Production energy consumption value prediction method and system based on artificial intelligence
Through artificial intelligence-based methods, screening and constructing key feature parameters, the problem that traditional energy consumption prediction methods are difficult to capture complex nonlinear relationships and dynamic changes is solved, and higher energy consumption prediction accuracy and accuracy are achieved.
Patent Information
- Application Number
- CN202411589432.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-08
- Publication Date
- 2025-02-28
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional energy consumption prediction methods are difficult to accurately capture the complex nonlinear relationships and dynamic changes in industrial production processes, resulting in insufficient prediction accuracy and cannot meet the demand for energy consumption control in modern industries.
Using an artificial intelligence-based method, by collecting and preprocessing real-time data, building feature vectors, using correlation coefficients and mutual information values to filter key feature parameters, reducing noise components, and improving the prediction accuracy of the model.
By removing feature parameters that are not related to energy consumption or are less correlated, the model can more clearly capture the real signals related to energy consumption, improve the accuracy and accuracy of energy consumption prediction, and better reflect the changing trend of actual energy consumption.
Smart Images
Figure CN119539162A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of energy consumption prediction, and in particular to a method and system for predicting production energy consumption values based on artificial intelligence. Background Art
[0002] In the field of modern industrial production and energy management, accurately predicting energy consumption in the production process is of great significance for improving energy efficiency, reducing costs and reducing environmental pollution. Traditional energy consumption prediction methods mainly rely on statistical analysis, empirical formulas or simple mathematical models. However, with the improvement of industrial automation and the development of Internet of Things technology, traditional methods often only consider a small number of static parameters. Due to the failure to fully consider the complex nonlinear relationships and various dynamic change factors in the production process, traditional energy consumption prediction models can only give a rough trend prediction, but cannot achieve a high degree of accuracy. This is far from enough for industries that have strict requirements on energy consumption control. For example, in the fine chemical industry, even small fluctuations in energy consumption can lead to significant differences in product quality. Therefore, the lack of the ability to capture and model these subtle changes makes it difficult for traditional methods to meet the needs of modern industry for energy consumption prediction accuracy. Summary of the invention
[0003] The purpose of this application is to provide a method and system for predicting production energy consumption values based on artificial intelligence, remove characteristic parameters that are irrelevant to energy consumption and have low correlation, and screen out key features to reduce the noise components in the input data, so that the model can more clearly capture the real signals related to energy consumption. The focus on key features makes the model more focused when predicting, improves the prediction accuracy, and more accurately reflects the changing trend of actual energy consumption.
[0004] On the one hand, according to an embodiment of the present application, a method for predicting production energy consumption value based on artificial intelligence is proposed, step 1, collecting real-time data in the production process;
[0005] Step 2: preprocess the real-time data;
[0006] Step 3, constructing a feature vector based on the preprocessed real-time data;
[0007] Step 31, obtaining dynamic data according to the preprocessed real-time data, and obtaining candidate feature parameters according to the dynamic data;
[0008] Step 32, calculating the correlation coefficient between the candidate characteristic parameter and the energy consumption value;
[0009] Step 33, constructing a first predetermined threshold value for the correlation coefficient, determining whether the absolute value of the correlation coefficient is greater than or equal to the first predetermined threshold value, if so, defining the candidate feature parameter corresponding to the current correlation coefficient as a relevant feature parameter; if not, defining the candidate feature parameter corresponding to the current correlation coefficient as a non-relevant feature parameter, and proceeding to step 34;
[0010] Step 34, calculating the mutual information value for the non-correlated feature parameters;
[0011] Step 35, constructing a second predetermined threshold for the mutual information value, and determining whether the mutual information value is greater than or equal to the second predetermined threshold, if so, defining the non-correlated feature parameters corresponding to the current mutual information value as relevant feature parameters; if not, the non-correlated feature parameters corresponding to the current mutual information value remain unchanged;
[0012] Step 36, constructing a feature vector according to the relevant feature parameters in step 33 and step 35;
[0013] Step 4, inputting the feature vector into a pre-trained machine learning model to predict the energy consumption value in the production process;
[0014] Step 5: predict future energy consumption values based on the output results of the machine learning model.
[0015] According to one aspect of the embodiment of the present application, the calculation of the correlation coefficient includes:
[0016] Step 321, calculating the covariance between the candidate parameter and the energy consumption value;
[0017] Step 322, calculating the standard deviation of the candidate parameter;
[0018] Step 323, calculating the standard deviation of the energy consumption value;
[0019] Step 324, obtaining a correlation coefficient according to the covariance, the standard deviation of the candidate parameters, and the standard deviation of the energy consumption value.
[0020] According to one aspect of an embodiment of the present application, constructing a second predetermined threshold for the mutual information value specifically includes:
[0021] Step 351, obtaining a second threshold;
[0022] Step 352, modifying the second threshold to obtain a modified second threshold;
[0023] Step 353: Optimize the modified second threshold to obtain a second predetermined threshold.
[0024] According to one aspect of an embodiment of the present application, the second threshold is corrected according to the first predetermined threshold to obtain the corrected second threshold, specifically including:
[0025] Step 3521, calculating the mutual information value for the relevant feature parameters in step 33;
[0026] Step 3522, obtaining a mutual information value mean according to the mutual information value of step 353, and obtaining a mutual information value standard deviation according to the mutual information value mean as a correction item;
[0027] Step 3523, obtaining a corrected second threshold value according to the second threshold value and the correction item.
[0028] According to one aspect of an embodiment of the present application, optimizing the modified second threshold to obtain a predetermined second threshold includes:
[0029] Step 3531, dividing the value range of the mutual information value of the unrelated characteristic parameters into a plurality of intervals of equal width; the width of the interval is N;
[0030] Step 3532, determining the number of mutual information values in each interval and the number of all mutual information values; constructing a distribution density function according to the width of each interval and the number of all mutual information values;
[0031] Step 3533, defining the interval with the largest number of mutual information values as the pending interval;
[0032] Step 3534, calculating the distribution density value for each mutual information value in the to-be-determined interval according to the distribution density function, if the distribution density values of 3 to (Q-1) mutual information values in the to-be-determined interval reach three times the distribution density values of other mutual information values in the to-be-determined interval, the to-be-determined interval is defined as a determined interval; Q represents the maximum number of mutual information values in the to-be-determined interval;
[0033] Step 3535, searching for a mutual information value that is the same as the modified second threshold value in the determined interval. If a mutual information value that is the same as the modified second threshold value is found, the distribution density value of the mutual information value that is the same as the modified second threshold value is used as the optimization value. If no mutual information value is found, returning to step 3531;
[0034] Step 3536, obtaining a predetermined second threshold value based on the modified second threshold value and the optimized value.
[0035] According to one aspect of the embodiment of the present application, the distribution density function is:
[0036]
[0037] In the formula, f i represents the distribution density function of the ith interval, Q i represents the number of mutual information values in the i-th interval, ΔI represents the interval width, and Z represents the maximum number of intervals.
[0038] According to one aspect of the embodiment of the present application, the calculation formula of the correlation coefficient is:
[0039]
[0040] R XY represents the correlation coefficient between the candidate parameter X and the energy consumption value Y, Cov(X, Y) represents the covariance between the candidate parameter X and the energy consumption value Y, σ X represents the standard deviation of the candidate parameter X, σ Y Represents the standard deviation of the energy consumption value Y.
[0041] According to one aspect of the embodiment of the present application, the calculation formula of the predetermined second threshold is:
[0042]
[0043] In the formula, represents a predetermined second threshold value, represents the modified second threshold, Represents the modified second threshold The density function value of .
[0044] On the other hand, the embodiment of the present application proposes a production energy consumption value prediction system based on artificial intelligence, which is used to execute the above-mentioned production energy consumption value prediction method based on artificial intelligence, including the following modules:
[0045] Collection module: used to collect real-time data during the production process;
[0046] Preprocessing module: connected with the collection module, used to preprocess the real-time data;
[0047] Feature vector construction module: connected with the preprocessing module, used to construct feature vectors according to the preprocessed real-time data;
[0048] Energy consumption value prediction module: connected to the feature vector construction module, used to input the feature vector into a pre-trained machine learning model to predict the production process and future energy consumption values.
[0049] In summary, the beneficial technical effects of this application are:
[0050] Based on the existing prediction of energy consumption through prediction models, the present invention screens the input data of the model and removes some characteristic parameters that are irrelevant to or have low correlation with energy consumption. The screened key features reduce the noise components in the input data, so that the model can more clearly capture the real signals related to energy consumption. The focus on key features makes the model more focused when predicting, improves the prediction accuracy, and more accurately reflects the changing trend of actual energy consumption.
[0051] The present invention divides the candidate parameters into relevant characteristic parameters and non-relevant characteristic parameters by using the correlation coefficient between the candidate parameters and the energy consumption value and the first predetermined threshold value, and uses the mutual information value and the second predetermined threshold value to screen out the relevant characteristic parameters from the non-relevant characteristic parameters, and finally combines all the relevant characteristic parameters into a characteristic vector as the input feature of the model. Through the feature screening step, the features that are irrelevant or weakly correlated with the energy consumption value are removed, and the feature information that is highly correlated with the energy consumption value is retained, including linear relationship information and nonlinear relationship information. The characteristic vector comprehensively describes the change law of the energy consumption value, better adapts to various dynamic changes that occur in the production process, and the model can focus on the features that are highly correlated with the energy consumption value during the prediction process, thereby improving the prediction accuracy of the energy consumption value. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 is a flow chart of this application;
[0053] Figure 2 It is a schematic diagram of the structure of this application. DETAILED DESCRIPTION
[0054] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.
[0055] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0056] In the embodiments of the present application, the same reference numerals represent the same components, and for the sake of brevity, detailed descriptions of the same components are omitted in different embodiments. It should be understood that the thickness, length, width and other dimensions of various components in the embodiments of the present application shown in the drawings, as well as the overall thickness, length, width and other dimensions of the integrated device are only exemplary descriptions and should not constitute any limitation to the present application.
[0057] The term "plurality" used in the present application refers to two or more (including two).
[0058] On the one hand, the following combination Figure 1 to Figure 2 A production energy consumption value prediction method and system based on artificial intelligence according to an embodiment of the present application are described in detail.
[0059] See attached Figure 1 and attached Figure 2 , a production energy consumption value prediction method based on artificial intelligence, comprising the following steps:
[0060] Step 1: Collect real-time data during the production process.
[0061] In this embodiment, real-time data in the production process is collected through sensors on the machine equipment, such as temperature sensors, pressure sensors, flow meters, etc. The collected real-time data includes but is not limited to:
[0062] Equipment parameters: equipment model, serial number, manufacturer information, production line layout and design parameters, etc.;
[0063] Process parameters: product specifications, process formula, etc.;
[0064] Production batch information: production batch number, start time, end time, etc.;
[0065] Sensor data: temperature sensor data, pressure sensor data, flow meter data, displacement sensor data, vibration sensor data, etc.;
[0066] Control system data: PLC controller status information, DCS controller status information, operator console operation records, etc.;
[0067] Environmental data: temperature and humidity in the workshop, ambient noise level, air quality index, etc.;
[0068] Fault alarm information: equipment fault alarm records, production line shutdown event records, etc.;
[0069] Production progress data: production progress of the current production line, completion status of each process, etc.
[0070] Step 2: preprocess the real-time data.
[0071] In industrial production and manufacturing environments, preprocessing real-time data is a very important step, which directly affects the quality of subsequent data analysis and decision-making. Preprocessing mainly includes data cleaning, format conversion, standardization, missing value processing and other steps.
[0072] Step 3: construct a feature vector based on the preprocessed real-time data.
[0073] Step 31, obtaining dynamic data according to the preprocessed real-time data, and obtaining candidate feature parameters according to the dynamic data.
[0074] Dynamic data refers to data that is constantly changing during the production process, reflecting the operating status of the production equipment and the status of the production process. Dynamic data includes the sensor data, control system data, environmental data, fault alarm information, etc. mentioned in step 1. Candidate feature parameters refer to data features that may be associated with energy consumption in the production process. Aggregate dynamic data by time window, such as summarizing statistical indicators such as average, maximum, minimum, etc. by minute, hour, or shift. Extract useful feature parameters from the aggregated data, for example, temperature change rate, pressure fluctuation range, vibration intensity, displacement speed, fault frequency, and downtime.
[0075] Step 32, calculating the correlation coefficient between the candidate characteristic parameters and the energy consumption value.
[0076] Step 321, calculating the covariance between the candidate parameter and the energy consumption value.
[0077] Step 322, calculating the standard deviation of the candidate parameters.
[0078] Step 323, calculating the standard deviation of the energy consumption value.
[0079] Step 324, obtaining a correlation coefficient according to the covariance, the standard deviation of the candidate parameters, and the standard deviation of the energy consumption value.
[0080] The calculation formula of the correlation coefficient is:
[0081]
[0082] R XY represents the correlation coefficient between the candidate parameter X and the energy consumption value Y, Cov(X, Y) represents the covariance between the candidate parameter X and the energy consumption value Y, σ X represents the standard deviation of the candidate parameter X, σ Y Represents the standard deviation of the energy consumption value Y.
[0083] Step 33, construct a first predetermined threshold value for the correlation coefficient, and determine whether the absolute value of the correlation coefficient is greater than or equal to the first predetermined threshold value. If so, define the candidate feature parameters corresponding to the current correlation coefficient as relevant feature parameters; if not, define the candidate feature parameters corresponding to the current correlation coefficient as non-relevant feature parameters, and enter step 34.
[0084] The absolute value of the correlation coefficient between the candidate parameter and the energy consumption value is close to 1, indicating that the candidate parameter has a strong correlation with the energy consumption value, and the absolute value of the correlation coefficient between the candidate parameter and the energy consumption value is close to 0, indicating that the candidate parameter has a low correlation with the energy consumption value. Therefore, in this embodiment, the data point whose correlation coefficient value range is close to the middle value is used as the first predetermined threshold of the correlation coefficient, and the first predetermined threshold is 0.6. The candidate feature parameter whose absolute value of the correlation coefficient is greater than or equal to the first predetermined threshold is defined as a relevant feature parameter related to the energy consumption value. The candidate feature parameter whose absolute value of the correlation coefficient is less than the first predetermined threshold is defined as a non-relevant relevant feature parameter that is not related to the energy consumption value.
[0085] Step 34, calculating the mutual information value for the non-correlated feature parameters.
[0086] Since the first predetermined threshold is a threshold obtained through statistical analysis, the non-correlated feature parameters determined by the first predetermined threshold are likely to include relevant feature parameters, and therefore the mutual information value is calculated for the non-correlated feature parameters.
[0087] The correlation coefficient can be used to initially screen out some characteristic parameters that may be misjudged as non-correlated due to noise or the limitations of the statistical method itself. The mutual information value can capture nonlinear relationships and further confirm the relationship between these characteristic parameters and energy consumption values, reducing the risk of misjudgment. The mutual information value is calculated using the joint probability distribution function.
[0088] Step 35, constructing a second predetermined threshold for the mutual information value, and determining whether the mutual information value is greater than or equal to the second predetermined threshold, if so, defining the non-correlated feature parameters corresponding to the current mutual information value as relevant feature parameters; if not, the non-correlated feature parameters corresponding to the current mutual information value remain unchanged;
[0089] Step 351, obtaining a second threshold.
[0090] By analyzing the historical candidate parameters, the second threshold is designed based on any candidate parameter and the standard specification corresponding to the device. For example, the upper temperature limit recommended by the manufacturer is 80°C. When the temperature exceeds this upper limit, the energy consumption of the device will increase significantly. Therefore, this embodiment selects a balance point of the mutual information value as the second threshold, which can avoid selecting too high and resulting in too few features, or selecting too low and resulting in the introduction of a large number of irrelevant features.
[0091] Step 352, correct the second threshold to obtain a corrected second threshold.
[0092] Since the second threshold is obtained based on experience and has a certain error, if the relevant characteristic parameters in the non-relevant characteristic parameters are screened by the second threshold, the accuracy is obviously not high. Therefore, this step optimizes the second threshold, specifically including:
[0093] Step 3521, calculating the mutual information value for the relevant feature parameters in step 33.
[0094] Step 3522, according to the mutual information value of step 353, obtain the mean of the mutual information value, and according to the mean of the mutual information value, obtain the standard deviation of the mutual information value as a correction item.
[0095] Step 3523, obtaining a corrected second threshold value according to the second threshold value and the correction item.
[0096]
[0097] represents the modified second threshold, T2 represents the second threshold, k·σ I represents the correction term, σ I Represents the standard deviation of the mutual information value of the relevant feature parameters.
[0098] The modified second threshold is determined based on statistical analysis, which reflects the mean and standard deviation of the mutual information value, thereby ensuring that the screened feature parameters are statistically significant and closer to the characteristics of the actual data.
[0099] Step 353: Optimize the modified second threshold to obtain a second predetermined threshold.
[0100] The modified second threshold is obtained through statistical analysis and has only a statistically significant effect, but is not reflected in the distribution of actual data. It cannot be guaranteed that the threshold is accurate. Therefore, this embodiment continues to optimize the modified second threshold, specifically including:
[0101] Step 3531, dividing the value range of the mutual information value of the unrelated characteristic parameters into a number of intervals of equal width; the interval width is N.
[0102] Step 3532, determining the number of mutual information values in each interval and the number of all mutual information values; constructing a distribution density function according to the width of each interval and the number of all mutual information values;
[0103] Although the width of each interval is the same, the number of mutual information values in each interval is not equal. The distribution density value of the mutual information value in each interval is obtained by the following distribution density function.
[0104] The distribution density function is:
[0105]
[0106] In the formula, f i represents the distribution density function of the ith interval, Q i represents the number of mutual information values in the i-th interval, ΔI represents the interval width, and Z represents the maximum number of intervals.
[0107] Step 3533, defining the interval with the largest number of mutual information values as the pending interval.
[0108] First, we need to use the distribution density of mutual information values to determine which intervals have high or low distribution density of mutual information values. If the distribution density of mutual information values in a certain interval is high, it means that there are related feature parameters in the non-related feature parameters corresponding to the mutual information values in the interval. The ultimate goal is to filter out the related feature parameters from the non-related feature parameters.
[0109] Then, from the intervals in step 3531, an interval with the highest mutual information value distribution density is selected, which helps to screen out the relevant feature parameters included in the non-relevant feature parameters corresponding to the mutual information values.
[0110] The interval with the highest distribution density needs to meet two conditions at the same time: the number of mutual information values in the interval is the largest and the distribution density of some mutual information values in the interval is much higher than the distribution density of other mutual information values in the interval. Only when both conditions are met at the same time can the interval be defined as the interval with the highest distribution density.
[0111] Therefore, in this step, the interval with the largest number of mutual information values is defined as the pending interval.
[0112] Step 3534, calculate the distribution density value for each mutual information value in the pending interval according to the distribution density function. If the distribution density value of 3 to (Q-1) mutual information values in the pending interval reaches three times the distribution density value of other mutual information values in the pending interval, the pending interval is defined as a determined interval.
[0113] The distribution density value of some mutual information values in the determined interval far exceeds the distribution density values of other mutual information values in the interval. In this embodiment, preferably, if the distribution density value of 3 to (Q-1) mutual information values in the pending interval reaches three times the distribution density value of other mutual information values in the pending interval, then the current pending interval is defined as a determined interval, indicating that the distribution density of the interval is high, and there are relevant feature parameters in the non-relevant feature parameters corresponding to the mutual information values in the interval.
[0114] Step 3535, search for the mutual information value that is the same as the modified second threshold from the determined interval. If the mutual information value that is the same as the modified second threshold is found, the distribution density value of the mutual information value that is the same as the modified second threshold is used as the optimization value. If not found, return to execute step 3531.
[0115] This embodiment finds the mutual information value that is the same as the modified second threshold value in the determination interval in order to verify whether the modified second threshold value exists in the determination interval. If it exists in the determination interval, it is considered that the modified second threshold value is correlated with the energy consumption value. After successful verification, the distribution density value of the mutual information value that is the same as the modified second threshold value is calculated. On the one hand, the distribution position of the modified second threshold value in the determination interval can be seen. Such a design reflects the actual distribution of the modified second threshold value. On the other hand, the relative importance of the modified second threshold value in the determination interval can be evaluated, and it can also be used as an optimization value to optimize the modified second threshold value to obtain the second predetermined threshold value.
[0116] If the mutual information value that is the same as the modified second threshold value cannot be found in the determined interval, then return to recalculate the mutual information value of the irrelevant feature parameters, reduce the interval width and increase the number of intervals, so that the mutual information value can be more finely expressed, and re-determine the new determined interval, and find the mutual information value that is the same as the modified second threshold value from the new determined interval, and calculate the distribution density value of the mutual information value found as the optimization value.
[0117] In another embodiment, if the mutual information value identical to the modified second threshold value is still not found after multiple re-executions, the distribution density value of the modified second threshold value within the determined interval can be calculated by interpolation method to estimate the distribution density value of the threshold value within the determined interval and obtain the optimized value. Although the second predetermined threshold value obtained by this embodiment deviates from that of the previous embodiment, compared with the modified second threshold value, this embodiment can effectively improve the optimization accuracy of the modified second threshold value and improve the screening accuracy of subsequent related feature parameters.
[0118] Step 3536, obtaining a predetermined second threshold value based on the modified second threshold value and the optimized value.
[0119] The calculation formula of the predetermined second threshold is:
[0120]
[0121] In the formula, represents a predetermined second threshold value, represents the modified second threshold, Represents the modified second threshold The density function value of .
[0122] In this embodiment, the second threshold is first corrected by statistical analysis to obtain a corrected second threshold, and then the corrected second threshold is optimized by data distribution to obtain the second predetermined threshold. The optimized threshold not only ensures statistical significance, but also ensures that the threshold can reflect the distribution of actual data through data distribution analysis and optimization of the corrected second threshold. Combining the two, the accuracy of the second predetermined threshold is improved. The whole process is scientific and reasonable, and provides data support for the subsequent screening of relevant characteristic parameters.
[0123] Finally, the mutual information value of the non-correlated feature parameters is greater than or equal to the second predetermined threshold, and the non-correlated feature parameters corresponding to the current mutual information value are defined as relevant feature parameters; if the mutual information value of the non-correlated feature parameters is less than the second predetermined threshold, the non-correlated feature parameters corresponding to the current mutual information value do not change. This embodiment eliminates the non-correlated feature parameters whose mutual information values are less than the second predetermined threshold, and does not consider this parameter in the subsequent energy consumption value prediction process.
[0124] Step 36, constructing a feature vector according to the relevant feature parameters in steps 33 and 35.
[0125] Quickly eliminate those characteristic parameters that are almost irrelevant to the energy consumption value, use the second predetermined threshold to further eliminate those characteristic parameters with weak correlation with the target variable, retain those characteristic parameters with high correlation, and finally construct a characteristic vector with these characteristic parameters. The characteristic vector is expressed as: X = (x1, x2, x3, ..., x n ), x n Represents the value of the nth characteristic parameter.
[0126] Step 4: input the feature vector into a pre-trained machine learning model to predict the energy consumption value in the production process.
[0127] Load a pre-trained machine learning model that has been trained on historical data and can predict energy consumption values in the production process. Common machine learning models include linear regression, support vector machine (SVM), decision tree, random forest, gradient boosted tree (GBT), neural network, etc. Input the feature vector X into the model, and the model will make predictions based on the values in the feature vector. The model will calculate the predicted energy consumption value based on the values in the feature vector and the weights, biases and other parameters in the model. Since the feature vector is based on real-time data to obtain parameter values, the energy consumption value at the current moment in the production process is obtained through the pre-trained machine learning prediction. According to the predicted energy consumption value at the current moment in the production process, corresponding measures can be taken to optimize the production process.
[0128] Step 5: predict future energy consumption values based on the output results of the machine learning model.
[0129] The feature vector X and the predicted energy consumption value at the current moment are input into the pre-trained machine learning model, and the energy consumption value at the next moment is output. The real-time data at the next moment is obtained and a new feature vector is constructed based on it. The new feature vector and its corresponding energy consumption value are input into the pre-trained machine learning model, and the energy consumption at the next moment is output. The above steps are repeated until the energy consumption values of all required future time points are predicted.
[0130] The filtered feature vector contains the characteristic parameters most relevant to energy consumption and removes irrelevant or redundant information, so that the model can more accurately capture the essential laws of energy consumption changes, which helps to improve the accuracy of the prediction model and reduce noise interference caused by irrelevant features.
[0131] The filtered feature vectors are more stable and less susceptible to individual outliers, making the model more robust when facing different data distributions.
[0132] The filtered feature vectors focus on factors that are closely related to energy consumption, helping the model to more accurately capture key information in the production process and improve the prediction accuracy of energy consumption values.
[0133] Refer to the attached Figure 2 This embodiment also provides a production energy consumption value prediction system based on artificial intelligence, including the following modules:
[0134] Collection module: used to collect real-time data during the production process;
[0135] Preprocessing module: connected with the collection module, used to preprocess the real-time data;
[0136] Feature vector construction module: connected with the preprocessing module, used to construct feature vectors according to the preprocessed real-time data;
[0137] Energy consumption value prediction module: connected to the feature vector construction module, used to input the feature vector into a pre-trained machine learning model to predict the production process and future energy consumption values.
[0138] The above description is only a preferred specific implementation manner of the present application, and is not intended to limit the protection scope of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the protection scope of the present application.
Claims
1. A production energy consumption value prediction method based on artificial intelligence, characterized in that: The following steps are involved: Step 1, collect real-time data during the production process; Step 2: preprocess the real-time data; Step 3, constructing a feature vector based on the preprocessed real-time data; Step 31, obtaining dynamic data according to the preprocessed real-time data, and obtaining candidate feature parameters according to the dynamic data; Step 32, calculating the correlation coefficient between the candidate characteristic parameter and the energy consumption value; Step 33, constructing a first predetermined threshold value for the correlation coefficient, determining whether the absolute value of the correlation coefficient is greater than or equal to the first predetermined threshold value, and if so, defining the candidate feature parameter corresponding to the current correlation coefficient as a relevant feature parameter; If not, the candidate feature parameter corresponding to the current correlation coefficient is defined as a non-correlated feature parameter, and the process goes to step 34; Step 34, calculating the mutual information value for the non-correlated feature parameters; Step 35, constructing a second predetermined threshold for the mutual information value, and determining whether the mutual information value is greater than or equal to the second predetermined threshold, if so, defining the non-correlated feature parameters corresponding to the current mutual information value as relevant feature parameters; if not, the non-correlated feature parameters corresponding to the current mutual information value remain unchanged; Step 36, constructing a feature vector according to the relevant feature parameters in step 33 and step 35; Step 4, inputting the feature vector into a pre-trained machine learning model to predict the energy consumption value in the production process; Step 5: predict future energy consumption values based on the output results of the machine learning model.
2. The method for predicting production energy consumption based on artificial intelligence according to claim 1, characterized in that: Calculation of correlation coefficients, including: Step 321, calculating the covariance between the candidate parameter and the energy consumption value; Step 322, calculating the standard deviation of the candidate parameter; Step 323, calculating the standard deviation of the energy consumption value; Step 324, obtaining a correlation coefficient according to the covariance, the standard deviation of the candidate parameters, and the standard deviation of the energy consumption value.
3. The method for predicting production energy consumption based on artificial intelligence according to claim 1, characterized in that: Constructing a second predetermined threshold for the mutual information value specifically includes: Step 351, obtaining a second threshold; Step 352, modifying the second threshold to obtain a modified second threshold; Step 353: Optimize the modified second threshold to obtain a second predetermined threshold.
4. The method for predicting production energy consumption based on artificial intelligence according to claim 3 is characterized in that: Correcting the second threshold according to the first predetermined threshold to obtain a corrected second threshold specifically includes: Step 3521, calculating the mutual information value for the relevant feature parameters in step 33; Step 3522, obtaining a mutual information value mean according to the mutual information value of step 353, and obtaining a mutual information value standard deviation according to the mutual information value mean as a correction item; Step 3523, obtaining a corrected second threshold value according to the second threshold value and the correction item.
5. The method for predicting production energy consumption based on artificial intelligence according to claim 3 is characterized in that: Optimizing the modified second threshold to obtain a predetermined second threshold includes: Step 3531, dividing the value range of the mutual information value of the unrelated characteristic parameters into a plurality of intervals of equal width; the width of the interval is N; Step 3532, determining the number of mutual information values in each interval and the number of all mutual information values; constructing a distribution density function according to the width of each interval and the number of all mutual information values; Step 3533, defining the interval with the largest number of mutual information values as the pending interval; Step 3534, calculating the distribution density value for each mutual information value in the to-be-determined interval according to the distribution density function, if the distribution density values of 3 to (Q-1) mutual information values in the to-be-determined interval reach three times the distribution density values of other mutual information values in the to-be-determined interval, the to-be-determined interval is defined as a determined interval; Q represents the maximum number of mutual information values in the to-be-determined interval; Step 3535, searching for a mutual information value that is the same as the modified second threshold value in the determined interval. If a mutual information value that is the same as the modified second threshold value is found, the distribution density value of the mutual information value that is the same as the modified second threshold value is used as the optimization value. If no mutual information value is found, returning to step 3531; Step 3536, obtaining a predetermined second threshold value based on the modified second threshold value and the optimized value.
6. The method for predicting production energy consumption based on artificial intelligence according to claim 5 is characterized in that: The distribution density function is: In the formula, f i represents the distribution density function of the ith interval, Q i represents the number of mutual information values in the i-th interval, ΔI represents the interval width, and Z represents the maximum number of intervals.
7. The method for predicting production energy consumption based on artificial intelligence according to claim 2 is characterized in that: The calculation formula of the correlation coefficient is: R XY represents the correlation coefficient between the candidate parameter X and the energy consumption value Y, Cov(X, Y) represents the covariance between the candidate parameter X and the energy consumption value Y, σ X represents the standard deviation of the candidate parameter X, σ Y Represents the standard deviation of the energy consumption value Y.
8. The method for predicting production energy consumption based on artificial intelligence according to claim 5 is characterized in that: The calculation formula of the predetermined second threshold is: In the formula, represents a predetermined second threshold value, represents the modified second threshold, Represents the modified second threshold The density function value of .
9. A production energy consumption value prediction system based on artificial intelligence, used to execute the production energy consumption value prediction method based on artificial intelligence described in any one of claims 1 to 8, characterized in that: Includes the following modules: Collection module: used to collect real-time data during the production process; Preprocessing module: connected with the collection module, used to preprocess the real-time data; Feature vector construction module: connected with the preprocessing module, used to construct feature vectors according to the preprocessed real-time data; Energy consumption value prediction module: connected to the feature vector construction module, used to input the feature vector into a pre-trained machine learning model to predict the production process and future energy consumption values.