A production energy efficiency optimization method based on AI and industrial Internet

Abstract

The present invention discloses a production energy efficiency optimization method based on AI and industrial Internet, specifically related to the field of production energy efficiency optimization, including: installing vibration sensors on variable-frequency production equipment, constructing a device monitoring network and collecting vibration signals in the no-load stable state, and using time-frequency feature comparison and blind source separation methods to extract resonance coupling components with spatial coherence; establishing a resonance mode recognition model based on a graph convolutional neural network to realize the recognition of dangerous resonance modes and send frequency collision avoidance instructions; further constructing a multi-objective optimization function, using the standard particle swarm optimization algorithm to calculate the optimal working frequencies of each device, and storing the results in the device control network, and finally adjusting the frequencies through the edge controller to ensure that the devices work under the optimal frequency combination, effectively suppressing the energy loss and sudden increase in power consumption caused by resonance.

Description

Technical Field

[0001] The present invention relates to the technical field of production energy efficiency optimization. More specifically, the present invention relates to a production energy efficiency optimization method based on AI and industrial Internet. Background Art

[0002] In a high-density device environment, especially in industrial scenarios such as automotive welding workshops and semiconductor factories, multiple variable-frequency devices (such as air compressors and pump sets) often operate simultaneously, and their operating frequencies are controlled by different conditions. Under specific frequency combinations, some devices will exhibit subsonic resonance phenomena. The operating frequencies of multiple devices are close to the natural frequencies of the devices or there are resonance frequencies between them, and the vibrations will be superimposed on each other to generate low-frequency subsonic waves. This resonance effect may be transmitted to other devices or the environment through the structure of the device itself. At this time, the resonance between devices will trigger an amplification effect of mechanical vibration, resulting in significant ineffective power consumption and energy waste, thereby having an adverse impact on production efficiency and equipment life.

[0003] Therefore, how to accurately monitor the vibration characteristics of devices and suppress subsonic resonance through intelligent means has become an important issue in industrial production. In response to this problem, traditional monitoring methods usually rely on single vibration sensor monitoring and frequency screening, and it is difficult to accurately identify and analyze complex resonance phenomena, especially the hidden power consumption when multiple devices work together.

[0004] To solve the above problems, a technical solution is provided now. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a production energy efficiency optimization method based on AI and industrial Internet to solve the problems proposed in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] S1: Install vibration sensors on variable-frequency production devices to establish a device monitoring network, and collect the original vibration signal sequence under the no-load stable state of the devices;

[0008] S2: Construct a reference template of time-frequency characteristics based on the original vibration signal sequence, and judge whether the characteristics of the real-time vibration signal are effective by comparing the feature matching degree between the feature vector of the real-time vibration signal sequence and the reference template;

[0009] S3: Use a sliding window to perform blind source separation on the collected real-time signal sequence, and extract the resonance coupling components with spatial coherence characteristics in the independent vibration components;

[0010] S4: Construct a weighted graph data structure based on the independent vibration components after blind source separation, and establish a resonance mode recognition model using the graph convolutional neural network architecture;

[0011] S5: Connect the trained and verified resonance mode recognition model to the device monitoring network for dangerous resonance mode recognition, and send frequency collision avoidance instructions to the corresponding device nodes according to the recognition results;

[0012] S6: Construct a multi-objective optimization function containing the resonance energy transfer coefficient for the device that receives the frequency collision avoidance instruction, solve the optimal working frequency combination of each frequency conversion device and store it in the device control network;

[0013] S7: Obtain the optimal working frequency combination in the device monitoring network, and send a frequency regulation instruction to the edge controller of the target device to perform the frequency adjustment operation.

[0014] In a preferred embodiment, in S1, installing vibration sensors on the frequency conversion production equipment to establish a device monitoring network, and collecting the original vibration signal sequence under the no-load stable state of the equipment specifically includes:

[0015] Count all the frequency conversion production equipment in the production line, arrange piezoelectric vibration sensors at the installation positions of the frequency conversion production equipment, convert all the equipment into a node-based physical topology structure based on the position distribution of the sensors, and establish a device monitoring network;

[0016] Configure the time-sensitive network communication protocol, establish a clock synchronization link between device nodes, send periodic time calibration signals for time alignment, and construct a data transmission priority queue based on the device node topology structure;

[0017] Collect the three-axis vibration signal sequence under the no-load stable state of the equipment, apply a denoising algorithm to eliminate electromagnetic interference noise, and obtain the original vibration signal sequence of the equipment with a set time window size.

[0018] In a preferred embodiment, in S2, constructing a reference template of time-frequency characteristics based on the original vibration signal sequence, and judging whether the characteristics of the real-time vibration signal are effective by comparing the feature matching degree between the feature vector of the real-time vibration signal sequence and the reference template specifically includes:

[0019] Construct a multi-dimensional feature vector based on the time-domain kurtosis index and frequency-domain energy data of the original vibration signal sequence, and generate a reference template containing time-frequency characteristics;

[0020] The device monitoring network monitors the real-time vibration signal sequence of the device operation in real time, extracts the corresponding time-frequency feature vector and calculates the feature matching degree with the reference template through cosine similarity. When the average matching degree of consecutive set sampling periods exceeds the set matching degree threshold, send a feature valid confirmation signal.

[0021] In a preferred embodiment, in S3, a sliding window is used to perform blind source separation on the collected real-time signal sequence, and extracting the resonance coupling components with spatial coherence characteristics in the independent vibration components specifically includes:

[0022] When the device monitoring network receives a feature valid confirmation signal, a sliding window is used to perform blind source separation on the collected real-time signal sequence;

[0023] The real-time signal sequence is whitened by diagonalizing the covariance matrix, and independent component analysis is performed on the whitened signal using the FastICA algorithm. The separation matrix is iteratively solved based on the negative entropy maximization criterion to separate the independent vibration components of each device;

[0024] Based on the physical topology structure of the device monitoring network, the physical connection stiffness parameters of the device nodes are set to construct a transfer constraint matrix, where the calculation method of the physical connection stiffness parameters is:

[0025] ;

[0026] In the formula, is the physical connection stiffness parameter from device to device , is the propagation speed of the vibration signal in the medium connecting device and device , is the distance from device to device , is the vibration signal attenuation rate from device to device ;

[0027] The phase offset of the separated component is corrected through matrix projection operation, and the resonance coupling components with spatial coherence characteristics in the independent vibration components are extracted.

[0028] In a preferred embodiment, in S4, based on the independent vibration components after blind source separation, a weighted graph data structure is constructed, and using a graph convolutional neural network architecture to establish a resonance mode recognition model specifically includes:

[0029] The time-frequency features of the independent vibration components and resonance coupling components after blind source separation are fused with the device physical topology coordinate data to construct a weighted resonance mode graph, and the edge weights of the graph reflect the physical connection stiffness parameters between device nodes;

[0030] A resonance mode recognition model is established using a graph convolutional neural network architecture, aggregating the independent vibration components of neighboring device nodes layer by layer, calculating the resonance energy transfer coefficient between devices through a multi-head attention mechanism, and generating a resonance coupling strength matrix in the frequency-space dimension;

[0031] Set the resonance coupling strength threshold, mark the adjacent device node pairs with resonance coupling strength exceeding the threshold as dangerous resonance modes, and record their dominant resonance coupling strength and energy transfer paths.

[0032] In a preferred embodiment, in S5, connecting the trained and verified resonance mode recognition model to the device monitoring network for dangerous resonance mode recognition, and sending frequency collision avoidance instructions to the corresponding device nodes according to the recognition results specifically includes:

[0033] Select the historical operation data of the device network as the training set and validation set data and input them into the resonance mode recognition model, and adjust the model parameters based on the matching degree deviation between the predicted resonance coupling strength and the measured dominant resonance coupling strength of the validation set data;

[0034] When the prediction accuracy of the model is verified by a continuously set number of groups of historical operation data and reaches the preset index, connect the model to the device monitoring network for dangerous resonance mode recognition, and send frequency collision avoidance instructions to the adjacent device node pairs corresponding to the recognized dangerous resonance modes.

[0035] In a preferred embodiment, in S6, constructing a multi-objective optimization function containing the resonance energy transfer coefficient for the devices that receive the frequency collision avoidance instructions, and solving the optimal working frequency combinations of each frequency conversion device and storing them in the device control network specifically includes:

[0036] When the device receives the frequency collision avoidance instructions, initialize the device working frequency search space based on the resonance coupling strength matrix in the resonance mode diagram, define the adjustable frequency range and the minimum adjustment step size of the device, and construct a multi-objective optimization function containing the resonance energy transfer coefficient;

[0037] Introduce the Cauchy mutation mechanism into the standard particle swarm optimization algorithm, expand the search radius by dynamically adjusting the mutation probability, input the device's real-time independent vibration components and the coupling strength matrix parameters into the multi-objective optimization function, and solve the optimal working frequency combinations of each frequency conversion device and store them in the device control network.

[0038] In a preferred embodiment, in S7, obtaining the optimal working frequency combinations in the device monitoring network, and sending frequency regulation instructions to the edge controller of the target device to perform frequency adjustment operations specifically includes:

[0039] Obtain the optimal working frequency combinations in the device monitoring network, and send frequency regulation instructions to the edge controller of the target device based on the data transmission priority queue according to the preset routing table;

[0040] The device node stores the received frequency regulation instructions in the register, dynamically adjusts the instruction effective timing according to the current operating state of the device, and performs the frequency adjustment operation when the controller ready signal is aligned with the clock synchronization pulse.

[0041] Technical effects and advantages of a production energy efficiency optimization method based on AI and industrial Internet of this invention:

[0042] By installing vibration sensors on frequency conversion devices and establishing a device monitoring network, the original vibration signals under no-load stable state are collected in real time. The sliding window and blind source separation technologies are used to extract independent vibration components, and the resonance coupling components with spatial coherence characteristics are identified from the complex vibration signals. By constructing a weighted graph data structure and using graph convolutional neural network (GCN) for resonance mode identification, potential dangerous resonance modes can be accurately identified, and the resonance coupling between devices can be effectively avoided through frequency collision avoidance instructions, reducing the ineffective power consumption. Based on the multi-objective optimization solution strategy of the optimization function, the optimal working frequency combination can be determined for each frequency conversion device, thus improving the energy efficiency and stability of the device cluster. Through intelligent algorithms and accurate vibration monitoring, the whole system can achieve dynamic adjustment, reduce the hidden power consumption of devices, and improve the overall production efficiency. Brief Description of the Drawings

[0043] Figure 1 It is a schematic diagram of a production energy efficiency optimization method based on AI and industrial Internet of this invention. Detailed Embodiments

[0044] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. Embodiment 1:

[0045] Figure 1 A production energy efficiency optimization method based on AI and industrial Internet of this invention is given, which includes the following steps:

[0046] S1: Install vibration sensors on the frequency conversion production equipment to establish a device monitoring network, and collect the original vibration signal sequence under the no-load stable state of the equipment;

[0047] S2: Based on the original vibration signal sequence, construct a reference template of time-frequency characteristics, and judge whether the characteristics of the real-time vibration signal are effective by comparing the feature matching degree between the feature vector of the real-time vibration signal sequence and the reference template;

[0048] S3: Use a sliding window to perform blind source separation on the collected real-time signal sequence, and extract the resonance coupling components with spatial coherence characteristics in the independent vibration components;

[0049] S4: Based on the independent vibration components after blind source separation, construct a weighted graph data structure, and establish a resonance mode identification model using a graph convolutional neural network architecture;

[0050] S5: Connect the trained and verified resonance mode identification model to the device monitoring network for dangerous resonance mode identification, and send frequency collision avoidance instructions to the corresponding device nodes according to the identification results;

[0051] S6: For the devices that receive the frequency collision avoidance instructions, construct a multi-objective optimization function including the resonance energy transfer coefficient, solve the optimal working frequency combination of each frequency conversion device, and store it in the device control network;

[0052] S7: Obtain the optimal working frequency combination in the device monitoring network, and send frequency regulation instructions to the edge controller of the target device to execute the frequency adjustment operation.

[0053] In S1, install vibration sensors on the frequency conversion production equipment to establish a device monitoring network, and collect the original vibration signal sequence under the no-load stable state of the equipment.

[0054] This can be achieved by querying the database or integrating with the production line equipment management system (such as ERP, SCADA system) to identify and count all the frequency conversion production equipment (such as air compressors, pump groups, etc.) in the production line. Install piezoelectric vibration sensors at the key parts of each device (such as motors, bearings, pump bodies, etc.). These positions should be based on the structural characteristics of the device to ensure that the key vibration sources that may cause failures are monitored. Piezoelectric sensors are suitable for frequency conversion devices because their high sensitivity can effectively capture weak vibration signals, especially resonance signals at low frequencies and infrasonic frequencies.

[0055] Configure and implement the Time-Sensitive Networking (TSN) protocol. TSN supports real-time data transmission and ensures accurate alignment of communication times between devices. TSN can support high-precision clock synchronization, ensure time consistency of all device nodes, and avoid data inconsistency caused by time delays. Configure the device nodes to send time calibration signals within a fixed period, and perform clock correction through synchronization algorithms (such as offset adjustment algorithms). After each device node receives the synchronization signal, it adjusts its local clock to achieve global synchronization.

[0056] Under the no-load stable state of the device, collect the three-axis vibration signals (acceleration signals in the X, Y, and Z directions). The device should be in a constant operating state to ensure no interference from external load changes. These vibration signals will provide the basic vibration characteristic data of the device. Use a denoising algorithm (such as wavelet denoising or Kalman filtering) to denoise the collected signals. Electromagnetic interference noise usually appears in the high-frequency region, and the denoising algorithm restores the true characteristics of the signal by filtering out the high-frequency noise components. Set an appropriate time window (such as 1 second, 5 seconds, etc.) to segment and collect the vibration signals of the device, forming a vibration signal data sequence for each device at different time points.

[0057] In S2, based on the original vibration signal sequence, construct a reference template of time-frequency characteristics. By comparing the feature vectors of the real-time vibration signal sequence with the feature matching degree of the reference template, determine whether the characteristics of the real-time vibration signal are valid.

[0058] The time-domain kurtosis index is an index used to measure the peak characteristics of a signal. The larger the calculated time-domain kurtosis value of the vibration signal, the more spike components there are in the signal, which is usually related to the abnormal vibration of the device. The specific kurtosis calculation formula is as follows:

[0059] ;

[0060] In the formula, is the kurtosis value of the vibration signal sequence, is the number of signal samples within the set time window, is the signal value, representing the amplitude of the a-th signal sample, 、 are the mean and standard deviation of the signal samples within the set time window, respectively.

[0061] Convert the vibration signal to the frequency domain through the fast Fourier transform (FFT), calculate the energy spectrum of the signal, extract the energy distribution information of each frequency band through the spectrum data, identify the main frequency components and their corresponding energy values, and form frequency-domain characteristics.

[0062] Integrate the kurtosis value in the time domain and the energy data in the frequency domain to construct a multi-dimensional feature vector of the device. These feature vectors will reflect the vibration characteristics of the device and serve as the reference characteristics for the normal operating state of the device without electromagnetic interference, forming a feature reference template.

[0063] The device monitoring network monitors the real-time vibration signal sequence of the device in real time, extracts the corresponding time-frequency feature vectors and calculates the feature matching degree with the reference template through cosine similarity. Among them, the size of each sampling period is the same as the size of the original vibration signal sequence, and the calculation method of its matching degree is as follows:

[0064] ;

[0065] In the formula, is the result of matching degree calculation, , are the eigenvectors (including time-domain kurtosis index and frequency-domain energy characteristics) of the a-th original vibration signal and the real-time vibration signal respectively.

[0066] When the average matching degree of consecutive (more than 1) sampling periods exceeds the set matching degree threshold (set comprehensively based on the device operating environment and electromagnetic interference degree, with the default setting being 0.95), a feature valid confirmation signal is sent, indicating that the vibration characteristics of the device continuously conform to the normal working state and there is no complex environment interference at this time.

[0067] In S3, a sliding window is used to perform blind source separation on the collected real-time signal sequence, and the resonance coupling component with spatial coherence characteristics is extracted from the independent vibration components.

[0068] When the device monitoring network receives the feature valid confirmation signal, it means that the matching degree between the real-time vibration signal sequence and the benchmark template has exceeded the set threshold, and the signal can be regarded as valid. At this time, the monitoring network performs real-time processing on the collected original signal sequence through the sliding window technology. The role of the sliding window is to ensure the step-by-step analysis of the signal in the continuous data stream, avoiding the computational pressure brought by processing a large amount of data at one time. The window size is set to 100 ms or shorter to ensure the rapid response of the signal, and the sliding step is set to 20 ms or shorter to ensure the accuracy of the continuous analysis of the signal.

[0069] The real-time signal sequence is whitened by diagonalizing the covariance matrix. Whitening processing is to diagonalize the covariance matrix of the signal so that the various signal components are no longer correlated. Its main purpose is to eliminate the correlation in the signal and ensure that the subsequent blind source separation can be carried out more effectively.

[0070] The FastICA algorithm is used to perform independent component analysis (ICA) on the whitened signal. The purpose is to separate the independent vibration components of each device. The FastICA algorithm is based on the maximum negative entropy criterion. By continuously iteratively optimizing the separation matrix W to maximize the negative entropy (i.e., non-Gaussianity) of the signal, blind source separation is achieved. The specific expression of the process of realizing the maximum negative entropy criterion is:

[0071] ;

[0072] In the formula, w is the basis vector being optimized, is the updated weight vector, indicating the direction of each independent component. By gradually optimizing w, the most independent components can be extracted from the real-time signal. N is the number of signals within the sliding window, is the frequency-domain eigenvector of the i-th signal, is the inner product of the current weight vector w and the signal , representing the projection of the extracted signal component in a certain dimension. is the learning rate, used to control the amplitude of each update. g(y) is a non-linear function (where y is an arbitrary variable). Common choices include tanh(y), or other higher-order functions. The role of this function is to increase the non-linear components of the signal, making the independent components more independent. The maximization of negentropy depends on the non-Gaussianity of the signal. A non-linear function is used to enhance this non-Gaussianity, thereby effectively extracting independent components.

[0073] Negentropy is defined as the degree of non-Gaussianity of the signal and is used to measure the independence of the signal. Each separated signal component represents an independent vibration source, corresponding to different vibration modes of the device.

[0074] Based on the physical topology of the device monitoring network, set the physical connection stiffness parameters of the device nodes to construct a transfer constraint matrix. The calculation method of the physical connection stiffness parameters is as follows:

[0075] ;

[0076] In the formula, is the physical connection stiffness parameter from device to device . is the propagation speed of the vibration signal in the medium connecting device and device . is the distance from device to device . is the vibration signal attenuation rate from device to device . The above variable values are all obtained through actual tests. The magnitude of the physical connection stiffness parameter determines the coupling strength of the vibration signal between devices, and thus affects the degree of the resonance coupling effect. The transfer constraint matrix will serve as the basis for subsequent transfer function calculation and signal correction.

[0077] After independent component analysis, each separated vibration component will be affected by the physical topology, propagation delay, and phase shift between devices. To ensure that the separated vibration components can accurately reflect the vibration coupling relationship between devices, matrix projection operation is used to correct the phase shift of the separated signal.

[0078] By using the transfer constraint matrix to correct the phase of each independent vibration component, the vibration signals of each device node are adjusted to conform to the actual propagation path. The goal of matrix projection calculation is to minimize the phase shift of the signal and ensure the spatio-temporal consistency of the vibration signal. The frequency components that synchronously appear among multiple device nodes are identified. These frequency components exhibit spatial coherence (a common method is to calculate the Pearson correlation coefficient), that is, these frequencies are significantly manifested in the vibration signals of multiple devices. By setting a correlation threshold, those frequency components are screened out that have a high correlation in the vibration signals of multiple devices (for example, the correlation coefficient is greater than the set value), and these components are the resonance coupling frequencies. Energy extraction is performed on these resonance frequencies, the corresponding resonance coupling modes are identified, and a basis is provided for subsequent vibration control and energy optimization.

[0079] In S4, based on the independent vibration components after blind source separation, a weighted graph data structure is constructed, and a resonance mode identification model is established using the graph convolutional neural network architecture.

[0080] The time-frequency characteristics of the independent vibration components and resonance coupling components after blind source separation are fused with the device physical topology coordinate data to construct a weighted resonance mode graph, and the edge weights of the graph reflect the physical connection stiffness parameters between device nodes.

[0081] The constructed resonance mode graph is used as the input of the graph convolutional neural network (GCN). The node features include the vibration signal features (time-frequency features) of the device, and the edge weights are the physical connection stiffness parameters between devices. The vibration feature information of neighboring nodes is aggregated layer by layer. Through the convolution operation, the node features are weighted and summed with the features of their neighboring nodes, so as to obtain the resonance information between nodes.

[0082] Based on the graph convolutional network, a multi-head attention mechanism is added to strengthen the information aggregation of specific neighborhoods by calculating the attention weights between each device node and its neighboring nodes. Through the multi-head attention mechanism, different resonance coupling modes between device nodes can be captured, and the resonance energy transfer coefficient between each pair of device nodes is calculated. The specific expression is:

[0083] ;

[0084] In the formula, 、 are the feature vectors of node o and node p respectively, and are the weight matrices of the q-th dimension and the k-th dimension in the attention mechanism respectively, k is the dimension of the vector, is the calculated attention coefficient representing the resonance energy transfer coefficient. Based on the calculation results of the resonance energy transfer coefficient, a square matrix of H*H (H is the total number of nodes in the device monitoring network) is established The resonance coupling strength matrix in the frequency - space dimension.

[0085] Set the resonance coupling strength threshold (comprehensively set based on the coefficient magnitudes in the resonance coupling strength matrix and the magnitudes of the device resonance coupling components), mark the adjacent device node pairs with resonance coupling strength exceeding the threshold as dangerous resonance modes, and record their dominant resonance coupling strength and energy transfer paths.

[0086] In S5, connect the trained and verified resonance mode recognition model to the device monitoring network for dangerous resonance mode recognition, and send frequency collision avoidance instructions to the corresponding device nodes according to the recognition results.

[0087] Select the historical operation data of the device network as the training set and the verification set. These historical data include the vibration signals of device nodes in different operating states, especially the vibration data when the device experiences resonance phenomena. The data set should contain the vibration signals of the device under normal operation and dangerous resonance conditions to ensure that the model can recognize the diversity and variability of resonance modes. Select a certain proportion of the data from the historical operation data as the training set, and the data should cover the vibration signals of the device under different working conditions such as no - load, light - load, and heavy - load. Select the device operation data in different time periods as the verification set for model testing and parameter adjustment.

[0088] Input the selected training set data to train the resonance mode recognition model. This model mainly predicts the possible resonance frequencies and intensities based on the vibration signal characteristics of the device (time - frequency characteristics, kurtosis index). During training, use the supervised learning method. Through the labeled dangerous resonance mode data, the model continuously adjusts its parameters to learn the relationship between the vibration signal patterns and resonance phenomena.

[0089] Through the trained model, calculate the predicted resonance coupling strength, and compare it with the measured resonance coupling strength in the verification set data. Use the mean square error index to evaluate the prediction accuracy of the model, and adjust the model parameters to minimize the prediction error.

[0090] Connect the trained and verified resonance mode recognition model to the device monitoring network. The device monitoring network collects the vibration signals of the device in real - time and inputs them into the resonance mode recognition model for real - time analysis. The model identifies whether the device is in a dangerous resonance mode state based on the real - time vibration signals. The standard for a dangerous resonance mode means that the device produces an obvious resonance coupling effect at a certain operating frequency, resulting in excessive energy transfer, which may cause device damage or ineffective energy consumption. When the vibration signal of the device matches the resonance frequency predicted by the model and a dangerous resonance mode is recognized, the model will output a frequency collision avoidance instruction signal indicating which device is in a dangerous resonance state.

[0091] In S6, for the devices that receive the frequency collision avoidance instruction, a multi-objective optimization function including the resonance energy transfer coefficient is constructed, and the optimal working frequency combination of each frequency conversion device is solved and stored in the device control network.

[0092] After the device receives the frequency collision avoidance instruction, first, the frequency search space is initialized according to the working characteristics of the device. The working frequency range of the device is determined by the lowest frequency and the highest frequency adjustment range of the device, and the minimum step size of frequency adjustment is determined by the minimum frequency conversion scale of the device.

[0093] According to the resonance coupling strength matrix between devices and the real-time vibration characteristics of the devices, a multi-objective optimization function including the resonance energy transfer coefficient is constructed. The purpose of this function is to minimize the resonance coupling strength between devices while avoiding resonance between devices and reducing power consumption or improving working efficiency as much as possible. Therefore, the actual purpose is to minimize the resonance coupling strength between devices, that is, to make less than a set threshold to avoid generating dangerous resonance modes, and at the same time, or improve the working efficiency of the device as much as possible. The specific expression is:

[0094] ;

[0095] In the formula, is the set optimization function, is the coefficient for weighing power consumption and resonance coupling, which is specifically set based on the actual device energy consumption and the resonance energy transfer coefficient. is the actual power consumption of the o-th device, is the current working frequency of device o.

[0096] To improve the search ability of the particle swarm optimization algorithm, a Cauchy mutation mechanism is introduced. The Cauchy mutation mechanism dynamically adjusts the search radius of the particle by introducing a new mutation probability. Specifically, the Cauchy mutation mechanism increases the global search ability for the particle position. Especially in the case of a large search space, it can prevent falling into local optimal solutions. By dynamically adjusting the mutation probability to expand the search radius, the real-time independent vibration component of the device and the parameters of the coupling strength matrix are input into the multi-objective optimization function. After all particles are updated, by calculating the fitness (optimization objective) of each particle, the optimal frequency combination, that is, the device working frequency that can minimize the resonance coupling strength and power consumption to the greatest extent, is obtained, and the corresponding parameters are sent to the device monitoring network.

[0097] In S7, the optimal working frequency combination in the device monitoring network is obtained, and a frequency regulation instruction is sent to the edge controller of the target device to perform the frequency adjustment operation.

[0098] The device monitoring network adopts a time-sensitive network communication protocol and realizes communication scheduling between devices through a priority queue. The frequency regulation instructions of each device node will be queued in the data transmission queue, and the frequency regulation instructions will be sent to the edge controller of the target device according to the preset routing table.

[0099] The device node stores the received frequency regulation instructions in the register, dynamically adjusts the instruction effective timing according to the current operating state of the device, and performs the frequency adjustment operation when the controller ready signal is aligned with the clock synchronization pulse, ensuring that the device operates at the optimal frequency to avoid resonance and improve the operating energy efficiency of the device.

[0100] The above formulas are all dimensionless and take their numerical values for calculation. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters and threshold selection in the formula are set by those skilled in the art according to the actual situation.

[0101] 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 processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center containing one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0102] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0103] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and modules described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0104] In 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 merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of the devices or modules can be in electrical, mechanical, or other forms.

[0105] The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules. They can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0106] In addition, in each embodiment of the present application, the functional modules can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0107] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0108] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims described above.

[0109] Finally: The above description is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A production energy efficiency optimization method based on AI and industrial Internet, characterized in that: The steps include: S1: Install vibration sensors on variable frequency production equipment to establish an equipment monitoring network and collect the original vibration signal sequence of the equipment in a no-load stable state; S2: construct a reference template of time-frequency features based on the original vibration signal sequence, and judge whether the real-time vibration signal features are valid by comparing the feature vector of the real-time vibration signal sequence with the feature matching degree of the reference template; S3: Use a sliding window to perform blind source separation on the collected real-time signal sequence and extract the resonant coupling components with spatial coherence characteristics from the independent vibration components; S4: A weighted graph data structure is constructed based on the independent vibration components after blind source separation, and a resonant mode recognition model is established using a graph convolutional neural network architecture; S5: Connect the trained and verified resonance mode recognition model to the equipment monitoring network to identify dangerous resonance modes, and send frequency collision avoidance instructions to the corresponding equipment nodes according to the recognition results; S6: construct a multi-objective optimization function including a resonance energy transfer coefficient for the device that receives the frequency collision avoidance instruction, solve the optimal operating frequency combination of each frequency conversion device and store it in the device control network; S7: Obtain the optimal operating frequency combination in the device monitoring network, and send a frequency control instruction to the edge controller of the target device to perform a frequency adjustment operation; In S4, a weighted graph data structure is constructed based on the independent vibration components after blind source separation, and a resonant mode recognition model is established using a graph convolutional neural network architecture. Specifically, the following steps are involved: The time-frequency characteristics of the independent vibration components and resonant coupling components after blind source separation are integrated with the physical topological coordinate data of the equipment to construct a weighted resonant modal graph. The edge weights of the graph reflect the physical connection stiffness parameters between equipment nodes. A resonance mode recognition model is established using a graph convolutional neural network architecture. The constructed resonance mode graph is used as the input of the graph convolutional neural network. The independent vibration components of the neighboring device nodes are aggregated layer by layer. The resonance energy transfer coefficient between devices is calculated through a multi-head attention mechanism to generate a resonance coupling strength matrix in the frequency-space dimension. A resonance coupling strength threshold is set, and adjacent device node pairs whose resonance coupling strength exceeds the threshold are marked as dangerous resonance modes, and their dominant resonance coupling strength and energy transfer path are recorded.

2. According to claim 1, a production energy efficiency optimization method based on AI and industrial Internet is characterized in that: In S1, vibration sensors are installed on variable frequency production equipment to establish an equipment monitoring network, and the original vibration signal sequence of the equipment in the no-load stable state is collected, including: Count all the variable frequency production equipment in the production line, arrange piezoelectric vibration sensors at the installation locations of the variable frequency production equipment, convert all equipment into a node-based physical topology structure based on the location distribution of the sensors, and establish an equipment monitoring network; Configure the time-sensitive network communication protocol, establish a clock synchronization link between device nodes, send periodic timing signals for time alignment, and build a data transmission priority queue based on the device node topology structure; The three-axis vibration signal sequence of the equipment in the no-load stable state is collected, and the denoising algorithm is applied to eliminate the electromagnetic interference noise to obtain the original vibration signal sequence of the equipment with a set time window size.

3. According to claim 2, a production energy efficiency optimization method based on AI and industrial Internet is characterized in that: In S2, a reference template of time-frequency features is constructed based on the original vibration signal sequence, and the feature matching degree of the feature vector of the real-time vibration signal sequence is compared with that of the reference template to determine whether the real-time vibration signal feature is valid. Specifically, the following steps are performed: A multi-dimensional feature vector is constructed based on the time domain kurtosis index and frequency domain energy data of the original vibration signal sequence to generate a reference template containing time-frequency features; The equipment monitoring network monitors the real-time vibration signal sequence of the equipment in real time, extracts the corresponding time-frequency feature vector and calculates the feature matching degree with the reference template through cosine similarity. When the matching degree mean of consecutive set sampling periods exceeds the set matching degree threshold, a feature validity confirmation signal is sent.

4. According to claim 3, a production energy efficiency optimization method based on AI and industrial Internet is characterized in that: In S3, a sliding window is used to perform blind source separation on the collected real-time signal sequence, and the resonant coupling components with spatial coherence characteristics in the independent vibration components are extracted, specifically including: When the equipment monitoring network receives a feature valid confirmation signal, a sliding window is used to perform blind source separation on the collected real-time signal sequence; The real-time signal sequence is whitened by diagonalizing the covariance matrix, and the FastICA algorithm is used to perform independent component analysis on the whitened signal. The separation matrix is ​​iteratively solved based on the negative entropy maximization criterion to separate the independent vibration components of each device. Based on the physical topology of the equipment monitoring network, the physical connection stiffness parameters of the equipment nodes are set to construct the transfer constraint matrix, where the physical connection stiffness parameters are calculated as follows: ； In the formula, For equipment To device The physical connection stiffness parameters, For vibration signals in the device With equipment The propagation speed in the connecting medium, For equipment To device The distance For equipment To device The vibration signal attenuation rate; The phase offset of the separated components is corrected by matrix projection operation, and the resonant coupling components with spatial coherence characteristics in the independent vibration components are extracted.

5. The production energy efficiency optimization method based on AI and industrial Internet according to claim 4 is characterized in that: In S5, the trained and verified resonance mode recognition model is connected to the equipment monitoring network to identify dangerous resonance modes, and a frequency collision avoidance instruction is sent to the corresponding equipment node according to the recognition result, which specifically includes: The historical operation data of the equipment network is selected as the training set and the validation set data to input into the resonance mode identification model, and the model parameters are adjusted based on the matching degree deviation between the predicted resonance coupling strength and the measured dominant resonance coupling strength of the validation set data; When the prediction accuracy of the model verified by a set number of consecutive groups of historical operation data reaches a preset index, the model is connected to the equipment monitoring network to identify dangerous resonance modes, and frequency collision avoidance instructions are sent to adjacent equipment nodes corresponding to the identified dangerous resonance modes.

6. The production energy efficiency optimization method based on AI and industrial Internet according to claim 5 is characterized in that: In S6, a multi-objective optimization function including a resonance energy transfer coefficient is constructed for the device that receives the frequency collision avoidance instruction, and the optimal operating frequency combination of each frequency conversion device is solved and stored in the device control network, which specifically includes: When the device receives a frequency collision avoidance instruction, it initializes the device operating frequency search space based on the resonance coupling strength matrix in the resonance mode diagram, defines the device adjustable frequency range and the minimum adjustment step, and constructs a multi-objective optimization function including the resonance energy transfer coefficient. The Cauchy mutation mechanism is introduced into the standard particle swarm optimization algorithm. The search radius is expanded by dynamically adjusting the mutation probability. The real-time independent vibration components of the equipment and the coupling strength matrix parameters are input into the multi-objective optimization function. The optimal operating frequency combination of each frequency conversion equipment is solved and stored in the equipment control network.

7. The production energy efficiency optimization method based on AI and industrial Internet according to claim 6 is characterized in that: In S7, obtaining the optimal operating frequency combination in the device monitoring network and sending a frequency control instruction to the edge controller of the target device to perform a frequency adjustment operation specifically includes: Obtain the optimal operating frequency combination in the device monitoring network, and send frequency control instructions to the edge controller of the target device according to the preset routing table based on the data transmission priority queue; The device node stores the received frequency control instruction into the register, dynamically adjusts the instruction effectiveness timing according to the current operating status of the device, and executes the frequency adjustment operation when the controller ready signal is aligned with the clock synchronization pulse.

Images