CN113919574A - Prediction method of heavy metal emissions from waste incineration plant flue gas based on Keras - Google Patents

Abstract

The invention discloses a Keras-based prediction method for heavy metal emission in flue gas of a waste incineration plant, which relates to the technical field of heavy metal emission prediction and comprises the following steps: collecting continuous data and classified data related to the waste incineration process to form a data set; carrying out data preprocessing on the data set; inputting the data in the data set into a neural network model in a parallel mode; setting the structure of a neural network model, and selecting an optimal neural network model; dividing the preprocessed data set into a training set, a testing set and a verification set, training the neural network model by using the training set, verifying the precision of the neural network model by using the testing set, and performing cyclic iteration to obtain the neural network model meeting the set precision; and predicting the heavy metal emission of the flue gas by using a model. The method can obtain a high-precision neural network prediction model and realize accurate prediction of heavy metal emission of the flue gas.

Description

Keras-based prediction method for heavy metal emission in flue gas of waste incineration plant

Technical Field

The invention relates to the technical field of heavy metal emission prediction, in particular to a Keras-based prediction method for heavy metal emission in flue gas of a waste incineration plant.

Background

The safe and sustainable disposal of the household garbage is an important factor related to the quality of life of residents in China, the substance living level of the residents is continuously improved along with the rapid development of the social economy in China, and the cleaning amount of the household garbage is increased day by day.

Most of the current domestic garbage disposal modes are landfill technologies, and landfill refers to the process of compressing and stacking domestic garbage in a selected landfill site after anti-seepage materials are laid in the site. Due to the rapid development of cities, the land resources are increasingly in short supply, the application area of landfill sites is increasingly reduced, and the garbage disposal becomes an important problem in the development of the cities.

In the household garbage incineration power generation technology, the volume and the quality of the household garbage can be greatly reduced in the incineration process, the reduction of the household garbage is realized, meanwhile, the steam generated by incineration is used for pushing a steam turbine to generate power, the recycling of the household garbage disposal can also be realized, and a large amount of pathogenic microorganisms or pathogenic protozoa and the like in the garbage can be killed in the high-temperature environment in the incineration process, so that the harmlessness of the household garbage is realized.

The initial investment of the household garbage incineration power plant is huge, and compared with a simple operation mode of a landfill, the household garbage incineration power plant needs to be operated by professional personnel; meanwhile, most of the incinerator types are reciprocating grate furnaces, the incinerator has high requirement on the heat value of domestic garbage entering the incinerator, and the common garbage has high water content and low heat value, so that the combustion requirement of the incinerator cannot be met.

In the garbage disposal process, the pollution elements in the household garbage can be separated out in the high-temperature process and generate a large amount of pollutants including heavy metals which have serious harm to human bodies and ecological environment through complex chemical reactions in an incinerator. Particularly, in the process of burning electronic waste and other substances doped with heavy metals in a furnace in the manufacturing process, the heavy metals can be separated out, the heavy metals are divided into four categories of nonvolatile heavy metals, semi-volatile heavy metals, volatile heavy metals and gaseous heavy metals according to the difference of the boiling points of the heavy metals, the other three categories of heavy metals except the nonvolatile heavy metals can be removed in a mode of entering smoke or being attached to fly ash, and the heavy metals have 'triple effect' on human bodies, namely teratogenesis, carcinogenesis and mutagenesis.

Therefore, how to accurately predict the heavy metal emission in the flue gas of the waste incineration power plant is a problem which needs to be solved urgently by the technical personnel in the field.

Disclosure of Invention

In view of the above, the invention provides a Keras-based prediction method for heavy metal emission in flue gas of a waste incineration plant.

In order to achieve the above purpose, the invention provides the following technical scheme:

a Keras-based prediction method for heavy metal emission in flue gas of a waste incineration plant comprises the following steps:

step one, collecting continuous data and classified data related to a waste incineration process to form a data set;

secondly, data preprocessing is carried out on the data in the data set, and the data are unified into an input form of a machine learning model;

inputting the data in the data set into a neural network model in a parallel mode;

setting the structure of the neural network model, and selecting the optimal neural network model;

dividing the preprocessed data set into a training set, a testing set and a verification set, training the neural network model by using the training set, verifying the precision of the neural network model by using the testing set, and performing cyclic iteration to obtain the neural network model meeting the set precision;

and step six, predicting the heavy metal emission of the flue gas according to the neural network model meeting the set precision.

Optionally, the method further includes a final output precision detection step, where for the neural network model satisfying the set precision, a verification set is used to perform precision verification on the neural network model, so as to obtain the final output precision of the neural network model.

Optionally, the continuous data includes original household garbage components, economic development levels of various regions, heavy metal emission data and the like, and the classified data includes furnace types, flue gas purification processes and the like.

Optionally, the data preprocessing specifically refers to performing data conversion and normalization processing on continuous data, and performing one-hot code conversion on classified data.

Optionally, the data conversion specifically refers to converting the percentage content of the original household garbage components into the charging content of various heavy metals through multidimensional matrix conversion, and the specific method includes:

setting a certain component X of the domestic garbage of unit mass_iThe content of the heavy metal j is y, the proportion of elements precipitated from the percolate is g, and the main process of matrix conversion is as follows:

y_{into the furnace}＝y_{Enter the factory}×g_Percolate

In the formula, y_{Enter the factory}Expressing the content of heavy metals per unit mass of garbage entering the plant, HM_i,jIs a certain garbage component X of unit mass_iContent coefficient of heavy metal j in (L)_i,jIs the content coefficient of heavy metal j in unit mass of percolate.

The normalization processing refers to the situation that indexes with large data base numbers are compressed into indexes with small base numbers due to magnitude difference of economic development indexes of various regions, and the indexes with small base numbers cannot be normally expressed due to direct input, so that various economic development index variables are normalized, and data errors caused by the normalization processing are reduced.

Optionally, the typing data is subjected to numerical conversion (i.e., one-hot code conversion) in a one-hot code form, and the single-column typing data is processed into multi-dimensional matrix numerical input data.

Optionally, the setting of the structure of the neural network model specifically includes setting a hidden layer, an activation function, and the number of neurons in each layer of the neural network model based on the Keras machine learning model according to the data volume of the raw data.

Optionally, the optimal neural network model is selected, specifically, by comparing the accuracies of different neural network models, the neural network model structure with the highest model accuracy and the highest operation speed is selected.

Optionally, the loop iteration mode specifically includes that a root mean square is used as a check error index, the error index is fed forward to the front end of the hidden layer, and input coefficients of each neuron are redistributed;

and setting a root mean square error threshold, and stopping loop iteration when the root mean square error output by the neural network model is less than or equal to the root mean square error threshold to obtain the neural network model meeting the set precision.

According to the technical scheme, the invention discloses and provides a Keras-based prediction method for heavy metal emission in flue gas of a waste incineration plant, and compared with the prior art, the Keras-based prediction method for heavy metal emission in flue gas of the waste incineration plant has the following beneficial effects:

the invention collects continuous data and classified data, and carries out data preprocessing on different types of data to obtain the input form data of the machine learning model, thereby providing a good data base for the construction and optimization of the subsequent model and reducing data errors. Furthermore, the method and the device set the specific structure of the prediction model based on Keras, further select the optimal neural network model structure, and can improve the accuracy of data prediction. After the optimal prediction model is selected, the preprocessed data (the heavy metal charging content, the original household garbage component, the furnace type, the economic development level, the flue gas purification process, the heavy metal emission data and the like) are divided into a training set, a testing set and a verification set, and the training, testing and final precision verification of the model are respectively carried out, so that a high-precision neural network prediction model can be obtained, and the accurate prediction of the heavy metal emission of the flue gas is realized.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.

FIG. 1 is a schematic diagram of the process steps of the present invention;

FIG. 2 is a diagram of a network architecture for machine learning in accordance with the present invention;

FIG. 3 is a flow chart of the algorithm of the present invention;

FIG. 4 is a diagram illustrating fitting results according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The embodiment of the invention discloses a Keras-based prediction method for heavy metal emission in flue gas of a waste incineration plant, and the method is shown in figure 1 and comprises the following steps:

step one, collecting continuous data and classified data related to the waste incineration process to form a data set. The continuous data comprises original household garbage components, economic development levels of various regions, emission data of different types of heavy metals and the like, and the classified data comprises furnace types, flue gas purification processes and the like. In the subsequent training of the neural network, the original household garbage components, the economic development levels of all regions, the furnace types, the flue gas purification process and the like are used as the input characteristics of the model, and the emission data of different types of heavy metals are used as the corresponding output.

And step two, carrying out data preprocessing on the data in the data set, and unifying the data into an input form of a machine learning model. The data preprocessing specifically refers to data conversion and normalization processing on continuous data and one-hot code conversion on classified data.

The data conversion specifically refers to converting the percentage content of the original household garbage components into the charging content of various heavy metals through multi-dimensional matrix conversion, and the method comprises the following steps:

setting a certain component X of the domestic garbage of unit mass_iThe content of the heavy metal j is y, the proportion of elements precipitated from the percolate is g, and the main process of matrix conversion is as follows:

y_{into the furnace}＝y_{Enter the factory}×g_Percolate

In the formula, y_{Enter the factory}Expressing the content of heavy metals per unit mass of garbage entering the plant, HM_i,jIs a certain garbage component X of unit mass_iContent coefficient of heavy metal j in (L)_i,jIs the content coefficient of heavy metal j in unit mass of percolate.

The normalization processing means that due to the fact that economic development indexes of different places have magnitude difference, the indexes with large data base numbers are compressed into the indexes with small base numbers by direct input, and the indexes with small base numbers cannot be normally expressed, so that various economic development index variables are normalized, and data errors caused by the normalization processing are reduced.

The typing data is subjected to numerical conversion (namely, single-hot coding conversion) in a single-hot coding mode, and the single-column typing data is processed into multi-dimensional matrix numerical input data.

Therefore, the data preprocessing is adopted to obtain the data of the original household garbage components, the economic development levels of various regions, the furnace types, the flue gas purification process, the furnace charging contents of various heavy metals and the like in the input form of the machine learning model.

And step three, theoretically, the domestic garbage is fed into the furnace, incinerated in the furnace and purified by the smoke gas in a sequential manner in the operation process, but errors are accumulated because the input variables are connected in series and are respectively provided with the hidden layer and the neuron, so that the errors caused by the structural design of the model are reduced by adopting variable parallel input in the invention.

And step four, setting the structure of the neural network model, and selecting the optimal neural network model. The number of hidden layers has a greater influence on the model than the number of neurons in a single layer, and therefore the hidden layers, the activation functions, and the number of neurons in each layer of the neural network model are set based on the Keras machine learning model according to the data volume of the original data. And selecting a neural network model structure with highest model precision and highest running speed by comparing the precision of different neural network models.

And fifthly, dividing the preprocessed data set into a training set, a testing set and a verification set, training a neural network model by using the training set, wherein the original household garbage components, the economic development levels of various regions, the furnace types, the flue gas purification processes and the like are used as input characteristics of the model, the emission data of different types of heavy metals are used as corresponding output, and the testing set is used for verifying the precision of the neural network model and carrying out cyclic iteration, so as to obtain the neural network model meeting the set precision, referring to fig. 2.

The loop iteration mode is specifically that the root mean square is used as a check error index, the error index is fed forward to the front end of the hidden layer, and input coefficients of all neurons are redistributed; and setting a root mean square error threshold, and stopping loop iteration when the root mean square error output by the neural network model is less than or equal to the root mean square error threshold to obtain the neural network model meeting the set precision.

And step six, predicting the heavy metal emission of the flue gas according to the neural network model meeting the set precision, and referring to fig. 3. Taking the prediction of Cd + Tl as an example, the fitting of the actual value and the operation result of the predicted value is shown in FIG. 4, the fitting slope is close to 1, namely the coincidence degree of the actual value and the predicted value is good.

In another embodiment, the step seven further includes a step of detecting final output precision, and for the neural network model meeting the set precision, performing precision verification on the neural network model by using a verification set to obtain the final output precision of the neural network model.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (9)

Translated fromChinese

1.一种基于Keras的垃圾焚烧厂烟气重金属排放预测方法，其特征在于，包括以下步骤：1. a method for predicting the discharge of heavy metals in waste incineration plant flue gas based on Keras, is characterized in that, comprises the following steps:步骤一、收集垃圾焚烧过程涉及的连续型数据和分类型数据，构成数据集；Step 1. Collect continuous data and classified data involved in the waste incineration process to form a data set;步骤二、将所述数据集中的数据进行数据预处理；Step 2, performing data preprocessing on the data in the data set;步骤三、将所述数据集中的数据以并联的形式输入神经网络模型中；Step 3: Input the data in the data set into the neural network model in parallel;步骤四、设置神经网络模型的结构，并选取最优的神经网络模型；Step 4: Set the structure of the neural network model, and select the optimal neural network model;步骤五、将预处理之后的数据集分为训练集、测试集和验证集，使用训练集训练所述神经网络模型，使用所述测试集对所述神经网络模型的精度进行校验，并进行循环迭代，得到满足设定精度的神经网络模型；Step 5: Divide the preprocessed data set into a training set, a test set and a verification set, use the training set to train the neural network model, use the test set to verify the accuracy of the neural network model, and carry out Loop iteration to obtain a neural network model that meets the set accuracy;步骤六、依据满足设定精度的所述神经网络模型，对烟气重金属排放量进行预测。Step 6: Predict the emission of heavy metals in the flue gas according to the neural network model that meets the set accuracy.2.根据权利要求1所述的一种基于Keras的垃圾焚烧厂烟气重金属排放预测方法，其特征在于，还包括最终输出精度检测步骤，对于满足设定精度的所述神经网络模型，使用验证集对所述神经网络模型进行精度校验，得到神经网络模型的最终输出精度。2. a Keras-based method for predicting heavy metal emissions from waste incineration plant flue gas according to claim 1, further comprising a final output accuracy detection step, for the neural network model that satisfies the set accuracy, use verification Set the accuracy of the neural network model to obtain the final output accuracy of the neural network model.3.根据权利要求1所述的一种基于Keras的垃圾焚烧厂烟气重金属排放预测方法，其特征在于，所述连续型数据包括原始生活垃圾组分、各地经济发展水平、重金属排放数据，所述分类型数据包括炉型、烟气净化工艺。3. a Keras-based method for predicting heavy metal discharge in waste incineration plant flue gas according to claim 1, wherein the continuous data comprises original household waste components, local economic development levels, heavy metal discharge data, and the The classification data includes furnace type and flue gas purification process.4.根据权利要求1所述的一种基于Keras的垃圾焚烧厂烟气重金属排放预测方法，其特征在于，所述数据预处理具体指，对于连续型数据进行数据换算和归一化处理，对于分类型数据进行独热编码转换。4. a Keras-based method for predicting heavy metal emissions from waste incineration plant flue gas according to claim 1, wherein the data preprocessing specifically refers to performing data conversion and normalization for continuous data, and for One-hot encoding transformation of typed data.5.根据权利要求4所述的一种基于Keras的垃圾焚烧厂烟气重金属排放预测方法，其特征在于，所述数据换算是指通过多维矩阵转化，将原始生活垃圾组分的百分比含量转化为各类重金属入炉含量，具体方法为：5. a kind of Keras-based waste incineration plant flue gas heavy metal emission prediction method according to claim 4, is characterized in that, described data conversion refers to by multi-dimensional matrix transformation, the percentage content of original domestic waste component is transformed into The content of various heavy metals into the furnace, the specific methods are:设单位质量生活垃圾某组分X_i中重金属j含量为y，渗滤液析出后的元素占比为g，矩阵转换主要过程如下：Assuming that the content of heavy metal j in a certain component X_i of unit mass of domestic waste is y, and the proportion of elements after leachate precipitation is g, the main process of matrix transformation is as follows:

y_入炉＝y_入厂×g_渗滤液y_{entering the furnace} = y_{entering the factory} × g_leachate式中，y_入厂表示单位质量入厂垃圾中重金属含量，HM_i,j为单位质量某垃圾组分X_i中的重金属j的含量系数，L_i,j为单位质量渗滤液中重金属j的含量系数。In the formula, y_{entering the plant} represents the heavy metal content in the unit mass of incoming waste, HM_i,j is the content coefficient of the heavy metal j in a certain waste component X_i per unit mass, L_i,j is the heavy metal j in the unit mass leachate Content factor.6.根据权利要求4所述的一种基于Keras的垃圾焚烧厂烟气重金属排放预测方法，其特征在于，所述分类型数据通过独热编码的形式进行数值化转换，将单列分类型数据处理为多维矩阵性数值型输入数据。6. A Keras-based method for predicting heavy metal emissions from waste incineration plant flue gas according to claim 4, wherein the classification data is numerically converted by one-hot encoding, and the single-column classification data is processed Enter data for a multidimensional matrix numeric type.7.根据权利要求1所述的一种基于Keras的垃圾焚烧厂烟气重金属排放预测方法，其特征在于，所述设置神经网络模型的结构，具体包括，根据原始数据的数据量，基于Keras机器学习模型，设置神经网络模型的隐藏层、激活函数以及各层的神经元数量。7. a Keras-based method for predicting heavy metal emissions from waste incineration plant flue gas according to claim 1, wherein the described setting of the structure of the neural network model specifically includes, according to the data volume of the original data, based on the Keras machine To learn the model, set the hidden layer, activation function and the number of neurons in each layer of the neural network model.8.根据权利要求1所述的一种基于Keras的垃圾焚烧厂烟气重金属排放预测方法，其特征在于，所述选取最优的神经网络模型，具体为，通过对比不同神经网络模型的精度，选择出模型精度最高、运行速度最快的神经网络模型结构。8. a Keras-based method for predicting heavy metal emissions from waste incineration plant flue gas according to claim 1, wherein the optimal neural network model is selected, specifically, by comparing the accuracy of different neural network models, Select the neural network model structure with the highest model accuracy and the fastest running speed.9.根据权利要求1所述的一种基于Keras的垃圾焚烧厂烟气重金属排放预测方法，其特征在于，所述循环迭代的方式具体为，以均方根作为校验误差指标，将误差指标前馈至隐藏层前端，并重新分配各神经元输入系数；9. A Keras-based method for predicting heavy metal emissions from waste incineration plant flue gas according to claim 1, wherein the cycle iteration method is specifically, taking root mean square as a check error index, and using the error index Feed forward to the front end of the hidden layer and redistribute the input coefficients of each neuron;设置均方根误差阈值，当神经网络模型输出的均方根误差小于等于所述均方根误差阈值时，停止循环迭代，得到满足设定精度的神经网络模型。A root mean square error threshold is set, and when the root mean square error output by the neural network model is less than or equal to the root mean square error threshold, the loop iteration is stopped to obtain a neural network model that meets the set accuracy.

Images