invention discloses an expansive soil slope hydrothermal control active area identification method. The method comprises the following steps: acquiring a vertical deformation time sequence of each pixel of a target area and a rainfall time sequence of a corresponding time period; processing the two groups of time sequences, and extracting periodic response characteristics representing the moisture-driven deformation process; constructing a hydrothermal transfer model reflecting a swelling soil moisture diffusion mechanism and a depth integral deformation mechanism so as to establish a physical mapping relation between theoretical prediction response characteristics and normalized active area depth; and substituting the periodic response characteristics into a hydrothermal transfer model to carry out inversion matching solution, and calculating the total depth of the active area of each pixel in combination with a pre-calibrated moisture diffusion coefficient. According to the method, the theoretical upper limit of depth inversion of a traditional single-layer time domain model is effectively broken through, and accurate physical quantitative recognition of the depth of the active area under the complex hydrothermal boundary is achieved.

The embodiment of the invention discloses a service data processing method and device, equipment and a medium, and the method comprises the steps: obtaining an incremental data calculation task data packet of a target service, and subscribing a target data theme based on the task configuration information of the incremental data calculation task data packet, so as to start a corresponding incremental data calculation task; obtaining and identifying incremental business data of the target data theme; and performing time sequence processing on the incremental business data, and updating a data chart associated with the incremental business data based on a time sequence processing result so as to complete the incremental data calculation task. According to the technical scheme provided by the embodiment of the invention, the problems that the business data display updating is slow and the calculation amount in the data updating process is large are solved, the real-time dynamic updating aiming at the incremental business data is realized, and the real-time performance and the updating efficiency of the business data chart display updating are improved.

The invention relates to a method suitable for predicting the power of an electric vehicle charging station, belongs to the field of power prediction of charging stations, and solves the problems of the power prediction method of the electric vehicle charging station in the aspects of data acquisition and preprocessing methods, time sequence processing capability, prediction precision, stability and the like. The power prediction refers to predicting the power demand or supply of the charging station at a certain or certain time point in the future through a certain mathematical model and algorithm. The invention provides an electric vehicle charging station power short-term prediction method based on transformer historical data, which is characterized in that historical voltage, current and power data recorded by a transformer are read as prediction input signals, correlation coefficients of input features and output power are calculated, and a prediction model fusing a plurality of deep neural networks is constructed; in the first part, historical data of various transformers are calculated to serve as correlation coefficients of to-be-selected input characteristics and output power, the input characteristics which are most beneficial to improvement of prediction result accuracy are selected, interference is reduced, and calculation overhead is reduced; in the second part, effective representation in an input sequence is captured by using a convolutional neural network, so that the complexity of a prediction model is reduced, and model convergence is accelerated; in the third part, the time dependence characteristic of sequence data is captured by using a long short-term memory network, and information is stored and updated by using a memory unit; and in the fourth part, the attention weight of each time step is calculated for the output result of the previous layer by using an attention mechanism, and the time step which is most important for prediction is highlighted. According to the prediction method, the time sequence processing capability is optimized, and the power prediction precision and stability are improved.