Retracted: Construction of an Industrial Structure Analysis and Evaluation Model for Oil and Gas Resource-Based Cities Based on Deep Learning Model and Cluster Analysis

the


Introduction
For historical and national industrial policy reasons, the focus of the petrochemical industry in Daqing has been on the production of fuel oil alone, with a short industrial chain and shallow processing levels, which has led Daqing into a strange circle of a single industrial structure [1].+ere are generally two routes of oil processing, one is to process crude oil directly into refined oil products, and the other is to process crude oil into cracking materials [2].Daqing should accelerate the development of supporting industries, extend the industrial chain, take the route of fine chemicals and strengthen the petrochemical industry [3].
Increase investment in scientific research to improve the recovery rate of oil fields.Accelerate the development and application of high technology through cooperation with universities and research institutes to gain time for the development of successive industries [4].Pay attention to the protection of oil resources, improve the efficiency of resource development and utilization, and strive to create a "century-old oilfield", so as to provide the necessary accumulation and more time for the transformation of the city's economy [5].
Natural resources such as solar and wind energy have always been recognised worldwide as nonpolluting and costfree renewable resources [6].+e use of solar energy was proposed as early as 1968.Daqing has an abundance of light, with 2,726 hours of sunshine per year, a total annual solar radiation of 491.4 KJ/cm 2 and a high monsoon in spring and autumn, influenced by the cold inland Mongolian air and the warm ocean monsoon [7].Under the dual pressure of the conventional energy shortage and the deterioration of the global ecological environment, Daqing should increase its independent research and development efforts, accelerate the use of solar and wind energy, continuously optimise the energy structure, improve the efficiency of energy use, and drive the development of related industries to ensure that energy does not become an obstacle to Daqing's low-carbon transformation [8].
On the basis of the oil and petrochemical industry, cultivate new succession industries, gradually invest the large amount of capital accumulated from the development

R E T R A C T E D
of oil and gas resources and the petrochemical industry into succession industries, cultivate new leading industries, and thus complete the establishment of a diversified economic model as a whole [9].On the one hand, the development of deep processing industries for agricultural and livestock products.Daqing is rich in arable land and grassland resources, therefore, according to the needs of the market, we can implement the strategy of "conversion of agriculture to animal husbandry" and vigorously develop the farming industry, at the same time, we can support the deep processing of agricultural products enterprises and develop green agriculture.On the other hand, the tertiary industry should be vigorously developed.+e tertiary sector is a complex and mixed industry group, and it is important to scientifically determine the priorities and development of the tertiary sector, and to continue to exploit its advantages in order to build Daqing into a distribution and service centre in western Heilongjiang [10,11].

Basic Information about Daqing City
Daqing is an oil and chemical capital, located in the western part of Heilongjiang Province, with a total area of 22,200 square kilometres, 5 districts, 3 counties, 1 autonomous county and 1 national high-tech industrial development zone, and is the largest oil production base and an important petrochemical industry base in China [12].+e added value of the city's industry reached 305.91 billion yuan, an increase of 3.8% year-on-year, of which the added value of the regulated industry reached 241 billion yuan, an increase of 2.3% year-on-year; 40 million tons of crude oil and 3.51 billion cubic metres of natural gas were produced, with the added value of the oil industry reaching 178.8 billion yuan, a decrease of 0.1% year-on-year; the added value of the petrochemical industry reached 27.19 billion yuan, an increase of 7.1% year-on-year.+e ratio of the three industries was 4.5 : 77 : 18.5, with the secondary industry dominating and ranking 11th in terms of comprehensive strength among the prefecture-level cities in China, as shown in Table 1 [13].

Basic Characteristics of Daqing's Industrial Structure.
As a typical city of oil and gas resources, Daqing, while making a great contribution to the country, has also accumulated many problems of its own.As far as its industrial structure is concerned, the main problems in Daqing are [14].Due to the constraints of oil and gas resources generation and exploitation, the exploration, development and refining sites are unevenly distributed and the urban agglomeration is poor.Usually, the city is concentrated on the construction of certain areas in good condition and in the right location, making it a relatively concentrated central living area, such as the early mining area of Jean-Houlou, and then connected by certain traffic routes, forming a cluster of districts, counties and towns in a dotted pattern.+e overall layout of Daqing is therefore relatively fragmented, which does not facilitate the full use of some facilities and affects the urbanisation of the city [17].+ere are many other problems with the industrial structure of Daqing, such as the single structure of scientific and technological talents, the lack of awareness and the management system.

Empirical Analysis of Industrial Structure in Daqing.
In order to make the analysis of the industrial structure of Daqing accurate and specific, this paper further analyses the basic characteristics and problems of the industrial structure of Daqing from a quantitative perspective, in addition to the qualitative description.Taking into account the scope of the indicators and the conditions of application, the following indicators have been selected.

Industry Value Added
Share.+e value added share of industries in resource-based cities can be expressed by the following formula: where G ij is the value added of industry j in region i and G i is the total value added of industry in region i.Generally

R E T R A C T E D
speaking, it is considered that the industry is likely to become the leading industry when it is WI ij > 15%.As can be seen from Tables 2 and 3, the secondary industry accounts for 77% of the three industries in Daqing and 55% of the key industrial sectors above the scale, which is absolutely dominant.+e petroleum and petrochemical industry accounts for 50.6% of the total regional GDP and 65.7% of the secondary industry.It can be seen that the industrial structure of Daqing is dominated by the secondary industry, especially by the petroleum and petrochemical industry, indicating the overweight and homogeneous nature of Daqing's industrial structure.

Value of Change in Industrial
Structure.+e value of the change in the industrial structure of a resource-based city can be expressed by the following formula: where q i1 and q i0 are the shares of the output value of industry i in the total output value in the reporting period and the base period respectively.+e value of industrial structure change, K, is an indicator of the rate of change of industrial structure, the larger the value, the greater the rate of change of industrial structure; conversely, the smaller the value [18].Using the data in Table 4 to calculate the value of industrial structure change in Daqing City, we calculated the value of industrial structure change in Daqing City from 2011 to 2014 with the ratio of K2011 � 0.004, K2012 � 0.026, K2013 � 0.056 and K2014 � 0.104, which shows that the industrial structure of Daqing City tends to be constant.

Industrial Structure Entropy
Values.+e entropy value of the industrial structure of a resource-based city can be expressed by the following formula: where e t is the entropy value of the industrial structure in period t, W it is the proportion of the output value of industry i in period t, and n is the number of industrial sectors.+e larger the entropy value of the industrial structure, the more balanced the industrial development is and the more diversified the development structure tends to be.Table 4 was used to calculate the entropy value of the industrial structure of Daqing, see Table 5.From Table 6, it can be seen that the entropy value of the industrial structure of Daqing is on the rise between 2010 and 2014, but the value is small compared to the national entropy value of the industrial structure, indicating that the overall development of the industrial structure of Daqing tends to be homogeneous and uneven.

Cluster Analysis and Deep Learning-Based
Industrial Structure Optimisation +e precise and efficient allocation of the hundreds of associated oil manufacturing points is a key factor affecting the structure of the industry.In this paper, machine learning techniques are used to adjust the network configuration of oil manufacturing points in order to achieve uniform distribution of network service load among the capacity layer, coverage layer and wide and deep coverage layer according to the oil carrying capacity [19].At the same time, oil types are classified according to the characteristics of urban oil production distribution, and personalised optimisation schemes are generated for different types of oil, and the load optimisation schemes can be updated automatically as the network service development changes.+e process of industrial structure optimisation based on machine learning is shown in Figure 1.+e data obtained are petroleum-grade data for a certain period of time, including public petroleum manufacturing point data, measurement report (MR) data, configuration data and performance data.+e ETL (extract-transformload) processing of the data is done based on the storage and computing resources of the Big Data platform, providing useable formatted data for subsequent analysis and modelling [20].
+e MR data are segmented according to RSRP (reference signal receiving power) strength, and the data are classified according to RSRP strength, reference signal receiving power, reference signal receiving power, reference signal receiving power, and reference signal receiving power.+e two-dimensional curves are plotted according to RSRP strength and MR quantity, as shown in Figure 2.
+e key features for oil scene classification were extracted from the MR data.+e arithmetic mean, variance, kurtosis coefficient and skewness coefficient of each MR data were extracted.Using these four extracted features for scene delineation, the scene is automatically delineated using the K-means clustering algorithm, defining the number of oil as N, and the MR data of each oil has the above four features extracted, the data matrix is as follows:

R E T R A C T E D
x 11 x 12 x 13 x 14 x 21 x 22 x 23 x 24 . . .
where x ij denotes the jth attribute of the i-th sector.First, K points are randomly selected from the data set as the initial cluster centres, then the distance between each sample and the cluster centre is calculated and the sample is assigned to the class in which the nearest cluster centre is located.Euclidean distance is used to calculate the distance between data objects: Update the class cluster centre: the average value of all data objects in the corresponding class cluster, i.e. the updated class cluster centre of that class cluster.Define the class cluster centre of the kth class cluster as Center k , then the class cluster centre is updated as follows: where C k is the kth class cluster and |C k | is the number of data objects in the kth class cluster, the summation here is the sum of all elements in class cluster C k over each column of attributes, so Center k is also a vector with 4 attributes, denoted as Iterations are continuously performed to reclassify the class clusters and update the class cluster centres, and the number of iterations T is set, and the iteration is terminated after the T th iteration, at which point the resulting class clusters are the final clustering results.
Based on the scenarios classified in (2), the impact of configuration oil manufacturing points on performance indicators and the implied relationship between configuration and performance data are explored in each scenario to adjust the wireless oil manufacturing points, improve network performance and enhance the network KPI (key performance indicator).A DNN (deep neural network) model is used to learn the intrinsic relationship between the configured oil manufacturing points and the performance indicator oil manufacturing points to obtain a trained and stable neural network model.
Define the configuration of oil manufacturing points X to include the whole set of oil manufacturing points of the entire oil group, including the resource configuration oil manufacturing points, frequency configuration oil manufacturing points and boundary level configuration oil manufacturing points of each oil it contains, etc. +e matrix containing m frequency points and n oil manufacturing points per frequency point can be expressed as where x mn denotes the nth oil manufacturing point for the mth frequency point.If the network structure of F1D1D2FDD 900 MHz contains 4 frequency points and each frequency point has 12 oil manufacturing points configuration data, the size of the X matrix is (4,12).+e performance indicator oil manufacturing point Y is the determination indicator data of the industrial structure of the oil group, which may include: the maximum number of connected users of RRC (radio resource control), the utilization rate of uplink PRB (physical resource block) and downlink PRB utilization rate of each frequency point included in the oil group, including m frequency points, each frequency point has k oil manufacturing points, and the matrix size is (m, k), which can be expressed as where y mk is the kth load indicator data for the mth frequency point.If the network structure of F1D1D2FDD 900 MHz contains 4 frequency points, each with 3 indicator data, the size of the Y matrix is (4,3).+e relationship between performance metrics and configured oil manufacturing points is investigated and the problem is modelled as a multivariate regression problem.+e intrinsic relationship of the regression problem is learnt using a fully connected neural network algorithm model, using training data to train the neural network model.+e input is X. +e data in X is read in and converted to a column

R E T R A C T E D
matrix as the input to the neural network, and similarly the Y matrix is converted to a column matrix as the output of the neural network.
+e output of the regression problem, Y, contains multiple variables, so the cost function is designed based on this problem as where y i is the sample data and y i ′ is the predicted value calculated by the neural network.
+e average of MSE (mean squared error) of each variable is used as the loss function, and the cost function of the network is minimized by training the model, adjusting the weights and bias, thus obtaining a stable neural network model [21].

Model Training and Results Analysis
According to the oil scene classification method proposed above, the oil scene is classified using MR data.+e scene is automatically classified using the K-means clustering algorithm, and the time complexity of the algorithm is , where t is the number of iterations, k is the number of clusters, m is the number of features and n is the number of samples [22,23].Generally t, k and m can be considered as constants, so the time and space complexity can be simplified to O(n).+e mean value of MR statistics for each category is found, and the distribution of the mean value of MR statistics for each scene category is shown in Figure 3.
+e model of the relationship between configuration oil manufacturing points and performance oil manufacturing points was trained for each of the four types of oil scenarios, and the deep learning algorithm model was used to learn the intrinsic relationship between configuration oil manufacturing points and performance index oil manufacturing points.+e complexity of the DNN algorithm is mainly determined by the number of hidden layers, hidden units and samples.+e evaluation indexes of the DNN model are shown in Table 7, and the evaluation results show that the deep learning algorithm has a good fit to the data and the error between the predicted data and the real data is small.+e optimisation search is performed based on the trained DNN model, and the optimized configuration recommendations and tuning effect prediction results are output.+e problem is transformed into a hill-climbing algorithm to solve the multi-dimensional constrained optimisation problem, and the optimized configuration recommendations and tuning prediction results are shown in Table 8. +e oil scenario selected in this paper is F1D1D2FDD 900 MHz, where F1, D1 and D2 are all 20 MHz bandwidth in the existing network, and FDD 900 MHz is 5 MHz bandwidth.+erefore, the closer the amount of services carried by the four frequency points is to the bandwidth ratio, the better the service balance is.
Relying on the provincial big data platform and the resource capacity of the cloud resource pool, the above for the oil scenario segmentation, DNN model training and optimal configuration of oil manufacturing point solution are all done on the offline side, which is a light-weight operation.+e efficiency of the online deployment part depends mainly on the number of samples, which is much less complex compared to the offline side, and is more beneficial for real-time and efficient oil manufacturing point optimisation.+e current manual configuration solution is labour-and time-consuming and has a low percentage of improvement for load-unbalanced oil.Test results show that the machine learning-based load optimisation solution can achieve fine-grained load sharing according to the bandwidth capacity of the oil, and the improvement ratio of oil with unbalanced load reaches over 90%, saving a lot of manpower while improving the accuracy of load optimisation.

Conclusion
+is paper presents a cluster analysis and deep learning-based solution for the industrial structure of oil cities.At the same time, the development of the load optimisation system and the validation of the existing network have been carried out jointly with the manufacturer.In the selected test grid for the poor load optimisation of the existing network, the oil is tested and validated in the existing network, and the test improvement ratio can reach 90%.+e system truly combines machine learning with industrial economic optimisation, forming a technical solution for network oil manufacturing point tuning based on machine learning, and exploring a dynamic urban industrial adaptive load optimisation solution with more refined oil granularity and time granularity.

Figure 1 :
Figure 1: Machine learning-based industrial structure optimisation process.

Figure 2 :
Figure 2: Example of the distribution of MR sampling points in the RSRP intensity interval.

Figure 3 :
Figure 3: Mean distribution of MR statistics by scenario category.

Table 1 :
Gross regional product of Daqing city, 2014.

Table 2 :
Distribution of Daqing's three industries in 2014 Unit: RMB billion.

Table 3 :
Distribution of key industries in Daqing above-scale industry, 2014.

Table 8 :
Optimisation proposals and predicted results of adjustment effects.