WO2021093362A1 - 一种蒸发过程出料苛性碱浓度测量装置精度补偿方法 - Google Patents

一种蒸发过程出料苛性碱浓度测量装置精度补偿方法 Download PDF

Info

Publication number
WO2021093362A1
WO2021093362A1 PCT/CN2020/103203 CN2020103203W WO2021093362A1 WO 2021093362 A1 WO2021093362 A1 WO 2021093362A1 CN 2020103203 W CN2020103203 W CN 2020103203W WO 2021093362 A1 WO2021093362 A1 WO 2021093362A1
Authority
WO
WIPO (PCT)
Prior art keywords
value
compensation
caustic alkali
data
alkali concentration
Prior art date
Application number
PCT/CN2020/103203
Other languages
English (en)
French (fr)
Inventor
柴天佑
贾瑶
王良勇
Original Assignee
东北大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 东北大学 filed Critical 东北大学
Priority to US17/309,942 priority Critical patent/US11893472B2/en
Publication of WO2021093362A1 publication Critical patent/WO2021093362A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • C—CHEMISTRY; METALLURGY
    • C01—INORGANIC CHEMISTRY
    • C01F—COMPOUNDS OF THE METALS BERYLLIUM, MAGNESIUM, ALUMINIUM, CALCIUM, STRONTIUM, BARIUM, RADIUM, THORIUM, OR OF THE RARE-EARTH METALS
    • C01F7/00—Compounds of aluminium
    • C01F7/02—Aluminium oxide; Aluminium hydroxide; Aluminates
    • C01F7/04—Preparation of alkali metal aluminates; Aluminium oxide or hydroxide therefrom
    • C01F7/06—Preparation of alkali metal aluminates; Aluminium oxide or hydroxide therefrom by treating aluminous minerals or waste-like raw materials with alkali hydroxide, e.g. leaching of bauxite according to the Bayer process
    • C—CHEMISTRY; METALLURGY
    • C01—INORGANIC CHEMISTRY
    • C01F—COMPOUNDS OF THE METALS BERYLLIUM, MAGNESIUM, ALUMINIUM, CALCIUM, STRONTIUM, BARIUM, RADIUM, THORIUM, OR OF THE RARE-EARTH METALS
    • C01F7/00—Compounds of aluminium
    • C01F7/02—Aluminium oxide; Aluminium hydroxide; Aluminates
    • C01F7/04—Preparation of alkali metal aluminates; Aluminium oxide or hydroxide therefrom
    • C01F7/14—Aluminium oxide or hydroxide from alkali metal aluminates
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/17—Function evaluation by approximation methods, e.g. inter- or extrapolation, smoothing, least mean square method
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Definitions

  • the invention belongs to the technical field of compensation of measuring devices for production parameters of alumina evaporation process, and in particular relates to a method for compensating the accuracy of a device for measuring the concentration of caustic alkali in an evaporation process.
  • the concentration of caustic is a key process indicator of the alumina evaporation process, because the qualification of the concentration of caustic will directly affect the ratio of the alumina grinding process and the dissolution rate of the dissolution process, thereby affecting the final product of alumina production—— The quality of alumina. Therefore, online detection of caustic alkali concentration is a very important task in alumina production.
  • the present invention provides a method for compensating the accuracy of a device for measuring the concentration of caustic alkali in an evaporation process.
  • the main technical solutions adopted by the present invention include:
  • a method for compensating the accuracy of a caustic alkali concentration measuring device in an evaporation process including the following steps:
  • Step 1 Data collection: collect process data of lye refractive index, temperature, caustic alkali concentration meter value and laboratory value during evaporation;
  • Step 2 Data preprocessing: Perform moving average filtering, timing matching, and normalization processing on the process data collected in Step 1, to obtain preprocessed process data;
  • Step 3 Input the preprocessed process data into the accuracy compensation model of the caustic alkali concentration measuring device to obtain the compensation value;
  • Step 4 Add the caustic alkali concentration meter value and the compensation value to realize the online compensation of the caustic alkali concentration.
  • the length of the sliding filtering window is set, that is, the number of moving average filtering points is N, and the filtering formula is:
  • X(t) is the value at time t after filtering
  • X'(t) is the value at time t of the original data
  • N is the length of the sliding filter window.
  • the 2-hour process data is divided into three parts according to the optimized control period of 40 minutes, and each part takes the average value of the 40-minute process data, which corresponds to the laboratory data sampled last time;
  • timing matching formula is:
  • X(i) is the value at the i-th time after filtering
  • X(k) is the process data matching the k-th test value.
  • the input and output variable states used in the accuracy compensation model of the caustic alkali concentration measuring device are all normalized:
  • x n represents the state of the variable at the nth point
  • x max represents the maximum value of the variable in all historical data
  • x min represents the minimum value of the variable in all historical data
  • the input variable is the preprocessed process data
  • the output variable is the compensation value
  • the method further includes constructing a precision compensation model of the caustic alkali concentration measuring device, and training the model parameters.
  • constructing the accuracy compensation model of the caustic alkali concentration measuring device includes the following steps:
  • the deep learning algorithm uses a two-layer LSTM network to establish a precision compensation model for the caustic alkali concentration measurement device.
  • the method further includes: evaluating the accuracy compensation model of the caustic alkali concentration measuring device by using the root mean square error, the average absolute error, and the average absolute percentage error:
  • y i is the test value of the i-th group of samples, It is the compensated caustic concentration value of the i-th group of samples.
  • the method for compensating the accuracy of the caustic alkali concentration measuring device in the evaporation process provided by the present invention has the following beneficial effects:
  • the method for compensating the accuracy of the discharge caustic concentration measuring device in the evaporation process is to collect the input and output data of the internal model of the caustic concentration online measuring device and the test data to form a large data set, and use the deep learning algorithm to construct and train the caustic concentration measurement
  • the device accuracy compensation model is used to form a complete model of the caustic alkali concentration, and the real-time data of the production process is used to realize the online compensation of the caustic alkali concentration. On this basis, it guides the optimization of the feed volume, steam pressure, and the amount of new alkali added in the evaporation process, and controls the caustic alkali concentration within the qualified range to achieve the purpose of stabilizing alumina production and reducing production costs.
  • the method of the present invention can more accurately compensate the concentration value of the online meter, and the compensated concentration value can follow the trend of actual changes; and the measurement accuracy can meet the needs of actual production.
  • Fig. 1 is a structural diagram of the accuracy compensation algorithm of the caustic alkali concentration measuring device described in the embodiment of the present invention
  • Fig. 2 is a structural diagram of the cyclic neural network described in the embodiment of the present invention.
  • Fig. 3 is a structural diagram of LSTM described in an embodiment of the present invention.
  • 5 is a comparison diagram of the error comparison between the concentration value and the laboratory value of the caustic alkali concentration measuring device described in the embodiment of the present invention before and after the accuracy compensation;
  • Fig. 7 is an error distribution diagram before and after the accuracy compensation of the caustic alkali concentration measuring device described in the embodiment of the present invention.
  • This embodiment discloses a method for compensating the accuracy of a caustic alkali concentration measuring device in an evaporation process, which includes the following steps:
  • Step 1 Data collection: collect process data of lye refractive index, temperature, caustic alkali concentration meter value and laboratory value during evaporation;
  • Step 2 Data preprocessing: Perform moving average filtering, timing matching, and normalization processing on the process data collected in Step 1, to obtain preprocessed process data;
  • Step 3 Input the preprocessed process data into the accuracy compensation model of the caustic alkali concentration measuring device to obtain the compensation value;
  • Step 4 Add the caustic alkali concentration meter value and the compensation value to realize the online compensation of the caustic alkali concentration.
  • the length of the sliding filter window is set, that is, the number of moving average filtering points is N, and the filtering formula is:
  • X(t) is the value at time t after filtering
  • X'(t) is the value at time t of the original data
  • N is the length of the sliding filter window.
  • the 2-hour process data is divided into three parts according to the optimized control period of 40 minutes, and each part takes the average value of the 40-minute process data, which corresponds to the test data of the previous sample;
  • timing matching formula is:
  • X(i) is the value at the i-th time after filtering
  • X(k) is the process data that matches the k-th test value.
  • the input and output variable states used in the accuracy compensation model of the caustic alkali concentration measuring device are all normalized:
  • x n represents the state of the variable at the nth point
  • x max represents the maximum value of the variable in all historical data
  • x min represents the minimum value of the variable in all historical data
  • the input variable is the preprocessed process data
  • the output variable is the compensation value
  • This embodiment also includes constructing the accuracy compensation model of the caustic alkali concentration measuring device and training the model parameters.
  • the deep learning algorithm uses a two-layer LSTM network to establish a precision compensation model for the caustic alkali concentration measurement device.
  • This embodiment also includes: using the root mean square error, the average absolute error, and the average absolute percentage error to evaluate the accuracy compensation model of the caustic alkali concentration measuring device:
  • y i is the test value of the i-th group of samples, It is the compensated caustic concentration value of the i-th sample.
  • the method for compensating the accuracy of the caustic alkali concentration measuring device in the evaporation process of the present invention includes the following steps:
  • Process data is obtained by reading the evaporation process PHD database, with a time interval of 1 minute, including collecting the discharging lye refractive index R(k) and the discharging lye temperature T(k) And the caustic alkali concentration meter value; the test data y Nk (k) is obtained by accessing the enterprise MES system with a time interval of 2 hours.
  • the collected data is stored in the local disk with increasing time tags, and the storage format is a csv file.
  • Step 2 Data preprocessing: Perform moving average filtering on the 3 types of process data collected in Step 1.
  • X(t) is the value at time t after filtering
  • X'(t) is the value at time t of the original data
  • N is the length of the sliding filter window.
  • Time sequence matching As the sampling period of process data and laboratory data is different (sampling period: 1 minute for process data, 2 hours for test data), in order to make full use of process data and extract its characteristics, the 2 hours of process data is divided into an optimized control period of 40 minutes 3 copies, each taking the mean value of the 40-minute process data, corresponding to the laboratory data sampled last time.
  • the timing matching formula is:
  • X(i) is the value at the i-th time after filtering
  • X(k) is the process data that matches the k-th test value.
  • Normalization processing The input and output variable states used in the accuracy compensation model of the evaporation process are all normalized:
  • Step 3 Build the accuracy compensation model of the caustic alkali concentration measuring device, and train the model parameters
  • the left part of the above figure is a cyclic structure of the cyclic neural network, which represents that the input of the current hidden layer includes the output of the hidden layer at the previous moment. Expand the network structure to get the right part of the figure.
  • W, U, and V are weights
  • x t is the input at time t
  • st is the hidden state at time t, that is, the hidden layer output, which is the memory unit of the network.
  • o t is the output at time t.
  • the mathematical formula of cyclic neural network is as follows:
  • f is usually a nonlinear activation function, such as tanh and relu.
  • s t is obtained from the hidden output s t-1 at the previous moment and the input x t at the current moment.
  • the softmax function is the activation function of the output layer, which is often used in classification problems, and the output is mapped to the probability distribution of (0,1).
  • the traditional RNN model has the problem of gradient disappearance and gradient explosion, the problem is particularly serious when the sequence is very long. At this time, the traditional RNN model can not be used directly, and a special case of RNN-long short-term memory is more widely used. Network (LSTM).
  • LSTM Network
  • Long short-term memory network Long short-term memory, LSTM is a special type of RNN.
  • the difference between LSTM and RNN is that it adds a "processor" to the algorithm to determine whether the information is useful or not.
  • This The structure of the processor is called a cell, as shown in Figure 6. Three doors are placed in a cell, which are called input gate, forget gate and output gate. When a piece of information enters the LSTM network, it can be judged whether it is useful according to the rules. Only the information that meets the algorithm authentication will be left, and the non-compliant information will be forgotten through the forget door.
  • h t represents all the outputs of the LSTM unit at time t;
  • W f , W i , W C , W o are the weight matrix composed of coefficients, b f , b i , b c , and b o are the bias vectors of the corresponding weights , [sigma] is the activation function sigmoid, tanh is the activation function; * is a point multiplication;
  • C t represents calculation time t memory cells; f t, i t, o t are the point of time t input gate, forgetting and output gates Calculation method. It can be seen from Figure 6 that the outputs of the input gate, forget gate and output gate are connected to a multiplication element to control the input and output of the information flow and the state of the cell unit.
  • a double-layer LSTM network and a full-connected layer are used to construct the accuracy compensation model of the caustic concentration measurement device.
  • the accuracy compensation model is trained and tested using the previously divided training set and test set, and the root mean square error is used as the error function for The error back propagation learning of the neural network, the calculation formula of the mean square error is:
  • Nk(k) represents the output value of the caustic alkali concentration measuring device
  • e(k) represents the output value of the accuracy compensation model based on deep learning
  • Step 4 Real-time compensation prediction of caustic alkali concentration
  • the accuracy compensation algorithm caustic concentration prediction value and the output value of the caustic concentration measurement device are evaluated using root mean square error, average absolute error, and average absolute percentage error:
  • y i is the test value of the i-th group of samples, It is the caustic alkali concentration value before and after compensation of the i-th group of samples.
  • the index calculation results are shown in Table 1 and Figure 6.
  • the RMSE index before compensation is 9.332, the RMSE index after compensation is 3.006; the MAE index before compensation is 8.235, the MAE index after compensation is 2.264; the MAPE index before compensation is 3.562, and the MAPE index after compensation is 3.006
  • the index is 0.97; among them, from the RMSE index, after compensation, it is 67.8% higher than before compensation; from the MAE index, after compensation is 72.5% higher than before compensation; from the MAPE index, after compensation is 72.8% higher than before compensation .
  • the distribution of the error before and after compensation is shown in Table 2 and Figure 7.
  • the enterprise defines the allowable error between the instrument value and the test value within 3g/l, and use Comparing the compensated caustic alkali concentration value with the laboratory value, the error is within 3g/l and it is deemed qualified. It can be seen from the figure that the error before compensation is mostly concentrated above 5g/l, and the error after compensation is mostly concentrated below 3g/l.
  • the pass rate of the caustic alkali concentration value before compensation is 12.16%.
  • the pass rate of the caustic alkali concentration value after compensation is 71.36%, which is 59.2% higher than that before compensation.
  • the accuracy is compensated. The effect is obvious.
  • the accuracy of the accuracy compensation model of the caustic alkali concentration measuring device is high, and the reliability and accuracy of the model are high.
  • a model accuracy judgment module is also constructed.
  • the accuracy compensation model of the caustic alkali concentration measuring device cannot meet the requirements, the accuracy compensation model needs to be retrained and corrected.
  • new models are trained to realize the long-term, stable and accurate online compensation of the accuracy compensation model.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Chemical & Material Sciences (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Organic Chemistry (AREA)
  • Biochemistry (AREA)
  • Analytical Chemistry (AREA)
  • Medicinal Chemistry (AREA)
  • Immunology (AREA)
  • Food Science & Technology (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Algebra (AREA)
  • Inorganic Chemistry (AREA)
  • Geology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Operations Research (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Geochemistry & Mineralogy (AREA)

Abstract

一种蒸发过程出料苛性碱浓度测量装置精度补偿方法，包括以下步骤：步骤一：数据采集：采集蒸发过程碱液折光度、温度、苛性碱浓度仪表值及化验值的过程数据；步骤二：数据预处理：对步骤一采集的过程数据进行滑动平均滤波处理、时序匹配、归一化处理，获得经过预处理的过程数据；步骤三：将经过预处理的过程数据输入苛性碱浓度测量装置精度补偿模型，获得补偿值；步骤四：将苛性碱浓度仪表值与补偿值相加，实现苛性碱浓度的在线补偿。能够比较精确地补偿在线仪表浓度值，补偿后的浓度值能够跟随实际变化的趋势；测量精度能够满足实际生产的需要。

Description

一种蒸发过程出料苛性碱浓度测量装置精度补偿方法 技术领域
本发明属于氧化铝蒸发过程生产参数的测量装置的补偿技术领域，尤其涉及一种蒸发过程出料苛性碱浓度测量装置精度补偿方法。
背景技术
苛性碱浓度是氧化铝蒸发过程的关键工艺指标，因为苛性碱浓度的合格与否将直接影响氧化铝的磨矿工序的配比和溶出工序的溶出率，从而影响氧化铝生产的最终产品——氧化铝的质量。因此在线检测苛性碱浓度是氧化铝生产中非常重要的任务。
目前苛性碱浓度的检测手段通常有两种，一种是在蒸发过程碱液调配完成后，通过人工取样、化验的手段获得。由于取样间隔时间较长(通常是2小时或者4小时)，化验也需要很长的时间，因此苛性碱浓度的监测存在严重的滞后。利用这些严重滞后的信息来指导生产，容易导致苛性碱浓度波动较大，碱液合格率低，导致其余工艺控制困难，最终导致氧化铝品质较低。另一种是苛性碱浓度在线检测仪表，但这些在线仪表价格昂贵，而且通常仪表在安装完成后，其内部模型参数固定，由于国内氧化铝企业所使用的原矿成分波动较大，蒸发过程工况复杂多变，随着时间的推移，在线测量装置输出值可能发生漂移，仪表精度降低，仪表厂家维护需投入大量资金，且重新校准过程较为繁琐。
发明内容
(一)要解决的技术问题
针对现有存在的技术问题，本发明提供一种蒸发过程出料苛性碱浓度测量装置精度补偿方法。
(二)技术方案
为了达到上述目的，本发明采用的主要技术方案包括：
一种蒸发过程出料苛性碱浓度测量装置精度补偿方法，包括以下步骤：
步骤一：数据采集：采集蒸发过程碱液折光度、温度、苛性碱浓度仪表值及化验值的过程数据；
步骤二：数据预处理：对步骤一采集的过程数据进行滑动平均滤波处理、时序匹配、归一化处理，获得经过预处理的过程数据；
步骤三：将经过预处理的过程数据输入苛性碱浓度测量装置精度补偿模型，获得补偿值；
步骤四：将苛性碱浓度仪表值与补偿值相加，实现苛性碱浓度的在线补偿。
优选地，步骤二中采用的滑动平均滤波处理中，设置滑动滤波窗口长度，即滑动平均滤波点数为N，滤波算式为：
Figure PCTCN2020103203-appb-000001
式中，X(t)为滤波后t时刻的值，X′(t)为原数据t时刻的值，N为滑动滤波窗口长度。
优选地，步骤二中采用的时序匹配处理中，将2小时的过程数据按优化控制周期40分钟分成3份，每份取40分钟过程数据的均值，对应上一次采样的化验数据；
其中，时序匹配公式为：
Figure PCTCN2020103203-appb-000002
式中，X(i)为滤波后第i时刻的值，X(k)为匹配第k点化验值的过程 数据。
优选地，步骤二中采用的归一化处理，对苛性碱浓度测量装置精度补偿模型所用的输入输出变量状态均进行归一化处理：
Figure PCTCN2020103203-appb-000003
其中，对于某一变量数据的历史数据X＝[x 1,···,x n]，x n表示第n个点该变量状态；x max表示该变量在所有历史数据中的最大值；x min表示该变量在所有历史数据中的最小值；
其中，输入变量为经过预处理的过程数据，输出变量为补偿值。
优选地，还包括构建苛性碱浓度测量装置精度补偿模型，并训练模型参数。
优选地，构建苛性碱浓度测量装置精度补偿模型包括如下步骤：
将所选取的历史过程数据和历史苛性碱浓度仪表值与历史化验值的误差作为苛性碱浓度测量装置精度补偿模型的输入输出训练数据，采用深度学习算法，构建精度补偿模型；
深度学习算法采用双层LSTM网络建立苛性碱浓度测量装置精度补偿模型。
优选地，还包括：使用均方根误差、平均绝对误差、平均绝对百分比误差对所述苛性碱浓度测量装置精度补偿模型进行评价：
所述均方根误差的计算公式为：
Figure PCTCN2020103203-appb-000004
所述平均绝对误差的计算公式为：
Figure PCTCN2020103203-appb-000005
所述均方根误差的计算公式为：
Figure PCTCN2020103203-appb-000006
其中，y i为第i组样本的化验值，
Figure PCTCN2020103203-appb-000007
为第i组样本的补偿后的苛性碱浓度值。
(三)有益效果
本发明的有益效果是：本发明提供的一种蒸发过程出料苛性碱浓度测量装置精度补偿方法具有以下有益效果：
本发明提供的蒸发过程出料苛性碱浓度测量装置精度补偿方法通过采集苛性碱浓度在线测量装置内部模型的输入输出数据和化验数据所组成大数据集，采用深度学习算法构建并训练苛性碱浓度测量装置精度补偿模型，以形成苛性碱浓度完整模型，并使用生产过程的实时数据实现苛性碱浓度的在线补偿。在此基础上，从而指导蒸发过程进料量、蒸汽压力、新加碱量的优化，将苛性碱浓度控制在合格区间内，达到稳定氧化铝生产，降低生产成本的目的。本发明的方法能够比较精确地补偿在线仪表浓度值，补偿后的浓度值能够跟随实际变化的趋势；测量精度能够满足实际生产的需要。
附图说明
图1为本发明实施例中所述的苛性碱浓度测量装置精度补偿算法结构图；
图2为本发明实施例中所述的循环神经网络结构图；
图3为本发明实施例中所述的LSTM结构图；
图4为本发明实施例中所述的苛性碱浓度测量装置精度补偿前后浓度值与化验值对比图；
图5为本发明实施例中所述的苛性碱浓度测量装置精度补偿前后浓度值与化验值的误差对比图；
图6为本发明实施例中所述的苛性碱浓度测量装置精度补偿前后的误差评价指标图；
图7为本发明实施例中所述的苛性碱浓度测量装置精度补偿前后的误差分布图。
具体实施方式
为了更好的解释本发明，以便于理解，下面结合附图，通过具体实施方式，对本发明作详细描述。
本实施例公开了一种蒸发过程出料苛性碱浓度测量装置精度补偿方法，包括以下步骤：
步骤一：数据采集：采集蒸发过程碱液折光度、温度、苛性碱浓度仪表值及化验值的过程数据；
步骤二：数据预处理：对步骤一采集的过程数据进行滑动平均滤波处理、时序匹配、归一化处理，获得经过预处理的过程数据；
步骤三：将经过预处理的过程数据输入苛性碱浓度测量装置精度补偿模型，获得补偿值；
步骤四：将苛性碱浓度仪表值与补偿值相加，实现苛性碱浓度的在线补偿。
本实施例中步骤二中采用的滑动平均滤波处理中，设置滑动滤波窗口长度，即滑动平均滤波点数为N，滤波算式为：
Figure PCTCN2020103203-appb-000008
式中，X(t)为滤波后t时刻的值，X′(t)为原数据t时刻的值，N为滑动滤波窗口长度。
本实施例中步骤二中采用的时序匹配处理中，将2小时的过程数据按优化控制周期40分钟分成3份，每份取40分钟过程数据的均值，对应上一次采样的化验数据；
其中，时序匹配公式为：
Figure PCTCN2020103203-appb-000009
式中，X(i)为滤波后第i时刻的值，X(k)为匹配第k点化验值的过程数据。
本实施例中步骤二中采用的归一化处理，对苛性碱浓度测量装置精度补偿模型所用的输入输出变量状态均进行归一化处理：
Figure PCTCN2020103203-appb-000010
其中，对于某一变量数据的历史数据X＝[x 1,···,x n]，x n表示第n个点该变量状态；x max表示该变量在所有历史数据中的最大值；x min表示该变量在所有历史数据中的最小值；
其中，输入变量为经过预处理的过程数据，输出变量为补偿值。
本实施例中还包括构建苛性碱浓度测量装置精度补偿模型，并训练模型参数。
本实施例中构建苛性碱浓度测量装置精度补偿模型包括如下步骤：
将所选取的历史过程数据和历史苛性碱浓度仪表值与历史化验值的误差作为苛性碱浓度测量装置精度补偿模型的输入输出训练数据，采用深度学习算法，构建精度补偿模型；
深度学习算法采用双层LSTM网络建立苛性碱浓度测量装置精度补偿模型。
本实施例中还包括：使用均方根误差、平均绝对误差、平均绝对百分比误差对所述苛性碱浓度测量装置精度补偿模型进行评价：
所述均方根误差的计算公式为：
Figure PCTCN2020103203-appb-000011
所述平均绝对误差的计算公式为：
Figure PCTCN2020103203-appb-000012
所述均方根误差的计算公式为：
Figure PCTCN2020103203-appb-000013
其中，y i为第i组样本的化验值，
Figure PCTCN2020103203-appb-000014
为第i组样本的补偿后的苛性碱浓度值。
如图1所示，本方法发明的蒸发过程出料苛性碱浓度测量装置精度补偿方法，包括以下步骤：
步骤一：数据采集
采集过程生产参数数据及离线取样的化验数据，过程数据通过读取蒸发过程PHD数据库获得，时间间隔1分钟，包括采集出料碱液折光度R(k)、出料碱液温度T(k)以及苛性碱浓度仪表值；化验数据y Nk(k)通过访问企业MES系统获得，时间间隔2小时。同时将采集到的数据以时间标签递增的方式存储到本地磁盘中，存储格式为csv文件。
步骤二：数据预处理：对步骤一采集的3类过程数据进行滑动平均滤波处理。将过程数据组成数据矩阵X＝[x 1,x 2,x 3]。设置滑动滤波窗口长度，即滑动平均滤波点数为10，则每一列的滤波算式为：
Figure PCTCN2020103203-appb-000015
式中，X(t)为滤波后t时刻的值，X′(t)为原数据t时刻的值，N为滑动滤波窗口长度。
时序匹配：由于过程数据与化验数据采样周期不同(采样周期：过程数据1分钟，化验数据2小时)，为充分使用过程数据及提取其特征，将2小时的过程数据按优化控制周期40分钟分成3份，每份取40分钟过程数据的均值，对应上一次采样的化验数据。时序匹配公式为：
Figure PCTCN2020103203-appb-000016
式中，X(i)为滤波后第i时刻的值，X(k)为匹配第k点化验值的过程数据。
对于时序匹配完成的数据，划分蒸发过程出料苛性碱浓度测量装置精度补偿模型的输入输出变量，以折光度、温度为输入变量，苛性碱浓度仪表值与化验值的误差为输出变量。
归一化处理：对蒸发过程精度补偿模型所用的输入输出变量状态均进行归一化处理：
Figure PCTCN2020103203-appb-000017
其中，对于某变量数据的历史数据D＝[x 1,···,x n]，x n表示第n时刻该变量状态；x max表示该变量在所有历史数据中的最大值；x min表示该变量在所有历史数据中的最小值；
步骤三：构建苛性碱浓度测量装置精度补偿模型，并训练模型参数
经数据处理的输入输出数据共有10500组，选用8000组数据用于训练，2500组用于测试。将所选取的折光度、温度数据作为输入数据，浓度仪表值和化验值的误差作为输出数据，采用深度学习算法，构建深度学习精度补偿模型。处理时间序列数据较好的深度学习算法循环神经网络结构如图5所示。
上图中左边部分是循环神经网络的一个循环结构，代表着当前隐藏层的输入包括上一个时刻隐藏层的输出。将该网络结构展开，得到图中右边部分。其中W，U和V是权重，x t是t时刻的输入，s t是对应t时刻的隐藏状态，即隐藏层输出，是网络的记忆单元。o t是t时刻的输出。循环神经网络的数学公式如下：
s t＝f(Ux t+Ws t-1)       (16)
o t＝soft max(Vs t)         (17)
其中，f通常是非线性的激活函数，例如tanh和relu。s t通过前一时刻的隐藏输出s t-1和当前时刻的输入x t求得。softmax函数是输出层的激活函数，常用于分类问题，将输出映射为(0,1)的概率分布。
由于传统的RNN模型存在梯度消失和梯度爆炸的问题，当序列很长的时候问题尤其严重，此时传统的RNN模型一般不能直接使用，而较为广泛使用的是RNN的一个特例——长短期记忆网络(LSTM)。
长短期记忆网络(Long short-term memory，LSTM)是一种RNN特殊的类型，LSTM区别于RNN的地方，主要就在于它在算法中加入了一个判断信息有用与否的“处理器”，这个处理器作用的结构被称为cell，如图6所示。一个cell当中被放置了三扇门，分别叫做输入门、遗忘门和输出门。一个信息进入LSTM的网络当中，可以根据规则来判断是否有用。只有符合算法认证的信息才会留下，不符的信息则通过遗忘门被遗忘。
Cell结构的运作原理可以用式(18)～式(22)表示：
f t＝σ(W f·[h t-1,x t]+b f)        (18)
i t＝σ(W i·[h t-1,x t]+b i)         (19)
C t＝f t*C t-1+i t*tanh(W C·[h t-1,x t]+b C)       (20)
o t＝σ(W o·[h t-1,x t]+b o)      (21)
h t＝o t*tanh(C t)         (22)
其中，h t表示t时间点LSTM单元的所有输出；W f，W i，W C，W o为系数组成的权重矩阵，b f，b i，b c，b o为对应权重的偏置向量，σ是激活函数sigmoid，tanh为激活函数；·是点乘运算；C t表示t时刻记忆细胞的计算方法；f t，i t，o t分别是t时间点输入门、遗忘门和输出门的计算方法。由图6可以看出输入门、遗忘门和输出门3个控制门的输出各自连接到了一个乘法元件上，从而分别控制信息流的输入、输出以及细胞单元的状态。
采用双层LSTM网络、一层全连接层构建苛性碱浓度测量装置精度补偿模型，使用前面划分好的训练集和测试集对精度补偿模型进行训练和测试，采用均方根误差作为误差函数用于神经网络的误差反向传播学习，均方误差的计算公式为：
Figure PCTCN2020103203-appb-000018
将步骤三中苛性碱浓度测量装置精度补偿模型输出反归一化得到的补偿值与测量装置的输出值相加，得到苛性碱浓度预测值，具体如下式所示：
Figure PCTCN2020103203-appb-000019
式中，
Figure PCTCN2020103203-appb-000020
为精度补偿算法苛性碱浓度预测值；Nk(k)表示苛性碱浓度测量装置输出值；e(k)表示基于深度学习的精度补偿模型输出值；
步骤四：苛性碱浓度的实时补偿预测
预测效果如图4和5所示。由图4及图5可以看出，补偿后苛性碱浓度值较补偿前，跟随化验值效果较好，与化验值之间误差明显较小。
对精度补偿算法苛性碱浓度预测值、苛性碱浓度测量装置输出值使用均方根误差、平均绝对误差、平均绝对百分比误差对其进行评价：
所述均方根误差的计算公式为：
Figure PCTCN2020103203-appb-000021
所述平均绝对误差的计算公式为：
Figure PCTCN2020103203-appb-000022
所述均方根误差的计算公式为：
Figure PCTCN2020103203-appb-000023
其中，y i为第i组样本的化验值，
Figure PCTCN2020103203-appb-000024
为第i组样本的补偿前后的苛性碱浓度值。
指标计算结果如表1及图6所示，补偿前RMSE指标为9.332，补偿后RMSE指标为3.006；补偿前MAE指标为8.235，补偿后MAE指标为2.264；补偿前MAPE指标为3.562，补偿后MAPE指标为0.97；其中，从RMSE指标来看，补偿后较补偿前提高67.8％；从MAE指标来看，补偿后较补偿前提高72.5％；从MAPE指标来看，补偿后较补偿前提高72.8％。
表1苛性碱浓度测量装置补偿前后的误差评价指标表
  RMSE MAE MAPE
补偿前 9.332 8.235 3.562
补偿后 3.006 2.264 0.97
补偿前后误差的分布情况如表2及图7所示，企业鉴于化验室化验得到苛性碱浓度值本身可能存在一定误差，因此，将仪表值与化验值的允许误差定义在3g/l以内，用补偿后的苛性碱浓度值与化验值比较，误差在3g/l以内即视为合格。从图中可以看出，补偿前的误差多集中于5g/l以上，补偿后误差多集中于3g/l以下。补偿前苛性碱浓度值的合格率为12.16％，引 入本文建立的苛性碱浓度测量装置精度补偿模型后，补偿后的苛性碱浓度值的合格率为71.36％，较补偿前提升59.2％，精度补偿效果明显。
表2苛性碱浓度测量装置精度补偿前后的误差分布表
  |error|<＝3 3<|error|<＝5 |error|>5
补偿前 304 322 1874
补偿后 1784 549 167
综上所述，苛性碱浓度测量装置精度补偿模型的精度较高，模型可靠性与准确度高。
本实施例中还构建了模型精度判断模块，当苛性碱浓度测量装置精度补偿模型无法满足要求时，需要对精度补偿模型重新训练校正。通过近期所采集积累的过程数据和化验值，训练新的模型，实现精度补偿模型长期、稳定、准确的在线补偿。
以上结合具体实施例描述了本发明的技术原理，这些描述只是为了解释本发明的原理，不能以任何方式解释为对本发明保护范围的限制。基于此处解释，本领域的技术人员不需要付出创造性的劳动即可联想到本发明的其它具体实施方式，这些方式都将落入本发明的保护范围之内。

Claims (7)

  1. 一种蒸发过程出料苛性碱浓度测量装置精度补偿方法，其特征在于，包括以下步骤：
    步骤一：数据采集：采集蒸发过程碱液折光度、温度、苛性碱浓度仪表值及化验值的过程数据；
    步骤二：数据预处理：对步骤一采集的过程数据进行滑动平均滤波处理、时序匹配、归一化处理，获得经过预处理的过程数据；
    步骤三：将经过预处理的过程数据输入苛性碱浓度测量装置精度补偿模型，获得补偿值；
    步骤四：将苛性碱浓度仪表值与补偿值相加，实现苛性碱浓度的在线补偿。
  2. 根据权利要求1所述的补偿方法，其特征在于：
    步骤二中采用的滑动平均滤波处理中，设置滑动滤波窗口长度，即滑动平均滤波点数为N，滤波算式为：
    Figure PCTCN2020103203-appb-100001
    式中，X(t)为滤波后t时刻的值，X′(t)为原数据t时刻的值，N为滑动滤波窗口长度。
  3. 根据权利要求1所述的补偿方法，其特征在于：步骤二中采用的时序匹配处理中，将2小时的过程数据按优化控制周期40分钟分成3份，每份取40分钟过程数据的均值，对应上一次采样的化验数据；
    其中，时序匹配公式为：
    Figure PCTCN2020103203-appb-100002
    式中，X(i)为滤波后第i时刻的值，X(k)为匹配第k点化验值的过程 数据。
  4. 根据权利要求1所述的补偿方法，其特征在于：步骤二中采用的归一化处理，对苛性碱浓度测量装置精度补偿模型所用的输入输出变量状态均进行归一化处理：
    Figure PCTCN2020103203-appb-100003
    其中，对于某一变量数据的历史数据X＝[x 1,···,x n]，x n表示第n个点该变量状态；x max表示该变量在所有历史数据中的最大值；x min表示该变量在所有历史数据中的最小值；
    其中，输入变量为经过预处理的过程数据，输出变量为补偿值。
  5. 根据权利要求1所述补偿方法，其特征在于：还包括构建苛性碱浓度测量装置精度补偿模型，并训练模型参数。
  6. 根据权利要求5所述的补偿方法，其特征在于，构建苛性碱浓度测量装置精度补偿模型包括如下步骤：
    将所选取的历史过程数据和历史苛性碱浓度仪表值与历史化验值的误差作为苛性碱浓度测量装置精度补偿模型的输入输出训练数据，采用深度学习算法，构建精度补偿模型；
    深度学习算法采用双层LSTM网络建立苛性碱浓度测量装置精度补偿模型。
  7. 根据权利要求6所述的补偿方法，其特征在于：
    还包括：使用均方根误差、平均绝对误差、平均绝对百分比误差对所述苛性碱浓度测量装置精度补偿模型进行评价：
    所述均方根误差的计算公式为：
    Figure PCTCN2020103203-appb-100004
    所述平均绝对误差的计算公式为：
    Figure PCTCN2020103203-appb-100005
    所述均方根误差的计算公式为：
    Figure PCTCN2020103203-appb-100006
    其中，y i为第i组样本的化验值，
    Figure PCTCN2020103203-appb-100007
    为第i组样本的补偿后的苛性碱浓度值。
PCT/CN2020/103203 2019-11-11 2020-07-21 一种蒸发过程出料苛性碱浓度测量装置精度补偿方法 WO2021093362A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/309,942 US11893472B2 (en) 2019-11-11 2020-07-21 Accuracy compensation method for discharge caustic alkali concentration measuring device in evaporation process

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911096987.0 2019-11-11
CN201911096987.0A CN110794093B (zh) 2019-11-11 2019-11-11 一种蒸发过程出料苛性碱浓度测量装置精度补偿方法

Publications (1)

Publication Number Publication Date
WO2021093362A1 true WO2021093362A1 (zh) 2021-05-20

Family

ID=69443935

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/103203 WO2021093362A1 (zh) 2019-11-11 2020-07-21 一种蒸发过程出料苛性碱浓度测量装置精度补偿方法

Country Status (3)

Country Link
US (1) US11893472B2 (zh)
CN (1) CN110794093B (zh)
WO (1) WO2021093362A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113494527A (zh) * 2021-07-30 2021-10-12 哈尔滨工业大学 一种基于电磁辅助式恒力弹簧支架的恒力控制方法

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110794093B (zh) * 2019-11-11 2021-12-03 东北大学 一种蒸发过程出料苛性碱浓度测量装置精度补偿方法
CN113467295B (zh) * 2021-06-18 2022-10-25 东北大学 一种氧化铝溶出过程的自动控制方法
CN113608560B (zh) * 2021-06-18 2023-08-25 东北大学 一种氧化铝碱液调配过程的控制系统
CN113608506B (zh) * 2021-06-18 2022-10-25 东北大学 一种氧化铝运行指标的智能检测装置
CN113671011A (zh) * 2021-08-20 2021-11-19 中煤科工集团重庆研究院有限公司 一种抗压强突变影响氧气浓度高精度测量方法
CN116467911B (zh) * 2023-04-13 2023-12-15 深圳职业技术学院 基于多工况信息融合的蒸发过程出口溶液浓度估算方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101281182A (zh) * 2008-05-22 2008-10-08 沈阳东大自动化有限公司 铝酸钠溶液组分浓度软测量方法
CN102809966A (zh) * 2012-07-30 2012-12-05 上海交通大学 基于半闭环的软测量仪表及其软测量方法
US20140054464A1 (en) * 2012-08-20 2014-02-27 Sabic Innovative Plastics Ip B.V. Real-time online determination of caustic in process scrubbers using near infrared spectroscopy and chemometrics
CN107168255A (zh) * 2017-05-16 2017-09-15 浙江工业大学 一种基于集成神经网络的聚丙烯熔融指数混合建模方法
CN110096755A (zh) * 2019-04-08 2019-08-06 沈阳工业大学 固体蓄热炉内高温加热元件在线温度软测量方法及系统
CN110208453A (zh) * 2019-05-22 2019-09-06 沈阳博宇科技有限责任公司 一种铝酸钠溶液中苛性碱浓度的软测量方法
CN110794093A (zh) * 2019-11-11 2020-02-14 东北大学 一种蒸发过程出料苛性碱浓度测量装置精度补偿方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9212989B2 (en) * 2005-10-06 2015-12-15 Jp3 Measurement, Llc Optical determination and reporting of gas properties
CN105821170A (zh) * 2016-05-11 2016-08-03 东北大学 一种高炉多元铁水质量指标软测量系统及方法
CN108304823B (zh) * 2018-02-24 2022-03-22 重庆邮电大学 一种基于双卷积cnn和长短时记忆网络的表情识别方法
CN109740742A (zh) * 2019-01-14 2019-05-10 哈尔滨工程大学 一种基于lstm神经网络的目标跟踪方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101281182A (zh) * 2008-05-22 2008-10-08 沈阳东大自动化有限公司 铝酸钠溶液组分浓度软测量方法
CN102809966A (zh) * 2012-07-30 2012-12-05 上海交通大学 基于半闭环的软测量仪表及其软测量方法
US20140054464A1 (en) * 2012-08-20 2014-02-27 Sabic Innovative Plastics Ip B.V. Real-time online determination of caustic in process scrubbers using near infrared spectroscopy and chemometrics
CN107168255A (zh) * 2017-05-16 2017-09-15 浙江工业大学 一种基于集成神经网络的聚丙烯熔融指数混合建模方法
CN110096755A (zh) * 2019-04-08 2019-08-06 沈阳工业大学 固体蓄热炉内高温加热元件在线温度软测量方法及系统
CN110208453A (zh) * 2019-05-22 2019-09-06 沈阳博宇科技有限责任公司 一种铝酸钠溶液中苛性碱浓度的软测量方法
CN110794093A (zh) * 2019-11-11 2020-02-14 东北大学 一种蒸发过程出料苛性碱浓度测量装置精度补偿方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113494527A (zh) * 2021-07-30 2021-10-12 哈尔滨工业大学 一种基于电磁辅助式恒力弹簧支架的恒力控制方法
CN113494527B (zh) * 2021-07-30 2022-06-24 哈尔滨工业大学 一种基于电磁辅助式恒力弹簧支架的恒力控制方法

Also Published As

Publication number Publication date
US11893472B2 (en) 2024-02-06
US20220083839A1 (en) 2022-03-17
CN110794093B (zh) 2021-12-03
CN110794093A (zh) 2020-02-14

Similar Documents

Publication Publication Date Title
WO2021093362A1 (zh) 一种蒸发过程出料苛性碱浓度测量装置精度补偿方法
CN110705692B (zh) 一种时空注意力网络的非线性动态工业过程产品预测方法
CN116757534B (zh) 一种基于神经训练网络的智能冰箱可靠性分析方法
Wu et al. Modeling and analysis of tool wear prediction based on SVD and BiLSTM
CN106845796A (zh) 一种加氢裂化流程产品质量在线预测方法
CN111626785A (zh) 一种基于结合注意力的cnn-lstm网络基金价格预测方法
CN110490496B (zh) 一种基于分步约简筛选复杂工业过程中影响产品质量的敏感变量的方法
CN112949894B (zh) 一种基于简化型长短期记忆神经网络的出水bod预测方法
CN113447879A (zh) 基于多应力影响的电能表计量误差预测方法、系统及装置
CN113393057A (zh) 一种基于深度融合机器学习模型的小麦产量集成预测方法
Zhang et al. A two-stage data-driven approach to remaining useful life prediction via long short-term memory networks
CN111103420B (zh) 一种原料不确定下的酚醛树脂产品质量预测方法
CN115096357A (zh) 一种基于ceemdan-pca-lstm的室内环境质量预测方法
CN103279030B (zh) 基于贝叶斯框架的动态软测量建模方法及装置
CN117520784A (zh) 基于卷积注意力长短期神经网络的地下水位多步预测方法
CN107704944A (zh) 一种基于信息论学习的股市波动区间预测方法
CN116842358A (zh) 一种基于多尺度卷积和自适应特征融合的软测量建模方法
CN116757078A (zh) 一种基于作用力的煤粉流速测量方法及系统
CN116306803A (zh) 一种基于wsfa-afe的ilstm神经网络的出水bod浓度预测方法
CN105787507A (zh) 基于预算支持向量集的LS-SVMs在线学习方法
CN115688865A (zh) 面向脱硫工艺烟气的长短期记忆网络工业软测量方法
CN115563579A (zh) 一种电容式传感器测量含沙量的数据融合方法
Wu et al. Time series online prediction algorithm based on least squares support vector machine
CN114971022A (zh) 基于D-value-LSTM神经网络模型的风电功率预测方法
Ling et al. A soft sensor based on influent mode discrimination neural network for a wastewater treatment process

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20888245

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20888245

Country of ref document: EP

Kind code of ref document: A1