Enterprise device performance dynamic prediction method and device, electronic device, and storage medium

Abstract

The application relates to the technical field of computer application, and particularly discloses an enterprise equipment efficiency dynamic prediction method and device, electronic equipment and a storage medium, wherein the method comprises the following steps: obtaining original data of different types of equipment, and extracting original feature vectors of the equipment; calculating weight coefficients of the equipment based on historical contribution degrees, current health states and real-time loads of the equipment, and performing weighted aggregation on the original feature vectors to obtain dynamic weighted feature vectors at an equipment level; aggregating to form a production line layer feature vector; and inputting the production line layer feature vector into a pre-constructed equipment efficiency prediction model to generate an equipment efficiency prediction result; the processing procedure of the method involves hierarchical aggregation and dynamic weighting, can effectively utilize multi-source heterogeneous equipment data, and can realize dynamic prediction of equipment efficiency on the edge side with limited resources, so that dynamic and accurate prediction of equipment efficiency can be realized, and technical support can be provided for intelligent upgrading of a factory.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer application, in particular to an enterprise equipment efficiency dynamic prediction method and device, electronic equipment and storage medium. BACKGROUND

[0002] The traditional enterprise equipment efficiency prediction method has long dominated the industrial field, and its core relies on centralized servers for large-scale data processing and complex model calculation.

[0003] However, the existing equipment efficiency prediction method largely ignores the dynamic change characteristics of the equipment running state. For example, on a complex production line, different equipment undertakes different tasks and has different importance, and the real-time load of the equipment itself also fluctuates with the change of production tasks. However, the traditional prediction method often fails to fully consider these key factors, resulting in a deviation between the prediction model and the actual running state of the equipment, and the prediction result is difficult to accurately reflect the real situation of the equipment efficiency, and the prediction accuracy and real-time performance are difficult to meet the needs of actual application.

[0004] At present, there is no effective technical solution to the above problems. SUMMARY

[0005] The purpose of the present application is to provide an enterprise equipment efficiency dynamic prediction method, device, electronic equipment and storage medium, so as to realize dynamic and accurate prediction of equipment efficiency under the premise of paying attention to the dynamic change characteristics of the equipment running state.

[0006] In a first aspect, the present application provides an enterprise equipment efficiency dynamic prediction method for equipment efficiency prediction, comprising the following steps:

[0007] S1, obtaining raw data of different types of equipment, and performing data cleaning, format unification processing on the raw data, extracting the original feature vector of each equipment, the original feature vector representing the key feature parameters of the corresponding equipment running state;

[0008] S2, calculating the weight coefficient of each equipment based on the historical contribution degree, current health state and real-time load of each equipment, and weighting and aggregating the original feature vectors of the equipment at the same level on the production line according to the weight coefficient, to obtain the dynamic weighted feature vector of the equipment level;

[0009] S3, according to the process flow of the production line, aggregating the dynamic weighted feature vectors of the equipment on the same production line to form a production line layer feature vector;

[0010] S4, inputting the production line layer feature vector into a pre-constructed equipment efficiency prediction model to generate an equipment efficiency prediction result.

[0011] The enterprise equipment efficiency dynamic prediction method of the present application realizes dynamic configuration of the weight coefficient by comprehensively considering the equipment historical contribution, the current health state and the real-time load to calculate the equipment weight coefficient, and according to the production line process, aggregates the dynamic weighted feature vectors of the equipment on the same production line to form a production line layer feature vector, so as to abstract the overall running state of the production line at a high level, and then inputs the production line layer feature vector into a pre-constructed equipment efficiency prediction model to generate an equipment efficiency prediction result, so as to finally realize the prediction output of the equipment efficiency. The overall processing process involves hierarchical aggregation and dynamic weighting, can effectively utilize multi-source heterogeneous equipment data, and thus can realize dynamic and accurate prediction of the equipment efficiency, thereby providing technical support for intelligent upgrading of the factory.

[0012] The enterprise equipment efficiency dynamic prediction method, wherein the historical contribution degree is determined based on the following steps:

[0013] A1, obtaining running data of each equipment in a preset historical period, the running data including yield data, energy consumption data and running time data;

[0014] A2, calculating the unit time job quantity corresponding to each equipment according to the yield data of each equipment;

[0015] A3, calculating the unit time energy consumption intensity corresponding to each equipment according to the energy consumption data and the running time data of each equipment;

[0016] A4, calculating the historical contribution degree corresponding to each equipment according to the unit time job quantity and the unit time energy consumption intensity.

[0017] The above processing mode clearly defines the acquisition approach of the historical contribution degree of each equipment, which quantifies and standardizes the historical running data of each equipment by calculating the unit time job quantity and the unit time energy consumption intensity, and the historical contribution degree obtained by combining the unit time job quantity and the unit time energy consumption intensity reflects the energy consumption intensity when the equipment produces unit job quantity. By defining the calculation method of the historical contribution degree, the historical performance of the equipment can be more objectively evaluated, thereby providing a more accurate basis for subsequent calculation of the equipment weight coefficient, and finally improving the accuracy of the equipment efficiency prediction.

[0018] The enterprise equipment efficiency dynamic prediction method, wherein the step of calculating the weight coefficient of each equipment based on the historical contribution degree, the current health state and the real-time load comprises:

[0019] S21, obtaining depreciation life data and actual working time data of each equipment, and calculating the current health state of each equipment according to the depreciation life data and the actual working time data;

[0020] S22, calculating real-time loads of each device according to actual operating powers and rated powers of the devices;

[0021] S23, calculating dynamic adjustment coefficients of each device according to the current health states and the real-time loads;

[0022] S24, calculating weight coefficients of each device in combination with the dynamic adjustment coefficients and the historical contribution degrees.

[0023] The above processing process combines the dynamic adjustment coefficients with the historical contribution degrees, and finally obtains the weight coefficients which not only consider historical efficiencies of the devices, but also take into account current operating states of the devices, so that the device weight coefficients can more accurately and dynamically reflect the device efficiencies. Thus, using the weight coefficients for device efficiency prediction can improve the accuracy and reliability of the prediction, and solve the problem of low prediction accuracy caused by unclear quantification of the weight coefficients.

[0024] The enterprise device efficiency dynamic prediction method, wherein step S2 further comprises the following steps:

[0025] S25, obtaining operating parameters of each device, the operating parameters comprising actual working voltage information, actual working current information and actual environment temperature information;

[0026] S26, calculating parameter adjustment coefficients of each device according to the actual working voltage information, the actual working current information and the actual environment temperature information;

[0027] S27, compensating and adjusting the weight coefficients of each device based on the parameter adjustment coefficients.

[0028] The enterprise device efficiency dynamic prediction method, wherein step S1 comprises:

[0029] S11, obtaining original data of different types of devices, evaluating qualities of the original data of each device based on data missing rates, and determining the original data as high-quality data if the data missing rate is lower than a preset threshold, or determining the original data as low-quality data otherwise;

[0030] S12, for the high-quality data, performing data cleaning by using an abnormal value detection method based on a statistical principle, and performing format unification processing to obtain cleaned high-quality data;

[0031] S13, for the low-quality data, performing data cleaning by using a missing value filling method based on data mining, and performing format unification processing to obtain cleaned low-quality data;

[0032] S14, respectively extract the time domain statistical features, frequency domain statistical features and wavelet transform features of the cleaned high-quality data and the cleaned low-quality data, to form the original feature vectors of each device.

[0033] The enterprise device performance dynamic prediction method, wherein step S3 comprises:

[0034] S31, obtain a device relationship graph, the device relationship graph represents devices as nodes and process flow relationships between devices as edges, and the weight of an edge represents the material transfer amount between devices;

[0035] S32, search for a production line chain based on the device relationship graph and the weight size relationship of the edges;

[0036] S33, extract a dynamic weighted feature vector of a corresponding device according to the production line chain, and integrate the dynamic weighted feature vector using the weight of the edge to generate a production line layer feature vector.

[0037] The enterprise device performance dynamic prediction method, wherein the device performance prediction model comprises a teacher model and a student model compressed and simplified based on the teacher model, and step S4 comprises:

[0038] S41, monitor the resource utilization rate of the edge device, the resource utilization rate comprising the CPU utilization rate, the memory utilization rate and the storage utilization rate;

[0039] S42, if any resource utilization rate exceeds a preset threshold, input the production line layer feature vector to the student model to generate a device performance prediction result; if all resource utilization rates do not exceed the preset threshold, input the production line layer feature vector to the teacher model to generate a device performance prediction result.

[0040] In a second aspect, the present application also provides an enterprise device performance dynamic prediction device for performing device performance prediction, comprising:

[0041] An acquisition module is configured to acquire original data of different types of devices, perform data cleaning and format unification processing on the original data, and extract original feature vectors of the devices, wherein the original feature vectors represent key feature parameters of the running state of the corresponding devices;

[0042] A first integration module is configured to calculate weight coefficients of devices based on historical contribution degrees, current health states and real-time loads of the devices, and perform weighted aggregation on original feature vectors of devices at the same level of a production line according to the weight coefficients to obtain dynamic weighted feature vectors of the devices at the level.

[0043] a second integration module configured to aggregate the dynamic weighted feature vectors of the devices on the same production line according to a production line process to form a production line layer feature vector;

[0044] a prediction module configured to input the production line layer feature vector into a pre-constructed device performance prediction model to generate a device performance prediction result.

[0045] The enterprise device performance dynamic prediction device provided in the present application calculates the weight coefficient of the device by comprehensively considering the historical contribution, current health state and real-time load of the device, so as to realize dynamic configuration of the weight coefficient, highlight the influence of important devices and state abnormal devices on the production line performance, reduce the influence of non-important devices and state good devices, aggregate the dynamic weighted feature vectors of the devices on the same production line according to the production line process to form a production line layer feature vector, abstract the overall running state of the production line at a high level, input the production line layer feature vector into a pre-constructed device performance prediction model to generate a device performance prediction result, and finally realize prediction output of the device performance. The overall processing process involves hierarchical aggregation and dynamic weighting, can effectively utilize multi-source heterogeneous device data, and can realize dynamic and accurate prediction of the device performance, thereby providing technical support for intelligent upgrading of the factory.

[0046] In a third aspect, the present application further provides an electronic device including a processor and a memory, wherein the memory stores computer readable instructions, and when the computer readable instructions are executed by the processor, the steps in the method provided in the first aspect are executed.

[0047] In a fourth aspect, the present application further provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the method provided in the first aspect are executed.

[0048] It can be known from the above that the enterprise equipment efficiency dynamic prediction method, device, electronic equipment and storage medium provided in the application, compared with the processing mode of the traditional method relying on a centralized server for data processing and model calculation, realizes dynamic configuration of the weight coefficient by comprehensively considering the device historical contribution, current health state and real-time load to calculate the device weight coefficient, highlights the influence of important devices and state abnormal devices on the production line efficiency, reduces the influence of non-important devices and state good devices, aggregates the dynamic weighted feature vectors of the devices on the same production line according to the production line process to form a production line layer feature vector, abstracts the overall running state of the production line at a high level, inputs the production line layer feature vector into a pre-constructed device efficiency prediction model to generate a device efficiency prediction result, and finally realizes the prediction output of the device efficiency. The overall processing process involves hierarchical aggregation and dynamic weighting, can effectively utilize multi-source heterogeneous device data, and realizes dynamic prediction of the device efficiency on the edge side with limited resources, so that the dynamic and accurate prediction of the device efficiency can be realized, and technical support can be provided for intelligent upgrading of the factory. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 A flowchart of an enterprise equipment efficiency dynamic prediction method provided in an embodiment of the application.

[0050] Figure 2 A structural schematic diagram of an enterprise equipment efficiency dynamic prediction device provided in an embodiment of the application.

[0051] Figure 3 A structural schematic diagram of an electronic equipment provided in an embodiment of the application.

[0052] The drawings show that: 201, an acquisition module; 202, a first integration module; 203, a second integration module; 204, a prediction module; 301, a processor; 302, a memory; and 303, a communication bus. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the application will be clearly and completely described in combination with the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. The components of the embodiments of the application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the application provided in the drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the application. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the application.

[0054] It should be noted that like reference numerals and characters refer to like items throughout the attached drawings and alternative embodiments thereof. Note that the first, second, etc. terminology can be used in the description for the sake of brevity but is not intended to limit the scope of the application.

[0055] In a first aspect, referring to Figure 1 Some embodiments of the present application provide a dynamic prediction method for enterprise equipment performance, which is used for equipment performance prediction, and comprises the following steps:

[0056] S1, obtaining original data of different types of equipment, and performing data cleaning, format unification processing on the original data, and extracting original feature vectors of each equipment, the original feature vectors representing key feature parameters of the corresponding equipment running state;

[0057] S2, calculating the weight coefficient of each equipment based on the historical contribution degree, current health state and real-time load of each equipment, and weighting and aggregating the original feature vectors of the equipment at the same level on the production line according to the weight coefficient, to obtain the dynamic weighted feature vector of the equipment level;

[0058] S3, according to the process flow of the production line, aggregating the dynamic weighted feature vectors of the equipment on the same production line to form a production line layer feature vector;

[0059] S4, inputting the production line layer feature vector into a pre-constructed equipment performance prediction model to generate an equipment performance prediction result.

[0060] Specifically, in step S1, the original data comes from different types of equipment in a large factory workshop, and the data cleaning process can include operations such as removing duplicate values, handling missing values and abnormal values, etc. to ensure data quality. Format unification processing can convert data of different formats into a unified format, such as timestamp format, numerical format, etc. to facilitate subsequent processing. The extraction of original feature vectors can be based on time domain, frequency domain or time-frequency domain analysis methods, such as extracting mean, variance, spectral energy, etc. These original feature vectors represent key feature parameters of the corresponding equipment running state, which can lay a data foundation for subsequent performance prediction.

[0061] More specifically, step S2 is used to calculate the weight coefficient of each device, which comprehensively considers the historical contribution degree, the current health status and the real-time load of the device, so as to dynamically reflect the importance and running condition of the device in the production process; wherein the historical contribution degree can be calculated according to the yield, energy consumption and other data of the device in the historical period, reflecting the historical performance of the device, the current health status can be evaluated according to the depreciation period, actual working time and other data of the device, reflecting the current aging degree of the device, the real-time load can be calculated according to the actual running power, rated power and other data of the device, reflecting the current running pressure of the device; the calculation of the weight coefficient can be realized by using weighted average method, exponential weighting method and other methods, and the calculated weight coefficient is then used to weight and aggregate the original feature vectors of the devices at the same level on the production line (such as multiple sets of parallel devices at the same processing stage), thereby obtaining the dynamic weighted feature vector of the device level. The weighting and aggregation process can use linear weighting, nonlinear weighting and other methods to fuse the original feature vectors of the devices at the same level, and this weighting and aggregation method can highlight the influence of important devices and state abnormal devices on the efficiency of the production line.

[0062] More specifically, step S3 aggregates the dynamic weighted feature vectors of the devices on the same production line according to the production line process, to form a production line layer feature vector, wherein the aggregation of the production line layer feature vector can be realized by summation, averaging, splicing and other methods, and the production line layer feature vector can represent the running state of the production line as a whole and provide input for the efficiency prediction at the production line level.

[0063] More specifically, in step S4, the device efficiency prediction model can be a machine learning model, a deep learning model, etc., such as support vector machine, recurrent neural network, etc. The model can be trained using the production line layer feature vectors and device efficiency data determined from historical device running data. The device efficiency prediction result output by the model can provide quantitative evaluation of the device running state for the enterprise, assisting managers in making production decisions and device maintenance.

[0064] Compared with the processing mode of relying on a centralized server for data processing and model calculation in a traditional method, the enterprise device efficiency dynamic prediction method of the embodiment of the application realizes dynamic configuration of the weight coefficient by comprehensively considering the historical contribution of the device, the current health state and the real-time load to calculate the weight coefficient of the device, so as to highlight the influence of important devices and state abnormal devices on the efficiency of the production line, reduce the influence of non-important devices and state good devices, and aggregate the dynamic weighted feature vectors of the devices on the same production line according to the process flow of the production line to form a production line layer feature vector, so as to abstract the overall running state of the production line at a high level, then input the production line layer feature vector into a device efficiency prediction model constructed in advance to generate a device efficiency prediction result, and finally realize the prediction output of the device efficiency. The overall processing process involves hierarchical aggregation and dynamic weighting, can effectively utilize multi-source heterogeneous device data, and realizes dynamic prediction of the device efficiency on the edge side with limited resources, so as to realize dynamic and accurate prediction of the device efficiency and provide technical support for intelligent upgrading of the factory.

[0065] In some preferred embodiments, the historical contribution degree is determined based on the following steps:

[0066] A1, obtaining running data of each device in a preset historical period, the running data including yield data, energy consumption data and running time data;

[0067] A2, calculating the corresponding unit time work load of each device according to the yield data of each device;

[0068] A3, calculating the corresponding unit time energy consumption intensity of each device according to the energy consumption data and the running time data of each device;

[0069] A4, calculating the corresponding historical contribution degree of each device according to the unit time work load and the unit time energy consumption intensity.

[0070] Specifically, the preset historical period can be set according to actual needs; in step A2, the unit time work load is calculated by dividing the total yield of the device in the preset historical period by the total running time, so as to quantify the production capacity of the device in unit time. The work load unit can be the number of products, the number of processed parts and other quantifiable indicators.

[0071] More specifically, in step A3, the unit time energy consumption intensity is calculated by dividing the total energy consumption of the device in the historical period by the total running time, so as to evaluate the energy consumption level of the device in unit time. The energy consumption unit can be kilowatt-hour, joule, etc.

[0072] More specifically, in step A4, the historical contribution degree is obtained by dividing the unit time operation amount by the unit time energy consumption intensity, i.e., historical contribution degree = unit time operation amount / unit time energy consumption intensity, which reflects the operation amount generated by the device per unit energy consumption, and the higher the ratio, the higher the historical contribution degree of the device, which can quantify the efficiency level of the device in the historical period.

[0073] More specifically, the above processing mode clearly shows the acquisition approach of the historical contribution degree of each device, which quantifies and standardizes the historical operation data of each device by calculating the unit time operation amount and the unit time energy consumption intensity, and the historical contribution degree obtained by combining the unit time operation amount and the unit time energy consumption intensity reflects the energy consumption intensity when generating unit operation amount, and by clearly defining the calculation method of the historical contribution degree, the historical performance of the device can be more objectively evaluated, thereby providing a more accurate basis for subsequent calculation of the weight coefficient of the device, and finally improving the accuracy of the device efficiency prediction.

[0074] In some preferred embodiments, the step of calculating the weight coefficient of each device based on the historical contribution degree, the current health state, and the real-time load of each device comprises:

[0075] S21, obtaining depreciation life data and actual working time data of each device, and calculating a current health state of each device according to the depreciation life data and the actual working time data;

[0076] S22, calculating a real-time load of each device according to an actual running power and a rated power of each device;

[0077] S23, calculating a dynamic adjustment coefficient of each device according to the current health state and the real-time load;

[0078] S24, calculating a weight coefficient of each device in combination with the dynamic adjustment coefficient and the historical contribution degree.

[0079] Specifically, in step S21, the current health state quantifies the health level of the device by the ratio of the used time length of the device to the theoretical service life time length of the device, the depreciation life data represents the theoretical service life of the device, and the preset annual working time length is the planned working time of the device per year, and the product of the two is the total theoretical working time length of the device, wherein the calculation formula of the current health state is preferably: current health state = 1-(actual working time data / (depreciation life data* preset annual working time length)). Through the formula calculation, the health condition of the device can be quantified, and the longer the working time, the lower the current health state value.

[0080] More specifically, in step S22, the real-time load is calculated by the actual running power and the rated power of the device, which reflects the current running intensity of the device. Wherein, the calculation formula of the real-time load is preferably: real-time load = actual power / rated power.

[0081] More specifically, in step S23, the dynamic adjustment coefficient is preferably obtained by weighted summation of the current health status and the real-time load, and is preferably calculated by the following formula: dynamic adjustment coefficient = a * current health status + b * real-time load, wherein the weight coefficients a and b are used to adjust the relative importance of the health status and the real-time load in the dynamic adjustment coefficient, and can be adjusted and set according to actual application requirements.

[0082] More specifically, in step S24, the weight coefficient is preferably obtained by multiplying the historical contribution degree and the dynamic adjustment coefficient, wherein the historical contribution degree reflects the performance output capability of the device in the historical period, and the dynamic adjustment coefficient is obtained by adjusting the contribution degree according to the current state (current health status and real-time load) of the device, so that the final weight coefficient takes into account the historical performance and current state of the device.

[0083] More specifically, the above processing process combines the dynamic adjustment coefficient with the historical contribution degree, and the final weight coefficient takes into account both the historical performance of the device and the current running state of the device, so that the device weight coefficient can more accurately and dynamically reflect the device performance. Therefore, using the weight coefficient for device performance prediction can improve the accuracy and reliability of the prediction, and solve the problem of low prediction accuracy caused by unclear weight coefficient quantization method.

[0084] In some preferred embodiments, step S2 further comprises the following steps:

[0085] S25, obtaining the running parameters of each device, the running parameters including actual working voltage information, actual working current information, and actual environmental temperature information;

[0086] S26, calculating the parameter adjustment coefficient of each device according to the actual working voltage information, the actual working current information, and the actual environmental temperature information;

[0087] S27, compensating and adjusting the weight coefficient of each device based on the parameter adjustment coefficient.

[0088] Specifically, in step S25, the running parameters can be obtained by real-time collection via sensors, and the sensors are configured to monitor the actual working voltage, the actual working current, and the actual environmental temperature of the device, which can directly reflect the real-time state of the device running.

[0089] More specifically, in step S26, the calculation process of the parameter adjustment coefficient can include parameter standardization and weighted summation processing, so that different devices have standardized parameter adjustment coefficients, which can quantify the influence degree of the running parameters on the device performance.

[0090] More specifically, step S27 is configured to compensate and adjust the weight coefficients of each device according to the parameter adjustment coefficient, so as to realize dynamic adjustment of the weight coefficients based on the operating parameters, and make the adjusted weight coefficients reflect the real-time changes of the operating parameters of the devices.

[0091] More specifically, the above processing process further considers the influence of the operating parameters of the devices on the weight coefficients, wherein the parameter adjustment coefficient obtained in combination with step S25 and step S26 is intended to quantify the comprehensive influence of the voltage, current and temperature on the efficiency of the devices, and these parameters can more deeply reflect the real-time state of the device operation and the changes of the external environment, and the compensated and adjusted weight coefficients in step S27 can more comprehensively and accurately reflect the actual efficiency state of the devices.

[0092] More specifically, in step S27, the weight coefficients are preferably adjusted based on the following formula: compensated and adjusted weight coefficients = weight coefficients before compensation and adjustment * (1 + parameter adjustment coefficient).

[0093] In some preferred embodiments, step S26 comprises:

[0094] S261, converting the actual working voltage information, the actual working current information and the actual environmental temperature information into standardized voltage, standardized current and standardized temperature respectively based on a preset standardized conversion formula;

[0095] S262, weighting and summing the standardized voltage, the standardized current and the standardized temperature according to a preset voltage influence factor, a current influence factor and a temperature influence factor respectively to obtain the parameter adjustment coefficient corresponding to each device.

[0096] Specifically, in step S261, the preset standardized conversion formula is used to convert the actual working voltage information, the actual working current information and the actual environmental temperature information into dimensionless standardized values, and the standardized conversion formula is preferably: parameter standardized value = (actual parameter value - parameter mean value) / parameter standard deviation. The standardized conversion formula can eliminate the influence caused by different physical dimensions and numerical ranges. By subtracting the parameter mean value and dividing by the parameter standard deviation, the original data is converted into a standard normal distribution range with a mean value of 0 and a standard deviation of 1. Thus, the voltage, current and temperature data of different dimensions and numerical ranges can be compared and calculated under the same standard.

[0097] More specifically, in step S262, the preset voltage influence factor, current influence factor and temperature influence factor represent the relative importance of the voltage, current and temperature on the equipment performance, which can be a fixed value preset according to the equipment characteristics and operation experience; after obtaining the normalized voltage, normalized current and normalized temperature, the products of the corresponding influence factors are summed up, and the parameter adjustment coefficient is obtained, which comprehensively considers the comprehensive influence of voltage, current and temperature on equipment performance.

[0098] More specifically, the above processing process standardizes the actual working voltage information, actual working current information and actual environmental temperature information, eliminates the influence of dimension and numerical range difference on subsequent calculation, and then realizes the weighted summation of different operating parameters by introducing the preset influence factor, so as to obtain the comprehensive parameter adjustment coefficient, which can more scientifically and reasonably reflect the comprehensive influence of operating parameters on equipment performance, and then be used for compensation adjustment of equipment weight coefficient, so that the adjustment of equipment weight coefficient is more refined and accurate.

[0099] In some preferred embodiments, step S1 comprises:

[0100] S11, obtaining raw data of different types of equipment, evaluating the quality of raw data of each equipment based on data missing rate, if the data missing rate of raw data is lower than a preset threshold, determining it as high-quality data, otherwise determining it as low-quality data;

[0101] S12, for high-quality data, using an abnormal value detection method based on statistical principles for data cleaning, and performing format unification processing to obtain cleaned high-quality data;

[0102] S13, for low-quality data, using a missing value filling method based on data mining for data cleaning, and performing format unification processing to obtain cleaned low-quality data;

[0103] S14, extracting the time domain statistical features, frequency domain statistical features and wavelet transform features of the cleaned high-quality data and cleaned low-quality data respectively to form the original feature vectors of each equipment.

[0104] Specifically, in step S11, the data missing rate evaluation is used for preliminary judgment of the quality of raw data of each equipment, which provides a basis for subsequent data processing, which can be realized by collecting communication data packets in a preset time window for data acquisition frequency analysis.

[0105] More specifically, steps S12 and S13 adopt differentiated data cleaning strategies for different quality data to ensure the effectiveness and pertinence of data cleaning.

[0106] More specifically, step S12 adopts a statistical outlier detection method for data cleaning of high-quality data, which can use the 3σ principle or the box plot method to identify and process outliers. The format unification process can use timestamp alignment, data type conversion, dimension unification, etc. to ensure data format consistency.

[0107] More specifically, step S13 adopts a data mining missing value filling method for data cleaning of low-quality data, which can use K-nearest neighbor filling, regression filling or multiple imputation to restore the information of missing data. The format unification process is the same as that of high-quality data.

[0108] More specifically, in step S14, the time domain statistical features can include mean, standard deviation, peak value, kurtosis, etc. The frequency domain statistical features can be obtained by Fourier transform, such as energy spectrum, power spectrum, etc. The wavelet transform features can be extracted by discrete wavelet transform, such as wavelet coefficients of db4 wavelet basis. These features are combined into an original feature vector to comprehensively represent the equipment operating state.

[0109] More specifically, the above processing process distinguishes data by quality evaluation and adopts different data cleaning methods, which ensures the effectiveness and pertinence of data cleaning in the case of differences in original data quality, provides high-quality feature data support for subsequent equipment efficiency prediction, and further improves the accuracy of equipment efficiency prediction.

[0110] In some preferred embodiments, the data missing rate acquisition process includes:

[0111] B1, collecting communication data packets of each device within a preset time window, and parsing the communication data packets to extract the data acquisition frequency;

[0112] B2, if the data acquisition frequency is lower than the preset frequency threshold, determining that the data is missing, and calculating the data missing rate according to the data acquisition frequency and the preset frequency threshold.

[0113] Specifically, in step S1, the data acquisition frequency can be extracted by parsing the communication data packets using conventional existing data parsing and extraction means, which will not be described here.

[0114] More specifically, in step B2, the preset frequency threshold is the data transmission speed that the device should have in the normal working state, which can be adjusted according to the device type, data application scenario and requirement for data integrity. When the actual data acquisition frequency is lower than this threshold, the system determines that the data is missing. In the embodiments of the present application, the data missing rate is preferably calculated based on the following formula: data missing rate = (1-actual acquisition frequency / preset acquisition frequency)*100%.

[0115] More specifically, the above formula quantifies the degree of deviation between the actual acquisition frequency and the preset frequency threshold, thereby obtaining the degree of data loss, providing data support for the selection of subsequent differentiated data cleaning and feature extraction strategies.

[0116] In some preferred embodiments, step S3 comprises:

[0117] S31, acquire a device relationship graph, the device relationship graph represents devices as nodes and process flow relationships between devices as edges, and the weight of an edge represents the material transfer amount between devices;

[0118] S32, search for a production line chain based on the device relationship graph and the weight size relationship of the edges;

[0119] S33, extract a dynamic weighted feature vector of the corresponding device according to the production line chain, and integrate the dynamic weighted feature vector using the weight of the edge to generate a production line layer feature vector.

[0120] Specifically, the device relationship graph in step S1 is a pre-constructed relationship graph, which can be obtained by the following method: determining all devices in a factory workshop as nodes of the graph, then determining the connection relationship between the devices according to the actual production process, and the connection relationship constitutes the edges of the graph. The weight of the edge can be configured and set according to the average material transfer amount between the devices or the closeness of production cooperation between the devices. Thus, the mutual dependence relationship and material flow of the devices on the production line can be represented by the device relationship graph.

[0121] More specifically, in step S32, the process of searching for a production line chain can be: starting from the starting device of the device relationship graph, selecting the device with the largest relationship connected to the current device based on the weight size relationship of the edges using a preset search algorithm, reaching the next device along the edge, and adding the device to the production line chain. Repeat this process until the end device of the production line is reached or the weight of the edge is below a set threshold.

[0122] More specifically, in step S33, the process of generating a production line layer feature vector can be: obtaining the dynamic weighted feature vector of each device in the production line chain, then performing weighted averaging or weighted integration on the feature vectors according to the position of the device in the production line and the weight of the edge to generate the final production line layer feature vector. The weight of the edge can reflect the importance of the device in the production line and the material correlation. Thus, the core features of the production line can be more effectively extracted, reducing the interference of non-critical devices or low material transfer devices on the production line layer feature vector, thereby improving the representativeness of the production line layer feature vector and the accuracy of the prediction model.

[0123] In some preferred embodiments, step S32 comprises:

[0124] S321, initializing the production line chain as an empty linked list according to the device relationship graph, and taking the weight of the edge in the device relationship graph as the material transfer efficiency between devices;

[0125] S322, selecting a starting device from the device relationship graph, and based on the weight size relationship of the edge, using a greedy algorithm to preferentially select a downstream device connected to the current device and having the highest material transfer efficiency, adding the downstream device to the production line chain, and updating the current device to the downstream device, repeating the step until reaching an end device or the material transfer efficiency being lower than a preset threshold;

[0126] S323, if the number of devices in the production line chain is less than a preset number threshold, reducing the material transfer efficiency threshold and re-executing S322, otherwise, taking the current production line chain as the searched production line chain.

[0127] Specifically, in step S321, the device relationship graph can be constructed as a data structure, such as an adjacency list or an adjacency matrix, to represent the connection relationship and material transfer efficiency between devices. The weight of the edge can be quantified as a specific value, such as the amount of material transferred per unit time. The purpose of initializing an empty linked list is to store the devices on the searched production line chain, laying the foundation for subsequent search.

[0128] More specifically, in step S322, the greedy algorithm is introduced, after selecting a starting device from the device relationship graph, the downstream device connected to the current device and having the highest material transfer efficiency is preferentially selected, and the downstream device is added to the production line chain, and the step is repeated until reaching an end device or the material transfer efficiency being lower than a preset threshold, thereby ensuring the effectiveness and efficiency of device connection during the construction of the production line chain, and ensuring the integrity and rationality of the production line chain.

[0129] More specifically, in step S323, the number of devices in the production line chain is considered, if the number of devices in the production line chain is less than a preset number threshold, the material transfer efficiency threshold is reduced and S322 is re-executed, to ensure that the production line chain that may be missed when the material transfer efficiency threshold is high is searched again, improving the integrity of the production line chain search; if the number of devices in the production line chain is not less than the preset number threshold, the current production line chain is taken as the searched production line chain, to ensure the effectiveness of the search result, wherein the preset number threshold can be set according to the actual number of devices in the production line.

[0130] More specifically, the above processing mode adopts a greedy algorithm to search the production line chain, which can efficiently find a production line chain meeting the material transfer efficiency requirement in the case of complex device relationship, ensure the efficiency and quality of the production line chain by preferentially selecting the path with high material transfer efficiency, and improve the completeness of the search result by dynamically adjusting the material transfer efficiency threshold when the number of devices in the production line chain is insufficient, so as to ensure that potential production line chains are not missed.

[0131] In some preferred embodiments, step S33 comprises:

[0132] S331, obtain the dynamic weighted feature vector of the device corresponding to the production line chain, and the production line chain is denoted as C k =(e k,1 ,e k,2 ,…e k,i ,…,e k,mk ), wherein C k is the kth production line chain composed of a device node sequence, e k,i represents the ith device node on the kth production line chain, and mk is the total number of devices;

[0133] S332, obtain the static connection weight k,i between the device nodes e k,i+1 according to the weight of the edge;

[0134] S333, generate a dynamic normalization factor S k according to the device node, which satisfies:

[0135] (1)

[0136] S334, calculate the production line layer feature vector L k , which satisfies:

[0137] (2)

[0138] wherein, is the dynamic weighted feature vector of the device node e k,i , and is the dynamic weighted feature vector of the last device node.

[0139] Specifically, step S331 is used to extract the dynamic weighted feature vector of the device corresponding to the device node, which lays a data foundation for the subsequent feature integration process.

[0140] More specifically, in step S332, the static connection weight is obtained according to the weight of the edge in the device relationship graph, and the weight represents the material transfer amount between devices.

[0141] ​More specifically, in step S333, the dynamic normalization factor is calculated by summing up all the adjacent equipment connection weights in the production line chain and adding 1, which is used for subsequent vector normalization processing.

[0142] More specifically, in step S334, for each pair of adjacent equipment in the production line chain, the dynamic weighted feature vector of the previous equipment is multiplied by the static connection weight between them, all the product results are accumulated, and the dynamic weighted feature vector of the last equipment in the production line chain is added, and the final result is divided by the normalization factor, so as to obtain the production line layer feature vector.

[0143] More specifically, in the above processing manner, for each pair of adjacent equipment, the dynamic weighted feature vector of the previous equipment is multiplied by the static connection weight between them, and these products are accumulated, and the accumulated result is divided by the dynamic normalization factor to obtain the production line layer feature vector. Step S33 can effectively integrate the connection strength information between equipment into the production line layer feature vector by using this weighted summation strategy, wherein the greater the connection weight, the stronger the correlation between equipment, so as to more accurately reflect the running state of the entire production line, so that the subsequent equipment efficiency prediction model can make more accurate prediction based on the production line layer feature vector, overcoming the problem that the connection relationship between equipment is not fully considered in the traditional method, and improving the accuracy and reliability of the prediction.

[0144] In some preferred embodiments, the equipment efficiency prediction model includes a teacher model and a student model based on compression and simplification of the teacher model, and step S4 includes:

[0145] S41, monitoring the resource utilization rate of the edge device, the resource utilization rate including CPU utilization rate, memory utilization rate and storage utilization rate;

[0146] S42, if any resource utilization rate exceeds the preset threshold, inputting the production line layer feature vector into the student model to generate the equipment efficiency prediction result; if all resource utilization rates do not exceed the preset threshold, inputting the production line layer feature vector into the teacher model to generate the equipment efficiency prediction result.

[0147] Specifically, the monitoring of the resource utilization rate can be realized by periodically collecting system performance index data by deploying a monitoring agent on the edge device. The preset threshold can be configured as a critical value reflecting the resource load state of the edge device, which can be set according to the use demand.

[0148] More specifically, step S42 dynamically selects a device performance prediction model based on the resource utilization monitored in step S41, wherein the device performance prediction model includes a teacher model and a student model, the teacher model can be a complex model structure, for example, a deep neural network model, which has high prediction accuracy but consumes more computing resources, and the student model is a lightweight model obtained by simplifying and compressing the teacher model, for example, a shallow neural network model or a linear regression model, which may have slightly reduced prediction accuracy but significantly reduced resource consumption. When any of the resource utilizations monitored in step S41 exceeds the preset threshold, it indicates that the edge device is in a resource shortage state, and the system switches to the student model for prediction. When none of the resource utilizations exceeds the preset threshold, it indicates that the edge device is in a resource sufficient state, and the system selects the teacher model for prediction. In this way, by dynamically switching between the teacher model and the student model, the technical solution can adaptively select a suitable prediction model under different resource load states, ensure the prediction accuracy of the device performance, effectively reduce the resource consumption of the model deployed and running on the edge side, and be more suitable for application scenarios with limited edge computing resources.

[0149] More specifically, the student model learns knowledge from the teacher model through model compression and simplification techniques such as knowledge distillation, model pruning, etc., and significantly reduces the model complexity. The student model significantly reduces the computational load and resource consumption while ensuring a certain prediction accuracy, so that it can run quickly on resource-limited edge devices and reduce resource occupancy.

[0150] In a second aspect, referring to Figure 2 Some embodiments of the present application also provide an enterprise device performance dynamic prediction device for device performance prediction, comprising:

[0151] The acquisition module 201 is configured to acquire raw data of different types of devices, and perform data cleaning and format unification processing on the raw data, and extract original feature vectors of each device, wherein the original feature vectors represent key feature parameters of the running state of the corresponding device.

[0152] The first integration module 202 is configured to calculate weight coefficients of each device based on historical contribution degrees, current health states and real-time loads of each device, and aggregate the original feature vectors of the devices at the same level on the production line according to the weight coefficients to obtain dynamic weighted feature vectors of the device level.

[0153] The second integration module 203 is configured to aggregate the dynamic weighted feature vectors of the devices on the same production line to form a production line layer feature vector according to the process flow of the production line.

[0154] The prediction module 204 is configured to input the production line layer feature vector into a pre-constructed device performance prediction model to generate a device performance prediction result.

[0155] The enterprise equipment performance dynamic prediction device provided in the embodiments of the present application can realize dynamic configuration of the weight coefficient by comprehensively considering the historical contribution of the equipment, the current health state and the real-time load to calculate the weight coefficient of the equipment, so as to highlight the influence of important equipment and state abnormal equipment on the production line performance, reduce the influence of non-important equipment and state good equipment, and aggregate the dynamic weighted feature vectors of the equipment on the same production line according to the production line process to form a production line layer feature vector, thereby abstracting the overall running state of the production line at a high level, inputting the production line layer feature vector into the equipment performance prediction model constructed in advance to generate an equipment performance prediction result, and finally realizing the prediction output of the equipment performance. The overall processing process involves hierarchical aggregation and dynamic weighting, can effectively utilize multi-source heterogeneous equipment data, and realizes dynamic prediction of the equipment performance on the edge side with limited resources, thereby realizing dynamic and accurate prediction of the equipment performance and providing technical support for intelligent upgrading of the factory.

[0156] In some preferred embodiments, the enterprise equipment performance dynamic prediction device is used to execute the enterprise equipment performance dynamic prediction method provided in the first aspect.

[0157] In the third aspect, referring to Figure 3 Some embodiments of the present application also provide a structural schematic diagram of an electronic device. The present application provides an electronic device, which comprises a processor 301 and a memory 302. The processor 301 and the memory 302 are interconnected and communicate with each other through a communication bus 303 and / or other forms of connection mechanism (not marked). The memory 302 stores computer readable instructions executable by the processor 301. When the electronic device is running, the processor 301 executes the computer readable instructions to execute the method in any optional implementation manner of the above-mentioned embodiments.

[0158] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to perform the method in any of the optional implementation manners of the above-mentioned embodiments. The computer readable storage medium can be implemented by any type of volatile or nonvolatile storage devices or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or a compact disk.

[0159] In the embodiments of the present application, it should be understood that the disclosed device and method can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed coupling or direct coupling or communication connection between the units can be indirect coupling or communication connection through some communication interfaces, and can be electrical, mechanical or other forms.

[0160] In addition, the units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, some or all of the units can be selected to achieve the purpose of the present embodiment.

[0161] Furthermore, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0162] In this paper, the relationship terms such as first and second 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 the entities or operations.

[0163] The above merely provides an example of the present application, and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for dynamic prediction of enterprise equipment performance, used for predicting equipment performance, characterized in that, The method includes the following steps: S1. Obtain raw data from different types of devices, and perform data cleaning and format unification on the raw data to extract the raw feature vectors of each device. The raw feature vectors represent the key feature parameters of the corresponding device's operating status. S2. Calculate the weight coefficient of each device based on its historical contribution, current health status and real-time load, and aggregate the original feature vectors of devices at the same level on the production line according to the weight coefficients to obtain the dynamic weighted feature vector of the device level. S3. Based on the production line process flow, aggregate the dynamic weighted feature vectors of equipment on the same production line to form the production line layer feature vector; S4. Input the production line layer feature vector into the pre-built equipment performance prediction model to generate equipment performance prediction results; Step S3 includes: S31. Obtain the equipment relationship graph, wherein the equipment relationship graph represents equipment with nodes, the process flow relationship between equipment with edges, and the weight of the edge represents the amount of material transfer between equipment; S32. Based on the equipment relationship graph and the weight relationship of the edges, search to obtain the production line chain; S33. Extract the dynamic weighted feature vectors of the corresponding equipment according to the production line chain, and integrate them using the weights of the edges to generate the feature vector of the production line layer.

2. The method for dynamic prediction of enterprise equipment efficiency according to claim 1, characterized in that, The historical contribution was determined based on the following steps: A1. Obtain the operating data of each device within a preset historical period. The operating data includes output data, energy consumption data, and operating time data. A2. Based on the output data of each piece of equipment, calculate the workload per unit time for each piece of equipment; A3. Calculate the energy consumption intensity per unit time for each device based on the energy consumption data and operating time data of each device; A4. Calculate the historical contribution of each piece of equipment based on the workload per unit time and the energy consumption intensity per unit time.

3. The method for dynamic prediction of enterprise equipment efficiency according to claim 1, characterized in that, The steps for calculating the weighting coefficient of each device based on its historical contribution, current health status, and real-time load include: S21. Obtain the depreciation period data and actual working time data of each device, and calculate the current health status of each device based on the depreciation period data and the actual working time data; S23. Calculate the dynamic adjustment coefficient of each device based on the current health status and the real-time load; S24. Calculate the weight coefficient of each device by combining the dynamic adjustment coefficient and the historical contribution.

4. The method for dynamic prediction of enterprise equipment efficiency according to claim 3, characterized in that, Step S2 also includes the following steps: S25. Obtain the operating parameters of each device, including actual operating voltage information, actual operating current information, and actual ambient temperature information; S26. Calculate the parameter adjustment coefficients for each device based on the actual operating voltage information, actual operating current information, and actual ambient temperature information; S27. Adjust the weight coefficients of each device based on the parameter adjustment coefficients.

5. The method for dynamic prediction of enterprise equipment efficiency according to claim 1, characterized in that, Step S1 includes: S11. Obtain raw data from different types of devices, evaluate the quality of the raw data of each device based on the data missing rate, and determine the data as high quality if the data missing rate of the raw data is lower than a preset threshold; otherwise, determine the data as low quality. S12. For the high-quality data, an outlier detection method based on statistical principles is used for data cleaning, and the format is standardized to obtain cleaned high-quality data. S13. For the low-quality data, a missing value imputation method based on data mining is used to clean the data and perform format unification processing to obtain cleaned low-quality data. S14. Extract the time-domain statistical features, frequency-domain statistical features, and wavelet transform features of the cleaned high-quality data and the cleaned low-quality data respectively to form the original feature vector of each device.

6. The method for dynamic prediction of enterprise equipment efficiency according to claim 1, characterized in that, The equipment performance prediction model includes a teacher model and a student model compressed and simplified based on the teacher model. Step S4 includes: S41. Monitor the resource utilization rate of edge devices, wherein the resource utilization rate includes CPU utilization rate, memory utilization rate and storage utilization rate; S42. If any resource utilization rate exceeds a preset threshold, the production line layer feature vector is input into the student model to generate equipment performance prediction results; if all resource utilization rates do not exceed the preset threshold, the production line layer feature vector is input into the teacher model to generate equipment performance prediction results.

7. A dynamic prediction device for enterprise equipment performance, used for predicting equipment performance, characterized in that, include: The acquisition module is used to acquire raw data from different types of devices, and to perform data cleaning and format unification processing on the raw data, and to extract the raw feature vectors of each device. The raw feature vectors represent the key feature parameters of the corresponding device's operating status. The first integration module is used to calculate the weight coefficient of each device based on its historical contribution, current health status and real-time load, and to perform weighted aggregation of the original feature vectors of devices at the same level on the production line according to the weight coefficients to obtain the dynamic weighted feature vector of the device level. The second integration module is used to aggregate the dynamic weighted feature vectors of equipment on the same production line according to the production line process flow, and form the production line layer feature vector. The prediction module is used to input the feature vector of the production line layer into a pre-built equipment performance prediction model to generate equipment performance prediction results; The step of aggregating the dynamic weighted feature vectors of equipment on the same production line according to the production line process flow to form the production line layer feature vector includes: S31. Obtain the equipment relationship graph, wherein the equipment relationship graph represents equipment with nodes, the process flow relationship between equipment with edges, and the weight of the edge represents the amount of material transfer between equipment; S32. Based on the equipment relationship graph and the weight relationship of the edges, search to obtain the production line chain; S33. Extract the dynamic weighted feature vectors of the corresponding equipment according to the production line chain, and integrate them using the weights of the edges to generate the feature vector of the production line layer.

8. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions that, when executed by the processor, perform the steps of the method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it performs the steps of the method as described in any one of claims 1-6.

Images