Visual design of green information in urban environment based on global similarity calculation and multi-dimensional visualization technology

Knowledge extraction framework

To visualize green and low-carbon information in urban environmental design, this article integrates data mining and visualization technologies to extract knowledge from low-carbon information on the network. Figure 2 illustrates the entire process, which uses keywords such as low-carbon, carbon neutral, carbon cycle, and footprint and employs automatic data mining algorithms to develop corresponding models based on existing data. Through interpretation and validation of the models, analysis results or embedded knowledge is extracted. Figure 3 depicts visual mapping based on this knowledge, using visualization techniques to represent the input data in a certain visual form. The user can explore and extract features and patterns that meet their requirements from the visualization results via visual channels. In this interactive visual model, the extracted knowledge results generate new visualization outcomes by receiving interactive feedback from the user.

Similarity-based algorithm for arranging green low-carbon information dimensions

Cluster analysis, a statistical technique that has been extensively researched for many years, has mainly focused on improving distance-based cluster analysis, while cluster analysis tools based on K-means, K-medoids, and other methods have been integrated into statistical analysis packages or systems. In visualization implementations, observational learning using cluster analysis can eliminate the influence of example-based learning. In contrast to traditional clustering, which uses geometric distance to measure similarity, concept clustering groups objects together only if they can be described by a common concept. Cluster analysis can be applied to perform untrained observational learning on data when used in data mining for digital visualization. This article uses a memory-based clustering algorithm to mine information from a dataset of low-carbon information containing n data objects. The K-means algorithm is used to randomly select K initial cluster centers from known data points, and then iteratively assign each data point to the nearest cluster center, and update the location of the cluster center until the convergence condition is met. During this iteration, all data points are stored in memory. A similarity matrix is used to store the proximity between two of the n objects. First, for each number of points, calculate the similarity between them. By calculating the similarity between all data points, a similarity matrix is constructed. Each element of the matrix represents a similarity score between two data points. On the basis of the similarity matrix, the logarithmic data points are sorted using a suitable ranking algorithm. Finally, the optimal sorting result is obtained by adjusting the parameters based on the clustering algorithm. Let the similarity matrix be as follows:

(1) $$\left[\begin{array}{cccc}0& & & \\ d(2,1)& 0& & \\ ⋮& ⋮& 0& \\ d(n,1)& d(n,2)& \dots & 0\end{array}\right]$$

Here $d\left(i,j\right)$ is a quantitative representation of the similarity between $d_i(x_{i1},x_{i2},....,x_{in})$ and $d_j(x_{j1},x_{j2},....,x_{jn})$. Since $d\left(i,j\right)$ and $d\left(j,i\right)$ represent the same meaning, only the triangular array is written down, and the value of $d\left(i,j\right)$ is larger when objects i and j are more similar to each other and closer to 0, and vice versa. By calculating the similarity of data, it is possible to differentiate, classify and cluster the analysis of massive data. The simplest way to calculate the similarity (or difference) between objects is based on the distance between objects. The distance function is as follows:

(2) $$d(i,j)=\sqrt[q]{w_1{(x_{i1}-x_{j1})}^q+w_2{(x_{i2}-x_{j2})}^q+...+w_n{(x_{in}-x_{jn})}^q}$$where the values of $w_i(i\in \{1,2,...,n\})$ and $q$ are taken as follows

(3) $$\{\begin{array}{c}{w}_i=1,q=1\\ w_i=1,q=2\\ w_i=1,q\ge 2\\ w_i\in \text{ℤ} ,q=2\end{array}$$

After obtaining the similarity matrix between data objects, a similarity-based method can be employed to compare the similarity of each dimension in the parallel coordinate system. This will address the issue of data organization affecting the display effectiveness. An optimization algorithm can then be utilized to derive the best combination scheme for guiding the visualization display. The similarity between dimensions is obtained as follows:

(4) $$V_i=(x_{1i},x_{2i},...,x_{ni}{)}^T$$

The global similarity algorithm takes all the data as the object of study and computes a vector of attribute components of all the numbers choked in a certain dimension. In this way the laws implied within all the data can be represented quantitatively, let $mean(V_i)={\displaystyle \frac{1}{N}\sum _{j=0}^{N-1}{a}_{ij}}$, then we get:

(5) $$S(V_k,V_i)=\sqrt{\sum _{i=0}^{N-1}((a_{ik}-mean(V_k))-{((a_{il}-mean(V_l)))}^2}$$

This concludes the dimensional arrangement of the multidimensional data and enables clustering of the data for urban environment design.

Multi-dimensional information visualization

The visualization of green and low-carbon information is carried out in three main steps. Firstly, the information is collected from open platforms and networks using keywords related to low-carbon information. Secondly, the collected information is processed in a knowledge extraction framework, where data mining techniques are used to extract relevant information and present it in a simple and attractive form of knowledge representation, while discarding irrelevant information. Thirdly, the extracted multidimensional knowledge is arranged in dimensions, and the global similarity of images is determined to obtain the structure of the training model. After unsupervised learning of the training set, the design is evaluated based on the effectiveness of green low-carbon information visualization design criteria (as shown in Fig. 4). The design is implemented using the MVVM (Model-View-View-Model) pattern, utilizing the feature of two-way binding of Vue data, and enabling two-way interaction through the agreed logical relationship between View-Model. We set up a bidirectional data binding mechanism between the View and the view-model. This means that when the View’s data changes, the View-Model updates accordingly, and when the View-Model’s data changes, the View updates accordingly. When a user interacts with a View, the view passes the results of that interaction to the View-Model through a data binding mechanism. These interactions can be captured and processed using command patterns or other event handling mechanisms. When the View-Model’s data changes, it passes those changes to the View through a data binding mechanism. The view senses these changes and updates its display accordingly.

Data processing

For the experimental data utilized in this article, one portion was acquired through Python, while the other was obtained via DataFountain (DF platform), a preeminent collaborative innovation platform for data science and artificial intelligence in China, which furnishes artificial intelligence datasets, Big Data competitions, and open source sharing communities. An experimental dataset was procured from the DataFountain platform for the purposes of this article, and the efficacy of the proposed scheme was substantiated using this dataset. In addition, feature cleaning was executed on the dataset to ameliorate the effectiveness of the data. Details pertaining to the data cleaning process encompassed missing value processing (Rahman & Islam, 2011), outlier processing (Dong & Pan, 2021), normalization processing, Top-N category extraction (Wu, Macdonald & Ounis, 2020), and category merging. Carbon emission is an indispensable indicator in urban environmental design and low carbon analysis. The count of missing values in carbon emissions is generally trifling, and a definite value or interval of values is assigned to most industries. As evinced in Fig. 5, among the 8,720 records collected, merely 105 records in the carbon emissions field had missing values, which can be effortlessly expunged for analysis purposes owing to their scant quantity. Subsequently, the upper limit value of carbon emission, lower limit value of carbon emission, average carbon emission, and logarithmic average carbon emission were calculated and recorded in four novel columns, respectively. After once more cleaning the data features, the interrelationships between each feature in the dataset were clarified, followed by the selection of an appropriate graph for visual representation. The visualization was executed through multiple dimensions such as numerical distribution observation, single-factor analysis, multi-factor analysis, and word cloud diagram.

Comparative performance

As illustrated in Fig. 6, the comparative analysis of green low carbon information for clustering was conducted, wherein the results of literature (Song & Wu, 2022), literature (Shao et al., 2022), and literature (Mehbodniya et al., 2022) were contrasted. Ultimately, the cluster analysis dataflow anomaly detection group proposed in this article demonstrated a dataflow anomaly detection false alarm rate of merely 1.77% and a clustering difference of 1.34%, thereby realizing a 56% performance enhancement over literature (Shao et al., 2022), as well as a 46% performance improvement overall. As evidenced by its clustering effect, the model advanced in this article achieved the most exceptional performance with regard to the expression of urban green and low carbon information visualization.

Moreover, the stability of the interconnected relationship between relevant low-carbon information serves as a measure of information’s diversified display capability. Typically, the larger the result of the interconnection coefficient calculation, the stronger the diversified display capability of information, thereby enhancing the practicality of the platform system. Thus, an evaluation experiment is conducted to measure the diversified display capability, and the experimental results are presented in Fig. 7. Given that literature (Mehbodniya et al., 2022) exhibits the highest effect capability in terms of loss rate, the model presented in this article is compared to literature (Mehbodniya et al., 2022) for diversified display capability. The experimental training sets are categorized into three groups based on information metrics a, b, and c, respectively, and input into the built training model. It should be noted that the interconnection coefficient g represents the physical variable related to low-carbon information, and the calculation result of the interconnection coefficient g will vary with the change of the training data input. $g=D\cdot {\displaystyle \frac{\eta }{u}}$, $\eta (\eta \in \{a,b,c\})$ is the green low carbon information to be displayed after performing clustering, and $u$ represents the constant measurement coefficient.

This article includes a similarity-based dimension alignment algorithm, which effectively selects relevant dimensions for data mining and visualization display, thereby reducing the interference of irrelevant dimensions during mining and visualization tasks. Additionally, the proposed scheme performs correlation analysis on attributes, utilizing feature cleaning to filter out statistically irrelevant or weakly correlated attributes, while retaining the most pertinent attributes for mining or display purposes. As a result, the interconnection coefficient of the model presented in this article is consistently higher than that of literature (Mehbodniya et al., 2022) for different numbers of training iterations.

To demonstrate the superiority of the proposed model and provide a comprehensive evaluation, we conducted a thorough analysis with various experiments and data processing steps. Firstly, we applied a loss function analysis on both the Python collection and the dataset obtained from the DataFountain platform. This allowed us to evaluate the model’s performance on diverse datasets and determine its generalizability. Figure 8 provides a visual representation of the results obtained from this analysis, depicting the model’s performance and loss values.

Before proceeding with the analysis, we preprocessed the overall data to ensure its quality and consistency. We performed data cleaning, normalization, and feature extraction to eliminate noise and enhance the dataset’s representativeness. Subsequently, we focused on the “green low carbon” information from 2000 to 2015, a critical period marked by the evolution of multimedia technology and increasing emphasis on environmental concerns. For training and data mining, we carefully selected the relevant data, ensuring that it accurately represents the characteristics of the green low-carbon domain. This meticulous data selection process was crucial to training the model effectively and fostering its understanding of the complex relationships within the green low-carbon information.

During the training process, we monitored the change in loss values closely. The progressive reduction in loss values is a strong indicator of the model’s ability to learn and adapt to the given data. By utilizing multi-dimensional data processing and mining clustering techniques, the proposed model effectively exploited the intrinsic patterns and structures present in the data, leading to significant loss reduction. This demonstrated the model’s capacity to process and comprehend complex information, which is pivotal in achieving more effective control over the green low-carbon domain.

Discussion

The rapid advancement of information technology and network infrastructure has resulted in an exponential surge in data volume, particularly driven by the proliferation of multimedia content. Effectively processing and transforming this data into meaningful insights has emerged as a pivotal concern. To tackle the challenge of visualizing eco-friendly and low-carbon information within the context of urban environment design, this article employs cutting-edge data mining technology to construct a framework for knowledge extraction from information. Additionally, a similarity-based dimensional alignment algorithm is employed to process multidimensional data, discerning feature data that necessitates visualization display. Subsequently, a sophisticated visualization platform is meticulously constructed, employing Vue technology.

An intriguing finding from our analysis was the correlation between advancements in multimedia technology and the comprehensiveness of open green low-carbon information. The visualization in Fig. 8 vividly showcased the increasing completeness of open green low-carbon data over recent years. This observation highlighted the model’s adaptability to evolving data landscapes and its potential to contribute to real-world applications in green technology and environmental sustainability.

Moreover, we focused on enhancing the overall visualization structure of the model to improve its reliability and accuracy. A more interpretable and intuitive visualization not only helps researchers better understand the model’s inner workings but also boosts the model’s practical applicability. Users can confidently trust the model’s outputs and utilize them in decision-making processes, making it a valuable asset in various green technology domains.

The analysis of experimental results reveals that the proposed scheme demonstrates a remarkable enhancement in data information mining, thereby augmenting the availability of eco-friendly low-carbon information. This approach adeptly filters out redundant information, consequently minimizing the loss rate in model training. Moreover, it exhibits robust data diversification display capabilities, thereby effectively propelling the realization of sustainable and high-quality development. It further promotes the application of advanced green technology, bolsters eco-friendly low-carbon management capability, accelerates network green upgrade, and elevates the level of eco-friendly power usage. The implementation of this model fosters innovation-driven quality transformation, power utilization, and efficiency enhancement, thus facilitating the realization of sustainable and high-quality development within the industry.

In conclusion, the extensive analysis and experiments conducted in this study affirm the superiority of the proposed model. The incorporation of diverse datasets, rigorous data processing, and the use of advanced techniques have collectively contributed to the model’s exceptional performance. By efficiently reducing loss values, adapting to dynamic data trends, and providing reliable visualizations, the model exhibits significant potential for practical applications in the field of green low-carbon technology and beyond. Our research serves as a stepping stone for further advancements in this domain and offers valuable insights for researchers, policymakers, and industry professionals.