Mining with neural network

1. Fire coin global will launch DGD / BTC and DGD / eth transactions at 14:00 Beijing time on November 30, 2017

2. In Xiao's opinion, it is recommended to use light wallets first, such as geek wallets, imtoken, Cobo, etc. the operation is simple and easy to use. If the amount is relatively large, it is recommended to use cold wallet, but the operation is relatively complicated

3.

Li Yuwei

(China Institute of Geology and mineral resources information, Beijing 100812)

Li LinSong

(Beijing first computer factory, Beijing 100083)

Artificial Neural Network (ANN) technology is a new technology that can be used for pattern recognition. This paper introces the method of calculating mineral reserves by using ANN technology. This method is based on the information of the neighborhood rather than the whole study area, so it can ensure the success of the iterative process. This ANN method has been applied to a gold deposit with great grade variation and irregular sample distribution, and satisfactory results have been obtained. This paper also compares the application effect of ANN method and Kriging method

key words reserves estimation artificial neural network geostatistics neighborhood

1 Introction

so far, there are mainly two spatial interpolation methods for reserves calculation: distance reciprocal method and Kriging method. The former is simple and easy, but ignores the change of geological factors; The latter takes into account not only the spatial configuration of the samples, but also the geological conditions of the deposit. In the application of Kriging method, geologists often encounter some problems, such as unable to establish an appropriate variogram, the existence of non stationarity and so on. Wu Xiping and Zhou Yingxin (1993) used artificial neural network to estimate mineral reserves. This is obviously a good start, because ANN technology has the advantages of both reciprocal distance method and Kriging method: it is not only simple and easy, but also takes into account the changes of geological conditions. In addition, when using ANN method to estimate reserves, there is no need to consider such problems as variogram and stationarity. The range of ANN method used by Wu Xiping and Zhou Yingxin was the whole study area, with only 48 samples. If the number of samples increases, such as more than 100, or the data structure is complex, it will be difficult to achieve an ANN reserve estimation solution. In order to solve the above problems, the author designs a new Ann scheme for estimating reserves, which is based on the neighborhood information rather than the whole region information

2 domain definition

when using ANN method to estimate deposit reserves, there are two considerations for using neighborhood information instead of whole area information. One is that the estimation of blocks is only affected by the neighborhood samples, so it does not need to use the whole region samples; The other is that only with a small sample set can the ANN iteration be successful. When too many samples are used, the iterative process of ANN may be difficult to converge; When the number of samples is small, the possibility of convergence increases. According to the author's experience, when the number of samples is less than 50, Ann operation can be carried out smoothly. In short, the reason why we need a neighborhood is to extract an effective data subset from all sample sets, so that we can successfully estimate the block reserves with ANN method

the definition of neighborhood is similar to Kriging method when ANN method is used to estimate reserves. Firstly, a network system composed of regular blocks is designed. In this way, the problem is reced to estimating the value of a block according to the observations of several samples in the neighborhood of the block. For simplicity, the neighborhood can be defined as a rectangle (Figure 1). The samples in the rectangular neighborhood are used to estimate the reserves of the estimated block. In order to estimate the block more accurately, several estimation points can be set in the estimated block. The ANN program returns a value for each evaluation point of the block. The average value of these evaluation points is the evaluation of the block (Figure 2)

Figure 1 neighborhood definition

Figure 2 sample and estimate

the size of neighborhood can be adjusted until it contains 50 samples. There are three factors that affect the accuracy of block estimation. One is the data configuration of samples in the neighborhood. If the samples are evenly distributed, a satisfactory Ann estimation may be obtained. The second is the density of the sample. When there are more samples in the neighborhood, the ANN program will return an estimate with smaller error. The third is the spatial change of input variables. A variable with fluctuation of spatial height will lead to large estimation error

3ann model

artificial neural network method is one of the rapid development of artificial intelligence technology in recent years, it has been successfully used in pattern recognition. The attraction of this technique is that geologists using ANN only need to analyze the input and output layers. In other words, people don't need to study what happens between the input layer and the output layer, because those hidden layers can be regarded as a "black box". In this way, what people need to do is to understand the input and output events, which is quite familiar and easy for geologists; However, the extremely complex nonlinear relationship between input and output and the huge amount of computing tasks are left to the computer to deal with, which is just hard for people to undertake. From this point of view, geologists who use ANN technology to estimate reserves can avoid many problems that are difficult to solve when using Kriging method

The

block estimation is a spatial pattern recognition process. Considering the advantages of ANN and the previous work, this paper attempts to use this new pattern recognition technology for two-dimensional reserves estimation

the ANN structure consists of an input layer, two covert layers and an output layer. The input layer contains three variables: x-coordinate, y-coordinate and the average grade value of the neighborhood. The so-called neighborhood grade average refers to the average value calculated by the reciprocal distance method with the nearest four samples. Each layer contained five neurons. The output layer contains only one variable, the ore grade of the estimated point. Figure 3 shows the ANN structure

Fig. 3ann structure

let x < sub > I < / sub > be the input value of the i-th neuron in the next layer, x < sub > J < / sub > be the initial output value of the j-th neuron in the upper layer, and W < sub > ij < / sub > be the connection weight of the i-th neuron in the next layer and the j-th neuron in the upper layer, then the relationship between the m-th input neuron and the j-th output neuron can be established

Mathematical Geology and geological information

then use the following characteristic function

Mathematical Geology and geological information

to convert the initial output value x < sub > J < / sub > to the adaptive output value x < sub > J < / sub >

The learning algorithm used in

is simple back propagation method (BP), and its weight coefficient adjustment equation is

Mathematical Geology and geological information

this equation is used to modify the weight coefficient from the current value to the next value. Where k is the number of iterations, η For learning rate, δ< Sub > J < / sub > is the difference between the current value of the jth output neuron and its target value, and x < sub > J < / sub > is the adaptive output value of the jth neuron obtained in the kth iteration

although some improved BP algorithm or other more complex learning algorithms can be used, the simple BP algorithm can solve the problem of this paper well, so we do not intend to discuss and use other learning algorithms

The ANN method mentioned above is used to study J deposit in Henan Province. This is a quartz vein type gold deposit, which is dominated by pit exploration and controlled by a small number of boreholes. A total of 250 samples were obtained (Fig. 4). The distribution of samples is very uneven and the gold grade varies greatly. Therefore, this deposit is difficult to be described by general interpolation method. For the convenience of comparison, Kriging method and artificial neural network method are used to estimate the grade of the deposit according to the same block system. The block system consists of 25 × It is composed of 9 Kriging blocks, each of which is 50m in size × 50m The estimated block plus 50 samples constitute the neighborhood. Each estimated block has 3 blocks × Three valuation points. According to these neighborhood parameters and sample data, we can use the ANN model defined above to estimate the gold grade block by block. The number of iterations to achieve the specified accuracy varies widely, depending on the number of samples in the neighborhood, the configuration and the spatial variation of variables. The critical iteration error is 0.001. When the actual iteration error is less than the critical error, the iteration process ends. Most of the iterations end at 10000-30000 iterations, but the maximum number of iterations can reach 100000. Ann program for each block of 3 × Each of the three discrete valuation points returns a gold level valuation, and then the average value of the block is generated from them. Both discrete point estimation and block average are important results of ANN grade estimation

Fig. 4 sample point map

Fig. 2 shows a block of the gold deposit. In order to clearly understand the relationship between the sample and the evaluation point, only a part of its neighborhood is shown. This part of the neighborhood contains 11 samples, but their evaluation of the block is undoubtedly the most important. The nine small rectangles in the figure represent the evaluation points. It can be seen that Ann program returns a reasonable value for each evaluation point. The values of each evaluation point are in good agreement with the latest sample values. It can also be seen from Figure 2 that the estimation of the point obtained by ANN method is only determined by the nearest samples, and the influence of the samples far away from the estimation point on the estimation is negligible. This is the reason why when using ANN method to estimate a point or block, only the neighborhood is needed instead of the whole study area. The ANN learning information of the block neighborhood shown in Fig. 2 is listed in Table 1. After learning the weights, the X coordinate, y coordinate and the average value of neighborhood of each evaluation point are substituted into formula (1) to calculate the initial output value; Then the initial output value is substituted into formula (2) to obtain the adaptive output value. The point estimates for this block are shown in Table 2

Table 1 and Figure 2 show the learning information of the block

Figure 5 and Figure 6 show the point estimation and block estimation of the whole deposit, respectively. It can be seen that the point estimate in Figure 6 better depicts the gold grade distribution details of the deposit. This is very different from the result of Kriging method. Generally speaking, there is little difference between the point estimation of Kriging method and the block estimation, because Kriging method will proce a strong smooth effect in both point estimation and block estimation, but Ann block estimation will not proce such a strong smooth effect. Comparing Ann block estimation (Fig. 5) with Kriging block estimation (Fig. 7), although the average grade of the whole deposit is very close, which is 2.59375 according to ANN method and 2.49658 according to Kriging method, it is obvious that Kriging method's estimation image is greatly smoothed. We know that geostatistics provides two models for studying spatial data: one is Kriging model, which is used to estimate the local average value of a regionalized variable, but it is not faithful to its spatial variation details; The other is the conditional simulation model, which is used to describe the spatial variation of a regionalized variable carefully, but can not guarantee an optimal and unbiased local average. Ann seems to combine the advantages of these two models, especially for point estimation. The spatial variation of point estimation proced by an ANN mapping process is very similar to that of the actual sample. Of course, the denser the sample points are, the closer the characteristics of point estimation are to the actual characteristics

Table 2 point estimation of gold grade in block 2

Fig. 5 ANN method gold grade unit: g / T

Fig. 6ann method gold grade unit: g / T

Fig. 7 Kriging method gold grade unit: g / T

in order to illustrate the advantages of ANN estimation, the advantages of ANN method gold grade unit: g / T

Fig

4. 1. Buy fish and egg balls, go to 7-11 to buy chocolate, buy Haagen Dazs, buy Macao local fruits (the above are all single choices)

2. You can play in three rounds

5.

1. Yes,

2. The floating-point computing power of gtx1070 is lower than that of gtx1080, which is generally between gtx980 and gtxtitanx

6. mining is the most popular thing in the past two years, so the most popular mine cards are the high-end cards in the past two years, that is, 97098010701080 of n family and rx480 of a family.

7. Artificial neural networks (ANNs) is also referred to as neural networks (NNs) or connection model. It is an algorithmic mathematical model that imitates the behavior characteristics of animal neural networks and carries out distributed parallel information processing. This kind of network depends on the complexity of the system, by adjusting the relationship between a large number of internal nodes, so as to achieve the purpose of processing information
neural network can be used in pattern recognition, signal processing, knowledge engineering, expert system, optimal combination, robot control, etc. With the continuous development of neural network theory and related theories and technologies, the application of neural network will be more in-depth

the research of neural network can be divided into theoretical research and application research
theoretical research can be divided into the following two categories:
1
2. Based on the research results of neural basic theory, the neural network model with more perfect function and better performance is explored with mathematical methods, and the network algorithm and performance are deeply studied, such as stability, convergence, fault tolerance, robustness, etc; Develop new network mathematical theory, such as neural network dynamics, nonlinear neural field, etc
application research can be divided into the following two categories:
1. Research on software simulation and hardware implementation of neural network
2. Research on the application of neural network in various fields.

8.

Domestic ICO platforms are as follows:

1, currency crowdfunding

1, online time: 2015

2. Website: bizhongchou.com

Background: the ICO website of Babbitt, a blockchain media company

Office location: Hangzhou, Zhejiang Province

5. Introction: the earliest existing and active ICO platform in China was crowdfunding in 2015. Yuanbao.com and jucoin.com did similar crowdfunding in those years, but now they are out of business. However, although the activity of crowdfunding in recent years is not strong enough and its influence has been surpassed by the latecomers, its historical position is here after all

2. Currency investment

1. Online time: 2017

2. Website: bitouzi.com

3. Background: ICO website of the blockchain asset trading platform yuanjiu.com

Office location: Nanchang, Jiangxi Province

5. Introction: after coin investment opened ICO in May 2017, the coin investment and the traffic of coin Jiuwang soared, which was once inaccessible. Currency investment has opened a precedent for fast trading after ICO in the instry. However, because of this, there are too many and too fast projects, resulting in the weak rise of several subsequent projects, and even a burst. Now money investment has slowed down the pace of ICO, in the digestion of early projects

(3) icoage.com

1) online time: 2017

2. Website: icoage.com

Background: ICO website of Shanghai qukuai Information Technology Co., Ltd

4. Office location: Shanghai

Introction: icoage.com has a great influence in China. Although it doesn't have the support of exchange like currency investment, it is in the forefront of mining foreign projects. Then it can invest directly in RMB (buy bitcoin and Ethereum on the website), which is also convenient for investors< 4. Ico365.com

1. Online time: 2017

2. Website: icoage.com

Background: ICO website of Shenzhen Kedian Technology Co., Ltd

Office location: Shenzhen, Guangdong Province

5. Ico.info

1. Online time: 2017

2, website: ico.info

Background: ICO website of Beijing cloud coin Technology Co., Ltd

Office location: Beijing

9. AMD is the most efficient. Mining with 1070 or 1080 is because AMD's cards are no longer available. Now the mining income is too high, and it will be back in two months. 1070 or 1080 is not as efficient as AMD, but it's not bad. Anyway, it can make money.