A thermodynamic mechanism constrained chiller fault data generation method

Abstract

The application discloses a kind of thermodynamic mechanism constraints chiller fault data generation method, first acquisition data and pre-processing;Second data generation model training, generator and discriminator are designed and the loss function of the thermodynamic mechanism constraint generative adversarial network model is designed, and update training is carried out;Finally, the trained generative adversarial network model is used to generate fault data for each class of fault, and form the fault data set with large data volume.The application designs the loss function of the thermodynamic mechanism constraint when training data generation model, and the thermodynamic mechanism is punished, and the authenticity and reliability of generated data are enhanced.

Description

Technical Field

[0001] This invention belongs to the field of chiller unit fault diagnosis technology, specifically relating to a method for generating chiller unit fault data based on a thermodynamic mechanism-constrained generative adversarial network model. Background Technology

[0002] In real-world buildings, refrigeration and air conditioning systems are prone to malfunctions. Due to a general lack of professional skills and knowledge among maintenance personnel, and simplistic management methods, malfunctions are often not detected and addressed promptly, leading to inefficient or even faulty system operation and increasing energy consumption by an additional 30%. Reliable operation, proper control, timely diagnosis, and accurate prediction can achieve energy savings of over 14%. Clearly, fault diagnosis is crucial for achieving the system's "low-carbon, high-efficiency, and energy-saving" goals and improving the level of intelligent system operation and maintenance.

[0003] Thanks to the rapid development of computers and artificial intelligence, an increasing number of fault diagnosis models are built using machine learning algorithms. Machine learning-based diagnostic models require a large amount of labeled data for training. However, in actual field operations, chiller units operate mostly in a normal (fault-free) state. Therefore, the amount of normal sample data collected and stored in the historical datasets of chiller units far exceeds the amount of fault sample data, resulting in data imbalance and making it difficult to ensure that machine learning models achieve excellent diagnostic performance. To address the imbalance between normal and fault sample data, and considering the high cost and risk of simulating faults in actual units, an effective technical approach is to artificially generate sufficient fault sample data. However, purely data-driven models often neglect the thermodynamic principles that sample data should adhere to when generating fault sample data, such as energy conservation and mass conservation. This, to some extent, introduces the risk that the generated data violates thermodynamic principles.

[0004] Therefore, a more realistic and reliable method for generating fault data is needed, which takes into account thermodynamic mechanisms. Summary of the Invention

[0005] In view of this, in order to overcome the shortcomings of the prior art, the purpose of this invention is to provide a method for generating fault data of chiller units with thermodynamic mechanism constraints. In the training process of the data generation model, a loss function with thermodynamic mechanism constraints is designed to enhance its authenticity and reliability.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for generating fault data of a chiller unit constrained by thermodynamic mechanisms includes the following steps: S1. Acquire sensor measurement data of the chiller unit under various fault operating conditions and perform steady-state filtering to obtain steady-state data of the chiller unit under fault operating conditions. S2. Standardize the steady-state data of various faults; S3. Train a generative adversarial network model using standardized steady-state data, and design a loss function for the generative adversarial network model constrained by thermodynamic mechanisms to penalize violations of thermodynamic mechanisms. S4. Use the trained generative adversarial network model to generate fault data for each type of fault, forming a fault dataset with increased data volume.

[0007] Furthermore, step S3 specifically includes the following steps: a. Select random noise for the generative adversarial network model; b. Design the generator and discriminator for the generative adversarial network model; c. Design the loss function for a generative adversarial network model constrained by thermodynamic mechanisms; d. Based on the loss function designed in step c, update and train the model parameters of the generator and discriminator in step b to obtain the generator and discriminator model parameters that minimize the loss function calculation result, and obtain the trained generative adversarial network model.

[0008] Furthermore, in step a, the random noise can be selected from random noise including uniform distribution, normal distribution, and mixed multi-distribution, depending on the specific task and the characteristics of the data.

[0009] Furthermore, the generator and discriminator in step b can be designed as a deep network structure including multiple hidden layers.

[0010] Furthermore, in step c, the loss function of the generative adversarial network model constrained by thermodynamic mechanisms is designed. The specific mathematical expression is as follows: In the formula, These are the loss functions for generative adversarial networks (GANs), under thermodynamic mechanism constraints, generator loss function, discriminator loss function, mass conservation constraint under thermodynamic mechanism constraints, energy conservation constraint under thermodynamic mechanism constraints, and physical common sense constraint under thermodynamic mechanism constraints; λ gan ,λ physics ,λ gen ,λ disc These represent the weights of the loss function values ​​for the generative adversarial network model, the loss function values ​​under thermodynamic constraints, the generator loss function values, and the discriminator loss function values, respectively; r represents the exponent, N and P represent the number of features and samples, respectively, G(z) represents the samples generated by the generator from the input random noise (z), D() represents the output of the discriminator, x represents the real sample data, D(x) represents the discriminator's output on the real data, MSE represents the mean squared error, Mass Conservation Equation represents the mass conservation equation satisfied between features, and Energy Conservation Equation represents the energy conservation equation satisfied between features; X is a specific feature, and Y is another specific feature with a value less than that of feature X.

[0011] The beneficial effects of this invention are: This invention provides a method for generating fault data of chiller units constrained by thermodynamic mechanisms. When training the generated data model, a loss function constrained by thermodynamic mechanisms is designed to penalize violations of thermodynamic mechanisms, thereby enhancing the authenticity and reliability of the generated data and effectively avoiding the risk of data generated by purely data-driven models violating physical consistency. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of the chiller unit fault data generation method in Embodiment 1 of the present invention. Detailed Implementation

[0014] Specific embodiments are given below to further clarify, completely, and in detail the technical solution of the present invention. These embodiments are the preferred embodiments based on the technical solution of the present invention, but the scope of protection of the present invention is not limited to the following embodiments.

[0015] Example 1 A method for generating fault data of chiller units constrained by thermodynamic mechanisms, as shown in the attached document. Figure 1 As shown, the specific implementation steps are as follows: S1. Data preprocessing, including the following sub-steps: S11. Obtain characteristic data of sensor measurements of the chiller unit under various fault operating conditions by using the historical dataset stored on site; S12. Use existing steady-state filtering methods to filter fault data, filter out abnormal and dynamic data, and obtain steady-state data of the chiller unit under various fault modes. S13. Perform 0-1 standardization on the steady-state data obtained in S12 to eliminate the influence between different characteristic dimensions and obtain standardized steady-state data.

[0016] S2. Data generation model training, including the following sub-steps: S21. Select random noise for the generative adversarial network model, such as normally distributed random noise; S22. Design the generator and discriminator of the generative adversarial network model, for example, choose an autoencoder neural network as the generator and discriminator of the generative adversarial network model; S23. Loss function for generative adversarial network models constrained by thermodynamic mechanisms. The specific mathematical expression is as follows: In the formula: The total loss function of the generative adversarial network model constrained by thermodynamic mechanisms; These are the loss functions for the generative adversarial model (calculated by Equation 2), the loss function under thermodynamic mechanism constraints (calculated by Equation 5), the generator loss function (calculated by Equation 3), the discriminator loss function (calculated by Equation 4), the loss function under mass conservation constraints in thermodynamic mechanism (calculated by Equation 6), the loss function under energy conservation constraints in thermodynamic mechanism (calculated by Equation 7), and the loss function under physical common sense constraints in thermodynamic mechanism (calculated by Equation 8); λ gan ,λphysics ,λ gen ,λ disc These are the weights for the loss function value of the generative adversarial network model, the weights for the loss function value under thermodynamic mechanism constraints, the weights for the generator loss function value, and the weights for the discriminator loss function value; the specific values ​​of these weights can be flexibly selected according to actual needs.

[0017] γ is an exponent, used to severely penalize outputs that violate thermodynamic principles. N is the number of sensor-measured features in step S11; P is the number of generated data samples; G(z) is the sample generated by the generator using the input random noise (z); x is the real sample data; D() is the output of the discriminator; D(x) is the discriminator's output on the real data; MSE is the mean squared error. Mass Conservation Equation is the mass conservation equation satisfied between features; Energy Conservation Equation is the energy conservation equation satisfied between features; the subscript regulation is the loss term that violates common sense; X is a specific feature, and Y is another specific feature whose value is less than that of feature X.

[0018] S24. Based on the loss function designed in step S23, update and train the model parameters of the generator and discriminator in step S22 to obtain the generator and discriminator model parameters that minimize the loss function calculation result, and obtain the trained generative adversarial network model.

[0019] Step 3: Use the generative adversarial network model trained in Step 2 to generate fault data for each type of fault, forming a fault dataset with increased data volume.

[0020] Alternatively, in this embodiment, the exponent γ is set to 2.

[0021] Alternatively, in this embodiment, if all losses are considered equally important, then λ is set. gan ,λ physics ,λ gen ,λ disc All weights are equal to 1.

[0022] Alternatively, in this embodiment, the number of samples P is 1000.

[0023] As an alternative, in this embodiment, the Mass Conservation Equation has three characteristics: the chilled water flow rate m1 of subsystem 1, the chilled water flow rate m2 of subsystem 2, and the total chilled water flow rate m3 of the system, satisfying the following law of conservation of mass: m3 = m1 + m2, then: Mass Conservation Equation=m3-m1-m2.

[0024] As an alternative, in this embodiment, the Energy Conservation Equation has four characteristics: the chilled water absorber Q, the total chilled water flow rate m3, and the chilled water inlet and outlet temperatures T1 and T2, satisfying the following law of conservation of energy: Q = m3c p (T1-T2), then: Energy Conservation Equation=Q-m3c p (T1-T2).

[0025] As an alternative, in this embodiment, the subscript regulation has two characteristics: condensation temperature X and evaporation temperature Y. According to thermodynamic principles, these two characteristics should satisfy X > Y. Therefore, a loss term that violates this thermodynamic principle is set.

[0026] In summary, this invention proposes a method for generating fault data of chiller units constrained by thermodynamic mechanisms. It designs a loss function constrained by thermodynamic mechanisms and imposes strict penalties on violations of thermodynamic mechanisms, thereby enabling the generated data to have physical consistency and enhancing its authenticity and reliability.

[0027] The foregoing has shown and described the main features, basic principles, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention based on actual circumstances without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating fault data of a chiller unit constrained by thermodynamic mechanisms, characterized in that, Includes the following steps: S1. Acquire sensor measurement data of the chiller unit under various fault operating conditions and perform steady-state filtering to obtain steady-state data of the chiller unit under fault operating conditions. S2. Standardize the steady-state data of various faults; S3. Train a generative adversarial network (GAN) model using standardized steady-state data, and design a loss function for the GAN model constrained by thermodynamic mechanisms to penalize violations of thermodynamic mechanisms; specifically including the following steps: a. Select random noise for the generative adversarial network model; b. Design the generator and discriminator for the generative adversarial network model; c. Design the loss function for a generative adversarial network model constrained by thermodynamic mechanisms. ; d. Based on the loss function designed in step c, update and train the generator and discriminator model parameters from step b to obtain the generator and discriminator model parameters that minimize the loss function calculation result, thus obtaining the trained generative adversarial network model. S4. Use the trained generative adversarial network model to generate fault data for each type of fault, forming a fault dataset with increased data volume. loss function The specific mathematical expression is as follows: In the formula, These are the loss functions of the generative adversarial network model, the loss function under thermodynamic mechanism constraints, the generator loss function, the discriminator loss function, the loss function under mass conservation constraints in thermodynamic mechanism, the loss function under energy conservation constraints in thermodynamic mechanism, and the loss function under physical common sense constraints in thermodynamic mechanism. These are the weights of the loss function value of the generative adversarial network model, the weights of the loss function value under thermodynamic mechanism constraints, the weights of the generator loss function value, and the weights of the discriminator loss function value, respectively. Indicates an index. These represent the features and the number of samples, respectively. This indicates that the generator uses random noise as input. The generated samples, This represents the output of the discriminator. Represents real sample data. This represents the discriminator's output on the real data. Indicates the mean squared error. This represents the mass conservation equation satisfied between the characteristics. The energy conservation equations satisfied between the characteristics; For a specific feature, For values ​​less than the feature Another specific characteristic of the value.

2. The method for generating chiller unit fault data constrained by thermodynamic mechanisms according to claim 1, characterized in that, In step a, the random noise can be selected from uniform distribution, normal distribution, and mixed multi-distribution random noise, depending on the specific task and the characteristics of the data.

3. The method for generating chiller unit fault data constrained by thermodynamic mechanisms according to claim 1, characterized in that, In step b, the generator and discriminator can be designed as a deep network structure with multiple hidden layers.

Images