The construction of cross-regional cooperation innovation network governance system based on data fusion technology

In recent years, with the development of network computer science and the increasing maturity of artificial intelligence technology, data fusion technology has been applied more and more widely. In order to respond to real-life needs and prevent and resolve risks, it helps to further broaden the research path of network governance. This paper focuses on network governance in the era of Internet big data


Introduction
Data fusion is a new direction of systematic scientific research arising from the intersection, synthesis, and extension of modern information technology and multiple disciplines [1][2].As early as the late 1970s, concepts or terms related to information synthesis began to appear in some publicly published literature.The term "data fusion" has been commonly used for a longer period of time afterwards [3][4].Although a lot of results have been achieved in data fusion research, no unified theoretical framework has been formed in theory, and there is a lack of effective guiding principles in practical system design.It is a formal framework whose process involves the synthesis of multiple sources of information using mathematical methods and technical tools with the aim of obtaining high quality useful information, with a precise definition of high quality dependent on the application [5][6].Therefore, various types and levels of fusion exist, such as data fusion, image fusion, feature fusion, decision fusion, sensor fusion, classifier fusion, etc. [7].Fusion mainly synthesizes information from different sources, modes, media, times, places, and representations, which can finally lead to a more accurate description of the perceived object [8][9][10].
With the widespread use of the Internet in economic, political and social life around the world, the integration of Internet space and social space in the traditional sense is occurring in a realistic sense.On the one hand, more and more real social life takes place in the Internet, and social actions that originally need to take place in the physical scene can be realized in the Internet scene [11][12][13][14].On the other hand, online actions that originally took place only in the Internet scene have increasingly important real social impacts, in some cases directly related to real public issues, people-state relations.With the continuous development of society and the change of media environment, the network ecology has also changed, which inevitably brings some network chaos [15].For example, due to the drive of economic interests and imperfect market competition mechanism, the phenomenon of one-sided pursuit of commercialization and perfunctory fulfillment of social responsibilities by various interest subjects in cyberspace has occurred from time to time.In addition, the popularity of network terminals also allows more subjects to participate in the dissemination of network information, and in the massive amount of network information, many information is difficult to distinguish true from false, and some wrong values are circulated from time to time, etc. [16][17][18].The unhealthy development of the network ecosystem is detrimental to the spread of socialist core values, the formation of social cohesion, and the development of socioeconomic culture [19][20].Therefore, in the system of online social governance, governance for online communities should be given more attention, and online communities themselves should play a more important role in the system of online social governance [21][22].In response to several prominent contradictions in the current cyberspace, namely, the contradiction between the development of cyberspace and cybersecurity, the contradiction between the demand for free flow of information and social stability, the contradiction between the demand for new technologies and new business changes in the Internet and the existing system, and the contradiction between the exploitation of big data resources and the protection of citizens' privacy, there is an urgent need to accelerate the construction and build a comprehensive network governance system.Therefore, the construction of a network ecological governance system is of far-reaching significance for the sustainable development of the network ecological environment [23][24][25].
In the current network ecological environment, the concept of rule of law is gradually strengthened, and the literature [26] points out that the governance of cyberspace under the rule of law is an inevitability of the times, an inevitable requirement for the implementation of the rule of law strategy, and an inevitable choice for the governance of the network society.However, there are some real challenges and problems in the Internet governance in the current environment, such as the lack of system, the legal system is not sound, and the awareness of the rule of law has not been fully formed.It is recommended to build the Internet, use the Internet and manage the Internet according to the law.
The literature [27] is devoted to network governance, adopting a dynamic perspective on how networks act, change, and can be governed.To this end, it examines network dynamics and evolution.The literature [28] shows that building an integrated network governance system requires a holistic perspective and an integrated governance paradigm.It can be explored in four aspects: generative logic, problem logic, thinking logic and action logic.Among them, time evolution, contradiction transformation, kinetic energy conversion and pattern evolution constitute the generation logic of governance, presenting the historical inevitability of building an integrated network governance system; the whole picture of the cyberspace problem, the essential attributes of the integrated system and the mirror image of reality constitute the problem logic, presenting the practical necessity of an integrated system; holistic thinking, correlation thinking, big data thinking, mass thinking and crossborder integration thinking constitute the thinking logic, presenting the system science methodology; the leading ideas, goal orientation and means standardization based on the principle of subjective cogovernance constitute the action logic, reflecting the feasibility of building such an integrated system.The comprehensive network governance system is an important part of the social governance structure [29].It gives full play to the advantages of socialism, pools forces, and uses a variety of means to comprehensively manage network conflicts and problems.The core tasks of government in integrated cyber governance are responsibility management, uncertainty management, structure management and legitimacy management, which put new demands on government officials in the integrated cyber governance system.Therefore, regulating the functions of government officials in the comprehensive network governance system and tapping into the character of government officials therein have important implications for creating a clear cyberspace environment and building a social governance pattern of shared governance.Online public sphere is a new public sphere and a new normative body whose existence and development imply a radical change in the spatial structure of government governance in the national governance system [30].When we face the new public management issues arising from online public opinion and its formation, we must go beyond the Habermasian base and re-establish and maintain public management based on the "state-network" in the online public sphere, and build a "government-network domain" relationship.Under the concept of rule of law, we will build a new order based on the public domain of the network and government governance.At the present stage, China's network governance process is facing a serious situation, there are many problems, showing the complexity, enormity and urgency of the governance process.The government, market and society alone can hardly achieve an effective response, and must be tripartite participation and cooperation to build a modern comprehensive network governance mechanism with multiple subjects, diverse means, strong coordination and practical use.However, the various scenarios proposed above do not allow for the complexities of cross-regional governance.
Therefore, this paper introduces the Bayes-algorithm based on the principle of data fusion to solve the above problems.Starting from the comprehensive network governance system, we take the crossregional cooperation innovation network as the research object and carry out the research of collaborative governance model according to its characteristics.Based on Internet governance and social governance, a Bayes algorithm based on a data fusion algorithm is used to compute estimates and posterior means.And by using the simulation research model to construct the network governance system, the above values which are derived help us understand the operation mechanism of cross-regional cooperation innovation network in a deeper level, which can guide the clustering and flow of innovation factors between regions, promote the diffusion and spillover of innovation performance, and maximize the benefits of cooperative innovation.From the simulation experiment results, it is feasible enough to build a diversified, collaborative and intelligent cross-regional cooperation innovative network social governance system.

Construction of data fusion technology system
The structure of the system based on data fusion techniques is shown in Figure 1.In this paper, the information fusion algorithm estimated by Bayes inference shows that in this system the local detector , makes a local decision based on its sampled data, and sends it to the fusion center.
The task of the fusion center is to complete the mapping of the local decision space to the output space .Here, a cost is specified for each judgment.That is, assume that is the cost paid for judgments.

Figure 1. Data fusion system architecture
According to Bayes, the prior probability of Hr is assumed to be P(Hr).A set of local judgments is obtained, in which both local decisions and posterior probabilities have average cost.The role of data fusion is to divide the local space into m hypothesis subspaces R1, R2,…,Rm that do not overlap each other, i.e., to divide the intersection of m subspaces according to the threshold determined by the criterion of minimizing the average cost, and when falls into which subspace, the corresponding hypothesis is determined to hold, thus realizing the nonlinear transformation from input to output.

Convergent computing
The Bayes algorithm is a classical method for solving the inference used to solve uncertainty, and it is based on the Bayes formula in probability theory, which expresses the confidence level of each hypothesis in terms of probabilities, and it updates the likelihood function of the hypothesis given the previous likelihood estimates of the hypothesis and the added evidence.The technique can be performed based on classical probabilities or on subjective probabilities.

Bayes formula description:
Let the probability events A and B belong to the event domain F and the of the occurrence of event A. Then it is said that. ( The conditional probability of event B occurring under the condition that event A occurs.In Equation (1), P(AB) is the probability of the event A and B when they both occur.

P AB P B A P A =
If is a division of the basic event space, and .Then for anyone A, , there are. ( Where the is obtained from the analysis of existing data, called the prior probability; is the probability of re-correction after obtaining new information, called the posterior probability.Equation ( 2) is the Bayes formula.

Eigenvectors
Let the state to be measured of the system be the vector X, and the measured value measured by the sensor is I.The measurement equation of the sensor is (3) where is a function of I and X, and V is the random error.
For data fusion, it is the measured value obtained from N sensors, from which the true value of the state is estimated according to some estimation criterion function to determine the probability of giving evidence that the hypothesis is true.
For a single sensor , let its measurements be I and the estimated value of state X be and define as the loss function, and according to Bayes estimation, the corresponding risk expression is In this equation, detects the distribution pattern of the data, is the posterior probability of state X.
The estimation criterion that takes the least risk must be such that: (5) In order to obtain the estimate of the state XI.
From Equation (4), it can be seen that defining different , will give different estimation results.The commonly used has the following three forms.( ) where A is the positive definite weight matrix.
where is an arbitrarily small positive number.
And corresponding to the above three forms, the corresponding state-optimal estimates are the posterior mean estimates.
Posterior median estimates.
Maximum posterior estimate. ( Combined with the a priori knowledge of the likelihood that the assumption is indeed true, an independent sensor with measurement is added to the single sensor, and the measurement of the original sensor is denoted as , then the optimal estimate sought based on is the value after data fusion. So far, extending Bayes' inference to N independent sensors , similarly, the fused values of the measured data from N sensors based on the loss function defined in Equation ( 8) can be obtained as (12) At this point, the multi-sensor fusion problem is transformed into the problem of how to obtain the posterior probability of state X and find the corresponding maximum posterior estimate , using the subjective probability as the prior probability of the hypothesis and the probability of the evidence under the given hypothesis condition.According to Bayes' formula there are.When the measurements of N sensors are statistically independent, then.(14) From equations ( 13) and ( 14) and Bayes' theorem we have (15) where both and are independent of X and can be considered as normalization factors, which can be disregarded in finding the maximum a posteriori estimate .Therefore, equation ( 15) can be changed to (16) Where P(X | Ii) is the a posteriori estimate of the state after getting the sensor measurements.At this point, the multi-sensor information fusion problem is converted into the problem of how to get the state a posteriori probability, find the corresponding maximum a posteriori estimate, and derive the need to build an innovative network social governance system through cross-regional cooperation by the maximum a posteriori estimate.

Load identification process
The specific steps of the system simulation are as follows: firstly, a multi-degree-of-freedom model is established according to the given values of m, k and c.The displacement response calculated by the method of impulse response function convolution is used as the measured response, and the green function matrix between the load action point and the response measurement point is established by the system parameters, and then the external load is solved by using Bayes method and Tikhonov regularization method respectively.The flow chart of the method is shown in Figure 2.

P X P I I I X P X I I I P I I I
, , ,

P X I P I P X I I I P X P I I I
Flow of load recognition in physical space of a multi-degree-of-freedom system The following will be divided into three aspects from the recognition, repeatability and performance of the detection system.The system is in zero initial state, i.e., the displacement and velocity of the system at zero moment are zero, the simulation time step is 0.001s, and the total time is 0.5s.
3 System simulation study

Identification of the detection system
In the four-degree-of-freedom system shown in Figure 3, the amplitude of the action on the mass m is 100N, and its load time course is F=100*sin(50*t), the response on the first degree of freedom is measured, and the signal-to-noise ratio of the system response is set to 40dB and 30dB to correspond to the noise environment of lower noise and medium noise, respectively, to observe the recognition effect under different noise conditions.The recognition results of the Tikhonov regularized sums are shown in Figure 3(a) and Figure 3(b) when the signal-to-noise ratios of the responses are 40dB and 30dB.
(a) Identification results of sinusoidal loads in physical space at 40d B for a four-degree-of-freedom system From Figure 3(a) and (b), it can be seen that the recognition effect is the best for Bayesian regularization of sinusoidal load under two noise conditions, and the recognition effect of Tikhonov method is the second best.The Tikhonov curve method has the worst recognition effect at low noise, and the recognition effect is better at higher noise.The load recognized by Tikhonov regularization method fluctuates more around the real load and the smoothness is poor, which affects the effect of load recognition.The load curve identified by Bayesian regularization method is smooth, with little fluctuation, and is basically identical to the real load at low noise.
The correlation coefficient r and the relative error RE of the identified loads are compared in Table 1.The regularization parameters selected by the two regularization parameter methods are listed in Table 2.  From comparing the regularization parameters of the two methods in Table 1, Table 2, it can be seen that the regularization parameters chosen for the L-curve method and the GCV method are smaller than those of the Bayesian method.This is in line with the difference in load identification results, where the Bayesian regularization method has a better suppression effect on the noise in the response and identifies a smooth load curve, while the Tikhonov curve and the Tikhonov method have a poorer suppression effect on the noise, indicating that the regularization parameters chosen by both the Lcurve method and the GCV method are on the small side.The experiments show that it is shown that the algorithms can be used to construct a diversified and innovative network social governance system across regional cooperation.

Repeatability of the detection system
At the temperature of room temperature, air humidity of 17% RH and gas flow rate of 80 ml/min.The NO end and NC end of the three-way solenoid valve are each connected to a gas bag, and the gas bag connected to the NO end is filled with 99.99% nitrogen and the gas bag connected to the NC end is filled with NH3 at a concentration of 30 ppm.NH3, and after 80 seconds of feeding the sample, the NC end was closed and the NO end was opened to pass nitrogen, and the experiment was repeated twice while all conditions were kept constant.From the experimental curve Figure 4, it can be seen that except for the 4NE/H2S sensor which does not respond obviously, the other three sensors all responded to NH3, and the response curve remained basically the same after three injections, which shows that the detection system has a good repeatability.The experimental results after three injections are shown in Table 3. From the above experimental curve Figure 3 and the experimental data table 3, the experimental results show that the maximum difference in the response of each sensor to the same gas with the same concentration at a fixed flow rate does not exceed 0.02 V.This indicates that the detection system has good repeatability in detecting the same sample gas.The above two experiments demonstrate that the multi-sensor detection system is able to detect unknown gases stably and reliably, and show that the algorithm evaluation model can be used to build a collaborative and innovative network social governance system across regional cooperation.

Performance of the detection system
The experiment uses a small corpus downloaded on CNLP as test data, which has 1022 documents in five categories: computer, art, education, politics, and sports.The tests were conducted on the data mining tool WEKA, a tool software for performing data mining, which allows experiments on classification, clustering, etc.It includes a variety of classifiers, Bayesian being one of them.We programmed the new Bayesian classifier to be implemented under the same conditions and environment, and compared it with the original classifier in terms of classification accuracy, recall and f-measure through experiments.All data were converted from text format to feature matrix format after data preprocessing, i.e., the raff format suitable for the WEKA system, where a record represents a text.The corpus used in the experiments serves both as a training and a test set.From Figure 5, it can be seen that when the threshold c, which measures the relevance of feature items, takes different values, it affects the effect of text classification.The smaller the value of c, the more relevant feature items are derived when examining the relevance of features, so that the efficiency of the algorithm is lower, and the smaller the value of c, it may lead to strong twisting between feature items (i.e., feature items that are originally not highly relevant are considered to be related to each other), resulting in a decrease in the classification effect.The best performance can be seen in Figure 4 for c = 0.3, but considering the efficiency issue, c = 0.4, which has a better overall effect, was used during the experiments.From Figure 6, it can be seen that the classification effect of Bayes algorithm for sports, political category improved from about 60% to about 70%, which is a big improvement of about 10%, and the classification effect of computer and art category improved less, only about 1%, while in education category there is a certain degree of decline, about 1%, although the percentage of decline is not large, but this may be due to the feature This may be caused by the strong twist between the items.In the categories of sports, politics, and medicine, the number of relevant feature terms derived by examining the relevance of feature terms is greater than that of the other categories, and it is conceivable that this method of merging relevant feature terms is practical and feasible.This experiment shows that Bayes' algorithm is useful for improving the performance of Bayesian text classification system, and suggests that the algorithm model can be used to build an intelligent and innovative online social governance system across regional cooperation.
The above experiments show that the Bayes algorithm can be better applied in the network governance system in terms of recognition, repeatability and performance.

Conclusion
The construction of a comprehensive network governance system should be based on the source governance of network platforms, and the general idea is to normalize, rule of law and systematize network content governance.The basic model is the benign interaction between the government and the platform, i.e., the government supervises the platform, the platform supervises the individual, innovates the comprehensive network governance system, and optimizes the ecological environment of cyberspace.This paper proposes a reliability algorithm evaluation model based on a multi-stage experimental data fusion data method.The simulation results from the judgment data, calculation formula and posteriori estimates show that when the threshold c=0.4, the model is simple and easy to implement, and the realization idea is consistent with the gradual process of people's knowledge of unknown parameters from nothing to something, from less to more.And the experimental data is a gradual refinement process in each stage, which helps to build a diversified, collaborative and intelligent innovative network social governance system.Thus, it can be seen that data fusion technology has good application prospects and promotion value in the future, with far-reaching significance.

Figure 3 .
Figure 3. Identification results of sinusoidal load in physical space for the four-degree-of-freedom system

Figure 4 .
Figure 4. Experimental curve of system repeatability

Figure 5 .
Figure 5. Relationship between parameter c and accuracy

Figure 6 .
Figure 6.Comparison of regularization parameters for half-sine impact load recognition in physical space of four degree of freedom system

Table 1 .
Comparison of sinusoidal load identification effects in physical space of four-degree-of-freedom system

Table 2 .
Comparison of regularization parameters for sinusoidal load identification in physical space of four-degree-of-freedom system