Carbon measurement data filling method and system suitable for casting production of colored aluminum

Abstract

The invention discloses a carbon measurement data filling method suitable for colored aluminum casting production, and the method comprises the steps: obtaining the original carbon measurement data of the colored aluminum casting production through a data obtaining module, and then carrying out the preprocessing of the original carbon measurement data, including missing value detection and abnormal value detection; and abnormal value processing is carried out on the detected abnormal value. A data set is subjected to noise processing by using a segmented amplitude limiting cleaning method, so that the influence of noise on data is reduced, and the data quality is improved. After abnormal value processing and noise reduction processing are performed on a data set, missing carbon measurement data are filled by using a memory selection neural network model, and then the filled carbon measurement data are verified, so that the accuracy and rationality of the filled data are ensured. The missing data can be reasonably inferred and filled, and the completeness and accuracy of the carbon measurement data are ensured.

Description

Technical Field

[0001] The present invention belongs to the field of filling missing data in carbon metering, and more specifically, relates to a method and system for filling carbon metering data applicable to non-ferrous aluminum melting and casting production. Background Art

[0002] In the process of non-ferrous aluminum melting and casting production, accurate measurement of carbon content is a key link to ensure product quality and production efficiency. Real-time monitoring and control of carbon content are of great significance for optimizing production processes, improving product quality, and reducing production costs. However, in actual production processes, carbon metering data often has missing or abnormal data due to reasons such as sensor failures, data transmission interruptions, or environmental interferences. These missing or abnormal data not only affect the real-time monitoring of the production process but may also lead to deviations in subsequent data analysis and process optimization. Therefore, to solve the above problems, it is of great significance to provide a method and system for filling carbon metering data applicable to non-ferrous aluminum melting and casting production. Summary of the Invention

[0003] In view of the above defects or improvement requirements of the prior art, the present invention discloses a method and system for filling carbon metering data applicable to non-ferrous aluminum melting and casting production, thereby solving the problem of missing data in carbon metering data due to various reasons.

[0004] To achieve the above object, according to one aspect of the present invention, a method for filling carbon metering data applicable to non-ferrous aluminum melting and casting production is provided. The specific method is as follows:

[0005] A1. Obtain the original carbon metering data in the non-ferrous aluminum melting and casting production process, where the original carbon metering data includes carbon content measurement values at multiple time points;

[0006] A2. Preprocess the original carbon metering data, including missing value detection and outlier detection;

[0007] A3. Process the detected abnormal data for outliers and process the noise data at the same time;

[0008] A4. Based on the data after denoising and outlier processing, use a memory selection neural network model to fill the missing carbon metering data;

[0009] A5. Verify the filled carbon metering data to ensure the rationality and accuracy of the filled data.

[0010] Preferably, the preprocessing of the original carbon metering data, including missing value detection and outlier detection, is as follows:

[0011] Step 1: Identify the missing values (including null values and NaN values) in the original carbon measurement data, mark the positions of the missing values, and record the positions of all missing values for subsequent filling or elimination of the data at the missing value positions;

[0012] Step 2: Perform format standardization on the numerical values in the same attribute. For time series data, the timestamp format needs to be unified. For non-time series data, a suitable numerical unit needs to be selected to unify the numerical units of the data with the same attribute, thereby ensuring data consistency;

[0013] Step 3: Use the standard deviation method to detect outliers in the original carbon measurement data. The specific steps are as follows:

[0014] Step 3-1: Calculate the mean value for the data X = {x1, x2, …, x n} in the same attribute of the dataset. Calculate the average value of the data with the same attribute through the following formula, that is

[0015]

[0016] where n is the number of data with the same attribute in the dataset, x i is the numerical value of the data with the same attribute in the dataset, and μ is the average value of the data with the same attribute in the dataset;

[0017] Step 3-2: Calculate the standard deviation for the data X = {x1, x2,..., x n} in the same attribute of the dataset. Calculate the standard deviation of the data with the same attribute through the following formula, that is

[0018]

[0019] where n is the number of data with the same attribute in the dataset, x i is the numerical value of the data with the same attribute in the dataset, μ is the average value of the data with the same attribute in the dataset, and σ is the standard deviation of the data with the same attribute in the dataset;

[0020] Step 3-3: Calculate the boundaries of the outlier range based on the average value and standard deviation of the data with the same attribute obtained in the dataset. Calculate the upper and lower limits of the outliers of the data with the same attribute through the following formula, that is

[0021]

[0022] where μ is the average value of the data with the same attribute in the dataset, σ is the standard deviation of the data with the same attribute in the dataset, L is the lower limit of the outliers of the data with the same attribute, and U is the upper limit of the outliers of the data with the same attribute;

[0023] Step 3-4: Traverse the data with the same attribute in the dataset, and judge whether each data point exceeds the upper and lower limit ranges. If the data x i < L or x i > U, then x i is an outlier.

[0024] Preferably, perform outlier processing on the detected abnormal data, and at the same time process the noise data. The process is as follows:

[0025] Step 1: Analyze the detected outliers to judge whether the outliers are caused by data acquisition errors, equipment failures, process abnormalities or other factors. In the case where the number of outliers is small and the impact on the overall data is small, for outliers that are obviously wrong or meaningless, they can be directly removed;

[0026] Step 2: Remove the outliers that are obviously wrong or meaningless. For other abnormal data, use the linear interpolation method to correct the abnormal data. The specific steps are as follows:

[0027] Step 2-1: Sort the abnormal data to be processed and its data with the same attribute in chronological order, and at the same time mark the positions of the abnormal data to be processed;

[0028] Step 2-2: For each position y to be interpolated, find the two nearest known data points (y0, z0) and (y1, z1) before and after it, and ensure that y0 < y < y1. Use the linear interpolation formula to calculate the value z at position y. The linear interpolation calculation formula is as follows, that is

[0029]

[0030] where y is the position of the abnormal value to be processed, (y0, z0) and (y1, z1) are the two points closest to the abnormal value before and after, and y0 < y < y1, and z is the value after processing the abnormal value at position y;

[0031] Step 2-3: Replace the abnormal values with the data obtained by the linear interpolation calculation formula, and perform statistical analysis on the processed data. Verify whether the processed data is reasonable by calculating the mean and standard deviation. For unreasonable data, treat it as missing data to be filled in later;

[0032] Step 3: Use the piecewise amplitude limiting cleaning method to denoise the dataset, and regard the data outside the amplitude limiting range as noise and remove it. The specific steps are as follows:

[0033] Step 3-1: Sort the data with the same attribute in the dataset in chronological order, denoted as Q = {q1, q2, q3,..., q n}, the number of data is n, and k(k <n)个分段,计算每个分段的数据的均值和标准差,每个分段的均值和标准差计算公式如下,即

[0034]

[0035] in, is the mean of segment j (j=1,2,...,k), Q j is segment j, q i is the data value in the segment, η j is the standard deviation of segment j; if is an integer, then the number of data in each segment is Right now Otherwise, let n modulo k be m, and the segmentation is, the number of data in the first m segments is Right now The number of data in the last km segment is Right now in, Indicates the floor symbol;

[0036] Step 3-2: Calculate the limit range of each segment. The calculation formula for the limit range of each segment is as follows:

[0037]

[0038] Among them, Ran j is the limiting range of segment j, is the mean of segment j, η j is the standard deviation of segment j;

[0039] Step 3-3: The data outside the clipping range in each segment is regarded as noise data, and the noise data is removed.

[0040] Preferably, based on the data after denoising and outlier processing, a memory selection neural network model is used to fill in the missing carbon measurement data, and the process is as follows:

[0041] Step 1: Normalize the data after denoising and outlier processing, and use the Min-Max normalization method to scale the data to the range of [0,1]. The normalization formula is as follows:

[0042]

[0043] Among them, x is the actual value of the data to be normalized, x' is the normalized value, and x max 、x min is the maximum and minimum value of the data;

[0044] Step 2: Construct the input information, memory information, and output information of the memory selection network, and then construct the memory selection network framework. The specific steps are as follows:

[0045] Step 2-1: Construct the input of the memory selection network. The main function of this part is to construct the input information of the memory selection network. The calculation formula is as follows, that is

[0046] I(t) = δ(M i [x(t), H(t - 1)] + b i ) (9)

[0047] where I(t) is the constructed input information, δ is the sigmoid activation function, M i represents the input weight matrix, x(t) is the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, and b i is the input bias signal;

[0048] Step 2-2: Construct the memory information of the memory selection network. The memory information of the memory selection network is constructed in two stages. The memory information in the first stage is the input of the memory information in the second stage. The calculation formulas for the two-stage memory information of the memory selection network are as follows, that is

[0049] F(t) = δ(M f [x(t), H(t - 1)] + b f ) (10)

[0050]

[0051] where F(t) is the output of the memory information in the first stage, δ is the sigmoid activation function, M f is the memory parameter matrix, representing the matrix weight, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, and b f is the memory bias signal, R(t) is the output of the memory information in the second stage, that is, the total memory information output, R(t - 1) is the memory information output of the previous time node, I(t) is the input, is the intermediate memory information, and its calculation formula is as follows:

[0052]

[0053] where is the intermediate memory information, tanh is the hyperbolic tangent function, M c is the parameter matrix, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, and b c is the bias signal;

[0054] Step 2-3: Construct the output information of the memory selection network. The calculation formula for the output information of the memory selection network is as follows, namely

[0055]

[0056] where \(O(t)\) represents the output result of the output information, \(\delta\) is the sigmoid activation function, \(M\) o represents the weight matrix of the output gate, \(x(t)\) represents the input of the current node, \(H(t - 1)\) represents the hidden state output of the previous memory selection network node, \(b\) o is the output bias matrix, \(H(t)\) is the hidden state output of the current memory selection network node, \(\tanh\) is the hyperbolic tangent function, and \(R(t)\) represents the memory information;

[0057] Step 3: Divide the data set into a training set and a test set. The training set is used for model training, and the test set is used for model verification. Input the normalized data values into the trained memory selection neural network model to obtain output values, and then denormalize the output values and fill them into the corresponding data missing positions.

[0058] Preferably, verify the filled carbon accounting data to ensure the rationality and accuracy of the filled data. The process is as follows:

[0059] Step 1: Judge the rationality of the filled carbon accounting data, and confirm whether the filled data is within the range of \((\mu - 3\sigma,\mu + 3\sigma)\) according to the 3σ criterion (where \(\mu\) is the mean of the carbon accounting data and \(\sigma\) is the standard deviation of the carbon accounting data). Mark the data that exceeds the reasonable range;

[0060] Step 2: Draw a curve graph of the carbon accounting data before and after filling to check whether the filled data is continuous with the previous and subsequent data and whether there are mutations or unreasonable phenomena. Draw a box plot of the data before and after filling to check the distribution range and outliers of the data. For the mutation positions in the curve graph and the outliers in the box plot, judge whether they are the values to be filled. If they are the values to be filled, mark them;

[0061] Step 3: Adjust the parameters of the memory selection neural network model, and re-predict the marked values to be filled until the data to be filled is within the reasonable range, and then fill the reasonable predicted values into the corresponding data missing positions.

[0062] According to another aspect of the present invention, there is provided a carbon accounting data filling system applicable to non-ferrous aluminum casting production, including:

[0063] A data acquisition module for acquiring the original carbon accounting data in the non-ferrous aluminum casting production process;

[0064] A data preprocessing module for preprocessing the original carbon measurement data, including missing value detection and outlier data detection;

[0065] A data denoising and outlier processing module for processing outliers in the detected abnormal data and simultaneously processing noise data;

[0066] A data filling module for filling the missing carbon measurement data using a memory selection neural network model;

[0067] A data verification module for verifying the filled carbon measurement data.

[0068] Preferably, the process of the data preprocessing module preprocessing the original carbon measurement data is as follows:

[0069] Step 1: Determine the missing values (including null values and NaN values) in the original carbon measurement data, mark the positions of the missing values, and record the positions of all missing values for subsequent filling or elimination of the data at the missing value positions;

[0070] Step 2: Perform format standardization processing on the numerical values in the same attribute. For time series data, the time stamp format needs to be unified, and for non-time series data, a suitable numerical unit needs to be selected to unify the numerical units of the data with the same attribute, thereby ensuring data consistency;

[0071] Step 3: Use the standard deviation method to detect outliers in the original carbon measurement data. The specific steps are as follows:

[0072] Step 3-1: Calculate the mean value for the data X = {x1, x2,..., x n} in the same attribute in the dataset. Calculate the average value of the data with the same attribute through the following formula, that is

[0073]

[0074] where n is the number of data with the same attribute in the dataset, x i is the numerical value of the data with the same attribute in the dataset, and μ is the average value of the data with the same attribute in the dataset;

[0075] Step 3-2: Calculate the standard deviation for the data X = {x1, x2,..., x n} in the same attribute in the dataset. Calculate the standard deviation of the data with the same attribute through the following formula, that is

[0076]

[0077] where n is the number of data with the same attribute in the dataset, x iFor the values of data with the same attribute in the dataset, μ is the average value of the data with the same attribute in the dataset, and σ is the standard deviation of the data with the same attribute in the dataset;

[0078] Step 3-3: Calculate the boundaries of the outlier range based on the average value and standard deviation of the data with the same attribute in the obtained dataset. Calculate the upper and lower limits of the outliers of the data with the same attribute through the following formula, that is

[0079]

[0080] where μ is the average value of the data with the same attribute in the dataset, σ is the standard deviation of the data with the same attribute in the dataset, L is the lower limit of the outliers of the data with the same attribute, and U is the upper limit of the outliers of the data with the same attribute;

[0081] Step 3-4: Traverse the data with the same attribute in the dataset and determine whether each data point exceeds the upper and lower limit ranges. If the data x i <L or x i >U, then x i is an outlier.

[0082] Preferably, the process of the data denoising and outlier processing module for processing the detected outlier data and noise data is as follows:

[0083] Step 1: Analyze the detected outliers and determine whether the outliers are caused by data acquisition errors, equipment failures, process abnormalities or other factors. In the case where the number of outliers is small and the impact on the overall data is small, for outliers that are obviously incorrect or meaningless, they can be directly removed;

[0084] Step 2: Remove the outliers that are obviously incorrect or meaningless. For other outlier data, use the linear interpolation method to correct the outlier data. The specific steps are as follows:

[0085] Step 2-1: Sort the outlier data to be processed and its data with the same attribute in chronological order, and at the same time mark the positions of the outlier data to be processed;

[0086] Step 2-2: For each position y to be interpolated, find the nearest known data points (y0, z0) and (y1, z1) before and after it, and ensure that y0 < y < y1. Use the linear interpolation formula to calculate the value z at position y. The linear interpolation calculation formula is as follows, that is

[0087]

[0088] where y is the position of the outlier to be processed, (y0, z0) and (y1, z1) are the two points closest to the outlier before and after, and y0 < y < y1, and z is the value of the outlier after processing at position y;

[0089] Step 2-3: Replace the outliers with the data obtained by the linear interpolation formula, perform statistical analysis on the processed data, and verify whether the processed data is reasonable by calculating the mean and standard deviation. For unreasonable data, treat it as missing data to be filled later;

[0090] Step 3: Use the segmented clipping cleaning method to reduce the noise of the data set, treat the data outside the clipping range as noise and remove it. The specific steps are as follows:

[0091] Step 3-1: Sort the data of the same attribute in the data set in chronological order, denoted as Q = {q1, q2, q3, …, q n}, the number of data is n, and k(k <n)个分段,计算每个分段的数据的均值和标准差,每个分段的均值和标准差计算公式如下,即

[0092]

[0093] in, is the mean of segment j (j=1,2,...,k), Q j is segment j, q i is the data value in the segment, η j is the standard deviation of segment j; if is an integer, then the number of data in each segment is Right now Otherwise, let n modulo k be m, and the segmentation is, the number of data in the first m segments is Right now The number of data in the last km segment is Right now in, Indicates the floor symbol;

[0094] Step 3-2: Calculate the limit range of each segment. The calculation formula for the limit range of each segment is as follows:

[0095]

[0096] Among them, Ran j is the limiting range of segment j, is the mean of segment j, η j is the standard deviation of segment j;

[0097] Step 3-3: The data outside the clipping range in each segment is regarded as noise data, and the noise data is removed.

[0098] Preferably, the process of the data filling module using the memory selection neural network model to fill in the missing carbon measurement data is:

[0099] Step 1: Normalize the data based on the denoised and outlier - processed data. Use the Min - Max normalization method to scale the data to the range [0, 1]. The normalization formula is as follows, that is

[0100]

[0101] where \(x\) is the actual value of the data to be normalized, \(x'\) is the value after normalization, \(x_{max}\) max and \(x_{min}\) min are the maximum and minimum values of the data;

[0102] Step 2: Construct the input information, memory information, and output information of the memory selection network, and then construct the memory selection network framework. The specific steps are as follows:

[0103] Step 2 - 1: Construct the input of the memory selection network. The main function of this part is to construct the input information of the memory selection network. The calculation formula is as follows, that is

[0104] \(I(t)=\delta(M i [x(t),H(t - 1)]+b i )\ (9)

[0105] where \(I(t)\) is the constructed input information, \(\delta\) is the sigmoid activation function, \(M i represents the input weight matrix, \(x(t)\) is the input of the current node, \(H(t - 1)\) is the hidden - state output of the previous memory selection network node, and \(b i is the input bias signal;

[0106] Step 2 - 2: Construct the memory information of the memory selection network. The memory information of the memory selection network is constructed in two stages. The memory information in the first stage is the input of the memory information in the second stage. The calculation formulas for the two - stage memory information of the memory selection network are as follows, that is

[0107] \(F(t)=\delta(M f [x(t),H(t - 1)]+b f )\ (10)

[0108]

[0109] where \(F(t)\) is the output of the memory information in the first stage, \(\delta\) is the sigmoid activation function, \(M f is the memory parameter matrix, representing the matrix weight, \(x(t)\) represents the input of the current node, \(H(t - 1)\) is the hidden - state output of the previous memory selection network node, and \(b fis the memory bias signal, R(t) is the memory information output in the second stage, i.e., the total memory information output, R(t - 1) is the memory information output at the previous time node, and I(t) is the input. is the intermediate memory information, and its calculation formula is as follows:

[0110]

[0111] where is the intermediate memory information, tanh is the hyperbolic tangent function, M c is the parameter matrix, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, and b c is the bias signal;

[0112] Step 2 - 3: Construct the output information of the memory selection network. The calculation formula for the output information of the memory selection network is as follows, i.e.,

[0113]

[0114] where O(t) represents the output result of the output information, δ is the sigmoid activation function, M o represents the weight matrix of the output gate, x(t) represents the input of the current node, H(t - 1) represents the hidden state output of the previous memory selection network node, and b o is the output bias matrix, H(t) is the hidden state output of the current memory selection network node, tanh is the hyperbolic tangent function, and R(t) represents the memory information;

[0115] Step 3: Divide the dataset into a training set and a test set. The training set is used for model training, and the test set is used for model verification. Input the normalized data values into the trained memory selection neural network model to obtain output values, and then denormalize the output values and fill them into the corresponding data missing positions.

[0116] Preferably, the process of the data verification module verifying the filled carbon accounting data is as follows:

[0117] Step 1: Judge the rationality of the filled carbon accounting data, and confirm whether the filled data is within the range of (μ - 3σ, μ + 3σ) according to the 3σ criterion (where μ is the mean of the carbon accounting data and σ is the standard deviation of the carbon accounting data). Mark the data that exceeds the reasonable range.

[0118] Step 2: Plot the curve graph of carbon measurement data before and after filling over time, check whether the filled data is continuous with the data before and after, whether there are mutations or unreasonable phenomena, plot the box plot of the data before and after filling, check the distribution range and outliers of the data. For the mutation positions in the curve graph and the outliers in the box plot, determine whether they are values to be filled. If they are values to be filled, mark them;

[0119] Step 3: Adjust the parameters of the memory selection neural network model, re-predict the marked values to be filled until the data to be filled is within a reasonable range, and then fill the reasonable predicted values into the corresponding data missing positions.

[0120] According to another aspect of the present invention, there is provided a computer-readable storage medium, on which program instructions are stored, and when the program instructions are executed by a processor, the carbon measurement data filling method applicable to non-ferrous aluminum melting and casting production as described in any one of the above is implemented.

[0121] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

[0122] A carbon measurement data filling method and system applicable to non-ferrous aluminum melting and casting production disclosed by the present invention obtain the original carbon measurement data in the process of non-ferrous aluminum melting and casting production by using a carbon measurement data acquisition device, and then preprocess the original carbon measurement data, including missing value detection and outlier detection, and at the same time process the outliers and noise data. The processed data is used to fill the missing carbon measurement data by using a memory selection neural network model, and the filled carbon measurement data is verified to ensure the rationality and accuracy of the filled data. This data filling method and system use a memory selection neural network model to fill the missing carbon measurement data, which can effectively solve the deficiencies of traditional methods in processing non-linear data and time series data, and at the same time have the advantages of strong adaptability and real-time performance. The present invention can accurately and reasonably fill the carbon measurement data of non-ferrous aluminum melting and casting production, which is of great significance for the popularization and wide application of the carbon measurement data filling method applicable to non-ferrous aluminum melting and casting production. BRIEF DESCRIPTION OF THE DRAWINGS

[0123] Figure 1 is a device connection diagram provided by an embodiment of the present invention;

[0124] Figure 2 is a schematic flow chart of a carbon measurement data filling method applicable to non-ferrous aluminum melting and casting production provided by an embodiment of the present invention;

[0125] Figure 3 is a memory selection neural network architecture diagram of a carbon measurement data filling method applicable to non-ferrous aluminum melting and casting production provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0126] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0127] Example 1: As Figure 1 shown, a carbon metering data filling system applicable to non-ferrous aluminum melting and casting production provided by an embodiment of the present invention includes:

[0128] A data acquisition module, configured to acquire the original carbon metering data during the non-ferrous aluminum melting and casting production process;

[0129] A data preprocessing module, configured to preprocess the original carbon metering data, including missing value detection and outlier data detection;

[0130] A data denoising and outlier processing module, configured to perform outlier processing on the detected abnormal data and at the same time process the noise data;

[0131] A data filling module, configured to fill the missing carbon metering data by using a memory selection neural network model;

[0132] A data verification module, configured to verify the filled carbon metering data.

[0133] Example 2: As Figure 2 shown, a flowchart of a carbon metering data filling method applicable to non-ferrous aluminum melting and casting production provided by an embodiment of the present invention includes the following steps:

[0134] Step A1: Acquire the original carbon metering data during the non-ferrous aluminum melting and casting production process, where the original carbon metering data includes carbon content measurement values at multiple time points;

[0135] Step A2: Preprocess the original carbon metering data, including missing value detection and outlier detection;

[0136] Step A3: Perform outlier processing on the detected abnormal data and at the same time process the noise data;

[0137] Step A4: Based on the data after denoising and outlier processing, use a memory selection neural network model to fill the missing carbon metering data;

[0138] Step A5: Verify the filled carbon metering data to ensure the rationality and accuracy of the filled data;

[0139] Preferably, preprocess the original carbon measurement data, including missing value detection and outlier detection. The process is as follows:

[0140] Step 1: Determine the missing values (including null values and NaN values) in the original carbon measurement data, mark the positions of the missing values, and record the positions of all missing values for subsequent filling or elimination of the data at the missing value positions;

[0141] Step 2: Perform format standardization on the numerical values in the same attribute. For time series data, the timestamp format needs to be unified. For non-time series data, a suitable numerical unit needs to be selected to unify the numerical units of the data with the same attribute, thereby ensuring data consistency;

[0142] Step 3: Use the standard deviation method to detect outliers in the original carbon measurement data. The specific steps are as follows:

[0143] Step 3-1: Calculate the mean value for the data X = {x1, x2, …, x n} in the same attribute of the dataset. Calculate the average value of the data with the same attribute through the following formula, that is

[0144]

[0145] where n is the number of data with the same attribute in the dataset, x i is the numerical value of the data with the same attribute in the dataset, and μ is the average value of the data with the same attribute in the dataset;

[0146] Step 3-2: Calculate the standard deviation for the data X = {x1, x2,..., x n} in the same attribute of the dataset. Calculate the standard deviation of the data with the same attribute through the following formula, that is

[0147]

[0148] where n is the number of data with the same attribute in the dataset, x i is the numerical value of the data with the same attribute in the dataset, μ is the average value of the data with the same attribute in the dataset, and σ is the standard deviation of the data with the same attribute in the dataset;

[0149] Step 3-3: Calculate the boundaries of the outlier range based on the average value and standard deviation of the data with the same attribute obtained in the dataset. Calculate the upper and lower limits of the outliers of the data with the same attribute through the following formula, that is

[0150]

[0151] Among them, μ is the average value of the data of the same attribute in the dataset, σ is the standard deviation of the data of the same attribute in the dataset, L is the lower limit of the outlier of the data of the same attribute, and U is the upper limit of the outlier of the data of the same attribute;

[0152] Step 3-4: Traverse the data of the same attribute in the dataset, and judge whether each data point exceeds the upper and lower limit ranges. If the data x i < L or x i > U, then x i is an outlier.

[0153] Preferably, perform outlier processing on the detected abnormal data and at the same time process the noise data. The process is as follows:

[0154] Step 1: Analyze the detected outliers, and judge whether the outliers are caused by data acquisition errors, equipment failures, process anomalies or other factors. In the case where the number of outliers is small and has little impact on the overall data, for outliers that are obviously wrong or meaningless, they can be directly removed;

[0155] Step 2: Remove the outliers that are obviously wrong or meaningless. For other abnormal data, use the linear interpolation method to correct the abnormal data. The specific steps are as follows:

[0156] Step 2-1: Sort the abnormal data to be processed and its data of the same attribute in chronological order, and at the same time mark the positions of the abnormal data to be processed;

[0157] Step 2-2: For each position y to be interpolated, find the two nearest known data points (y0, z0) and (y1, z1) before and after it, and ensure that y0 < y < y1. Use the linear interpolation formula to calculate the value z at position y. The linear interpolation calculation formula is as follows, that is

[0158]

[0159] Among them, y is the position of the abnormal value to be processed, (y0, z0) and (y1, z1) are the two points nearest to the abnormal value before and after, and y0 < y < y1, and z is the value of the abnormal value at position y after processing;

[0160] Step 2-3: Replace the abnormal value with the data obtained by the linear interpolation calculation formula, and perform statistical analysis on the processed data. Verify whether the processed data is reasonable by calculating the mean value and standard deviation. For unreasonable data, treat it as missing data to be filled later;

[0161] Step 3: Use the segmented amplitude-limiting cleaning method to perform noise reduction processing on the dataset, and regard the data outside the amplitude-limiting range as noise and remove it. The specific steps are as follows:

[0162] Step 3-1: Sort the data of the same attribute in the data set in chronological order, denoted as Q = {q1, q2, q3, ..., q n}, the number of data is n, and k(k <n)个分段,计算每个分段的数据的均值和标准差,每个分段的均值和标准差计算公式如下,即

[0163]

[0164]

[0165] in, is the mean of segment j (j = 1, 2, ..., k), Qj is the mean of segment j, q i is the data value in the segment, η j is the standard deviation of segment j; if is an integer, then the number of data in each segment is Right now Otherwise, let n modulo k be m, and the segmentation is, the number of data in the first m segments is Right now The number of data in the last km segment is Right now in, Indicates the floor symbol;

[0166] Step 3-2: Calculate the limit range of each segment. The calculation formula for the limit range of each segment is as follows:

[0167]

[0168] Among them, Ran j is the limiting range of segment j, is the mean of segment j, η j is the standard deviation of segment j;

[0169] Step 3-3: The data outside the clipping range in each segment is regarded as noise data, and the noise data is removed.

[0170] Preferably, based on the data after denoising and outlier processing, a memory selection neural network model is used to fill in the missing carbon measurement data, and the process is as follows:

[0171] Step 1: Normalize the data after denoising and outlier processing, and use the Min-Max normalization method to scale the data to the range of [0,1]. The normalization formula is as follows:

[0172]

[0173] Among them, x is the actual value of the data to be normalized, x' is the value after normalization, and x max and x min are the maximum and minimum values of the data;

[0174] Step 2: Construct the input information, memory information, and output information of the memory selection network, and then construct the memory selection network framework. The specific steps are as follows:

[0175] Step 2-1: Construct the input of the memory selection network. The main function of this part is to construct the input information of the memory selection network. The calculation formula is as follows, that is

[0176] I(t) = δ(M i [x(t), H(t - 1)] + b i ) (9)

[0177] Among them, I(t) is the constructed input information, δ is the sigmoid activation function, M i represents the input weight matrix, x(t) is the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, and b i is the input bias signal;

[0178] Step 2-2: Construct the memory information of the memory selection network. The memory information of the memory selection network is constructed in two stages. The memory information in the first stage is the input of the memory information in the second stage. The calculation formulas for the two-stage memory information of the memory selection network are as follows, that is

[0179] F(t) = δ(M f [x(t), H(t - 1)] + b f ) (10)

[0180]

[0181] Among them, F(t) is the output of the memory information in the first stage, δ is the sigmoid activation function, M f is the memory parameter matrix, representing the matrix weight, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, and b f is the memory bias signal, R(t) is the output of the memory information in the second stage, that is, the total memory information output, R(t - 1) is the memory information output of the previous time node, I(t) is the input, is the intermediate memory information, and its calculation formula is as follows:

[0182]

[0183] Among them, is the intermediate memory information, tanh is the hyperbolic tangent function, M c is the parameter matrix, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, b c is the bias signal;

[0184] Step 2 - 3: Construct the output information of the memory selection network. The calculation formula for the output information of the memory selection network is as follows, that is

[0185]

[0186] where, O(t) represents the output result of the output information, δ is the sigmoid activation function, M o represents the weight matrix of the output gate, x(t) represents the input of the current node, H(t - 1) represents the hidden state output of the previous memory selection network node, b o is the output bias matrix, H(t) is the hidden state output of the current memory selection network node, tanh is the hyperbolic tangent function, and R(t) represents the memory information;

[0187] Step 3: Divide the data set into a training set and a test set. The training set is used for model training, and the test set is used for model verification. Input the normalized data values into the trained memory selection neural network model to obtain output values, and then denormalize the output values and fill them into the corresponding data missing positions.

[0188] Preferably, verify the filled carbon accounting data to ensure the rationality and accuracy of the filled data. The process is as follows:

[0189] Step 1: Judge the rationality of the filled carbon accounting data. According to the 3σ criterion, confirm whether the filled data is within the range of (μ - 3σ, μ + 3σ) (where μ is the mean of the carbon accounting data and σ is the standard deviation of the carbon accounting data). Mark the data that exceeds the reasonable range;

[0190] Step 2: Draw a curve graph of the carbon accounting data before and after filling to check whether the filled data is continuous with the previous and subsequent data and whether there are mutations or unreasonable phenomena. Draw a box plot of the data before and after filling to check the distribution range and outliers of the data. For the mutation positions in the curve graph and the outliers in the box plot, judge whether they are the values to be filled. If they are the values to be filled, mark them;

[0191] Step 3: Adjust the parameters of the memory selection neural network model, re - predict the marked values to be filled until the data to be filled is within the reasonable range, and then fill the reasonable predicted values into the corresponding data missing positions.

[0192] Example 3: As Figure 3As shown in the figure, the memory selection neural network architecture diagram of a carbon metering data filling method applicable to colored aluminum melting and casting production provided by an embodiment of the present invention is constructed as follows:

[0193] Step 1: Construct the input of the memory selection network. The main function of this part is to construct the input information of the memory selection network. The calculation formula is as follows, that is

[0194] I(t) = δ(M i [x(t), H(t - 1)] + b i )

[0195] where I(t) is the constructed input information, δ is the sigmoid activation function, M i represents the input weight matrix, x(t) is the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, and b i is the input bias signal;

[0196] Step 2: Construct the memory information of the memory selection network. The memory information of the memory selection network is constructed in two stages. The memory information in the first stage is the input of the memory information in the second stage. The calculation formulas for the two-stage memory information of the memory selection network are as follows, that is

[0197] F(t) = δ(M f [x(t), H(t - 1)] + b f )

[0198]

[0199] where F(t) is the output of the memory information in the first stage, δ is the sigmoid activation function, M f is the memory parameter matrix, representing the matrix weight, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, and b f is the memory bias signal, R(t) is the output of the memory information in the second stage, that is, the total memory information output, R(t - 1) is the memory information output of the previous time node, I(t) is the input, is the intermediate memory information, and its calculation formula is as follows:

[0200]

[0201] where is the intermediate memory information, tanh is the hyperbolic tangent function, M c is the parameter matrix, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, and b c is the bias signal;

[0202] Step 3: Construct the output information of the memory selection network. The calculation formula for the output information of the memory selection network is as follows, namely

[0203] O(t) = δ(M o [x(t), H(t - 1)] + b o )

[0204] H(t) = O(t)·tanh(R(t))

[0205] where, O(t) represents the output result of the output information, δ is the sigmoid activation function, M o represents the weight matrix of the output gate, x(t) represents the input of the current node, H(t - 1) represents the hidden state output of the previous memory selection network node, b o is the output bias matrix, H(t) is the hidden state output of the current memory selection network node, tanh is the hyperbolic tangent function, and R(t) represents the memory information.

[0206] In another embodiment of the invention, a computer-readable storage medium is further provided, on which program instructions are stored. When the program instructions are executed by a processor, a carbon metering data filling method and system applicable to colored aluminum melting and casting production as described above are implemented.

[0207] It should be noted that according to the needs of implementation, each step / component described in this application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0208] The method according to the present invention described above can be implemented in hardware, firmware, or be implemented as software or computer code that can be stored in a recording medium (such as a CD ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or be implemented as computer code originally stored in a remote recording medium or a non-transitory machine-readable medium and downloaded through a network and to be stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component (such as a RAM, ROM, flash memory, etc.) that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, the processor, or the hardware, the processing method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the processing shown herein, the execution of the code converts the general-purpose computer into a dedicated computer for executing the processing shown herein.

[0209] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A carbon measurement data filling method applicable to the production of colored aluminum melting and casting, characterized in that, The method includes the following steps: A1. Obtain the original carbon measurement data in the production process of colored aluminum casting. The original carbon measurement data includes carbon content measurement values at multiple time points; A2. Preprocess the original carbon measurement data, including missing value detection and outlier detection; A3. Process the detected outlier data and process the noise data at the same time; A4. Based on the data after denoising and outlier processing, use a memory selection neural network model to fill in the missing carbon measurement data; A5. Verify the filled carbon measurement data to ensure the rationality and accuracy of the filled data.

2. The carbon measurement data filling method applicable to the production of colored aluminum melting and casting according to claim 1, characterized in that Preprocess the original carbon measurement data, including missing value detection and outlier detection. The process is as follows: Step 1: Determine the missing values (including null values and NaN values) in the original carbon measurement data, mark the positions of the missing values, and record the positions of all missing values; Step 2: Standardize the format of the numerical values in the same attribute. For time series data, the time stamp format needs to be unified. For non-time series data, a suitable numerical unit needs to be selected to unify the numerical units of the same attribute data, so as to ensure the consistency of the data; Step 3: Use the standard deviation method to detect outliers in the original carbon measurement data. The specific steps are as follows: Step 3-1: For the data X = {x1, x2,..., x n} of the same attribute in the dataset, calculate the mean value. Calculate the average value of the data of the same attribute through the following formula, that is where n is the number of data with the same attribute in the dataset, x i is the value of the data with the same attribute in the dataset, and μ is the average value of the data with the same attribute in the dataset; Step 3-2: For the data X = {x1, x2,..., x n} of the same attribute in the dataset, calculate the standard deviation. Calculate the standard deviation of the data of the same attribute through the following formula, that is where n is the number of data with the same attribute in the dataset, x i is the value of the data with the same attribute in the dataset, μ is the average value of the data with the same attribute in the dataset, and σ is the standard deviation of the data with the same attribute in the dataset; Step 3-3: Calculate the boundaries of the outlier range based on the average value and standard deviation of the data in the same attribute in the obtained data set. Calculate the upper and lower limits of the outliers of the data in the same attribute through the following formula, that is where L is the lower limit of the outliers of the data in the same attribute, and U is the upper limit of the outliers of the data in the same attribute; Step 3-4: Traverse the data with the same attribute in the dataset, and determine whether each data point exceeds the upper and lower limit ranges. If the data x i < L or x i > U, then x i is an outlier.

3. The carbon metering data filling method applicable to the production of colored aluminum melting and casting according to claim 1, characterized in that, Process the detected outlier data and process the noise data at the same time. The process is as follows: Step 1: Analyze the detected outliers to determine whether the outliers are caused by data acquisition errors, equipment failures, process anomalies or other factors. When the number of outliers is small and has little impact on the overall data, directly remove the outliers that are obviously wrong or meaningless; Step 2: Exclude the outliers that are obviously wrong or meaningless. For other outlier data, use the linear interpolation method to correct the outlier data. The specific steps are as follows: Step 2-1: Sort the outlier data to be processed and its data in the same attribute in chronological order, and mark the positions of the outlier data to be processed at the same time; Step 2-2: For each position y to be interpolated, find the nearest known data points (y0, z0) and (y1, z1) before and after it, and ensure that y0 < y < y1. Use the linear interpolation formula to calculate the value z at position y. The linear interpolation calculation formula is as follows, that is where y is the position of the outlier to be processed, (y0, z0) and (y1, z1) are the two points closest to the outlier before and after, and y0 < y < y1, and z is the value of the outlier after processing at position y; Step 2-3: Replace the outlier with the data obtained through the linear interpolation calculation formula, and perform statistical analysis on the processed data. Verify whether the processed data is reasonable by calculating the mean value and standard deviation. For unreasonable data, treat it as missing data to be filled later; Step 3: Use the segmented clipping cleaning method to reduce the noise of the data set, treat the data outside the clipping range as noise and remove it. The specific steps are as follows: Step 3-1: Sort the data with the same attribute in the dataset in chronological order, denoted as Q = {q1, q2, q3,..., q n}, the number of data is n, randomly select k (k < n) segments, and calculate the mean and standard deviation of the data in each segment. The calculation formulas for the mean and standard deviation of each segment are as follows, that is Among them, is the mean of segment j (j = 1, 2,..., k), Q j is segment j, q i is the data value in the segment, η j is the standard deviation of segment j; if is an integer, then the number of data in each segment is That is Otherwise, let the remainder of n divided by k be m. The segmentation situation is that the number of data in the first m segments is That is The number of data in the subsequent k - m segments is That is Among them, represents the floor symbol; Step 3-2: Calculate the limit range of each segment. The calculation formula for the limit range of each segment is as follows: Among them, Ran j is the clipping range of segment j, is the mean of segment j, and η j is the standard deviation of segment j; Step 3-3: The data outside the clipping range in each segment is regarded as noise data, and the noise data is removed.

4. The carbon measurement data filling method applicable to the production of colored aluminum melting and casting according to claim 1, characterized in that, Based on the data after denoising and outlier processing, the memory selection neural network model is used to fill in the missing carbon measurement data. The process is as follows: Step 1: Normalize the data after denoising and outlier processing, and use the Min-Max normalization method to scale the data to the range of [0,1]. The normalization formula is as follows: Among them, x is the actual value of the data to be normalized, x' is the value after normalization, x max and x min are the maximum and minimum values of the data; Step 2: Construct the input information, memory information and output information of the memory selection network, and then construct the memory selection network framework. The specific steps are as follows: Step 2-1: Construct the input of the memory selection network. The main function of this part is to construct the input information of the memory selection network. The calculation formula is as follows: I(t) = δ(M i [x(t), H(t - 1)] + b i ) (9) Among them, I(t) is the input information after construction, δ is the sigmoid activation function, M i represents the input weight matrix, x(t) is the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, b i is the input bias signal; Step 2-2: Construct the memory information of the memory selection network. The memory information of the memory selection network is constructed in two stages. The memory information of the first stage is the input of the memory information of the second stage. The calculation formula of the two-stage memory information of the memory selection network is as follows, that is, F(t) = δ(M f [x(t), H(t - 1)] + b f ) (10) Among them, F(t) is the memory information output in the first stage, δ is the sigmoid activation function, M f is the memory parameter matrix, representing the matrix weight, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, b f is the memory bias signal, R(t) is the memory information output in the second stage, that is, the total memory information output, R(t - 1) is the memory information output of the previous time node, I(t) is the input, is the intermediate memory information, and its calculation formula is as follows: Among them, is the intermediate memory information, tanh is the hyperbolic tangent function, M c is the parameter matrix, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, b c is the bias signal; Step 2-3: Construct the output information of the memory selection network. The calculation formula of the output information of the memory selection network is as follows: Among them, O(t) represents the output result of the output information, δ is the sigmoid activation function, M o represents the weight matrix of the output gate, x(t) represents the input of the current node, H(t - 1) represents the hidden state output of the previous memory selection network node, b o is the output bias matrix, H(t) is the hidden state output of the current memory selection network node, tanh is the hyperbolic tangent function, and R(t) represents the memory information; Step 3: Divide the data set into a training set and a test set. The training set is used for model training, and the test set is used for model verification. Input the normalized data values ​​into the trained memory selection neural network model to obtain the output values. Denormalize the output values ​​and fill them into the corresponding data missing positions.

5. The carbon metering data filling method applicable to the production of colored aluminum melting and casting according to claim 1, characterized in that Verify the filled carbon measurement data to ensure the rationality and accuracy of the filled data. The process is as follows: Step 1: Make a reasonable judgment on the filled carbon measurement data, and confirm whether the filled data is within the range of (μ-3σ,μ+3σ) according to the 3σ criterion (where μ is the mean of the carbon measurement data and σ is the standard deviation of the carbon measurement data), and mark the data that exceeds the reasonable range; Step 2: Draw a graph of the carbon measurement data before and after filling over time, check whether the filled data is continuous with the previous and next data, whether there is a sudden change or unreasonable phenomenon, draw a box plot of the data before and after filling, check the distribution range and abnormal values ​​of the data, and determine whether the sudden change position of the graph and the abnormal value in the box plot are the values ​​to be filled. If so, mark them; Step 3: Adjust the parameters of the memory selection neural network model, re-predict the marked values ​​to be filled until the data to be filled is within a reasonable range, and then fill the corresponding data missing positions with reasonable predicted values.

6. A carbon metering data filling system applicable to the production of colored aluminum melting and casting, characterized in that, Used to implement the carbon metering data filling method applicable to non-ferrous aluminum casting production as described in any one of claims 1-5, The filling system comprises: Data acquisition module, used to obtain the original carbon measurement data in the non-ferrous aluminum casting production process; A data preprocessing module for preprocessing the original carbon measurement data, including missing value detection and outlier data detection; A data denoising and outlier processing module for processing the detected outlier data and denoising the noise data simultaneously; A data filling module for filling the missing carbon measurement data using a memory selection neural network model; A data verification module for verifying the filled carbon measurement data.

7. The carbon metering data filling system applicable to the production of colored aluminum melting and casting according to claim 6, characterized in that The process of the data preprocessing module preprocessing the original carbon measurement data is as follows: Step 1: Determine the missing values (including null values and NaN values) in the original carbon measurement data, mark the positions of the missing values, and record the positions of all missing values for subsequent filling or elimination of the data at the missing value positions; Step 2: Perform format standardization on the numerical values in the same attribute. For time series data, the time stamp format needs to be unified. For non-time series data, a suitable numerical unit needs to be selected to unify the numerical units of the data with the same attribute, thereby ensuring data consistency; Step 3: Use the standard deviation method to detect outliers in the original carbon measurement data. The specific steps are as follows: Step 3-1: For the data X = {x1, x2,..., x n} of the same attribute in the dataset, calculate the mean value. Calculate the average value of the data with the same attribute through the following formula, that is where n is the number of data with the same attribute in the dataset, x i is the value of the data with the same attribute in the dataset, and μ is the average value of the data with the same attribute in the dataset; Step 3-2: For the data X = {x1, x2,..., x n} of the same attribute in the dataset, calculate the standard deviation. Calculate the standard deviation of the data of the same attribute through the following formula, that is where n is the number of data with the same attribute in the dataset, x i is the value of the data with the same attribute in the dataset, μ is the average value of the data with the same attribute in the dataset, and σ is the standard deviation of the data with the same attribute in the dataset; Step 3-3: Calculate the boundaries of the outlier range based on the mean and standard deviation of the data with the same attribute in the obtained dataset. Calculate the upper and lower limits of the outliers of the data with the same attribute through the following formula, that is where μ is the mean of the data with the same attribute in the dataset, σ is the standard deviation of the data with the same attribute in the dataset, L is the lower limit of the outliers of the data with the same attribute, and U is the upper limit of the outliers of the data with the same attribute; Step 3-4: Traverse the data of the same attribute in the dataset, and judge whether each data point exceeds the upper and lower limit ranges. If the data x i < L or x i > U, then x i is an outlier.

8. The carbon metering data filling system applicable to the production of colored aluminum melting and casting according to claim 6, characterized in that, The process of the data denoising and outlier processing module processing the detected outlier data and noise data is as follows: Step 1: Analyze the detected outliers to determine whether the outliers are caused by data acquisition errors, equipment failures, process anomalies, or other factors. In the case where the number of outliers is small and has little impact on the overall data, obvious or meaningless outliers can be directly eliminated; Step 2: Excluding the obvious or meaningless outliers, use the linear interpolation method to correct the other outlier data. The specific steps are as follows: Step 2-1: Sort the outlier data to be processed and its data with the same attribute in chronological order, and mark the positions of the outlier data to be processed; Step 2-2: For each position y to be interpolated, find the nearest known data points (y0, z0) and (y1, z1) before and after it, and ensure that y0 < y < y1. Use the linear interpolation formula to calculate the value z at position y. The linear interpolation calculation formula is as follows, that is where y is the position of the outlier to be processed, (y0, z0) and (y1, z1) are the two points closest to the outlier before and after, and y0 < y < y1, and z is the value of the outlier after processing at position y; Step 2-3: Replace the outlier with the data obtained through the linear interpolation calculation formula, and perform statistical analysis on the processed data. Verify whether the processed data is reasonable by calculating the mean and standard deviation. For unreasonable data, treat it as missing data to be filled later; Step 3: Use the segmented clipping cleaning method to reduce the noise of the data set, treat the data outside the clipping range as noise and remove it. The specific steps are as follows: Step 3-1: Sort the data of the same attribute in the dataset in chronological order, denoted as Q = {q1, q2, q3,..., q n}, the number of data is n, randomly select k (k < n) segments, calculate the mean and standard deviation of the data in each segment, and the calculation formulas for the mean and standard deviation of each segment are as follows, that is Among them, is the mean of segment j (j = 1, 2,..., k), Q j is segment j, q i is the data value in the segment, η j is the standard deviation of segment j; if is an integer, then the number of data in each segment is That is Otherwise, let the remainder of n divided by k be m. The segmentation situation is that the number of data in the first m segments is That is The number of data in the subsequent k - m segments is That is Among them, represents the floor symbol; Step 3-2: Calculate the limit range of each segment. The calculation formula for the limit range of each segment is as follows: Among them, Ran j is the clipping range of segment j, is the mean value of segment j, and η j is the standard deviation of segment j; Step 3-3: The data outside the clipping range in each segment is regarded as noise data, and the noise data is removed.

9. The carbon metering data filling system applicable to the production of colored aluminum melting and casting according to claim 6, characterized in that, The process of filling missing carbon measurement data using the memory selection neural network model in the data filling module is as follows: Step 1: Normalize the data after denoising and outlier processing, and use the Min-Max normalization method to scale the data to the range of [0,1]. The normalization formula is as follows: Among them, x is the actual value of the data to be normalized, x' is the value after normalization, x max and x min are the maximum and minimum values of the data; Step 2: Construct the input information, memory information and output information of the memory selection network, and then construct the memory selection network framework. The specific steps are as follows: Step 2-1: Construct the input of the memory selection network. The main function of this part is to construct the input information of the memory selection network. The calculation formula is as follows: I(t) = δ(M i [x(t), H(t - 1)] + b i ) (9) Among them, I(t) is the input information after construction, δ is the sigmoid activation function, M i represents the input weight matrix, x(t) is the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, b i is the input bias signal; Step 2-2: Construct the memory information of the memory selection network. The memory information of the memory selection network is constructed in two stages. The memory information of the first stage is the input of the memory information of the second stage. The calculation formula of the two-stage memory information of the memory selection network is as follows, that is, F(t) = δ(M f [x(t), H(t - 1)] + b f ) (10) Among them, F(t) is the memory information output in the first stage, δ is the sigmoid activation function, M f is the memory parameter matrix, representing the matrix weight, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, b f is the memory bias signal, R(t) is the memory information output in the second stage, that is, the total memory information output, R(t - 1) is the memory information output of the previous time node, I(t) is the input, is the intermediate memory information, and its calculation formula is as follows: Among them, is the intermediate memory information, tanh is the hyperbolic tangent function, M c is the parameter matrix, x(t) represents the input of the current node, H(t - 1) is the hidden state output of the previous memory selection network node, b c is the bias signal; Step 2-3: Construct the output information of the memory selection network. The calculation formula of the output information of the memory selection network is as follows: Among them, O(t) represents the output result of the output information, δ is the sigmoid activation function, M o represents the weight matrix of the output gate, x(t) represents the input of the current node, H(t - 1) represents the hidden state output of the previous memory selection network node, b o is the output bias matrix, H(t) is the hidden state output of the current memory selection network node, tanh is the hyperbolic tangent function, and R(t) represents the memory information; Step 3: Divide the data set into a training set and a test set. The training set is used for model training, and the test set is used for model verification. Input the normalized data values ​​into the trained memory selection neural network model to obtain the output values. Denormalize the output values ​​and fill them into the corresponding data missing positions. The process of the data verification module verifying the filled carbon measurement data is as follows: Step 1: Make a reasonable judgment on the filled carbon measurement data, and confirm whether the filled data is within the range of (μ-3σ,μ+3σ) according to the 3σ criterion (where μ is the mean of the carbon measurement data and σ is the standard deviation of the carbon measurement data), and mark the data that exceeds the reasonable range; Step 2: Draw a graph of the carbon measurement data before and after filling over time, check whether the filled data is continuous with the previous and next data, whether there is a sudden change or unreasonable phenomenon, draw a box plot of the data before and after filling, check the distribution range and abnormal values ​​of the data, and determine whether the sudden change position of the graph and the abnormal value in the box plot are the values ​​to be filled. If so, mark them; Step 3: Adjust the parameters of the memory selection neural network model, re-predict the marked values ​​to be filled until the data to be filled is within a reasonable range, and then fill the corresponding data missing positions with reasonable predicted values.

10. A computer-readable storage medium having program instructions stored thereon, wherein the program instructions, when executed by a processor, implement the carbon metering data filling method applicable to non-ferrous aluminum smelting and casting production as described in any one of claims 1 to 5.

Images