Physical mechanism embedded industrial AIGC basic model, training method, device, equipment, storage medium and program product

Abstract

The embodiment of the invention provides a physical mechanism embedded industrial AIGC basic model, a training method, a training device, equipment, a storage medium and a program product. Belongs to the technical field of machine learning. Comprising the steps of adding original noise to training data to obtain noise data; inputting the noise data and the training data label into a to-be-trained model to obtain predicted noise; determining constraint strength according to the training step number and a preset step number threshold value; determining physical loss according to the predicted noise and the physical constraint; determining comprehensive physical loss according to the constraint strength and the physical loss; determining an error value according to the predicted noise, the original noise and the comprehensive physical loss; and optimizing the to-be-trained model by adopting the error value to obtain a new to-be-trained model, and repeatedly executing the process from adding the original noise to determining the error value until the error value is smaller than a preset error threshold value or the new training step number reaches a step number threshold value to obtain a data generation model. The problem that the energy consumption of training the model conforming to the physical constraint is high is solved.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to an industrial AIGC basic model, training method, device, equipment, storage medium and program product with embedded physical mechanism. Background Art

[0002] Data generation has been applied to data expansion and data enhancement of industrial equipment operation data. Currently, data generation models such as AIGC (Artificial Intelligence Generated Content) models can be used for data generation.

[0003] At present, the AIGC model in related technologies is usually trained using pre-collected data, and only the patterns of the data are learned during the training process.

[0004] However, equipment operation data naturally follows specific physical constraints. If one needs to obtain a model that can generate equipment operation data that complies with physical constraints during current model training, a large amount of training data is required. The energy consumption generated by the computer during model training is high, and it is difficult to generate data that complies with physical constraints from the model generated by the trained data. Summary of the invention

[0005] The embodiments of the present application provide an industrial AIGC basic model, training method, device, equipment, storage medium and program product with embedded physical mechanism, so as to solve the problem that the energy consumption generated by the computer during the model training process is high and the data generation model obtained by training has difficulty in generating data that complies with physical constraints.

[0006] In a first aspect, an embodiment of the present application provides an industrial AIGC basic model training method with embedded physical mechanism, which is applied to a computer, including: obtaining multiple groups of training data and training data labels corresponding to the training data; adding original noise to the training data to obtain noise data; inputting the noise data and the training data labels into the model to be trained so that the model to be trained outputs predicted noise; adding 1 to the number of training steps to obtain a new number of training steps; determining the constraint strength according to the new number of training steps and a preset step threshold; determining the physical loss according to the predicted noise and the preset physical constraint; determining the comprehensive physical loss according to the constraint strength and the physical loss; determining the error value according to the predicted noise, the original noise and the comprehensive physical loss; optimizing the model to be trained using the error value to obtain a new model to be trained, and again executing the process of adding the original noise to the training data to determine the error value, until the error value is less than the preset error threshold or the new number of training steps reaches the step threshold, and the new model to be trained is determined as the industrial AIGC basic model.

[0007] In a possible implementation, the constraint strength is determined according to the new number of training steps and a preset step threshold, including: dividing the new number of training steps by the step threshold to obtain a training progress; and calculating the constraint strength using the training progress.

[0008] In a possible implementation, the training progress is used to calculate the constraint strength, including: determining the difference between a preset value and the training progress as the incomplete ratio; dividing the training progress by the incomplete ratio to obtain a progress ratio; taking the logarithm of the progress ratio with a preset constant as the base to obtain a logarithmic value; multiplying the logarithmic value by a preset coefficient and inputting the result into a Sigmoid function to obtain the constraint strength.

[0009] In one possible implementation, the physical loss is determined based on the predicted noise and the preset physical constraints, including: inputting the predicted noise into the constraint equations corresponding to the physical constraints to obtain the output results of the constraint equations; and adding the output results of the constraint equations to obtain the physical loss.

[0010] In one possible implementation, the physical constraints include change trend constraints; the predicted noise is input into the constraint equations corresponding to the physical constraints to obtain the output results of the constraint equations, including: the predicted noise is input into the constraint equations corresponding to the change trend constraints to obtain the output results of the constraint equations corresponding to the change trend constraints.

[0011] In a possible implementation, each training step includes at least two time steps; the training data includes at least two data types, and the data types include a baseline type and at least one associated type associated with the baseline type; the prediction noise is input into the constraint equation corresponding to the change trend constraint to obtain the output result of the constraint equation corresponding to the change trend constraint, including: using the prediction noise of the current time step of the baseline type to subtract the prediction noise of the previous time step to obtain the baseline prediction noise difference; using the baseline prediction noise difference, divided by the sum of the prediction noise of the previous time step of the baseline type and a preset value, to obtain the characteristic slope corresponding to the baseline type; using the prediction noise of the current time step of the associated type to subtract the prediction noise of the previous time step to obtain the associated prediction noise difference; using the associated prediction noise difference, divided by the sum of the prediction noise of the previous training step of the associated type and a preset value, to obtain the characteristic slope corresponding to the associated type; calculating the mean square error between the characteristic slope corresponding to each associated type and the characteristic slope corresponding to the baseline type; summing the mean square errors and dividing them by the total number of time steps to obtain the output result of the constraint equation corresponding to the change trend constraint.

[0012] In a possible implementation, an error value is determined based on the predicted noise, the original noise and the comprehensive physical loss, including: calculating the noise mean square error between the predicted noise and the original noise; adding the noise mean square error to the comprehensive physical loss to obtain the error value.

[0013] In one possible implementation, after the error value is used to optimize the model to be trained to obtain a new model to be trained, and the process of adding original noise to the training data to determine the error value is performed again until the error value is less than a preset error threshold or the new number of training steps reaches the step threshold, and the new model to be trained is determined as the industrial AIGC basic model, it also includes: in response to detecting a data label input by a user, inputting the data label into the industrial AIGC basic model so that the industrial AIGC basic model outputs device operating parameters; outputting device operating parameters.

[0014] In one possible implementation, in response to detecting a data label input by a user, the data label is input into the industrial AIGC basic model so that the industrial AIGC basic model outputs the equipment operating parameters, and it also includes: inputting the data label into a pre-trained denoising model so that the denoising model denoises the preset noise data for multiple time steps to obtain the equipment operating data corresponding to the data label; obtaining various intermediate data generated by the denoising model in the process of generating equipment operating data; determining the intermediate physical loss based on the intermediate data and preset physical constraints; for any time step, taking the partial derivative of the intermediate physical loss based on the intermediate data to obtain the error gradient; determining the operating data to be output based on the equipment operating data and the error gradient corresponding to each time step; and outputting the operating data to be output.

[0015] In the second aspect, an embodiment of the present application provides an industrial AIGC basic model with embedded physical mechanism, which is trained using the first aspect and / or various possible implementation methods of the first aspect above; the industrial AIGC basic model with embedded physical mechanism includes: a label detection module, a data generation module and a data output module; a label acquisition module for acquiring data labels input by users; a data generation module for using data labels to perform denoising on preset noise data to obtain equipment operating parameters that meet preset physical constraints; a data output module for outputting equipment operating parameters.

[0016] In a third aspect, an embodiment of the present application provides an industrial AIGC basic model training device with embedded physical mechanism, which is applied to a computer, including: a data acquisition module, used to acquire multiple groups of training data and training data labels corresponding to the training data; a noise adding module, used to add original noise to the training data to obtain noise data; a model output module, used to input the noise data and the training data label into the model to be trained so that the model to be trained outputs predicted noise; a step increase module, used to add 1 to the training step number to obtain a new training step number; a strength determination module, used to determine the constraint strength according to the new training step number and a preset step number threshold; a physical loss module, used to determine the physical loss according to the predicted noise and the preset physical constraint; a comprehensive loss module, used to determine the comprehensive physical loss according to the constraint strength and the physical loss; an error determination module, used to determine the error value according to the predicted noise, the original noise and the comprehensive physical loss; a model optimization module, used to optimize the model to be trained using the error value to obtain a new model to be trained, and again execute the process of adding the original noise to the training data to determine the error value, until the error value is less than the preset error threshold or the new training step number reaches the step threshold, and the new model to be trained is determined as the industrial AIGC basic model.

[0017] In a fourth aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;

[0018] Memory stores computer-executable instructions;

[0019] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0020] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementations of the first aspect.

[0021] In a sixth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0022] The embodiments of the present application provide an industrial AIGC basic model, training method, apparatus, equipment, storage medium and program product with embedded physical mechanism. In the process of training the model, the constraint strength is determined according to the number of training steps and the step threshold, and the physical loss is obtained by using the physical constraint. The comprehensive physical loss is determined by combining the constraint strength and the physical loss, and added to the error value. The model is optimized by using the error value with the comprehensive physical loss added, so that the model learns the physical constraints in the process of training, so that the data generated by the trained industrial AIGC basic model conforms to the physical constraints. During the training process, less training data is required and the training time is shorter than that without adding physical constraints. Moreover, since the constraint strength varies with the number of training steps, the data regularity can be learned first in the early stage of training, and the influence of the physical constraints on the model training process can be gradually increased in the later stage of training to increase the stability of the training. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0024] Figure 1 A schematic diagram of the scenario of the industrial AIGC basic model training method with embedded physical mechanism provided in this application;

[0025] Figure 2 A flow chart of the industrial AIGC basic model training method with embedded physical mechanism provided in the embodiment of the present application;

[0026] Figure 3 A schematic diagram of the structure of an industrial AIGC basic model training device with embedded physical mechanism provided in an embodiment of the present application;

[0027] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0028] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0029] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0030] Data generation technology is now widely used in the industrial field, especially playing an important role in expanding and enhancing equipment operation data.

[0031] In existing related technologies, the industrial AIGC basic model is generally trained based on pre-collected data. However, this training method can only enable the model to learn the surface rules of the data.

[0032] It should be pointed out that the equipment operation data itself has clear physical constraint characteristics. If we want to generate equipment operation data that conforms to these physical constraints through model training, we currently face many challenges. On the one hand, a large amount of training data is required to ensure that the model can fully learn the physical constraint characteristics; on the other hand, during the model training process, computers consume a lot of energy. More importantly, even if a large amount of data and energy are invested, it is still quite difficult for the trained industrial AIGC basic model to generate data that strictly follows physical constraints.

[0033] This application is used in the scenario of training the basic model of industrial AIGC. It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation portals for users to choose to authorize or refuse.

[0034] Figure 1 A schematic diagram of the scenario of the industrial AIGC basic model training method with embedded physical mechanism provided in this application. Figure 1 , this scenario includes: a computer 101 and a database 102 .

[0035] In a specific implementation process, the terminal device 101 may include a computer, a server, a tablet, a mobile phone, a PDA (Personal Digital Assistant), a notebook, etc., which can input data.

[0036] The database 102 may include a single database or a combination of multiple databases, such as Oracle database, MySQL (relational database management system), DRDS (Distribute Relational Database Service, distributed relational database) database, ES (Elasticsearch, elastic search) database, etc., and this application does not impose any special restrictions on this.

[0037] The connection between the computer 101 and the database 102 can be a wired connection or a wireless connection. The computer 101 is used to read the training data and the training data labels corresponding to the training data stored in the database 102, and use the training data and the training data labels corresponding to the training data to perform model training, and is also used to generate data after the training is completed.

[0038] It is understandable that the scenarios illustrated in the embodiments of the present application do not constitute a specific limitation on the training method of the industrial AIGC basic model embedded in the physical mechanism. In other feasible implementations of the present application, the above scenarios may include more or fewer components than shown in the figure, or combine certain components, or split certain components, or arrange the components differently, which can be determined according to the actual application scenario and is not limited here. Figure 1 The scenarios shown can be implemented by hardware, software, or a combination of software and hardware.

[0039] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0040] Figure 2 A flow chart of the method for training an industrial AIGC basic model with embedded physical mechanism provided in the embodiment of the present application. The execution subject of the embodiment of the present application may be Figure 1 Computer 101 in. Figure 2 As shown, the method includes:

[0041] S201: Obtain multiple groups of training data and training data labels corresponding to the training data.

[0042] In this step, it may include reading the training data and the associated new linked data tags from a preset location in a storage unit in the computer, and it may also include reading the training data and the associated new linked data tags from a database.

[0043] S202: Add original noise to the training data to obtain noise data.

[0044] In this step, it may include gradually adding Gaussian noise to the training data to obtain noise data.

[0045] Specifically, this step can be expressed as a Markov chain:

[0046]

[0047] in, Represents the time step The noise data at Represents the time step The noise data at is a preset parameter that controls the amount of noise added at each step. represents the normal distribution and I represents the identity matrix. At the end of the forward diffusion process, the data is severely corrupted by noise and transformed into a distribution that is closer to a standard Gaussian distribution.

[0048] S203: Input the noise data and the training data label into the model to be trained, so that the model to be trained outputs the predicted noise.

[0049] In this step, the model to be trained can be any neural network model. The model to be trained attempts to gradually restore the noisy data to the original data to obtain the predicted noise. The reverse denoising process aims to train the neural network to gradually restore the Gaussian noise to the original data distribution. The process is defined as:

[0050]

[0051] in, Represented by the time step The noise data at time is restored to the time step The noise data at Represents the time step The noise data at Represents the time step The noise data at represents a normal distribution, represents the mean of the prediction noise at time step t, represents the variance of the prediction noise at time step t.

[0052] S204: Add 1 to the number of training steps to obtain a new number of training steps.

[0053] For example, if the initial number of training steps is 0, then add 1 to the number of training steps, and the new number of training steps is 1. For another example, if the initial number of training steps is 5, then add 1 to the number of training steps, and the new number of training steps is 6.

[0054] S205: Determine the constraint strength according to the new training steps and the preset step threshold.

[0055] In this step, the training progress may be determined according to the training steps and the step threshold, and the corresponding relationship between the preset training progress and the constraint strength may be found according to the training progress to obtain the constraint strength.

[0056] Among them, the correspondence between the training progress and the constraint strength can be preset by the staff and stored in the format of table, key-value pair, etc.

[0057] S206: Determine the physical loss according to the predicted noise and the preset physical constraints.

[0058] In this step, the predicted noise is input into the constraint equation corresponding to the preset physical constraint, and the physical loss is calculated.

[0059] S207: Determine the comprehensive physical loss based on the constraint strength and the physical loss.

[0060] In this step, the constraint strength is multiplied by the physical loss to obtain the comprehensive physical loss.

[0061] S208: Determine an error value according to the predicted noise, the original noise and the comprehensive physical loss.

[0062] In this step, the noise loss is determined according to the predicted noise and the original noise, and the noise loss and the comprehensive physical loss are weighted and summed to obtain an error value.

[0063] The noise loss is determined according to the predicted noise and the original noise, including inputting the predicted noise and the original noise into a preset loss function to calculate the noise loss.

[0064] S209: Use the error value to optimize the model to be trained to obtain a new model to be trained, and again perform the process of adding original noise to the training data to determine the error value until the error value is less than the preset error threshold or the new number of training steps reaches the step threshold, and the new model to be trained is determined as the industrial AIGC basic model.

[0065] In this step, the loss function is minimized by using optimization algorithms such as stochastic gradient descent to optimize the model to be trained.

[0066] From the description of the above embodiments, it can be seen that the embodiments of the present disclosure determine the constraint strength according to the number of training steps and the step threshold during the model training process, and use physical constraints to obtain physical losses. The comprehensive physical losses are determined by combining the constraint strength and the physical losses, and added to the error values. The model is optimized using the error values ​​with the comprehensive physical losses added, so that the model learns the physical constraints during the training process, so that the data generated by the trained industrial AIGC basic model conforms to the physical constraints. During the training process, less training data is required and the training time is shorter than when no physical constraints are added. Moreover, since the constraint strength varies with the number of training steps, the data patterns can be learned first in the early stage of training, and the influence of physical constraints on the model training process can be gradually increased in the later stage of training to increase the stability of the training.

[0067] In a possible implementation, the step S205 determines the constraint strength according to the new training steps and the preset step threshold, including: step S2051 and step S2052.

[0068] S2051: Divide the new number of training steps by the step threshold to obtain the training progress.

[0069] In this step, for example, if the new number of training steps is 10 and the step threshold is 50, the training progress is 0.2. For another example, if the new number of training steps is 10 and the step threshold is 100, the training progress is 0.1 or 10%.

[0070] S2052: Calculate constraint strength using training progress.

[0071] In this step, the training progress may be multiplied by a preset value to obtain the constraint strength, or the training progress may be directly used as the constraint strength.

[0072] From the description of the above embodiments, it can be seen that the embodiments of the present disclosure calculate the training progress and use the training progress to calculate the constraint strength, so that the constraint strength changes with the training progress, so that the data patterns are learned first in the early stage of training, and the influence of physical constraints on model training is gradually increased during the training process, thereby increasing the stability of the model training process and reducing the possibility of training errors.

[0073] In a possible implementation, in the above step S2052, the constraint strength is calculated using the training progress, including: steps S521 to S524.

[0074] S521: Determine the difference between the preset value and the training progress as the incomplete ratio.

[0075] In this step, the preset value is, for example, 1, 2, 3, etc.

[0076] S522: Divide the training progress by the incomplete ratio to obtain a progress ratio.

[0077] In this step, for example, if the training progress is 0.4 and the unfinished ratio is 0.6, the progress ratio is 2:3. For another example, if the training progress is 0.2 and the unfinished ratio is 0.8, the progress ratio is 0.25.

[0078] S523: Taking the logarithm of the progress ratio with a preset constant as the base, a logarithm value is obtained.

[0079] In this step, preset constants such as natural constants e, 10, etc. may also be pre-calibrated by the staff.

[0080] S524: Multiply the logarithmic value by the preset coefficient and input the result into the Sigmoid function to obtain the constraint strength.

[0081] In this step, the preset coefficient may be preset by the staff.

[0082] The above steps S521 to S524 can be expressed as the following formula:

[0083] Φ

[0084] Where Φ represents the constraint strength, Sigmoid represents the Sigmoid function, α represents the preset coefficient, ln represents the logarithmic function, and x represents the training progress.

[0085] From the description of the above embodiments, it can be seen that the embodiments of the present disclosure use the Sigmoid function to smoothly map the input to the range of "[0,1]". When the input is small, the growth is slow, and when the input is large, the growth is accelerated, so as to reduce the impact of physical constraints on model training in the early stage of training, and increase the impact of physical constraints on model training in the later stage. The preset coefficient α controls the rate and amplitude of the change of constraint strength. A larger value will lead to a steeper change in strength, thereby increasing the effect of constraint strength.

[0086] In a possible implementation, in the above step S206, determining the physical loss according to the predicted noise and the preset physical constraints includes: step S2061 and step S2062.

[0087] S2061: Input the predicted noise into the constraint equations corresponding to each physical constraint to obtain the output results of each constraint equation.

[0088] In this step, there may be multiple types of prediction noises, and different types of prediction noises may correspond to different constraint equations. The prediction noises are input into the corresponding constraint equations to obtain output results of the constraint equations.

[0089] For example, the predicted noise includes the intake air pressure and the exhaust air pressure of the aircraft engine. The intake air pressure and the exhaust air pressure correspond to a preset constraint equation. Then, the intake air pressure and the exhaust air pressure are input into the corresponding constraint equation to obtain an output result.

[0090] S2062: Add the output results of each constraint equation to obtain the physical loss.

[0091] In this step, for example, there are currently 5 constraint equations, and the corresponding output results are 0.2, 0.5, 1, 0.3, and 0.5, respectively, then the physical loss is 2.5. For another example, there are currently 3 constraint equations, and the corresponding output results are 2, 5, and 10, respectively, then the physical loss is 17.

[0092] From the description of the above embodiments, it can be seen that the embodiments of the present disclosure input the predicted noise into the constraint equation to obtain the output results of each constraint equation, add the output results to obtain the physical loss, thereby realizing the simultaneous consideration of multiple physical constraints, so that the trained model conforms to various physical constraints, and the trained industrial AIGC basic model can generate a model that conforms to multiple physical constraints.

[0093] In a possible implementation, the physical constraint includes a change trend constraint.

[0094] The change trend constraint is used to determine whether the change trend between the associated physical quantities meets the preset change trend requirements.

[0095] In the above step S2061, the predicted noise is input into the constraint equation corresponding to each physical constraint to obtain the output result of each constraint equation, including:

[0096] S20611: Input the predicted noise into the constraint equation corresponding to the change trend constraint to obtain the output result of the constraint equation corresponding to the change trend constraint.

[0097] In this step, the prediction noise of adjacent time steps may be input into the constraint equation corresponding to the change trend constraint to obtain the output result of the constraint equation corresponding to the change trend constraint.

[0098] It can be seen from the description of the above embodiments that the embodiments of the present disclosure determine whether the changes between different predicted noises (equipment operation data) meet the preset physical constraints by adopting the change trend constraints.

[0099] In a possible implementation, each training step includes at least two time steps. The training data includes at least two data categories, and the data categories include a reference category and at least one associated category associated with the reference category.

[0100] The data change trend of the associated category should be related to the change trend of the benchmark category, showing a positive or negative correlation.

[0101] The above step S20611 predicts the constraint equation corresponding to the change trend constraint of the noise input to obtain the output result of the constraint equation corresponding to the change trend constraint, including: steps S6111 to S6116.

[0102] S6111: Subtract the prediction noise of the previous time step from the prediction noise of the current time step of the benchmark type to obtain a benchmark prediction noise difference.

[0103] In this step, for example, if the prediction noise of the benchmark category at the current time step is 100 and the prediction noise of the previous time step is 90, then the benchmark prediction noise difference is 10. For another example, if the prediction noise of the benchmark category at the current time step is 20 and the prediction noise of the previous time step is 8, then the benchmark prediction noise difference is 12.

[0104] S6112: Using the reference prediction noise difference, divide it by the sum of the prediction noise of the previous time step of the reference type and the preset value to obtain the characteristic slope corresponding to the reference type.

[0105] In this step, the preset value is a minimum value to avoid the denominator being zero.

[0106] In a possible implementation, the above step S6111 and step S6112 can be expressed as the following formula:

[0107]

[0108] In the formula, represents the characteristic slope, represents the prediction noise at the current time step, represents the prediction noise of the previous time step, Indicates the preset value.

[0109] S6113: Subtract the prediction noise of the previous time step from the prediction noise of the current time step of the associated type to obtain the associated prediction noise difference.

[0110] This step is similar to the above step S6111 and will not be repeated here.

[0111] S6114: Using the associated prediction noise difference, divided by the sum of the prediction noise of the previous training step of the associated category and the preset value, obtain the characteristic slope corresponding to the associated category.

[0112] This step is similar to the above step S6112 and will not be repeated here.

[0113] S6115: Calculate the mean square error between the feature slope corresponding to each associated category and the feature slope corresponding to the reference category.

[0114] In this step, the characteristic slope corresponding to the target associated category is subtracted from the characteristic slope corresponding to the reference category to obtain the slope difference, the square of the slope difference is calculated to obtain the square of the difference corresponding to the target associated category, the square of the difference corresponding to each associated category is added together, and then divided by the number of associated categories to obtain the mean square error.

[0115] S6116: Sum the mean square errors and divide the sum by the total number of time steps to obtain the output result of the constraint equation corresponding to the change trend constraint.

[0116] In this step, the total number of time steps can be pre-calibrated by the staff, or can be equal to the number of data groups in each set of training data.

[0117] In a possible implementation, the above step S6115 and step S6116 can be expressed as the following formula:

[0118]

[0119] In the formula, Represents the data results, i represents the i-th time step, n represents the total number of time steps, represents the characteristic slope corresponding to the associated type, It represents the characteristic slope corresponding to the benchmark category, and MSN represents the mean square error function.

[0120] From the description of the above embodiments, it can be seen that the embodiments of the present disclosure calculate the characteristic slope of the baseline category and the characteristic slope of the associated category at different time steps, and calculate the mean square error of the characteristic slopes of the baseline category and the associated category, sum the mean square errors, and then divide them by the total number of time steps to obtain the output result corresponding to the change trend constraint. The output result corresponding to the change trend constraint can be used to determine the correlation between the changes in the associated category and the baseline category in the prediction noise, so that the change trend of the related data tends to be consistent.

[0121] The application scenarios of the above trend constraints, such as the total fan inlet pressure of an aircraft engine ( )、Fan outlet total pressure( )、Low pressure compressor outlet pressure( ), and physical fan speed ( ) between the physical constraints of the operating parameters. The Bernoulli equation describes the relationship between gas flow rate and pressure. This relationship can be expressed by the Bernoulli equation:

[0122]

[0123] in is the static pressure of the gas, and represent the gas density and gas flow rate respectively, is the total pressure of the gas. The total pressure of a gas is the static pressure of the gas Gas dynamic pressure At the air intake of an aircraft engine, the total pressure at the fan inlet is Indicates the total pressure of the gas , fan intake speed Indicates gas flow rate .therefore, and The relationship between can be expressed by the above Bernoulli equation:

[0124]

[0125] At the same time, the fan speed It is a key factor affecting the air compression efficiency. The increase in intake speed The rise improves the efficiency of the fan and increases the total pressure at the fan inlet. .

[0126] In all turbulence models, the inlet and outlet pressures and Following the isentropic compression relationship:

[0127]

[0128] in is the specific heat ratio, R is the gas constant, , is the speed of sound. In aircraft engines, the total pressure at the fan outlet and the total pressure at the fan inlet can be considered as the outlet and inlet pressures. and .therefore, and The relationship between can be expressed by the above isentropic compression equation:

[0129]

[0130] in exist middle, is the inlet temperature. Absolute speed V and fan speed The equation shows that increasing Can improve air compression efficiency and increase Mach number and outlet pressure .along with The increase in the low-pressure compressor leads to a higher pressure, which results in a greater compression ratio and increases the outlet pressure of the low-pressure compressor. Increase accordingly.

[0131] In summary, the pressures at each stage of compression are interdependent. Specifically, the total fan inlet pressure ( )、Fan outlet total pressure( )、Low pressure compressor outlet pressure( ), and physical fan speed ( ) show synchronous changes and are applicable to the above-mentioned change trend constraints.

[0132] In a possible implementation, in the above step S208, determining the error value according to the predicted noise, the original noise and the comprehensive physical loss includes:

[0133] S2081: Calculate the noise mean square error between the predicted noise and the original noise.

[0134] In this step, the prediction noise can be the prediction noise of all time steps, and the prediction noise of each time step can include data of one or more data types. The difference between the data of each data type in the prediction noise and the original noise is calculated, the square of the difference is calculated, and the squares of the differences of each data type are added and divided by the total number of data types to obtain the noise mean square error.

[0135] S2082: Add the noise mean square error and the comprehensive physical loss to obtain an error value.

[0136] In this step, for example, if the noise mean square error is 2 and the comprehensive physical loss is 5, the error value is 7. For another example, if the noise mean square error is 10 and the comprehensive physical loss is 7, the error value is 17.

[0137] In this step, the above steps S2081 and S2082 can be expressed as:

[0138]

[0139] In the formula, represents the error value, Represents the expected symbol, specifically the training data is collected from the data distribution D, Represents the original noise is sampled from a Gaussian distribution of the identity matrix, The overall representation is the noise mean square error, Overall represents the comprehensive physical loss.

[0140] From the description of the above embodiments, it can be seen that the embodiments of the present disclosure obtain the error value by combining the error of noise and the comprehensive physical loss, and thus optimize the model by combining the loss of noise prediction and the loss of physical constraints, so that the model not only learns the data laws during the training process, but also learns the physical laws, thereby increasing the speed at which the model learns the physical laws, reducing the energy consumption generated by the computer during the training process, and making the data generated by the trained model conform to the physical laws.

[0141] In a possible implementation, without using the error value to optimize the model to be trained in step S209 to obtain a new model to be trained, and again performing the process of adding original noise to the training data to determine the error value until the error value is less than a preset error threshold or the new number of training steps reaches the step threshold, and the new model to be trained is determined as the industrial AIGC basic model, it also includes: step S210 and step S211.

[0142] S210: In response to detecting a data tag input by a user, inputting the data tag into an industrial AIGC basic model so that the industrial AIGC basic model outputs equipment operating parameters.

[0143] This step may include detecting a data tag input by a user through a keyboard, or detecting a data tag when the method of an embodiment of the present application is executed.

[0144] S211: Output device operating parameters.

[0145] This step may include displaying the device operating parameters, or may include writing the device operating parameters into a database.

[0146] From the description of the above embodiments, it can be seen that the embodiments of the present disclosure use the industrial AIGC basic model to generate equipment operating parameters in the data label correspondence table after detecting the data label input by the user, thereby realizing the generation of equipment operating parameters that meet physical constraints.

[0147] In a possible implementation, after the above step S210 responds to detecting the data label input by the user and inputs the data label into the pre-trained industrial AIGC basic model so that the industrial AIGC basic model outputs the equipment operating parameters, it also includes: steps S220 to S225.

[0148] S220: Input the data label into a pre-trained denoising model, so that the denoising model denoises the preset noise data in multiple time steps to obtain the equipment operation data corresponding to the data label.

[0149] The denoising model used in this step is a model without adding physical constraint training compared to the above step S210.

[0150] Among them, the reverse sampling process without physical information can be defined as:

[0151]

[0152] in Indicates that at time step The data input to the denoising model is represents the noise schedule, , .also, Represents the noise estimate of the denoising model predictions.

[0153] S221: Acquire various intermediate data generated by the denoising model in the process of generating equipment operation data.

[0154] In this step, the intermediate data may include the prediction noise generated by the denoising model at different time steps during the denoising process.

[0155] S222: Determine the intermediate physical loss according to each intermediate data and the preset physical constraints.

[0156] In this step, the intermediate data is input into a physical constraint function corresponding to a preset physical constraint to obtain an intermediate physical loss.

[0157] S223: For any time step, the partial derivative of the intermediate physical loss is calculated based on the intermediate data to obtain the error gradient.

[0158] In this step, the intermediate data of each time step is used as the variable, the intermediate physical loss is used as the objective function, and the derivative function is called to obtain the partial derivative of the intermediate data with respect to the intermediate physical loss, that is, the error gradient.

[0159] S224: Determine the operation data to be output according to the equipment operation data and the error gradient corresponding to each time step.

[0160] In this step, the error gradients of each time step are summed, multiplied by a preset weight value, to obtain a weighted total error gradient, and the weighted total error gradient is subtracted from the device operation data to obtain the operation data to be output. It can also include subtracting the error gradients of each practice unit from the device operation data to obtain the operation data to be output.

[0161] In a possible implementation, the above steps S222 to S224 can be expressed as the following formula:

[0162]

[0163] In the formula, Indicates the equipment operation data. Indicates the preset weight value, represents the sum of the error gradients at each time step.

[0164] S225: Output the operation data to be output.

[0165] This step is similar to the above step S211 and will not be repeated here.

[0166] From the description of the above embodiments, it can be seen that the embodiments of the present disclosure use a pre-trained denoising model to predict the equipment operation data, and use the intermediate data generated in the prediction process to calculate the error gradient, remove the error gradient data in the equipment operation data, and obtain the operation data to be output, thereby achieving the goal of not adding physical constraints during the model training process, so that the final data conforms to the physical constraints while reducing the model training time, thereby reducing the energy consumption of the computer during the model training process.

[0167] An embodiment of the present application also provides an industrial AIGC basic model with embedded physical mechanism, and the industrial AIGC basic model is trained by using the training method of the industrial AIGC basic model with embedded physical mechanism provided by any of the above embodiments; the industrial AIGC basic model with embedded physical mechanism includes: a label acquisition module, a data generation module and a data output module; the label acquisition module is used to obtain the data label input by the user; the data generation module is used to use the data label to perform noise reduction processing on the preset noise data to obtain the equipment operating parameters that meet the preset physical constraints; the data output module is used to output the equipment operating parameters.

[0168] The tag acquisition module can obtain the content input by the user by reading the content input by the user, reading the content stored in the memory, or reading the value of a preset variable. The data output module can output the data to a file, to the memory, etc.

[0169] The industrial AIGC basic model with embedded physical mechanism provided in the embodiment of the present application is trained using the industrial AIGC basic model with embedded physical mechanism training method in the above embodiment, and can output equipment operating parameters that meet physical constraints.

[0170] Figure 3 This is a schematic diagram of the structure of the industrial AIGC basic model training device with embedded physical mechanism provided in the embodiment of the present application. Figure 3 As shown, the industrial AIGC basic model training device 300 with embedded physical mechanism includes: a data acquisition module 301, a noise addition module 302, a model output module 303, and a step increase module 304.

[0171] A data acquisition module 301 is used to acquire multiple sets of training data and training data labels corresponding to the training data;

[0172] A noise adding module 302 is used to add original noise to the training data to obtain noise data;

[0173] A model output module 303, used to input noise data and training data labels into the model to be trained, so that the model to be trained outputs predicted noise;

[0174] The step number increasing module 304 is used to increase the training step number by 1 to obtain a new training step number;

[0175] Strength determination module 305, used to determine the constraint strength according to the new training steps and the preset step threshold;

[0176] A physical loss module 306, for determining physical losses based on the predicted noise and the preset physical constraints;

[0177] A comprehensive loss module 307, for determining comprehensive physical losses based on the restraint strength and the physical losses;

[0178] An error determination module 308, for determining an error value based on the predicted noise, the original noise and the comprehensive physical loss;

[0179] The model optimization module 309 is used to optimize the model to be trained by using the error value to obtain a new model to be trained, and to re-execute the process of adding original noise to the training data to determine the error value until the error value is less than a preset error threshold or the new number of training steps reaches the step threshold, and the new model to be trained is determined as the industrial AIGC basic model.

[0180] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0181] In a possible implementation, the strength determination module 305 is specifically configured to divide the new number of training steps by the step threshold to obtain a training progress; and calculate the constraint strength using the training progress.

[0182] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0183] In a possible implementation, the strength determination module 305 is specifically used to determine the difference between a preset value and the training progress as the incomplete ratio; divide the training progress by the incomplete ratio to obtain a progress ratio; take the logarithm of the progress ratio with a preset constant as the base to obtain a logarithmic value; multiply the logarithmic value by a preset coefficient and input it into a Sigmoid function to obtain constraint strength.

[0184] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0185] In a possible implementation, the physical loss module 306 is used to input the predicted noise into the constraint equation corresponding to each physical constraint to obtain the output result of each constraint equation; and add the output results of each constraint equation to obtain the physical loss.

[0186] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0187] In one possible implementation, the physical constraint includes a change trend constraint; in one possible implementation, the physical loss module 306 is used to input the predicted noise into a constraint equation corresponding to the change trend constraint to obtain an output result of the constraint equation corresponding to the change trend constraint.

[0188] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0189] In one possible implementation, each training step includes at least two time steps; the training data includes at least two data types, and the data types include a baseline type and at least one associated type associated with the baseline type; the physical loss module 306 is specifically used to subtract the predicted noise of the previous time step from the predicted noise of the current time step of the baseline type to obtain a baseline predicted noise difference; the baseline predicted noise difference is divided by the sum of the predicted noise of the previous time step of the baseline type and a preset value to obtain a characteristic slope corresponding to the baseline type; the predicted noise of the current time step of the associated type is subtracted from the predicted noise of the previous time step to obtain the associated predicted noise difference; the associated predicted noise difference is divided by the sum of the predicted noise of the previous training step of the associated type and a preset value to obtain the characteristic slope corresponding to the associated type; the mean square error between the characteristic slope corresponding to each associated type and the characteristic slope corresponding to the baseline type is calculated; the mean square errors are summed and then divided by the total number of time steps to obtain the output result of the constraint equation corresponding to the change trend constraint.

[0190] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0191] In a possible implementation, the error determination module 308 is used to calculate the noise mean square error between the predicted noise and the original noise; and add the noise mean square error to the comprehensive physical loss to obtain an error value.

[0192] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0193] In one possible implementation, the model optimization module 309 is used to, in response to detecting a data label input by a user, input the data label into the industrial AIGC basic model so that the industrial AIGC basic model outputs the device operating parameters; output the device operating parameters.

[0194] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0195] In a possible implementation, the industrial AIGC basic model training device 300 with embedded physical mechanism also includes: a physical correction module 310.

[0196] The physical correction module 310 is used to input the data label into a pre-trained denoising model so that the denoising model can denoise the preset noise data for multiple time steps to obtain the equipment operation data corresponding to the data label; obtain the intermediate data generated by the denoising model in the process of generating the equipment operation data; determine the intermediate physical loss according to the intermediate data and the preset physical constraints; for any time step, calculate the partial derivative of the intermediate physical loss based on the intermediate data to obtain the error gradient; determine the operation data to be output according to the equipment operation data and the error gradient corresponding to each time step; and output the operation data to be output.

[0197] The device provided in this embodiment can be used to execute the technical solution of the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be repeated here.

[0198] In order to implement the above embodiment, the embodiment of the present application also provides an electronic device.

[0199] refer to Figure 4 , which shows a schematic diagram of the structure of an electronic device 400 suitable for implementing the embodiment of the present application, and the electronic device 400 may be a terminal device or a server. The terminal device may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, personal digital assistants (PDAs), tablet computers (Portable Android Devices, PADs), portable multimedia players (Portable Media Players, PMPs), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 4 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0200] like Figure 4As shown, the electronic device 400 may include a processor (such as a central processing unit, a graphics processing unit, etc.) 401, and a memory 402 connected to the processor in communication, which can perform various appropriate actions and processes according to the program stored in the memory 402, the computer execution instruction, or the program loaded from the storage device 408 to the random access memory (Random Access Memory, referred to as RAM) 403, to implement the physical mechanism embedded industrial AIGC basic model training method in any of the above embodiments, wherein the memory can be a read-only memory (Read Only Memory, referred to as ROM). In RAM403, various programs and data required for the operation of the electronic device 400 are also stored. The processing device 401, the memory 402, and the RAM 403 are connected to each other through the bus 404. The input / output (I / O) interface 405 is also connected to the bus 404.

[0201] Typically, the following devices may be connected to the I / O interface 405: an input device 406 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 407 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 408 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 409. The communication device 409 may allow the electronic device 400 to communicate with other devices wirelessly or by wire to exchange data. Figure 4 The electronic device 400 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0202] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 409, or installed from the storage device 408, or installed from the memory 402. When the computer program is executed by the processing device 401, the above-mentioned functions defined in the method of the embodiment of the present application are executed.

[0203] It should be noted that the computer-readable storage medium mentioned above in the present application may be a computer-readable signal medium or a computer storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable storage medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0204] The computer-readable storage medium may be included in the electronic device, or may exist independently without being installed in the electronic device.

[0205] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device executes the method shown in the above embodiment.

[0206] The computer program code for performing the operation of the present application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0207] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0208] The modules involved in the embodiments described in the present application may be implemented by software or hardware. The name of the unit does not limit the module itself in some cases. For example, the noise adding module may also be described as an "original noise adding module".

[0209] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0210] The present application also provides a computer-readable storage medium, which stores computer execution instructions. When the processor executes the computer execution instructions, the technical solution of the industrial AIGC basic model training method with embedded physical mechanism in any of the above-mentioned embodiments is implemented. The implementation principle and beneficial effects of the industrial AIGC basic model training method with embedded physical mechanism are similar to the implementation principle and beneficial effects of the industrial AIGC basic model training method with embedded physical mechanism. Please refer to the implementation principle and beneficial effects of the industrial AIGC basic model training method with embedded physical mechanism, which will not be repeated here.

[0211] In the context of the present application, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0212] The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the technical solution of the industrial AIGC basic model training method with embedded physical mechanism in any of the above-mentioned embodiments. Its implementation principle and beneficial effects are similar to the implementation principle and beneficial effects of the industrial AIGC basic model training method with embedded physical mechanism. Please refer to the implementation principle and beneficial effects of the industrial AIGC basic model training method with embedded physical mechanism, which will not be repeated here.

[0213] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

[0214] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0215] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A physical mechanism embedded industrial AIGC basic model training method, characterized in that: Applied to computers, including: Obtaining multiple sets of training data and training data labels corresponding to the training data; Adding original noise to the training data to obtain noise data; Inputting the noise data and the training data label into a model to be trained, so that the model to be trained outputs predicted noise; Add 1 to the number of training steps to get the new number of training steps; Determining the constraint strength according to the new number of training steps and a preset step threshold; Determining physical losses according to the predicted noise and preset physical constraints; determining a comprehensive physical loss based on the restraint strength and the physical loss; Determining an error value according to the predicted noise, the original noise and the comprehensive physical loss; The error value is used to optimize the model to be trained to obtain a new model to be trained, and the process of adding original noise to the training data to determine the error value is performed again until the error value is less than a preset error threshold or the new number of training steps reaches the step threshold, and the new model to be trained is determined as the industrial AIGC basic model.

2. The method according to claim 1, characterized in that The step of determining the constraint strength according to the new training step number and the preset step number threshold comprises: Divide the new number of training steps by the step threshold to obtain training progress; Using the training schedule, constraint strength is calculated.

3. The method according to claim 2, characterized in that The adopting the training progress to calculate the constraint strength includes: The difference between the preset value and the training progress is determined as the incomplete ratio; Dividing the training progress by the unfinished ratio to obtain a progress ratio; Taking the logarithm of the progress ratio with a preset constant as the base, a logarithm value is obtained; The logarithmic value is multiplied by a preset coefficient and then input into a Sigmoid function to obtain the constraint strength.

4. The method according to claim 1, characterized in that The determining of the physical loss according to the predicted noise and the preset physical constraint includes: Inputting the predicted noise into the constraint equations corresponding to the physical constraints to obtain output results of the constraint equations; The output results of each constraint equation are added together to obtain the physical loss.

5. The method according to claim 4, characterized in that The physical constraints include change trend constraints; The step of inputting the predicted noise into constraint equations corresponding to the physical constraints to obtain output results of the constraint equations includes: The predicted noise is input into a constraint equation corresponding to the change trend constraint to obtain an output result of the constraint equation corresponding to the change trend constraint.

6. The method according to claim 5, characterized in that Each training step includes at least two time steps; the training data includes at least two data types, the data types include a reference type and at least one associated type associated with the reference type; The step of inputting the predicted noise into the constraint equation corresponding to the change trend constraint to obtain the output result of the constraint equation corresponding to the change trend constraint comprises: Subtracting the prediction noise of the previous time step from the prediction noise of the current time step of the benchmark type to obtain a benchmark prediction noise difference; The reference prediction noise difference is divided by the sum of the prediction noise of the previous time step of the reference type and a preset value to obtain a characteristic slope corresponding to the reference type; Subtracting the prediction noise of the previous time step from the prediction noise of the current time step of the association type to obtain an association prediction noise difference; The associated prediction noise difference is divided by the sum of the predicted noise of the previous training step number of the associated type and the preset value to obtain a characteristic slope corresponding to the associated type; Calculating the mean square error between the characteristic slope corresponding to each associated category and the characteristic slope corresponding to the reference category; The mean square errors are summed and then divided by the total number of time steps to obtain the output result of the constraint equation corresponding to the change trend constraint.

7. The method according to claim 1, characterized in that The step of determining an error value according to the predicted noise, the original noise and the comprehensive physical loss comprises: Calculating a noise mean square error between the predicted noise and the original noise; The noise mean square error is added to the comprehensive physical loss to obtain the error value.

8. The method according to any one of claims 1 to 4, characterized in that: After the error value is used to optimize the model to be trained to obtain a new model to be trained, and the process of adding original noise to the training data to determine the error value is performed again, until the error value is less than a preset error threshold or the new number of training steps reaches the step threshold, and the new model to be trained is determined as the industrial AIGC basic model, it also includes: In response to detecting a data tag input by a user, inputting the data tag into the industrial AIGC base model so that the industrial AIGC base model outputs a device operating parameter; Output the device operating parameters.

9. The method according to claim 8, characterized in that After the step of inputting the data tag into the industrial AIGC basic model in response to detecting the data tag input by the user so that the industrial AIGC basic model outputs the equipment operation parameter, the step further includes: Inputting the data label into a pre-trained denoising model, so that the denoising model denoises the preset noise data in multiple time steps to obtain the equipment operation data corresponding to the data label; Acquire various intermediate data generated by the denoising model in the process of generating the equipment operation data; Determine the intermediate physical loss according to the intermediate data and the preset physical constraints; For any time step, the partial derivative of the intermediate physical loss is calculated based on the intermediate data to obtain an error gradient; Determining the operation data to be output according to the equipment operation data and the error gradient corresponding to each time step; The to-be-output operation data is outputted.

10. An industrial AIGC basic model with embedded physical mechanism, characterized in that: The industrial AIGC basic model is trained using the method described in any one of claims 1 to 9; The industrial AIGC basic model embedded in the physical mechanism includes: a label acquisition module, a data generation module and a data output module; The label acquisition module is used to acquire the data label input by the user; The data generation module is used to use the data label to perform noise reduction processing on the preset noise data to obtain equipment operating parameters that meet the preset physical constraints; The data output module is used to output the equipment operating parameters.

11. An industrial AIGC basic model training device with embedded physical mechanism, characterized in that: Applied to computers, including: A data acquisition module, used to acquire multiple sets of training data and training data labels corresponding to the training data; A noise adding module, used for adding original noise to the training data to obtain noise data; A model output module, used for inputting the noise data and the training data label into a model to be trained, so that the model to be trained outputs predicted noise; The step increase module is used to add 1 to the training step number to get a new training step number; A strength determination module, used to determine the constraint strength according to the new training step number and a preset step number threshold; A physical loss module, used to determine the physical loss according to the predicted noise and the preset physical constraints; A comprehensive loss module, used for determining a comprehensive physical loss according to the constraint strength and the physical loss; An error determination module, used to determine an error value according to the predicted noise, the original noise and the comprehensive physical loss; A model optimization module is used to optimize the model to be trained by using the error value to obtain a new model to be trained, and again execute the process of adding original noise to the training data to determine the error value, until the error value is less than a preset error threshold or the new number of training steps reaches the step threshold, and the new model to be trained is determined as the industrial AIGC basic model.

12. An electronic device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 9.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 9 when executed by a processor.

14. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 9 when being executed by a processor.

Images