Multi-source Heterogeneous Data Research Articles

Correlation among Composition, Microstructure and Hardness of 7xxx Aluminum Alloy Using Original Statistical Spatial-Mapping Method

The quantitative study of the relationship between material composition, microstructure and properties is of great importance for the improvement in material properties. In this study, the continuous data of elemental composition, recrystallization, hardness and undissolved phase distribution of the same sample in the range of 60 to 150 square millimeters were obtained by high-throughput testing instrument. The distribution characteristics and rules of a single data set were analyzed. In addition, each data set was divided into micro-areas according to the corresponding relationship of location, and the mapping between multi-source heterogeneous micro-area data sets was established to analyze and quantify the correlation between material composition, structure and hardness. The conclusions are as follows: (1) the average size of the insoluble phase in the middle of the two materials is larger than that of the surface, but due to the existence of central segregation, the average area of the T4 insoluble phase showed an abnormal decrease; (2) there was positive micro-segregation of Al, Cr, Ti, and Zr elements, and negative micro-segregation of Zn, Cu, and Fe elements in the recrystallized grains of the T5 middle segregation zone; (3) the growth process of the insoluble phase was synchronous with the recrystallization proportion and the size of the recrystallized grains; (4) the composition segregation and recrystallized coarse grains were the main reasons for the formation of low hardness zone in T4 and T5 materials, respectively.

A graph-based approach to multi-source heterogeneous information fusion in stock market

The stock market is an important part of the capital market, and the research on the price fluctuation of the stock market has always been a hot topic for scholars. As a dynamic and complex system, the stock market is affected by various factors. However, with the development of information technology, information presents multisource and heterogeneous characteristics, and the transmission speed and mode of information have changed greatly. The explanation and influence of multi-source and heterogeneous information on stock market price fluctuations need further study. In this paper, a graph fusion and embedding method for multi-source heterogeneous information of Chinese stock market is established. Relational dimension information is introduced to realize the effective fusion of multi-source heterogeneous data information. A multi-attention graph neural network based on nodes and semantics is constructed to mine the implied semantics of fusion graph data and capture the influence of multi-source heterogeneous information on stock market price fluctuations. Experiments show that the proposed multi-source heterogeneous information fusion methods is superior to tensor or vector fusion method, and the constructed multi-attention diagram neural network has a better ability to explain stock market price fluctuations.

How to improve urban transportation planning in big data era? A practice in the study of traffic analysis zone delineation

Traffic analysis zone (TAZ) is the basic unit of urban transportation planning. The appropriateness of TAZ delineation will affect the rationality of transportation planning and the accuracy of transportation analysis, and thus affect the final planning decision-making. However, in the big data era, there is still a lack of methods to integrate multi-source data for TAZ delineation. How to effectively fuse multi-source heterogeneous data in the TAZ delineation process is still an unsolved technical difficulty. To fill this gap, this paper designs a multi-source data-driven stepwise strategy to solve the TAZ delineation problem by creating a zoning system with zones showing homogeneous mobility behaviors and containing homogeneous land use characteristics. Firstly, mining the spatial-temporal travel features and land use information of transit stations and parcels from multi-source data, including transit smart card data, ride-hailing data, bike-sharing trip data, and point-of-interest (POI) data. Then, a core parcel determination algorithm which mainly consists of Fuzzy C-Means station clustering and a constructed parcel core degree function is proposed to generate the core parcels of each potential TAZ. After that, parcels are clustered around the core ones into TAZs through a multi-feature driven clustering algorithm, in the process of which the homogeneity within TAZs is guaranteed. Finally, the optimal zoning system is obtained by comparing the information loss calculation result of multiple zoning schemes. Taking Beijing as a case study area, 624 TAZs are obtained by applying the proposed method. The delineation result is comparatively analyzed with TAZs obtained by the traditional delineation method and a baseline method by applying multi-indicator measurement. The result reveals that the presented approach can promote the rationality of TAZ delineation.

Effect of Innovation Community on Innovation Performance in the Multi-layered Innovation Ecosystem

The existing literature on innovation ecosystems either focuses on whole-network at the macro-level or emphasizes a firm’s ego-network at the micro-level, while few studies have investigated network community analysis at the meso-level. Additionally, innovation ecosystem studies overlook the differences between knowledge ecosystems and business ecosystems and mostly focus on a single-layer analysis of the ecosystem. Therefore, the current paper aims to explore the mechanism of knowledge-business community connectivity in a two-layered innovation ecosystem on firm innovation performance. Based on multi-source heterogeneous data such as patents and VAT invoices, using social network analysis, Newman topological clustering and multiple regression, we analyze community connectivity and firm innovation performance. We find that the connectivity of within-knowledge communities and cross-business communities significantly positively affect firm innovation performance; the connectivity of cross-knowledge communities and within-business communities has a significant negative impact on firm innovation performance; in addition, the interaction of knowledge-business community connectivity has a significant negative effect on firm innovation performance. This study deepens and enriches the study of the mechanism of innovation communities on innovation performance at the meso level, and expands innovation ecosystem research from the perspective of knowledge-business two-layer networks; it provides insights for China to construct and improve innovation ecosystems and promote innovation-driven development.

A Meta-Knowledge Transfer-Based Differential Evolution for Multitask Optimization

Knowledge transfer plays a vastly important role in solving multitask optimization problems (MTOPs). Many existing methods transfer task-specific knowledge, such as the high-quality solution from one task to other tasks to enhance the optimization ability, which, however, may not work well or even have a negative effect if the tasks have very different task-specific knowledge. Hence, this article proposes a meta-knowledge transfer (MKT)-based differential evolution (MKTDE) algorithm by using a more general MKT method to solve MTOPs more efficiently. The meta-knowledge defined in this article refers to the knowledge that can evolve task-specific knowledge during the evolutionary search. That is, the meta-knowledge is a kind of “ <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">knowledge of knowledge</i> ,” which denotes the knowledge of “ <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">how to solve problem via evolution</i> ” and “ <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">the feature/way/method of evolving high-quality solution</i> .” The evolutionary search for solving different tasks can share common meta-knowledge even though these tasks involve heterogeneous data and have very different task-specific knowledge. Therefore, the MKT can associate the heterogeneous multisource data of different tasks via transferring the meta-knowledge to help solve MTOPs more efficiently in a more general way. Moreover, to further enhance the MKTDE, two novel and efficient methods are proposed. One is multiple populations for the multiple tasks framework using a unified search space for making knowledge transfer flexibly. The other is an elite solution transfer method for achieving positive high-quality solution transfer. The superior performance of the proposed MKTDE is verified via extensive numerical experiments on both widely used MTOP benchmark problems and real-world robot navigation problems, with comparisons with some state-of-the-art and the latest well-performing algorithms.

Meta-Prism 2.0: Enabling algorithm and web server for ultra-fast, memory-efficient, and accurate analysis among millions of microbial community samples.

BackgroundMicrobial community samples have been accumulating at a speed faster than ever, with hundreds of thousands of samples been sequenced each year. Mining such a huge amount of multisource heterogeneous data is becoming an increasingly difficult challenge, so efficient and accurate compare and search of samples is in urgent need: faced with millions of samples in the data repository, traditional sample comparison and search approaches fall short in speed and accuracy.FindingsHere we proposed Meta-Prism 2.0, a microbial community sample analysis method that has pushed the time and memory efficiency to a new limit without compromising accuracy. Based on sparse data structure, time-saving instruction pipeline, and SIMD optimization, Meta-Prism 2.0 has enabled ultra-fast, memory-efficient, flexible, and accurate search among millions of samples. Meta-Prism 2.0 was put to test on several data sets, with the largest containing 1 million samples. Results show that Meta-Prism 2.0’s 0.00001-s per sample pair compare speed and 8-GB memory needs for searching against 1 million samples have made it one of the most efficient sample analysis methods. Additionally, Meta-Prism 2.0 can achieve accuracy comparable with or better than other contemporary methods. Third, Meta-Prism 2.0 can precisely identify the original biome for samples, thus enabling sample source tracking. Finally, we have provided a web server for fast search of microbial community samples online.ConclusionsIn summary, Meta-Prism 2.0 has changed the resource-intensive sample search scheme to an effective procedure, which could be conducted by researchers every day even on a laptop, for insightful sample search, similarity analysis, and knowledge discovery. Meta-Prism 2.0 can be accessed at https://github.com/HUST-NingKang-Lab/Meta-Prism-2.0, and the web server can be accessed at https://hust-ningkang-lab.github.io/Meta-Prism-2.0/.

A framework for development of digital twin industrial robot production lines based on a mechatronics approach

During the past years, intelligent manufacturing has attracted enormous attention from both academia and industry. Smart factories/workshops/production lines are important carriers for realizing intelligent manufacturing. However, there are difficulties in smart factories in processing multi-source heterogeneous data, in integrating virtual and physical worlds, and in achieving continuous and iterative optimization with real-time feedback. Digital twin, which is able to deeply integrate physical and virtual worlds, provides an important technology for easing the difficulties mentioned above. The construction of a digital twin system can solve the problem of data fusion, analysis, processing, and applications in smart factory. Digital production lines are the foothold of a smart factory. Building a digital twin production line can facilitate rapid practice of digital twins and promote the development of intelligent manufacturing. To this end, this paper proposes a development framework for a digital twin production line with closed-loop control based on a mechatronics approach. The detailed development process is given based on the framework. The framework proposed may provide some inspiration and insights for design and development of digital twin production lines.

Research on intelligent station site selection method based on multi-source and heterogeneous data

The traditional station site selection work is mainly based on manual exploration by engineers whose shortcomings are timeconsuming, laborious, and uncertain. This paper studies the correlation between multi-source heterogeneous data, such as land spatial planning data, geographic data and power grid thematic data and power grid project station site selection. First of all, an adaptive particle swarm optimization (PSO) algorithm for processing the geographic information (GI_PSO) is proposed. The algorithm can not only adaptively adjust the particle&apos;s motion strategy and speed based on the particle diversity, but also adaptively adjust the number of particles based on the distance between extreme points. Second, the performance of the algorithm is evaluated in the extremum search of a PSO common test function. Then the multi-source heterogeneous data is processed to meet the needs of station selection, and the details of the algorithm used to find multiple sites is presented. Finally, an example of searching the optimal three station sites with the lowest cost in an area of about 60 km×45 km is used to verify the correctness of the method in the paper.

Spatio-Temporal Knowledge Graph Based Forest Fire Prediction with Multi Source Heterogeneous Data

Forest fires have frequently occurred and caused great harm to people’s lives. Many researchers use machine learning techniques to predict forest fires by considering spatio-temporal data features. However, it is difficult to efficiently obtain the features from large-scale, multi-source, heterogeneous data. There is a lack of a method that can effectively extract features required by machine learning-based forest fire predictions from multi-source spatio-temporal data. This paper proposes a forest fire prediction method that integrates spatio-temporal knowledge graphs and machine learning models. This method can fuse multi-source heterogeneous spatio-temporal forest fire data by constructing a forest fire semantic ontology and a knowledge graph-based spatio-temporal framework. This paper defines the domain expertise of forest fire analysis as the semantic rules of the knowledge graph. This paper proposes a rule-based reasoning method to obtain the corresponding data for the specific machine learning-based forest fire prediction methods, which are dedicated to tackling the problem with real-time prediction scenarios. This paper performs experiments regarding forest fire predictions based on real-world data in the experimental areas Xichang and Yanyuan in Sichuan province. The results show that the proposed method is beneficial for the fusion of multi-source spatio-temporal data and highly improves the prediction performance in real forest fire prediction scenarios.

Intelligent Prediction Model of CET-4 Passing Rate Based on Multisource Heterogeneous Data Mining

With the growing software of English in society, English gaining knowledge is turning into extra and greater important. The software and lookup of fact mining science in English educating and gaining knowledge will turn out to be a new improvement trend. This study is based on the information of non-English majors’ learning situation of Public English course and CET-4 scores. It selects the data related to CET-4 scores as the characteristics, takes CET-4 passing rate as the goal, and makes use of selection tree classification mannequin to predict the passing likelihood of CET-4. After comparison and analysis, it is observed that the multisource heterogeneous statistic mining science can precisely predict students’ grades, which verify the software of multisource heterogeneous information mining technological know-how in university English studying and have an impact on of the mannequin on the end result prediction. The results of College English final examination and gender are the main factors affecting the passing rate of CET4. The higher the score of College English final examination, the higher the passing rate of CET-4. The results of this study show that girls’ passing rate and average score of CET-4 are higher than boys’. Other factors have little influence on CET4 passing rate and can be ignored. There is a significant positive relationship between the scores of CET-4, CET-1, and Putonghua. After analysing the number attributes of the model, blended with the statistical consequences of information analysis, this paper offers some ideas and hints to enhance the instructing administration and the passing charge of CET-4.

On the determination and rank for the environmental risk aspects for ship navigating in the Arctic based on big Earth data.

The unique, ambiguous, and complex navigable environment determines the essential difference between Arctic shipping routes and conventional routes in regard to safety issues. To achieve a scientific understanding of the characteristics and variations of environmental risks involved in the Arctic shipping, it is essential to rationally address the uncertainty and incompleteness of environment-related risk information. In this study, fuzzy evidential reasoning is introduced to carry out multisource heterogeneous data fusion and spatiotemporal dynamic assessment of navigable environmental risks for Arctic shipping routes. Based on big Earth data collected from the European Center for Medium-Range Weather Forecasts, National Snow And Ice Data Center, National Center for Environmental Information, and University of Bremen from 2012 to 2019, a case study of the Northeast Passage is considered to demonstrate the feasibility of the proposed methodology. Finally, the results are described from three aspects: spatial distribution, temporal changes, and sensitivity analysis, with consideration of the entire passage and five marginal seas at the same time. Based on these findings, the prospect of application of big Earth data in risk assessment is further discussed from two aspects of knowledge acquisition by big data and risk analysis at different scales, to inspire sustainable development of Arctic shipping.

Construction of an Integrated Framework for Complex Product Design Manufacturing and Service Based on Reliability Data

With the application of new-generation information technology in the full life cycle process of a complex product, it is showing the characteristics of multi-source, real-time, heterogeneous, cross-domain transmission. Large data volume and low value density emerge in the process of complex product design manufacturing and services (DMS). This leads to “information islands” and insufficient utilization of cross-domain reliability data in the process of integration of DMS for complex product R&amp;D design data, manufacturing data and operation and maintenance services (O&amp;MS) data. This paper proposes and illustrates a framework of complex product DMS integration based on reliability data, including complex product design optimization based on manufacturing and service reliability data, complex product intelligent manufacturing process optimization based on real-time reliability data and complex product O&amp;MS optimization based on multi-source heterogeneous reliability data. Additionally, it then realizes complex product design reliability and optimization, manufacturing process reliability and optimization and O&amp;MS reliability and intelligent decision optimization based on reliability data. Finally, the DMS integration framework based on reliability-data-driven proposal is corrected through the case of engine MDS integration, which can effectively improve the cross-domain reliability data utilization and overall product reliability of complex products. The proposed framework extends the application of reliability theory in the process of complex product DMS integration and provides a reference for enterprises in the R&amp;D, manufacturing and O&amp;MS of complex products.

Research on the discourse power evaluation of academic journals from the perspective of multiple fusion: Taking Medicine, General and Internal journals as an example

In the open science environment, this article evaluates the discourse power of academic journals from the perspective of multiple integration. It is conducive to scientific research management and provides a reference for enriching and perfecting the evaluation theory and indicators system of academic journals. Based on the theory of evaluation science, discourse power theory and communication theory, first, this article discusses the basic issues of the discourse power evaluation for academic journals. It defines the academic discourse power and the discourse power of academic journals. It is proposed that the discourse power of academic journals is composed of discourse influence and discourse leading. Discourse influence is composed of discourse influence ability and discourse power, and discourse leading is composed of social media discourse, news and policy discourse, encyclopaedia discourse, peer-review discourse and video discourse leading. This article explores the formation process and operation mechanism of the discourse power for academic journals. Then, this article constructs the discourse power evaluation model of academic journals. Second, this article integrates multi-source heterogeneous data, then adopts correlation analysis, integrated factor analysis, entropy weight method, Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) method and two-dimensional four-quadrant mapping method to conduct empirical research on the discourse power evaluation of Medicine, General and Internal journals from the perspectives of multi-dimensions, multi-factors, multi-indicators and multi-methods fusion. The results show that the research on the discourse power evaluation for academic journals based on the theory, method and application logic is practical, comprehensive and reliable.

A convolutional neural network-based multi-sensor fusion approach for in-situ quality monitoring of selective laser melting

Selective laser melting (SLM) is an emerging and popular metal additive manufacturing (AM) technique to fabricate advanced metal components with complex geometries. However, its broad adoption in industry is still hampered by poor process repeatability and low part consistency. To overcome this limitation, there are growing efforts towards in-situ monitoring and control technologies for quality assessment of parts. In this paper, a convolutional neural network (CNN)-based multi-sensor fusion approach is proposed to integrate layer-wise images, acoustic emission signals, and photodiode signals for in-situ quality monitoring of SLM. An off-axial in-situ monitoring system equipped with three types of sensors, namely, a digital camera, a microphone, and a photodiode is first developed for signatures acquisition. Porosity and density measurements are then carried out to label the layer-based sensing data and measure the quality. Thereafter, an image processing method and a signal-to-image strategy are proposed to merge the multi-source heterogeneous sensor data. Finally, three CNN-based multi-sensor fusion models are developed from data-level fusion, feature-level fusion, and decision-level fusion respectively and compared in quality identification. Data-level fusion model is established through channel fusion, features are extracted and fused in the feature-level fusion model, and the decision-level fusion model is developed by fusing the classification results from individual models. Results showed that the proposed CNN-based multi-sensor fusion approach can significantly improve the classification accuracy compared to the three individual sensor-based models. Furthermore, the feature-level data fusion model exhibits the best classification performance among the three data fusion models.

English Teaching Ability Evaluation Algorithm based on management, leadership and entrepreneurship including education, Data Fusion and Notch Filtering

All universities and colleges has been offering English as one of the important courses. The effectiveness of English teaching is inextricably related to the quality of talent acquisition and student growth. The term evaluation of teaching quality describes the type of content knowledge in which the focus is to analyze quality teaching. One of the significant aims of every school and university is to improve the English teaching quality and promote the development of English teaching and entrepreneurship. As the existing evaluation algorithms have been facing various problems of fuzzy core dimension characteristics, which result in poor data integrity. Therefore, we proposed an English teaching ability evaluation algorithm based on data fusion, entrepreneurship, and notch filtering. Moreover, for the purpose, to achieve the goal of education entrepreneurship and quality teaching, strengthen of the English teaching quality algorithm is necessary, as well as the leadership and management qualities of the teachers also have a lot of impact on the performance of teaching quality and students working performance and it is a topic worthy of in-depth and significant investigation. We select the explicit indicators that are easy to measure, establish the hierarchical structure system of evaluation indicators, construct the evaluation model by data fusion and notch filtering, obtain multi-source heterogeneous data, extract the core dimension characteristics of English teaching ability, make rational use of teaching resources, convert variables into standard units, and set up the evaluation mechanism by using random walk algorithm. Experimental results: the average data integrity of the designed English teaching ability evaluation algorithm is 4.33, and the average data integrity of the other two evaluation algorithms is 1.921 and 1.885, which proves that the English teaching ability evaluation algorithm integrating data fusion and notch filtering technology has a broader application prospect.

Design of English Teaching Sharing System Combining Internet of Things and Memory Mechanism

In the era of Internet of Everything, English teaching has evolved continuously undergoing radical changes. Informatization and modern technologies are continuously promoting the reform of education, and the application of these technological tools makes information sharing a reality. Timely information sharing contributes to the development of English teaching as a part of curriculum requirements. As a consequence of this rapid progress, the theoretical and practical research on IoT-based English teaching sharing systems is also increasing. In this paper, we initiate an exhaustive research of the popular methodologies and then adopt LSTM network management platform data to address the shortcomings of the current teaching system has memory mechanism issues. At the outset, the standard English curriculum is discussed to establish the background, and then, data fusion among various types of information is considered. This is done by completely utilizing independent and scattered heterogeneous data in the information platform without changing or minimizing the information available in the shared system. Then, the information fusion is thoroughly analyzed resulting in the proposal of a multisource and multigranularity information fusion method based on a neural network. The method implements a study considering multisource heterogeneous data, which is followed by carrying out of functional design and process design. The paper also conducts a study on CNN-based multisource information association and designs information association algorithms to customize the dataset. Finally, the study of multigranularity feature fusion based on CNN and LSTM is implemented. The design relevant to granularity calculation algorithm helps in the customization of the dataset. The multigranularity feature fusion is conducted integrating CNN and LSTM. The experimental results show that the proposed method effectively integrates the scattered teaching information in the system, which is conducive to improving the utilization rate of teaching resources and simultaneously promoting the development of teaching information.

Research on the Transformer Intelligent Operation and Maintenance System Based on the Graph Neural Network

With the development of the smart grid and energy Internet, the power industry generates huge, multi-source, heterogeneous, and highly coupled data, which are difficult to utilize. The intelligent operation and maintenance system of the power transformer based on the knowledge graph and graph neural network is developed in this article. The multi-source heterogeneous data are structured and modeled by the constructed knowledge graph, and it presents the correlation among data more intuitively. On this basis, the graph neural network is designed to achieve the prediction and excavate the deep information hidden in the data. The testing results show that the system has fully used the multi-dimensional and interrelated heterogeneous data, achieving a deep information mine. It benefits the management and strategy implementation for the system scientifically and guides the operation and maintenance of the transformer. The system is of great significance on improving the efficiency of the transformer maintenance and safe operation.

Digital Twin-driven approach to improving energy efficiency of indoor lighting based on computer vision and dynamic BIM

Intelligent lighting systems and surveillance systems have become an important part of intelligent buildings. However, the current intelligent lighting system generally adopts independent sensor control and does not perform multi-source heterogeneous data fusion with other digital systems. This paper fully considers the linkage between the lighting system and the surveillance system and proposes a digital twin lighting (DTL) system that mainly consists of three parts. Firstly, a visualized operation and maintenance (VO&M) platform for a DTL system was established based on dynamic BIM. Secondly, the environment perception, key-frame similarity judgment, and multi-channel key-frame cut and merge mechanism were utilized to preprocess the video stream of the surveillance system in real-time. Lastly, pedestrians detected using YOLOv4 and the ambient brightness perceived by the environment perception mechanism were transmitted to the cloud database and were continuously read by the VO&M platform. The intent here was to aid timely adaptive adjustment of the digital twin and realistic lighting through the internet. The effectiveness of the proposed method was verified by experimenting with a surveillance video stream for 14 days. The key results of the experiments are as follows: (1) the accuracy rate of intelligent decision control reached 95.15%; (2) energy consumption and electricity costs were reduced by approximately 79%; and (3) the hardware cost and energy consumption of detection equipment and the time and cost of operation and maintenance (O&M) were greatly reduced.

A dual energy benchmarking methodology for energy-efficient production planning and operation of discrete manufacturing systems using data mining techniques

Adequate energy awareness is an essential prerequisite for energy-efficient production planning and operation in manufacturing systems. Nonetheless, a lack of awareness of the impact of managerial and operational energy-influencing factors on energy consumption can be observed in practice due to irrational and inexplicable energy benchmarks. To this end, this paper presents a dual energy benchmarking methodology that specifically promotes energy awareness in terms of managerial and operational factors. The presented methodology uses data mining techniques to excavate energy data as well as managerial and operational data, consisting of three steps: (i) data preparation to provide reliable multi-source heterogeneous data; (ii) dynamic energy benchmarking that uses the decision tree algorithm to classify different energy consumption patterns; (iii) load shape benchmarking that interprets energy use behaviors based on load shape features. To demonstrate the effectiveness and practicality, the methodology was implemented in a die-casting workshop. The results showed that a total of nine dynamic energy benchmarks were created primality under the impact of capacity utilization, two of which generated five representative load curves to characterize the workshop's daily energy use behaviors. It was concluded that the methodology could provide practical recommendations on energy-efficient production planning and operation in practical applications.

M2TNet: Multi-modal multi-task Transformer network for ultra-short-term wind power multi-step forecasting

Accurate wind power forecasting is crucial for the safe, stable, and economical operation of green power systems. Multi-step forecasting of wind power has received increasing attention in recent years. From the perspective of machine learning, multi-step prediction can be transformed into a multi-task learning problem, i.e., multiple single-step prediction tasks. From a data-driven perspective, it is often difficult to respect the differences and relationships among multi-source heterogeneous data that are typically used in multi-step forecasting research, e.g., wind power and numerical weather prediction (NWP) data. This work proposes a multi-modal multi-task transformer network (M2TNet) model that can achieve ultra-short-term wind power multi-step forecasting based on multi-source heterogeneous data. The M2TNet is a unified framework that integrates multiple feature extractors, including a Transformer, a feature fusion layer-based fully-connected network, and a prediction terminal layer. The developed model applies multi-modal and multi-task learning strategies to effectively fuse multi-modal information and enable knowledge sharing among multiple single-step prediction tasks. In addition, the Transformer’s computing efficiency and ability to mine complex dependent data is exploited for joint learning of multi-source heterogeneous data. Data, including NWP and wind power, from a wind farm in Northeast China were used to validate M2TNet. The correlations between the input variables were analyzed using maximal information coefficient method to control the scale of the model. For prediction accuracy, the experimental results showed that, compared with the existing model, M2TNet reduced the root mean square error by 0.19%, 0.99%, 1.05%, and 1.53% in 4-, 8-, 12-, and 16-step ahead predictions, respectively. Furthermore, for computational efficiency, the training time of the existing model at a wind farm is 1.66 times that of M2TNet. This confirmed that the M2TNet model performs better in terms of prediction accuracy and computational efficiency. Our work illustrates the potential of M2TNet for large-scale wind farm applications.