An intelligent prediction system for energy consumption in industrial production processes and its prediction algorithm

Abstract

The present invention discloses an intelligent prediction system for energy consumption in an industrial production process and a prediction algorithm thereof, and specifically relates to the technical field of production energy consumption prediction, and is used to solve the problem of inaccurate energy consumption prediction in industrial production. The present invention collects energy consumption characteristic variables and energy data information in the process of energy conversion, distribution and utilization of production equipment, organizes the collected energy data into a time-space matrix, uses a sliding window to control the number of data streams, and uses a K nearest neighbor algorithm to identify abnormal data in the energy data. An abnormal data is interpolated during the compression process of the energy data through a sliding window anomaly detection method of state changes, thereby improving the integrity of the data for prediction, and then analyzing the changing law of the energy consumption collected in the industrial production process, and timely adjusting the process where the problem occurs, thereby improving the accuracy and sensitivity of the prediction, effectively controlling energy consumption and production costs, and improving the benefits of the production enterprise.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial production energy consumption prediction, and more specifically, to an industrial production process energy consumption intelligent prediction system and a prediction algorithm thereof. Background Art

[0002] In the industrial production process, a mechanism is usually adopted to arrange tasks according to the specific production targets of production orders. However, this production order-driven approach may lead to huge equipment energy consumption, thereby significantly increasing the production costs of enterprises. Since the current mechanism mainly relies on specific production tasks and lacks a clear prediction of the energy consumption value of production equipment, it cannot effectively alleviate the problem of excessive energy consumption of production equipment, resulting in unnecessary energy waste.

[0003] By adopting a mechanism that drives production behavior based on the energy consumption value of production equipment, energy consumption can be effectively controlled, that is, the energy consumption value of production equipment can be accurately predicted. Enterprises can arrange production tasks and production process details more specifically, thereby effectively controlling the consumption of excess energy. The lack of clear and reliable predictions in the industrial production process makes it difficult for enterprises to effectively control energy consumption and production costs, which has an adverse impact on the benefits of the enterprise. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an intelligent prediction system for energy consumption in an industrial production process and a prediction algorithm thereof. By analyzing the energy data information collected from the production equipment, the collected energy data is organized into a space-time matrix, and then the energy data is interpolated for abnormal data through a sliding window anomaly detection method. The energy data is input into a differential autoregressive sliding average model for training to determine the training status. When a deviation analysis signal is generated for the energy consumption prediction in the industrial production process, the source device is traced and the energy consumption data at subsequent times is recorded and adjusted accordingly to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] An intelligent prediction algorithm for energy consumption in an industrial production process comprises the following steps:

[0007] In the industrial production process, the energy conversion, distribution and utilization process of production equipment is analyzed, the characteristic variables of energy consumption in the production process are extracted, and energy data is collected according to time;

[0008] The collected energy data is organized into a spatiotemporal matrix, a sliding window is used to control the number of data streams, a distance-based anomaly detection algorithm is used to analyze state mutations, and a K-nearest neighbor algorithm is used to identify abnormal data in energy data;

[0009] The abnormal data of energy data is interpolated during the compression process through the sliding window anomaly detection method of state change, and the energy data is input into the differential autoregressive sliding average model for training to determine the training status;

[0010] Analyze the changing patterns of energy consumption collected during industrial production, obtain energy consumption information during industrial production, and determine the state of energy consumption forecast;

[0011] When the energy consumption forecast in the industrial production process produces a deviation analysis signal, the traceability equipment records the energy consumption data at subsequent times for analysis and adjusts the equipment according to the analysis results.

[0012] In a preferred embodiment, the collected energy data is organized into a time-space matrix, and a sliding window is used to control the number of data streams. The specific process is as follows:

[0013] Collect energy data through sensors, obtain the energy data collected by each sensor according to time, and organize the energy data into a time-space matrix;

[0014] The spatiotemporal data matrix is ​​divided into blocks according to time and space, and each block is subjected to a two-dimensional wavelet transform;

[0015] The threshold is obtained according to the soft threshold or hard threshold calculation, and the coefficients in the wavelet coefficient matrix whose absolute values ​​are less than the threshold are set to zero;

[0016] The truncated wavelet coefficients are inversely transformed to reconstruct the compressed spatiotemporal data to obtain compressed spatiotemporal data blocks. A sliding window is used to control the number of data streams according to the time correlation.

[0017] In a preferred embodiment, a distance-based anomaly detection algorithm is used to analyze state mutations, and a K-nearest neighbor algorithm is used to identify abnormal data in energy data. The specific process is as follows:

[0018] Obtain point anomalies in noise mode during energy data compression;

[0019] Obtain a data set containing energy data, wherein the data set contains labels of waiting idle and normal working states during process switching;

[0020] Select features from the dataset for distance calculation, and use Manhattan distance for each data object;

[0021] Sort the distance of each data object and find the K neighbors closest to the data object;

[0022] According to the ratio of normal data to abnormal data in the K nearest neighbors, labels are assigned to data objects, and energy data with labels as abnormal are regarded as abnormal data.

[0023] In a preferred embodiment, the variation pattern of energy consumption collected during the industrial production process is analyzed to obtain energy consumption information during the industrial production process and determine the state of energy consumption prediction. The specific process is as follows:

[0024] Analyze the changing patterns of energy consumption collected during industrial production to obtain energy consumption information during industrial production. The energy consumption information includes energy consumption change information, forecast adjustment information, and forecast error information.

[0025] Energy consumption change information includes one-way energy consumption deviation value, forecast adjustment information includes adjustment frequency floating value, forecast error information includes energy consumption forecast variation index;

[0026] The prediction control coefficient is obtained by comprehensively calculating the obtained one-way energy consumption deviation value, the adjustment frequency floating value and the energy consumption prediction variation index;

[0027] The one-way energy consumption deviation value, adjustment frequency floating value, and energy consumption forecast variation index are proportional to the forecast control coefficient;

[0028] The generated prediction control coefficient is compared with the control threshold to generate a deviation analysis signal and a prediction stability signal to determine the state of the energy consumption prediction.

[0029] In a preferred embodiment, the generated prediction control coefficient is compared with the control threshold, a deviation analysis signal and a prediction stability signal are generated, and the state of energy consumption prediction is determined. The specific process is as follows:

[0030] If the predicted control coefficient is greater than or equal to the control threshold, a deviation analysis signal is generated;

[0031] If the predicted control coefficient is less than the control threshold, a predicted stability signal is generated;

[0032] The metabolic equivalents of the deviation analysis signals generated in the sleeping state are re-estimated, the data collection parameters are readjusted, and the metabolic equivalents are re-estimated according to the historical metabolic equivalent data to obtain a new metabolic equivalent value.

[0033] In a preferred embodiment, when the energy consumption forecast in the industrial production process generates a deviation analysis signal, the traceability device records the energy consumption data at subsequent times for analysis, and adjusts the equipment according to the analysis results. The specific process is as follows:

[0034] Trace the data that generates the deviation analysis signal, find the corresponding sensor device and mark it immediately, and generate the prediction control coefficient for the energy consumption data collected by the sensor at subsequent times, and establish a prediction control coefficient data set;

[0035] Calculate the mean and standard deviation of the predicted control coefficients in the data set;

[0036] For the element data in the data set, calculate the deviation between the element data and the mean value to obtain the outlier degree value;

[0037] Compare the outlier degree value of the data in the prediction control coefficient data set with the set outlier threshold. When the outlier degree value of the data in the data set is greater than or equal to the discrete threshold, the data is recorded as an outlier.

[0038] When the number of outliers is greater than or equal to the set outlier number threshold, it is determined that there is a problem with the energy consumption forecast, an early warning is performed, and maintenance personnel are notified to inspect and maintain the model or related equipment.

[0039] An industrial production process energy consumption intelligent prediction system, used for the above industrial production process energy consumption intelligent prediction algorithm, comprising:

[0040] A data collection module is used to collect energy consumption characteristic variables and energy data information during the energy conversion, distribution and utilization of production equipment according to time;

[0041] The data analysis module is used to organize the collected energy data into a spatiotemporal matrix, use a sliding window to control the number of data streams, use a distance-based anomaly detection algorithm to analyze state mutations, use a K-nearest neighbor algorithm to identify abnormal data in energy data, interpolate abnormal data in the energy data during compression using a sliding window anomaly detection method for state changes, and input the energy data into a differential autoregressive sliding average model for training to determine the training status;

[0042] The energy consumption comparison module is used to analyze the changing rules of energy consumption collected during the industrial production process, obtain energy consumption information during the industrial production process, and determine the state of energy consumption prediction;

[0043] The prediction and control module is used to trace the source equipment and record the energy consumption data at subsequent times and make corresponding adjustments when the energy consumption forecast in the industrial production process generates deviation analysis signals.

[0044] Technical effects and advantages of the present invention:

[0045] The present invention collects energy consumption characteristic variables and energy data information during the energy conversion, distribution and utilization of production equipment, organizes the collected energy data into a time-space matrix, uses a sliding window to control the number of data streams, and uses a K-nearest neighbor algorithm to identify abnormal data in the energy data. The energy data is interpolated for abnormal data during the compression process through a state-changing sliding window anomaly detection method, thereby improving the integrity of the data for prediction, and then analyzing the changing laws of the energy consumption collected during the industrial production process, and timely adjusting the process where problems occur, thereby improving the accuracy and sensitivity of the prediction, effectively controlling energy consumption and production costs, and improving the benefits of production enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The present invention is a flow chart of an intelligent prediction algorithm for energy consumption in an industrial production process.

[0047] Figure 2 The present invention is a schematic diagram of the structure of an intelligent prediction system for energy consumption in an industrial production process. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Example 1

[0050] like Figure 1 As shown, an intelligent prediction algorithm for energy consumption in industrial production processes includes:

[0051] In the industrial production process, the energy conversion, distribution and utilization process used by production equipment is analyzed, the energy consumption characteristic variables in the production process are extracted, and energy data is collected according to time, as follows:

[0052] Industrial production driven by orders is prone to frequent starting and stopping of production equipment. Because the production requirements of each order may be different, production equipment usually consumes more energy during the start-up and stop process, and this energy consumption may not be fully compensated by the realization of production goals in a short period of time, thus causing excessive energy consumption; changes in industrial production orders lead to large intervals between production tasks of different products, and also cause equipment to be idle between production tasks. Equipment may still consume a certain amount of energy in an idle state, but cannot generate corresponding output.

[0053] In the industrial production process, energy is transported through the energy supply system, including the use of electricity, gas, steam and other energy sources to complete specific production, analyze the energy conversion, distribution and utilization process used by production equipment, and determine the energy consumption of equipment in each process;

[0054] Extract energy consumption characteristic variables of a certain process or the entire process in the industrial production process, and predict the energy consumption of subsequent processes based on the extracted energy consumption characteristics;

[0055] For example, the computer industrial numerical control system is a program language that is compiled by the industrial CNC machine tool. It outputs the action instructions for the servo drive motors of the machine tool spindle and feed axis, and controls other auxiliary devices to perform reasonable and orderly processing actions. For different stages of the entire working cycle, including power-on, preheating, actual processing and cooling, real-time monitoring of the power consumption of the CNC machine tool system is an important energy consumption feature. By collecting data such as the current and voltage of the spindle, feed axis and other drive motors, the actual power consumption can be calculated.

[0056] Various sensors (voltage, current sensors, etc.) are installed on industrial energy equipment for data collection, and the collected energy data is transmitted for energy consumption analysis. In an environment where multiple sensor devices collect data, due to the high collection frequency and large amount of data collected, the data can be first transmitted to the edge computing for processing, and then transmitted to the summary analysis end for processing of data collected by multiple sensor devices.

[0057] When performing rated frequency acquisition, the amount of data transmitted by each sensor is large. However, when analyzing the collected energy data for a long time, the huge amount of data becomes the main reason for excessive pressure on network transmission and data storage. To alleviate this problem, data compression measures must be taken at the edge to effectively reduce the amount of data. In the process of data compression, wavelet transform is used for data compression. The specific process is as follows:

[0058] Energy data is collected through sensors. The energy data collected by each sensor according to time is obtained. The energy data is organized into a spatiotemporal matrix, where rows represent time points and columns represent different sensors. This matrix is ​​represented as DJ. The spatiotemporal data matrix DJ is divided into blocks according to time and space. Each block can be represented as B i，j , where i represents the time block and j represents the sensor block. Perform a two-dimensional wavelet transform on each block and transform the DJ matrix into a BH matrix, BH = Z*D*Z T , where Z is the wavelet transform matrix, Z T is the transpose of Z;

[0059] The threshold is calculated according to the soft threshold or hard threshold. The soft threshold is to set the coefficients less than the threshold to zero and scale the coefficients greater than the threshold; the hard threshold directly sets the coefficients less than the threshold to zero, retains the main part of the wavelet coefficients, truncates the wavelet coefficient matrix, and sets the coefficients in the wavelet coefficient matrix whose absolute values ​​are less than the threshold to zero according to the calculated threshold;

[0060] The truncated wavelet coefficients are inversely transformed to reconstruct the compressed spatiotemporal data and obtain compressed spatiotemporal data blocks.

[0061] By comparing the errors before and after compression, we can adjust the parameters of wavelet transform or consider other optimization methods according to the needs to balance the compression rate and data accuracy. However, in the actual production process, energy data may be in an abnormal state due to environmental factors such as noise. The wavelet transform lacks the ability to detect abnormal data, resulting in reduced compression performance.

[0062] Sensor nodes collect data periodically, and the data collected before and after show significant time correlation. When performing data anomaly analysis through edge devices, it is difficult to process all data streams collected by sensors at the same time, so sliding windows are used to control the number of data streams. The results of anomaly detection are affected by the previous and next data points. A distance-based anomaly detection algorithm is used. For data objects, in the process of entering and leaving all sliding windows, only when the data object behaves normally during this period of time, it is judged as a normal value; otherwise, it is identified as an anomaly.

[0063] In actual production, production equipment is divided into two main states: process switching, waiting for idling, and normal operation. Between these two states, energy data values ​​often undergo discontinuous changes, which are often easily misunderstood as abnormal values. In the process of energy data compression, the main impact on the compression effect comes from the point anomalies in the noise mode described in the analysis of energy data characteristics. Therefore, when dealing with this state mutation, special processing is required to avoid misidentifying it as an abnormality.

[0064] Anomaly detection processing is mainly based on different situations with or without state mutation. If it is based on state mutation anomaly detection processing, the K nearest neighbor algorithm can help identify anomalies in the data. The specific steps are as follows:

[0065] Obtain a dataset containing energy data and ensure that the dataset contains labels for waiting idle and normal working status during process switching;

[0066] Select features for distance calculation from the data set. The selected features can capture the mutations that occur between process switching and normal working state, so as to obtain data of the corresponding features.

[0067] For each data object, calculate the distance between it and all adjacent data objects and use Manhattan distance for measurement. Manhattan distance is the sum of the absolute values ​​of the distance differences on each coordinate axis. For two points (x1, y1) and (x2, y2), MH = |x1-x2|+|y1-y2|;

[0068] Sort the distances of each data object to find the K nearest neighbors to the data object;

[0069] According to the ratio of normal data to abnormal data in the K nearest neighbors, a label is assigned to the current data object. The majority voting method can be used, that is, the current data is marked as the category with which the data in the K neighbors is more, and the data objects marked as abnormal are processed.

[0070] In the K nearest neighbor algorithm, the label assignment of anomaly detection is usually based on the distance and proportion of neighbors. Assume that N is the set of K nearest neighbors, where N normal is a set of normal data points, N anormal is a collection of abnormal data points. Then they can be classified by the following label assignment rules: normal | / K≥threshold, the data object is marked as normal data; otherwise, it is marked as abnormal data;

[0071] It should be noted that the threshold is a parameter set manually according to the actual demand scenario, which is used to adjust the abnormal judgment standard. Different application scenarios may require the threshold to be adjusted according to specific circumstances.

[0072] The anomaly detection algorithm mainly realizes the state mutation judgment and anomaly detection processing under different states. The state mutation judgment method is to make a threshold judgment based on the difference between the rated energy data of each device in the process switching waiting idle state and the normal working state. At the same time, it prevents the point anomaly generated by the noise mode from interfering with the state mutation judgment. It is stipulated that only when the number of mutation data is equal to the sliding window size, it is judged as a real state mutation;

[0073] The performance evaluation of anomaly detection algorithms mainly depends on two key indicators: recognition rate and false alarm rate. Among them, the recognition rate represents the ratio of correctly detected abnormal data to all abnormal data actually contained in the data set, while the false alarm rate represents the ratio of data that are incorrectly detected as abnormal to all actual detected abnormal data. Generally speaking, a higher recognition rate and a lower false alarm rate indicate good performance of the algorithm. The calculation formula of the recognition rate is: DR = TN / (TN+FN), and the calculation formula of the false alarm rate is: FR = FP / (FP+FN), where TN represents true negative (correctly identified abnormal data), FN represents false negative (abnormal data that is incorrectly identified as normal data), and FP represents false positive (normal data that is incorrectly identified as abnormal data).

[0074] The K nearest neighbor algorithm is optimized for the energy data after wavelet transformation, and dynamic adjustments are made during the compression process according to the actual fluctuation characteristics of the energy data to achieve data compression within a certain error range;

[0075] Before compressing energy data, we first use a sliding window anomaly detection method based on state changes. By adjusting the optimal parameters, we record the abnormal values ​​at specific time points. Since energy data has a strong time correlation, if a certain time point is detected as an abnormal point, it is assumed that under normal circumstances, the data value at this time point should be similar to the data values ​​at the previous and next time points.

[0076] Therefore, during the data compression process, the following processing method is adopted for these abnormal time points: the data value of the abnormal time point is replaced by the average value of the data values ​​before and after the abnormal time point, that is, the actual situation of the abnormal time point is better reflected by maintaining similarity, thereby maintaining data consistency during the compression process. Such a processing strategy helps to retain abnormal information in the compressed data and reduce the impact of abnormalities on the overall data compression effect to a certain extent.

[0077] The main performance indicators of compression algorithms are compression ratio and compression error. Compression ratio is used to measure the ability to compress a set of data, while compression error is used to measure the error between the actual value and the decompressed value. Generally speaking, a higher compression ratio and a lower compression error indicate that the compression effect of the algorithm is better, but in the actual production process, the requirement is to increase the compression ratio as much as possible while ensuring the compression error. The compression ratio represents the ratio of the data size before compression to the data size after compression. The specific calculation formula is as follows: CR = original data size / compressed data size;

[0078] The compression error is the difference between the compressed data and the original data, calculated using methods such as the root mean square error. Assuming N is the total number of data points, YS i is the i-th data point of the original data, YH i is the i-th data point of the compressed data, and the calculation formula of the compression error is:

[0079] The energy data is input into the difference autoregressive moving average model for training, and the non-stationary series is transformed into a stationary series through the difference method to realize the prediction of short-term future values.

[0080] Analyze the changing patterns of energy consumption collected during industrial production;

[0081] Obtain energy consumption information during industrial production, including energy consumption change information, forecast adjustment information, and forecast error information;

[0082] The energy consumption change information includes the one-way energy consumption deviation value and is calibrated as DXH, the forecast adjustment information includes the adjustment frequency floating value and is calibrated as TZP, and the forecast error information includes the energy consumption forecast variation index and is calibrated as NHY;

[0083] The one-way energy consumption deviation value in the energy consumption change information refers to the energy consumption performance deviation value of the system or equipment evaluated by comparing the difference between the actual energy consumption and the expected energy consumption in energy consumption management and optimization. It is usually used to measure the energy utilization efficiency of the system. The one-way energy consumption deviation value will affect the following aspects:

[0084] Energy cost: The one-way energy consumption deviation value reflects the difference between actual energy consumption and expected energy consumption. For enterprises, energy cost is an important economic factor. A larger one-way energy consumption deviation value leads to additional energy costs.

[0085] Environmental impact: Large deviation values ​​may lead to an increase in the carbon footprint generated by energy consumption and increase the environmental burden. By reducing energy consumption deviation, enterprises can better achieve environmental protection goals and reduce negative impacts on the environment.

[0086] The one-way energy consumption deviation value is obtained as follows:

[0087] Obtain the input energy consumption SR of the industrial production process per unit time, obtain the output energy LC, calculate the deep energy consumption conversion ratio: SD=LC / SR, obtain the total energy consumption coverage NF, obtain the actual energy consumption E=SR-LC, calculate the one-way energy consumption deviation value, and the calculation expression is: DXH=E*NF*SD / SR.

[0088] It should be noted that when measuring input energy, appropriate measuring devices are used to measure other input energy sources, such as steam, heat, electricity, etc., according to the type of energy used in the deep processing process. For example, the electricity consumption of deep processing equipment is measured, and the total electricity consumption of the equipment during deep processing is recorded using an electricity meter or electricity monitoring device; the measurement of output energy is the measurement of the energy of the workpiece or product produced after deep processing, which involves the use of energy metering equipment or laboratory testing of the output energy. The energy of waste heat or waste generated during the deep processing process can also be measured, and it can also be achieved through a calorimeter, temperature sensor or other waste treatment equipment to obtain the output energy statistically; the total energy consumption coverage refers to the normal operating state, standby state, shutdown state and other states in the production process. Each state corresponds to different energy consumption levels and weight characteristics. For example, in a production, the normal operating state accounts for 80% of the time, and the standby state accounts for 20%. The weight of the normal operating state is set to 0.1, and the standby state is set to 0.2, and the total energy consumption coverage is 0.084.

[0089] The adjustment frequency floating value in the forecast adjustment information refers to the frequency fluctuation of adjusting the model structure when the total energy consumption is estimated based on different forecast data in the energy consumption forecast model. The estimated result is inconsistent with the actual result, which affects the following aspects:

[0090] Data stability requirements: Low adjustment frequency may require less stability of input data, because the model is not easily affected by instantaneous changes in data. This may be more advantageous when facing noisy or poor quality input data;

[0091] Stability: Low adjustment frequency floating values ​​indicate the stability and accuracy of the forecasting model. Frequent adjustments may cause the model to overfit to specific data, while low-frequency adjustments help maintain the stability of the model and prevent overfitting.

[0092] The method for obtaining the adjustment frequency floating value is as follows:

[0093] Get the actual number of predictions made by the prediction model within a unit time t in the production cycle UB, get the preset prediction standard number CY, get the accurate number of energy consumption predictions ZC, calculate the prediction correlation value, the expression is: YC = (|UB-CY|+UB-ZC) / t, get the prediction correlation value of each unit time in the production cycle, and establish the prediction correlation value set YC i ={YC1, YC2, ..., YC n}, n is a positive integer, get the predicted correlation average value YC avg , get the total predicted range d, calculate the adjustment frequency floating value, the calculation expression is:

[0094] It should be noted that the production cycle includes multiple unit times, and the unit time is set according to actual needs. The total prediction range refers to the output range of the predicted value or the overall interval of the results when making a prediction, which includes all possible result ranges that the model can predict under specific conditions; the preset standard number of predictions can be obtained based on historical data, simulation models, industry standards or other appropriate methods.

[0095] The energy consumption prediction variation index in the prediction error information is an indicator of the degree of variation of the energy consumption prediction results, which is used to evaluate the accuracy and stability of the energy consumption prediction model. The smaller the variation index, the more stable the prediction results are, and the smaller the deviation from the actual energy consumption; conversely, a larger variation index may mean that the prediction results are more unstable and the deviation from the actual energy consumption is larger. The energy consumption prediction variation index has the following effects:

[0096] Improve energy efficiency: Stable energy consumption forecast results help to plan and control energy use more accurately, thereby improving energy efficiency. Accurate energy consumption forecasts can guide energy consumption optimization in the production process, avoid energy waste and reduce production costs.

[0097] Guiding optimization decisions: In production planning and energy management, a smaller energy consumption forecast variation index can provide more stable forecast results, help to formulate more reliable production plans and energy optimization strategies, and stable forecast results can help reduce uncertainty and improve the accuracy of production decisions.

[0098] The energy consumption prediction variation index is obtained as follows:

[0099] Obtain the actual energy consumption and predicted energy consumption at each time point, calculate the energy consumption residual CN between the actual energy consumption and the predicted energy consumption, energy consumption residual = actual energy consumption - predicted energy consumption, calculate the average actual energy consumption The energy consumption residuals at all time points are combined into a sequence, and the average energy consumption residual of the sequence is calculated. Calculate the standard deviation of energy consumption of the sequence, the expression is: M represents the total number of time points. The total number of predictions for all time points is obtained, the energy consumption sensitivity frequency PZ is obtained, and the energy consumption prediction variation index is calculated. The calculation expression is:

[0100] It should be noted that energy consumption sensitivity refers to the sensitivity of a system or process to the energy consumption caused by changes in its input or internal parameters. In industrial production or other systems, energy consumption sensitivity is used to describe the changes in the energy efficiency of the system under different conditions. Energy consumption sensitivity can be expressed as the rate of change of the system output (usually energy consumption) relative to the input or parameter, that is, the update of real-time input data affects the change of energy consumption forecast value in real time.

[0101] Summarize and analyze energy consumption change information, forecast adjustment information, and forecast error information;

[0102] The predicted control coefficient is obtained by comprehensively calculating the one-way energy consumption deviation value DXH, the adjustment frequency floating value TZP and the energy consumption prediction variation index NHY, and the predicted control coefficient is calibrated as Y x , the expression is: x =ln(r1*DXH+r2*TZP+r3*NHT+1), where Y x is the prediction control coefficient, r1, r2, r3 are the preset proportional coefficients of the one-way energy consumption deviation value DXH, the adjustment frequency floating value TZP and the energy consumption prediction variation index NHY, and r1, r2, r3 are all greater than zero.

[0103] It should be noted that the size of the preset proportional coefficient is to quantify each parameter to obtain a specific numerical value. In order to facilitate subsequent comparison, the size of the coefficient depends on the amount of sample data and the preliminary setting of the corresponding preset proportional coefficient for each group of sample data by technical personnel in this field. It is not unique, as long as it does not affect the proportional relationship between the parameter and the quantized numerical value, such as adjusting the frequency floating value, the energy consumption prediction variation index and the proportional relationship between the prediction control coefficient.

[0104] The larger the one-way energy consumption deviation value, the larger the adjustment frequency floating value, and the larger the energy consumption forecast variation index, that is, the larger the performance value of the prediction control coefficient, it indicates that when making energy consumption forecasts, it is more likely that the predicted energy consumption data will be unstable; the smaller the one-way energy consumption deviation value, the smaller the adjustment frequency floating value, and the smaller the energy consumption forecast variation index, that is, the smaller the performance value of the prediction control coefficient, it indicates that when making energy consumption forecasts, the predicted energy consumption data will be more stable.

[0105] The generated predicted control coefficient is compared with the control threshold to generate a deviation analysis signal and a predicted stability signal.

[0106] After obtaining the predicted control coefficient, the predicted control coefficient is compared with the control threshold;

[0107] If the prediction control coefficient is greater than or equal to the control threshold, a deviation analysis signal is generated, indicating that an energy consumption forecast error occurred during the energy consumption forecast in the production process, and it is necessary to tilt resources and improve monitoring sensitivity to obtain more accurate data;

[0108] If the predicted control coefficient is less than the control threshold, a predicted stability signal is generated, indicating that the energy consumption data prediction is relatively stable.

[0109] When deviation analysis signals are generated in the industrial production process, the process is handled according to the following steps:

[0110] The data that generates the deviation analysis signal is traced to the source, the corresponding sensor equipment is found and marked immediately, and the prediction control coefficient generated by the energy consumption data collected by the sensor at subsequent moments is used to establish a prediction control coefficient data set. By calculating the mean and standard deviation in the data set, the outlier degree value of each prediction control coefficient is obtained to determine the energy consumption status of the production process;

[0111] Calculate the mean and standard deviation of the predicted control coefficients in the data set;

[0112] For each data, calculate its deviation from the mean to obtain the outlier degree value. The specific formula for obtaining the outlier degree value is: Where R is the data point in the prediction control coefficient data set, is the mean of the predicted control coefficient data set, σ is the standard deviation of the predicted control coefficient data set;

[0113] The outlier degree value of the data in the prediction and control coefficient data set is compared with the set outlier threshold. When the outlier degree value of the data in the data set is greater than or equal to the discrete threshold, it indicates that the outlier degree of the prediction and control coefficient of energy consumption is too large, and the data is recorded as an outlier. When the number of outliers is greater than or equal to the set outlier number threshold, it is determined that there is a problem with the energy consumption prediction, and early warning processing is performed, and maintenance personnel are notified to inspect and maintain the model or related equipment to prevent problems from occurring during subsequent use and affecting the accuracy of energy consumption prediction.

[0114] It should be noted that the threshold settings in this embodiment are set according to actual conditions and are not fixed values, so no further analysis will be performed here.

[0115] The present invention collects energy consumption characteristic variables and energy data information during the energy conversion, distribution and utilization of production equipment, organizes the collected energy data into a time-space matrix, uses a sliding window to control the number of data streams, and uses a K-nearest neighbor algorithm to identify abnormal data in the energy data. The energy data is interpolated for abnormal data during the compression process through a state-changing sliding window anomaly detection method, thereby improving the integrity of the data for prediction, and then analyzing the changing laws of the energy consumption collected during the industrial production process, and timely adjusting the process where problems occur, thereby improving the accuracy and sensitivity of the prediction, effectively controlling energy consumption and production costs, and improving the benefits of production enterprises.

[0116] Embodiment 2, this embodiment is a system embodiment of embodiment 1, which is used to implement an intelligent prediction algorithm for energy consumption in an industrial production process introduced in embodiment 1, such as Figure 2 As shown, specifically including:

[0117] A data collection module is used to collect energy consumption characteristic variables and energy data information in the energy conversion, distribution and utilization process of production equipment according to time;

[0118] The data analysis module is used to organize the collected energy data into a spatiotemporal matrix, use a sliding window to control the number of data streams, use a distance-based anomaly detection algorithm to analyze state mutations, use a K-nearest neighbor algorithm to identify abnormal data in energy data, interpolate abnormal data in the energy data during compression using a sliding window anomaly detection method for state changes, and input the energy data into a differential autoregressive sliding average model for training to determine the training status;

[0119] The energy consumption comparison module is used to analyze the changing rules of energy consumption collected during the industrial production process, obtain energy consumption information during the industrial production process, and determine the state of energy consumption prediction;

[0120] The prediction and control module is used to trace the source equipment and record the energy consumption data at subsequent times and make corresponding adjustments when the energy consumption forecast in the industrial production process generates deviation analysis signals.

[0121] The above formulas are all dimensionless and calculated numerically. Specific dimension removal can be achieved by various means such as standardization, which will not be elaborated here. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0122] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, an ATA hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.

[0123] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0124] Those of ordinary skill 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, or a combination of computer software and electronic hardware. 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.

[0125] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0126] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0127] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0128] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An intelligent prediction algorithm for energy consumption in industrial production processes, characterized in that: The steps include: In the industrial production process, the energy conversion, distribution and utilization process of production equipment is analyzed, the characteristic variables of energy consumption in the production process are extracted, and energy data is collected according to time; The collected energy data is organized into a spatiotemporal matrix, a sliding window is used to control the number of data streams, and a distance-based anomaly detection algorithm is used to analyze state mutations. If there is a state mutation, the K nearest neighbor algorithm is used to identify abnormal data in the energy data; The abnormal data of energy data are interpolated during the compression process by using a sliding window anomaly detection method based on state changes, and the interpolated energy data are input into a differential autoregressive sliding average model for training to determine the training status; Analyze the changing patterns of energy consumption collected during industrial production, obtain energy consumption information during industrial production, and determine the state of energy consumption forecast; When the energy consumption forecast in the industrial production process generates a deviation analysis signal, the traceability equipment is used to record the energy consumption data at subsequent times for analysis, and the equipment is adjusted according to the analysis results; Analyze the changing rules of energy consumption collected during the industrial production process, obtain energy consumption information during the industrial production process, and determine the state of energy consumption prediction. The specific process is as follows: Analyze the changing patterns of energy consumption in the processed energy data during the industrial production process to obtain energy consumption information during the industrial production process, including energy consumption change information, forecast adjustment information, and forecast error information; Energy consumption change information includes one-way energy consumption deviation value, forecast adjustment information includes adjustment frequency floating value, forecast error information includes energy consumption forecast variation index; The prediction control coefficient is obtained by comprehensively calculating the obtained one-way energy consumption deviation value, the adjustment frequency floating value and the energy consumption prediction variation index; The one-way energy consumption deviation value, adjustment frequency floating value, and energy consumption forecast variation index are proportional to the forecast control coefficient; The generated prediction control coefficient is compared with the control threshold to generate a deviation analysis signal and a prediction stability signal to determine the state of the energy consumption prediction.

2. The intelligent prediction algorithm for energy consumption in industrial production process according to claim 1 is characterized by: The collected energy data is organized into a spatiotemporal matrix, and a sliding window is used to control the number of data streams. The specific process is as follows: Collect energy data through sensors, obtain energy data collected by each sensor according to time, and organize the energy data into a time-space matrix; The spatiotemporal data matrix is ​​divided into blocks according to time and space, and each block is subjected to a two-dimensional wavelet transform; The threshold is calculated according to the soft threshold or hard threshold, and the coefficients in the wavelet coefficient matrix whose absolute values ​​are less than the threshold are set to zero; The truncated wavelet coefficients are inversely transformed to reconstruct the compressed spatiotemporal data to obtain compressed spatiotemporal data blocks. A sliding window is used to control the number of data streams according to the time correlation.

3. The intelligent prediction algorithm for energy consumption in industrial production process according to claim 2 is characterized by: The distance-based anomaly detection algorithm is used to analyze state mutations. If there is a state mutation, the K nearest neighbor algorithm is used to identify abnormal data in the energy data. The specific process is as follows: Obtain point anomalies in noise mode during energy data compression; Obtain a data set containing energy data, wherein the data set contains labels of waiting idle and normal working states during process switching; Select features from the dataset for distance calculation, and use Manhattan distance for each data object; Sort the distance of each data object and find the K nearest neighbors of the data object; According to the ratio of normal data to abnormal data in the K nearest neighbors, labels are assigned to data objects, and energy data with labels as abnormal are regarded as abnormal data.

4. The intelligent prediction algorithm for energy consumption in industrial production process according to claim 3 is characterized by: The generated prediction control coefficient is compared with the control threshold, a deviation analysis signal and a prediction stability signal are generated, and the state of energy consumption prediction is determined. The specific process is as follows: If the predicted control coefficient is greater than or equal to the control threshold, a deviation analysis signal is generated; If the predicted control coefficient is less than the control threshold, a predicted stability signal is generated; The metabolic equivalents of the deviation analysis signals generated in the sleeping state are re-estimated, the data collection parameters are readjusted, and the metabolic equivalents are re-estimated according to the historical metabolic equivalent data to obtain a new metabolic equivalent value.

5. The intelligent prediction algorithm for energy consumption in industrial production process according to claim 4 is characterized by: When the energy consumption forecast in the industrial production process generates a deviation analysis signal, the traceability equipment records the energy consumption data at subsequent times for analysis, and adjusts the equipment according to the analysis results. The specific process is as follows: Trace the data that generates the deviation analysis signal, find the corresponding sensor device and mark it immediately, and generate the prediction control coefficient for the energy consumption data collected by the sensor at subsequent times, and establish a prediction control coefficient data set; Calculate the mean and standard deviation of the predicted control coefficients in the data set; For the element data in the data set, calculate the deviation between the element data and the mean value to obtain the outlier degree value; Compare the outlier degree value of the data in the prediction control coefficient data set with the set outlier threshold. When the outlier degree value of the data in the data set is greater than or equal to the discrete threshold, the data is recorded as an outlier. When the number of outliers is greater than or equal to the set outlier number threshold, it is determined that there is a problem with the energy consumption forecast, an early warning is performed, and maintenance personnel are notified to inspect and maintain the model or related equipment.

6. An intelligent prediction system for energy consumption in industrial production processes, used to implement an intelligent prediction algorithm for energy consumption in industrial production processes according to any one of claims 1 to 5, characterized in that: include: A data collection module is used to collect energy consumption characteristic variables and energy data information during the energy conversion, distribution and utilization of production equipment according to time; The data analysis module is used to organize the collected energy data into a spatiotemporal matrix, use a sliding window to control the number of data streams, and use a distance-based anomaly detection algorithm to analyze state mutations. If there is a state mutation, the K nearest neighbor algorithm is used to identify abnormal data in the energy data. The energy data is interpolated during the compression process through a sliding window anomaly detection method of state changes, and the interpolated energy data is input into a differential autoregressive sliding average model for training to determine the training status; The energy consumption comparison module is used to analyze the changing rules of energy consumption collected during the industrial production process, obtain energy consumption information during the industrial production process, and determine the state of energy consumption prediction; The prediction and control module is used to trace the source equipment and record the energy consumption data at subsequent moments and make corresponding adjustments when the energy consumption forecast in the industrial production process generates deviation analysis signals; Analyze the changing rules of energy consumption collected during the industrial production process, obtain energy consumption information during the industrial production process, and determine the state of energy consumption prediction. The specific process is as follows: Analyze the changing patterns of energy consumption in the processed energy data during the industrial production process to obtain energy consumption information during the industrial production process, including energy consumption change information, forecast adjustment information, and forecast error information; Energy consumption change information includes one-way energy consumption deviation value, forecast adjustment information includes adjustment frequency floating value, forecast error information includes energy consumption forecast variation index; The prediction control coefficient is obtained by comprehensively calculating the obtained one-way energy consumption deviation value, the adjustment frequency floating value and the energy consumption prediction variation index; The one-way energy consumption deviation value, adjustment frequency floating value, and energy consumption forecast variation index are proportional to the forecast control coefficient; The generated prediction control coefficient is compared with the control threshold to generate a deviation analysis signal and a prediction stability signal to determine the state of the energy consumption prediction.

Images