Production process index prediction method and system based on industrial big data
By combining the LSTM model to optimize the GM model and utilizing the historical data volume and time series characteristics of the target indicator data, the development coefficient and gray action of the GM model are optimized, which solves the problems of low prediction accuracy and reliability in traditional methods and achieves more accurate prediction of production process indicators.
Patent Information
- Application Number
- CN202510794954.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-14
- Publication Date
- 2025-09-16
AI Technical Summary
Traditional production process indicator prediction methods have difficulty in fully exploring the potential patterns and characteristics of data when dealing with complex and nonlinear industrial production data, resulting in low accuracy and reliability of prediction results.
The GM model optimized based on the LSTM model is adopted. By combining the historical data volume and time series characteristics of the target indicator data, the development coefficient and gray action of the GM model are optimized, and the hidden state of the LSTM model is used for weighted calculation to improve the prediction accuracy.
It improves the accuracy and reliability of production process indicator prediction, can better adapt to the complex and changeable industrial production data environment, makes up for the shortcomings of the traditional GM model, and enhances the ability to capture data change trends.
Smart Images
Figure CN120654890A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of industrial data prediction technology, and more specifically, relates to a production process indicator prediction method and system based on industrial big data. Background Art
[0002] In the field of industrial production, accurate prediction of production process indicators is of vital importance for optimizing production processes, improving production efficiency, reducing production costs and ensuring product quality.
[0003] Traditional production process indicator forecasting methods are often based on simple statistical models or empirical formulas, resulting in significant limitations when processing complex, nonlinear industrial production data. For example, existing technologies struggle to fully exploit the underlying patterns and characteristics within the data, and inaccurately capture changing trends. This results in low accuracy and reliability in forecasts, making them unable to meet the high-precision forecasting requirements of modern industrial production.
[0004] Therefore, an accurate and reliable method for predicting production process indicators is needed. Summary of the Invention
[0005] The purpose of this application is to provide a production process indicator prediction method and system based on industrial big data to improve the accuracy of production process indicator prediction.
[0006] A first aspect of an embodiment of the present application provides a production process indicator prediction method based on industrial big data, comprising: Determining a prediction model based on the data volume of historical data of target indicator data and the time series characteristics of the historical data of target indicator data; the target indicator data is the generation process indicator data to be predicted; In response to the prediction model being a GM model optimized based on the LSTM model, performing a first operation; The first operation includes: Input the historical data of the target indicator data into the LSTM model to obtain the hidden state corresponding to the target indicator data and the first prediction result; Based on the hidden state corresponding to the target indicator data, the development coefficient and gray action of the GM model are optimized to obtain the target GM model; Input the historical data of the target indicator data into the target GM model to obtain the second prediction result; The first prediction result and the second prediction result are weighted and calculated to obtain the prediction result of the target indicator data.
[0007] A second aspect of the embodiments of the present application provides a production process indicator prediction system based on industrial big data, comprising: A model determination module is used to determine a prediction model based on the data volume of historical data of target indicator data and the time series characteristics of the historical data of the target indicator data; the target indicator data is the generation process indicator data to be predicted; an indicator prediction module, configured to perform a first operation in response to the prediction model being a GM model optimized based on the LSTM model; The first operation includes: Input the historical data of the target indicator data into the LSTM model to obtain the hidden state corresponding to the target indicator data and the first prediction result; Based on the hidden state corresponding to the target indicator data, the development coefficient and gray action of the GM model are optimized to obtain the target GM model; Input the historical data of the target indicator data into the target GM model to obtain the second prediction result; The first prediction result and the second prediction result are weighted and calculated to obtain the prediction result of the target indicator data.
[0008] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the above-mentioned production process indicator prediction method based on industrial big data are implemented.
[0009] In a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned production process indicator prediction method based on industrial big data are implemented.
[0010] The beneficial effects of the production process indicator prediction method and system based on industrial big data provided by the embodiments of the present application are: This application is based on industrial big data, and determines the prediction model by comprehensively considering the data volume and time series characteristics of the historical data of the target indicator data. It can better adapt to the complex and changeable industrial production data environment and provide a basis for accurate prediction. It solves the problem that the traditional method is based on simple statistical models or empirical formulas, has limitations in processing complex and nonlinear industrial production data, is difficult to fully explore the potential laws and characteristics of the data, and cannot accurately capture the trend of data changes, resulting in low accuracy and reliability of the prediction results. This application optimizes the development coefficient and gray action of the GM model based on the hidden state obtained by the LSTM model to obtain the target GM model, integrates the advantages of the LSTM model into the GM model, makes up for the shortcomings of the traditional GM model, and enables the target GM model to more accurately reflect the trend of data changes. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0012] Figure 1 A flowchart of a production process indicator prediction method based on industrial big data provided in one embodiment of the present application; Figure 2 This is a structural block diagram of a production process indicator prediction system based on industrial big data provided by one embodiment of the present application; Figure 3 A schematic block diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0013] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.
[0014] In order to make the purpose, technical solutions and advantages of this application clearer, specific embodiments will be described below with reference to the accompanying drawings.
[0015] Please refer to Figure 1 , Figure 1 A flowchart of a production process indicator prediction method based on industrial big data provided in an embodiment of the present application is provided. The method is executed by an electronic device and includes: S101-S102.
[0016] S101: Determine a prediction model based on the data volume of historical data of target indicator data and the time series characteristics of the historical data of the target indicator data; the target indicator data is generation process indicator data to be predicted.
[0017] In this embodiment, the target indicator data may include various parameters of equipment on a production line, such as energy consumption, raw material consumption, or output. The production line may be a beer or dairy product line. The historical data of the target indicator data refers to the historically collected data corresponding to that data type. The data volume of the historical data of the target indicator data refers to the quantity and size of the historical data of the target indicator data. The data characteristics of the historical data of the target indicator data may include time series correlation, stationarity, and other characteristics.
[0018] In this embodiment, the prediction model can be an LSTM model, a GM model, a GM model optimized based on the LSTM model, or an ARIMA model. This application takes into account that different models have different applicable scenarios and characteristics. Therefore, this application considers determining the prediction model based on two dimensions. Specifically, the time series characteristics of the historical data include: the decay rate of the autocorrelation function of the historical data; the prediction model is determined based on the data volume of the historical data of the target indicator data and the time series characteristics of the historical data of the target indicator data, including: In response to the data amount of the historical data of the target indicator data being less than the first data amount, determining the GM model as the prediction model; In response to the data volume of the historical data of the target indicator data being greater than the first data volume, and the time series characteristics of the historical data of the target indicator data satisfying a strong correlation condition, determining the LSTM model as the prediction model; wherein the strong correlation condition is that the decay rate of the autocorrelation function of the historical data of the target indicator data is less than a preset decay rate; In response to the data volume of the historical data of the target indicator data being greater than the first data volume and the time series characteristics of the historical data of the target indicator data not satisfying the strong correlation condition, the GM model optimized based on the LSTM model is determined as the prediction model.
[0019] In this embodiment, the first data volume can be determined based on experience. When the data volume of the historical data of the target indicator data is less than the first data volume, it means that the number of data samples is relatively small, and the GM model only needs 4-5 data points to build a model, while LSTM usually requires more than 50 samples to be effectively trained. Therefore, in this application, when the data volume is relatively small, the GM model is selected as the prediction model.
[0020] Secondly, consider the case where the target indicator's historical data volume indicates a relatively large amount of data. In this case, a model can be determined based on the temporal characteristics of the target indicator's historical data. This is because when the temporal characteristics of the target indicator's historical data meet the strong correlation condition, the autocorrelation function of the target indicator's historical data decays slower than the preset decay rate. In other words, the ACF value decays slowly with increasing lag order, indicating that the data to be predicted has strong temporal dependence and is suitable for LSTM to capture long-term correlations. Conversely, if the ACF decays rapidly, approaching white noise, it can be considered a weak temporal dependence, and the advantages of LSTM are weakened. Therefore, LSTM is used to optimize the GM model, using LSTM to filter noise and retain trend characteristics.
[0021] S102: In response to the prediction model being a GM model optimized based on the LSTM model, performing a first operation; The first operation includes: Input the historical data of the target indicator data into the LSTM model to obtain the hidden state corresponding to the target indicator data and the first prediction result; Based on the hidden state corresponding to the target indicator data, the development coefficient and gray action of the GM model are optimized to obtain the target GM model; Input the historical data of the target indicator data into the target GM model to obtain the second prediction result; The first prediction result and the second prediction result are weighted and calculated to obtain the prediction result of the target indicator data.
[0022] In this embodiment, the first prediction result is obtained by the LSTM model based on the historical data of the target indicator data. Essentially, it is also the target indicator data predicted some time later. The hidden state represents the intermediate feature output at each time step when processing sequence data in the LSTM model. This embodiment of the application considers that the development coefficient and gray action of the traditional GM model are fixed, making it only applicable in scenarios where the data exhibits a strong monotonic trend or simple periodicity. When the data exhibits non-stationarity, local noise, or multi-scale fluctuations, the parameter estimation error is large. The development coefficient determines the overall trend of the sequence, and the gray action reflects the external interference or internal fluctuations of the system. The LSTM extracts the long-term dependency features of the sequence through a gating mechanism. Its hidden state (especially the last hidden state and the hidden state at each moment) contains rich information. For example, the last hidden state can be used to capture the comprehensive trend and long-term dependencies of the entire historical sequence, while the hidden state at each moment can characterize the short-term fluctuations, periodicity, or outliers of the data at different time points. Therefore, this application fuses the two models to optimize the GM model based on the output data and process data of the LSTM model. This improves prediction accuracy while not introducing too many new variables and parameters, as all of them are model process data, thereby improving data utilization.
[0023] In an embodiment of the present application, when the amount of historical data of the target indicator data is greater than the first data amount, and the time series characteristics of the historical data of the target indicator data do not meet the strong correlation conditions, the reason for not only using the LSTM model is that the prediction accuracy of the LSTM model for data with weak time series dependence is not high, but the GM model itself is suitable for small sample data prediction. Therefore, the present application makes predictions based on the two models, and finally performs weighted calculations to obtain predicted data. On this basis, the present application optimizes the parameters of the GM model through the intermediate data and output data of the LSTM model.
[0024] In this embodiment, the target GM model is a GM model after parameter optimization. Based on the target GM model, target indicator data after a period of time can be predicted to obtain second prediction data.
[0025] It should be noted that the LSTM model and GM model in the embodiments of this application are both trained models. The model training process will not be repeated in this embodiment of the application. Those skilled in the art can perform training based on experience. Secondly, the prediction time of the first prediction result and the second prediction result should be the same. For example, if the data to be output is one week later, the output of the LSTM model and the GM model should both be the predicted data one week later, so as to perform weighted calculation.
[0026] From the above, it can be concluded that this application is based on industrial big data, and determines the prediction model by comprehensively considering the data volume and data characteristics of the historical data of the target indicator data, which can better adapt to the complex and changeable industrial production data environment and provide a basis for accurate prediction. It solves the problem that the traditional method is based on simple statistical models or empirical formulas, has limitations in processing complex and nonlinear industrial production data, is difficult to fully explore the potential laws and characteristics of the data, and cannot accurately capture the trend of data changes, resulting in low accuracy and reliability of the prediction results. This application optimizes the development coefficient and gray action of the GM model based on the hidden state obtained by the LSTM model to obtain the target GM model, integrates the advantages of the LSTM model into the GM model, makes up for the shortcomings of the traditional GM model, and enables the target GM model to more accurately reflect the trend of data changes.
[0027] In one embodiment of the present application, the hidden state corresponding to the target indicator data includes: the hidden state of the last layer of the LSTM model and the hidden state of the LSTM model at each moment; Based on the hidden state corresponding to the target indicator data, the development coefficient and gray action of the GM model are optimized to obtain the target GM model, including: The hidden state of the last layer of the LSTM model is used as the global feature, and the development coefficient of the GM model is optimized based on the global feature; Construct a time series feature matrix based on the hidden state of the LSTM model at each moment; The fully connected layer based on the LSTM model extracts features from the time series feature matrix to obtain local fluctuation features; Optimize the grey action of GM model based on local fluctuation characteristics.
[0028] In this embodiment, the hidden state of the last layer of the LSTM model is represented as , the hidden state of the LSTM model at each moment is expressed as , It can be used as a global feature to reflect the overall trend of the entire sequence. It is constructed as a time series feature matrix, and the local fluctuation features are extracted through the convolution layer or the fully connected layer to correct the real-time disturbance component of the gray action.
[0029] In this embodiment, the global features corresponding to the target indicator data can solve the problem of insufficient long-term trend modeling of the GM model, so that the development coefficient adapts to the overall evolution law of the data; the local features corresponding to the target indicator data can solve the problem of insufficient characterization of short-term fluctuations by the GM model, so that the gray action dynamically reflects the real-time disturbance of the system. Through the nonlinear mapping of features and parameters, the "data understanding ability" of LSTM is injected into the "mechanism modeling framework" of the GM model, ultimately improving the prediction accuracy and robustness in complex industrial scenarios.
[0030] In this embodiment, the gray action is essentially a quantification of non-trend fluctuations in the system (such as noise, periodic interference, and external shocks). If the data exhibits significant local fluctuations (such as equipment vibration during production or sudden changes in indicators caused by operating mode switching), the fixed gray action of the traditional GM model is difficult to adapt dynamically, resulting in large prediction errors. Therefore, the gray action can be optimized based on local eigenvectors that characterize the intensity, frequency, and location of short-term fluctuations. Specifically, the optimization method can establish a mapping relationship between the local eigenvectors and the gray action, allowing the gray action to dynamically adjust with local fluctuation characteristics, thereby enhancing the GM model's ability to fit data details. The mapping relationship can be a linear mapping, and the slope and intercept can be solved by minimizing the GM model's fitting error (such as the mean square error).
[0031] From the above, it can be concluded that the last hidden state of the LSTM model, as a global feature, can reflect the comprehensive trend of the entire sequence. It is used to optimize the development coefficient of the GM model so that the development coefficient can better adapt to the overall evolution law of the data. The traditional GM model has deficiencies in long-term trend modeling, and this embodiment solves this problem by introducing global features, allowing the GM model to more accurately grasp the long-term trend of the data during prediction, thereby improving prediction accuracy. This embodiment optimizes the gray action through local features, enabling it to dynamically reflect the real-time disturbances of the system, enhancing the GM model's ability to fit data details and further improving prediction accuracy.
[0032] In one embodiment of the present disclosure, specifically, optimizing the development coefficient of the GM model based on the global features may include: Perform feature extraction on the global features to obtain trend features corresponding to the target indicator data, change features corresponding to the target indicator data, and period features corresponding to the target indicator data; Determine the sign of the development coefficient of the GM model based on the trend characteristics corresponding to the target indicator data; Determine the absolute value of the development coefficient of the GM model based on the change characteristics corresponding to the target indicator data; The time-varying characteristics of the development coefficient of the GM model are determined based on the periodic characteristics corresponding to the target indicator data.
[0033] In this embodiment, the global feature Contains comprehensive information of the entire historical series, and is an abstract representation of the long-term dependencies, overall trends, and cyclical patterns of the data. Determines the trend type and rate of change of the forecast sequence, The data is growing exponentially. When , the data shows an exponential decay trend. The bigger it is, the more dramatic and bigger the trend change is. The smaller it is, the smoother the trend change.
[0034] In this embodiment, trend characteristics can be represented by the sign or specific dimension of the feature vector, change characteristics can be represented by the modulus of the feature vector or the value of the specific dimension, and periodic characteristics can be represented by frequency domain characteristics or periodic indicators. Trend characteristics can be trend direction, such as growth / decay; change characteristics can be change rate, such as growth / decay rate; and periodic characteristics can be the intensity of seasonal fluctuations.
[0035] In this embodiment, the trend characteristic represents the overall upward / downward trend of the data and can determine the sign of the development coefficient. When the trend characteristic is upward, the sign of the corresponding development coefficient is positive, and when the trend characteristic is downward, the sign of the corresponding development coefficient is negative. The change characteristic, that is, the rate of change, determines the absolute value of the development coefficient, and the rate of change is positively correlated with the absolute value of the development coefficient. The periodic characteristic determines the time-varying characteristics of the development coefficient (such as periodic adjustment) and can be expressed as a period. In this embodiment, the essence is to solve the problem that the traditional GM model parameters are fixed and cannot adapt to the data characteristics of the data to be predicted. In other words, the essence is to optimize the parameters of the GM model.
[0036] Specifically, the process of determining the absolute value of the development coefficient based on the change characteristics can be determined based on a simple linear relationship or a simple mapping relationship, where the slope of the linear relationship can be determined based on experience. However, it should be noted that a range needs to be set for the development coefficient to prevent the value from being too large or too small due to no restrictions. The range size can be determined based on prior knowledge in the field to set a range with wider applicability, or it can be set based on actual application scenarios, which will not be repeated in the embodiments of this application.
[0037] In this embodiment, the fully connected layer of the LSTM model is used to extract features from the time series feature matrix to obtain local fluctuation features, including: The time series feature matrix is flattened into a one-dimensional vector, and the one-dimensional vector is input into the fully connected layer to obtain the local fluctuation feature.
[0038] In this embodiment, the hidden state of LSTM at each moment is constitute The time series feature matrix ,in is the sequence length, is the hidden layer dimension, each Contains local context information at the current moment (such as the fluctuation direction and amplitude at the adjacent moments), the matrix The overall temporal dynamics of the sequence are characterized.
[0039] In this embodiment, flattening H into a one-dimensional vector essentially merges the time dimension and the feature dimension, enabling the fully connected layer to capture long-range dependencies across time steps and complex fluctuation patterns (such as continuous upward / downward trends across multiple moments). The fully connected layer linearly transforms the flattened vector using a weight matrix and bias, and introduces nonlinearity through an activation function to extract high-order local fluctuation features (such as outliers and the intensity of periodic oscillations).
[0040] From the above, it can be concluded that the present application determines the absolute value of the development coefficient based on the change characteristics corresponding to the target indicator data, and the rate of change is positively correlated with the absolute value of the development coefficient. The change characteristics reflect the rate of change of the data. Through this determination method, the development coefficient can more accurately describe the severity of the change in the data trend. For example, in the production process, if the rate of change of a certain indicator accelerates, the absolute value of the development coefficient increases accordingly, and the model can more sensitively capture the changes in the data, thereby improving the accuracy of the prediction. The present application flattens the time series feature matrix composed of the hidden states of the LSTM at each moment into a one-dimensional vector, merges the time dimension and the feature dimension, and enables the fully connected layer to process information on the time dimension and the feature dimension at the same time. The fully connected layer can capture long-distance dependencies and complex fluctuation patterns across time steps, such as trends of continuous rise or fall at multiple moments, thereby more comprehensively understanding the dynamic changes in the production process indicator data.
[0041] In one embodiment of the present application, performing weighted calculation on the first prediction result and the second prediction result to obtain the prediction result of the target indicator data includes: Determine a first prediction weight based on the data volume of the historical data of the target indicator data and the data characteristics of the historical data of the target indicator data, where the first prediction weight is a weight corresponding to the first prediction result; Determining a second prediction weight based on the first prediction weight, where the second prediction weight is a weight corresponding to the second prediction result; the sum of the first prediction weight and the second prediction weight is 1; A weighted calculation is performed based on the first prediction weight, the second prediction weight, the first prediction result and the second prediction result to obtain the prediction result of the target indicator data.
[0042] In this embodiment, the first prediction weight can be determined specifically in the following manner: Determining a first prediction weight based on the data volume of the historical data of the target indicator data and the data characteristics of the historical data of the target indicator data includes: The first prediction weight is determined based on the data volume of the historical data of the target indicator data and the decay rate of the autocorrelation function of the historical data of the target indicator data; wherein, the data volume of the historical data of the target indicator data is positively correlated with the first prediction weight, and the decay rate of the autocorrelation function of the historical data of the target indicator data is negatively correlated with the first prediction weight.
[0043] In this embodiment, the larger the amount of historical data of the target indicator data, the more fully the LSTM model is trained, and the larger its corresponding weight should be. The decay rate of the autocorrelation function is used to measure the strength of the data's temporal dependence. A slow decay rate indicates significant long-term dependence, and a fast decay rate indicates that short-term fluctuations dominate. The weaker the temporal dependence, the lower the advantage of the LSTM model, and therefore the corresponding weight is reduced.
[0044] In this embodiment, the first prediction weight may be determined based on the amount of historical data of the target indicator data, the decay rate of the autocorrelation function of the historical data of the target indicator data, and a first formula. The first formula may be:
[0045] in, represents the first prediction weight, The slope parameter is used to control the sensitivity of weight changes. In this embodiment, it can be set to 2, or it can be determined based on multiple experiments. Indicates the amount of historical data, The data volume benchmark value (which can be set as the minimum number of samples required for effective model training), represents the decay rate of the autocorrelation function, The decay rate benchmark value can be set to the median decay rate of typical industrial time series data, such as 0.1.
[0046] In the first formula, the larger the amount of historical data, The closer it is to 1, the larger the proportion of the prediction results output by the LSTM model in the weighted calculation, and the faster the decay rate of the autocorrelation function. The closer it is to 0, the smaller the proportion of the prediction results output by the LSTM model in the weighted calculation.
[0047] Based on the calculated first prediction weight, the second prediction weight can be determined, that is, the sum of the two weights can be equal to 1. Finally, the first prediction result and the second prediction result can be weighted calculated based on the aforementioned weights to finally obtain the prediction result.
[0048] From the above, it can be concluded that this application determines the weight corresponding to the LSTM model by comprehensively considering the amount of historical data of the target indicator data and the decay rate of the autocorrelation function, and then determines the weight corresponding to the second prediction result. This can more reasonably distribute the contributions of the two prediction models to the final prediction result. When the amount of historical data is large, the LSTM model can be more fully trained and its predictive ability is relatively stronger, so it is given a larger weight. When the decay rate of the autocorrelation function is fast, indicating weak data time series dependence and a decrease in the advantage of the LSTM model, its weight is appropriately reduced. This method of dynamically allocating weights based on data characteristics enables the final prediction result to combine the advantages of both models, improving the accuracy of the prediction.
[0049] Corresponding to the production process indicator prediction method based on industrial big data in the above embodiment, Figure 2 This is a structural block diagram of a production process indicator prediction system based on industrial big data provided by an embodiment of the present application. For ease of explanation, only the parts related to the embodiment of the present application are shown. Figure 2 The production process indicator prediction system 20 based on industrial big data includes: a model determination module 21 and an indicator prediction module 22.
[0050] The model determination module 21 is used to determine a prediction model based on the data volume of the historical data of the target indicator data and the time series characteristics of the historical data of the target indicator data; the target indicator data is the generation process indicator data to be predicted; The indicator prediction module 22 is configured to perform a first operation in response to the prediction model being a GM model optimized based on the LSTM model; The first operation includes: Input the historical data of the target indicator data into the LSTM model to obtain the hidden state corresponding to the target indicator data and the first prediction result; Based on the hidden state corresponding to the target indicator data, the development coefficient and gray action of the GM model are optimized to obtain the target GM model; Input the historical data of the target indicator data into the target GM model to obtain the second prediction result; The first prediction result and the second prediction result are weighted and calculated to obtain the prediction result of the target indicator data.
[0051] In one embodiment of the present application, the time series characteristics of the historical data include: an attenuation rate of an autocorrelation function of the historical data; the model determination module 21 is specifically configured to determine the GM model as the prediction model in response to the data volume of the historical data of the target indicator data being less than the first data volume; In response to the data volume of the historical data of the target indicator data being greater than the first data volume, and the time series characteristics of the historical data of the target indicator data satisfying a strong correlation condition, determining the LSTM model as the prediction model; wherein the strong correlation condition is that the decay rate of the autocorrelation function of the historical data of the target indicator data is less than a preset decay rate; In response to the data volume of the historical data of the target indicator data being greater than the first data volume and the time series characteristics of the historical data of the target indicator data not satisfying the strong correlation condition, the GM model optimized based on the LSTM model is determined as the prediction model.
[0052] In one embodiment of the present application, the hidden state corresponding to the target indicator data includes: the hidden state of the last layer of the LSTM model and the hidden state of the LSTM model at each moment; The indicator prediction module 22 is specifically used to use the hidden state of the last layer of the LSTM model as a global feature and optimize the development coefficient of the GM model based on the global feature; Construct a time series feature matrix based on the hidden state of the LSTM model at each moment; The fully connected layer based on the LSTM model extracts features from the time series feature matrix to obtain local fluctuation features; Optimize the grey action of GM model based on local fluctuation characteristics.
[0053] In one embodiment of the present application, the indicator prediction module 22 is further configured to extract features from the global features to obtain trend features corresponding to the target indicator data, change features corresponding to the target indicator data, and period features corresponding to the target indicator data; Determine the sign of the development coefficient of the GM model based on the trend characteristics corresponding to the target indicator data; Determine the absolute value of the development coefficient of the GM model based on the change characteristics corresponding to the target indicator data; The time-varying characteristics of the development coefficient of the GM model are determined based on the periodic characteristics corresponding to the target indicator data.
[0054] In one embodiment of the present application, the indicator prediction module 22 is further configured to flatten the time series feature matrix into a one-dimensional vector, and input the one-dimensional vector into a fully connected layer to obtain a local fluctuation feature.
[0055] In one embodiment of the present application, the indicator prediction module 22 is further configured to determine a first prediction weight based on the data volume of the historical data of the target indicator data and the data characteristics of the historical data of the target indicator data, where the first prediction weight is a weight corresponding to the first prediction result; Determining a second prediction weight based on the first prediction weight, where the second prediction weight is a weight corresponding to the second prediction result; the sum of the first prediction weight and the second prediction weight is 1; A weighted calculation is performed based on the first prediction weight, the second prediction weight, the first prediction result and the second prediction result to obtain the prediction result of the target indicator data.
[0056] In one embodiment of the present application, the indicator prediction module 22 is further used to determine a first prediction weight based on the data volume of the historical data of the target indicator data and the decay rate of the autocorrelation function of the historical data of the target indicator data; wherein, the data volume of the historical data of the target indicator data is positively correlated with the first prediction weight, and the decay rate of the autocorrelation function of the historical data of the target indicator data is negatively correlated with the first prediction weight.
[0057] See also Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided in one embodiment of the present application. Figure 3 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of the modules in the above-mentioned system embodiments, such as Figure 2 The functions of the model determination module 21 and the indicator prediction module 22 are shown.
[0058] It should be understood that in the embodiment of the present application, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.
[0059] The input device 302 may include a touchpad, a fingerprint collection sensor (for collecting user fingerprint information and fingerprint direction information), a microphone, etc. The output device 303 may include a display (LCD, etc.), a speaker, etc.
[0060] The memory 304 may include a read-only memory and a random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store a first amount of data.
[0061] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiment of the present application can execute the implementation method described in the embodiment of the production process indicator prediction method based on industrial big data provided in the embodiment of the present application, and can also execute the implementation method of the electronic device described in the embodiment of the present application, which will not be repeated here.
[0062] In another embodiment of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, all or part of the process of the method in the above embodiment is implemented. The computer program can also be used to instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above method embodiments are implemented. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium.
[0063] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Furthermore, the computer-readable storage medium can include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or is about to be output.
[0064] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0065] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0066] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or can be an electrical, mechanical or other form of connection.
[0067] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present application.
[0068] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.
[0069] The above are only specific embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.
Claims
1. A production process indicator prediction method based on industrial big data, characterized in that: include: Determine a prediction model based on the data volume of historical data of the target indicator data and the time series characteristics of the historical data of the target indicator data; The target indicator data is the generation process indicator data to be predicted; In response to the prediction model being a GM model optimized based on an LSTM model, performing a first operation; The first operation includes: Inputting historical data of the target indicator data into the LSTM model to obtain a hidden state corresponding to the target indicator data and a first prediction result; Optimizing the development coefficient and the grey action of the GM model based on the hidden state corresponding to the target indicator data to obtain a target GM model; Inputting historical data of the target indicator data into the target GM model to obtain a second prediction result; The first prediction result and the second prediction result are weightedly calculated to obtain the prediction result of the target indicator data.
2. The production process indicator prediction method based on industrial big data according to claim 1, characterized in that: The time series characteristics of the historical data include: the decay rate of the autocorrelation function of the historical data; The determining of the prediction model based on the data volume of the historical data of the target indicator data and the time series characteristics of the historical data of the target indicator data includes: In response to the data amount of the historical data of the target indicator data being less than the first data amount, determining the GM model as the prediction model; In response to the data volume of the historical data of the target indicator data being greater than the first data volume, and the time series characteristics of the historical data of the target indicator data satisfying a strong correlation condition, determining the LSTM model as the prediction model; wherein the strong correlation condition is that the decay rate of the autocorrelation function of the historical data of the target indicator data is less than a preset decay rate; In response to the data volume of the historical data of the target indicator data being greater than the first data volume, and the time series characteristics of the historical data of the target indicator data not meeting the strong correlation condition, the GM model optimized based on the LSTM model is determined as the prediction model.
3. The production process indicator prediction method based on industrial big data according to claim 1, characterized in that: The hidden state corresponding to the target indicator data includes: the hidden state of the last layer of the LSTM model and the hidden state of the LSTM model at each moment; The step of optimizing the development coefficient and the gray action of the GM model based on the hidden state corresponding to the target indicator data to obtain the target GM model includes: Using the hidden state of the last layer of the LSTM model as a global feature, and optimizing the development coefficient of the GM model based on the global feature; Constructing a time series feature matrix based on the hidden state of the LSTM model at each moment; Extract features from the time series feature matrix based on the fully connected layer of the LSTM model to obtain local fluctuation features; The grey action of the GM model is optimized based on the local fluctuation characteristics.
4. The production process indicator prediction method based on industrial big data according to claim 3 is characterized in that: Optimizing the development coefficient of the GM model based on the global feature includes: Extracting features from the global features to obtain trend features corresponding to the target indicator data, change features corresponding to the target indicator data, and period features corresponding to the target indicator data; Determining the sign of the development coefficient of the GM model based on the trend characteristics corresponding to the target indicator data; Determining the absolute value of the development coefficient of the GM model based on the change characteristics corresponding to the target indicator data; The time-varying characteristics of the development coefficient of the GM model are determined based on the periodic characteristics corresponding to the target indicator data.
5. The production process indicator prediction method based on industrial big data according to claim 3 is characterized in that: The fully connected layer based on the LSTM model performs feature extraction on the time series feature matrix to obtain local fluctuation features, including: The time series feature matrix is flattened into a one-dimensional vector, and the one-dimensional vector is input into a fully connected layer to obtain a local fluctuation feature.
6. The production process indicator prediction method based on industrial big data according to claim 1, characterized in that: The step of performing weighted calculation on the first prediction result and the second prediction result to obtain the prediction result of the target indicator data includes: determining a first prediction weight based on the data volume of the historical data of the target indicator data and the data characteristics of the historical data of the target indicator data, where the first prediction weight is a weight corresponding to the first prediction result; determining a second prediction weight based on the first prediction weight, where the second prediction weight is a weight corresponding to the second prediction result; and the sum of the first prediction weight and the second prediction weight is 1; A weighted calculation is performed based on the first prediction weight, the second prediction weight, the first prediction result, and the second prediction result to obtain a prediction result of the target indicator data.
7. The production process indicator prediction method based on industrial big data according to claim 6, characterized in that: The determining of the first prediction weight based on the data volume of the historical data of the target indicator data and the data characteristics of the historical data of the target indicator data includes: The first prediction weight is determined based on the data volume of the historical data of the target indicator data and the decay rate of the autocorrelation function of the historical data of the target indicator data; wherein, the data volume of the historical data of the target indicator data is positively correlated with the first prediction weight, and the decay rate of the autocorrelation function of the historical data of the target indicator data is negatively correlated with the first prediction weight.
8. A production process indicator prediction system based on industrial big data, characterized in that: include: A model determination module is used to determine a prediction model based on the data volume of the historical data of the target indicator data and the time series characteristics of the historical data of the target indicator data; The target indicator data is the generation process indicator data to be predicted; An indicator prediction module, configured to perform a first operation in response to the prediction model being a GM model optimized based on an LSTM model; The first operation includes: Inputting historical data of the target indicator data into the LSTM model to obtain a hidden state corresponding to the target indicator data and a first prediction result; Optimizing the development coefficient and the gray action of the GM model based on the hidden state corresponding to the target indicator data to obtain a target GM model; Inputting historical data of the target indicator data into the target GM model to obtain a second prediction result; The first prediction result and the second prediction result are weightedly calculated to obtain the prediction result of the target indicator data.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.