2.1

Overall framework of accounting big data analysis platform

Cloud computing is a service related to information technology and software, relying on the Internet. Through servers, storage and networks, cloud computing can be used by users according to their needs in the form of services. Network information retrieval involves artificial intelligence, database technology, information science, natural language processing, machine learning and many other knowledge and disciplines. The accounting big data formed by the combination of big data and accounting has played a strong role in promoting the development of accounting. And accounting big data is also generated on the basis of big data and original accounting data. It is not only pure accounting data, but also information and data after processing, so the form of data is not fixed ^7-8.

In the existing traditional accounting system, the accounting is single, that is, only the information that can be measured in money can be processed, but the processing ability of intangible assets and human capital and other non-financial information is very limited. When an enterprise develops to a certain scale, the application of new technology will inevitably produce a huge amount of information, and how to use the accounting system to combine experience with information and refine it through deduction, comparison and induction to realize the transformation of value into knowledge will be a big challenge for enterprises. After building the basic cloud service platform, enterprises only need to pay for the services provided by service providers, and no longer pay for the maintenance of hardware and equipment at all, saving costs again.

The construction of accounting big data analysis platform driven by cloud computing technology must follow the requirements of accounting standards, accounting system and other relevant laws and regulations, and integrate the effective original accounting data and processed accounting data into accounting big data, and collect, store, analyze and apply it to explore its value. Generally speaking, its analysis model includes three layers: the top layer, the middle layer and the data center layer; It is subdivided into five modules, the top layer includes two modules, the workflow and user interface subsystem, and the middle layer includes parallel data and data preprocessing subsystem. In addition, there is a data center, cloud computing data center, which is not included.

Traditional auditing often limits the scope of auditing to the financial information reflected in the electronic account set, and the processed data information is often incomplete, which makes it impossible to fully understand the audited object. The establishment of financial sharing center enables enterprises to see the dynamic situation of capital flow in real time, and the application of cloud computing and other technologies enables enterprises to calculate the stock or demand of funds. The docking of financial cloud platform and online banking enables the group to timely dispatch funds to units in need, and centralize the remaining idle funds for unified management to increase profitability⁹. The construction framework of accounting big data analysis platform driven by cloud computing technology mainly includes data source, data processing, data storage, data warehouse and data visualization, and the standardization and security mechanism of accounting big data run through the platform construction, as shown in Figure 1.

Figure 1.

Cloud computing technology-driven accounting big data analysis platform architecture

Among them, the cloud service platform layer: network devices, storage devices, operating systems, etc. built by system management, stores the processed data in different databases, human resource accounting data, product sales accounting data, customer accounting data, etc. are uniformly stored in the basic database, and method base, model base, knowledge base, etc. are uniformly stored in the analysis database, providing convenience for later analysis and use.

In financial financing decision-making, models such as capital cost, financial risk and capital demand prediction are called to analyze the capital structure and use of enterprises at first. Secondly, the weighted average cost of capital is used to analyze the effect of capital use, and then the demand for funds is predicted by the sales percentage. Finally, the financing decision is made according to the knowledge law in the knowledge base. Activity-based costing is usually used to predict operating costs, and cluster analysis is used to test the influence degree of different factors on costs, and then the corresponding model is automatically selected according to the changing trend of historical output and cost data, so as to realize the requirement of forecasting future costs.

Financial forecasting is to predict and manage the financial risk, operating conditions and costs of enterprises by using linear regression method through financial analysis results and accounting information of industries. After analyzing and processing the collected information, we can find the indicators of risk early warning, and then quantify the indicators to form a comprehensive monitoring system of financial risks. On the basis of the monitoring system, a comprehensive early warning and monitoring model can be established in a scientific way, and a comprehensive system for the detection, management and evaluation of enterprise financial risks can be realized ¹⁰. The decision-makers of enterprises only need to input relevant keywords in the search bar by using instant query on the mobile terminal, and the system will automatically present the corresponding accounting information according to the captured keywords, and explain the relevant accounting information by using the existing models in the system, so as to provide intelligent decision support for information users.

2.2

Task scheduling in accounting big data cloud platform

In the big data environment, too extensive data sources and diverse data forms, in the process of data collection, the collection of the same kind of data often requires multiple channels to operate together to obtain. And the data obtained from different channels are often quite different. According to the correlation of many events, data mining technology excavates the set of general rules to explain data, and uses training and self-learning to extract new relationships hidden in data. Based on this service model, enterprises can build an accounting big data analysis platform driven by cloud computing technology, without defining the storage scalability of accounting data, and without managing and controlling the infrastructure and network of the cloud, but they can control the deployment of applications and configure the corresponding environment to improve the return on investment of enterprises in software. At the same time, it should be noted that building software based on PaaS mode will be a long-term development trend.

PaaS implemented by cloud computing technology provides services such as automatic deployment, running monitoring and flexible scheduling for applications and middleware. With the application of VM virtual resource pool and Open Stack resource pool¹¹, it realizes multi-tenancy, service isolation, high availability and flexible expansion of public resources. With the continuous growth of business on the cloud and the migration of old systems to the cloud, it is becoming more and more important for resource managers to schedule tasks reasonably and use resources efficiently. In this paper, a task scheduling algorithm for accounting big data cloud computing based on improved FA(firefly algorithm) is proposed. Aiming at the problem that traditional FA is easy to fall into local optimization, chaotic disturbance strategy is introduced to disturb the part with lower target value, and fireflies flying out of the search space are bounced back according to the theory of real physical rebound in firefly location update, so as to maintain the optimization ability of the population.

In FA algorithm, the moving operation caused by mutual attraction between fireflies determines the optimization ability of the algorithm. Brightness and attraction are the two most basic concepts in FA algorithm, which are the basis for solving optimization problems. β represents the attraction between fireflies X_i, X_j, which is a measure of the distance r_ij between them, and it changes with r_ij. The smaller r_ij, the greater the attraction β, whereas the smaller the attraction β, the specific form is shown in (1).

Where γ stands for absorption factor and is a constant.

FA is applied to accounting big data cloud computing task scheduling. In this paper, fireflies are coded by natural numbers, and the dimension of fireflies’ position is equal to the number of cloud computing tasks to be scheduled. Assuming that there are m tasks and n assignable resources in the accounting cloud environment, fireflies can be encoded into m -dimensional vectors:

Where: 1 ≤ x_i ≤ n,1 ≤ i ≤ m, each one-dimensional subscript of firefly represents a task number, and each one-dimensional component x_i represents the resource number assigned to this task by the cloud environment.

In the initial stage of FA iteration, fireflies keep a large step size and a good global optimization ability, so as to avoid the premature maturity of fireflies and falling into local minima. When the neighbor set is empty, fireflies i try to move 20 times by random step size, and choose an optimal position to update the position of fireflies i during these 20 times, which can avoid the situation that a group of fireflies lose their mobility with great probability.

t represents the number of iterations; t_max stands for the maximum number of iterations; s_max, s_min represents the maximum moving step size and the minimum moving step size respectively; p is the step change amplitude adjustment factor; D stands for the dimension of fireflies.

Chaos is a common phenomenon in nature, and most nonlinear systems will have different degrees of chaotic motion ¹². In order to prevent the convergence rate from slowing down due to the lack of population diversity in the late iteration, chaotic sequences are introduced to disturb fireflies and enhance their search ability. In this paper, Tent mapping is selected to generate chaotic sequences, and its formula is:

When the convergence speed of the population begins to slow down, the fireflies are sorted according to the fitness function value, and the 20% fireflies with low fitness value in the current population are disturbed by Tent chaotic map. The basic execution steps based on the improved FA are shown in Figure 2 below.

Figure 2.

Process based on improved FA