A complex system fault location method based on Ising model

Abstract

The present invention discloses a complex system fault location method based on the Ising model, comprising: corresponding each sensor in the complex system to a node in the Ising model, using spin variables to represent the sensor state; determining the interaction parameters in the Ising model based on the correlation between different sensors and the changing trend of sensor readings; defining an energy function, which includes interaction terms, external field terms, and terms related to sensor readings; defining a loss function; randomly initializing parameters; minimizing the loss function using an optimization algorithm and updating the parameters; iteratively updating the node state using the Monte Carlo method, updating the node spin state with a preset probability based on the current energy function and temperature parameters; and determining the sensor fault state based on the final node spin state. The present invention can achieve rapid fault location in complex systems.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and in particular relates to a complex system fault location method based on an Ising model. Background Art

[0002] With the continuous development of modern industry and technology, complex systems are increasingly being used in various fields. For example, complex systems such as air conditioning units are often equipped with multiple sensors to monitor various physical parameters, such as flow, temperature, and humidity. The data from these sensors is crucial for the proper operation and fault diagnosis of the system. However, traditional fault diagnosis methods often have limitations when dealing with complex systems. On the one hand, for systems with a large number of sensors and complex structures, traditional methods can struggle to accurately locate the source of a fault, requiring significant time and manpower to troubleshoot. On the other hand, traditional methods may not fully exploit the correlations between sensors and the physical characteristics of the system, resulting in low accuracy and efficiency in fault diagnosis. Therefore, there is an urgent need for a new fault location method that can fully utilize the sensor data in complex systems and combine it with advanced mathematical models and algorithms to achieve rapid and accurate fault location. As a physical model, the Ising model has considerable potential in addressing complex system problems.

[0003] Patent publication number CN118487315A discloses a hybrid quantum-classical solution method, system, device, and storage medium for power system unit combinations designed to withstand extreme events. Patent publication number US10539468B2 discloses an anomaly detection device, an anomaly detection method, and a non-transitory computer-readable medium. Patent publication number CN113691311B discloses a fault location method, electronic device, and computer-readable storage medium for optical networks. All of these studies propose the possibility of applying the Ising model to the field of faults. However, these studies either require the construction of extremely complex system operation models, resulting in a significant increase in computational complexity and poor real-time performance, or are only applicable to simple system structures and cannot solve the problem of fault location in complex systems. Summary of the Invention

[0004] Technical problem solved: In response to the technical problem of fault location in complex systems, the present invention proposes a complex system fault location method based on the Ising model. By mapping the sensors in the complex system to nodes in the Ising model and defining interaction parameters according to the correlation and physical characteristics between the sensors, the characteristics of the Ising model are used to analyze the system status, thereby achieving rapid fault location.

[0005] Technical solution:

[0006] A complex system fault location method based on the Ising model, the complex system fault location method comprising the following steps:

[0007] S1 collects the readings of each sensor in the complex system and records the corresponding timestamps and device operating status tags; cleans the collected sensor data to remove outliers and noise; and normalizes the data from different sensors to make them have the same scale range;

[0008] S2, each sensor in the complex system corresponds to a node in the Ising model, and the state of the sensor is represented by a spin variable, where spin-up represents a normal state and spin-down represents a fault state;

[0009] S3, determining the interaction parameters in the Ising model based on the correlation between different sensors and the changing trend of sensor readings;

[0010] S4, defining an energy function, which includes interaction terms, external field terms, and terms related to sensor readings;

[0011] S5, defining a loss function based on the actual device operating state label, the state predicted by the Ising model, and the energy function;

[0012] S6, randomly initializes spin variables, interaction parameters, external field parameters, and parameters related to sensor readings;

[0013] S7, using an optimization algorithm to minimize the loss function and update the interaction parameters, external field parameters, and parameters related to the sensor readings;

[0014] S8, using the Monte Carlo method to iteratively update the node state, and update the node's spin state with a preset probability according to the current energy function and temperature parameters;

[0015] S9, repeat the iteration until the system reaches thermal equilibrium or meets certain convergence conditions;

[0016] S10, determine the fault state of the sensor based on the final node spin state

[0017] Furthermore, in step S1, the sensor data includes but is not limited to flow, temperature, and humidity data.

[0018] Furthermore, in step S1, the process of cleaning the collected sensor data includes:

[0019] Set a threshold to filter out abnormal values ​​in the sensor data, and then use statistical methods to remove noise in the sensor data.

[0020] Furthermore, in step S1, the data from different sensors are normalized using the following formula:

[0021]

[0022] Where x is the original data, x min and x max are the minimum and maximum values ​​of the data respectively.

[0023] Furthermore, in step S4, the energy function is:

[0024]

[0025] Among them J ij is the interaction parameter, s i is the node spin variable, h i is the external field parameter, d i is the difference between the sensor reading and the expected value, and λ is the adjustment parameter.

[0026] Furthermore, in step S5, the loss function is:

[0027] L total =αL CE +(1-α)E

[0028]

[0029] The weight parameter α balances the cross entropy loss L CE and the contribution E of the energy function; C is the total number of categories, y ic is an indicator variable, is the predicted probability that sample i belongs to category c.

[0030] Furthermore, in step S7, the gradient descent method is used to calculate the gradients of the interaction parameters, the external field parameters, and the parameters related to the sensor readings and update them. The updating process includes the following steps:

[0031] S71, calculate the loss function for the interaction parameter J ij Gradient:

[0032]

[0033] S72, calculate the loss function and the external field parameter h i Gradient:

[0034]

[0035] S73, calculate the gradient of the loss function with respect to the parameters related to the sensor readings:

[0036]

[0037] S74, updating interaction parameters, external field parameters, and parameters related to sensor readings;

[0038] For the interaction parameter J ij , the update formula is: Where α is the learning rate;

[0039] For the external field parameter h i , the update formula is:

[0040] For the parameter λ related to the sensor reading, the update formula is:

[0041] Furthermore, in step S8, the process of updating the spin state of the node includes the following steps:

[0042] S81, calculate sensor reading x i and the expected value μ i The difference d i :d i =x i -μ i ;

[0043] S82, calculating the terms in the energy function related to the sensor readings

[0044] S83, randomly initialize the node spin state to up or down, using Indicates the initial state;

[0045] S84, using the Monte Carlo method to iteratively update the node state; in each iteration, a node is randomly selected, and the spin state of the node is updated with a certain probability based on the current energy function and temperature parameter T;

[0046] The probability of updating the node spin state is calculated using the Metropolis algorithm:

[0047] P(s i →-s i )=min(1,e -ΔE / T )

[0048] Where ΔE is the energy change and T is the temperature parameter.

[0049] Furthermore, in step S10, determining the fault state of the sensor according to the final node spin state means that if the spin variable s corresponding to a certain sensor is i =-1, it is considered that the sensor is faulty.

[0050] Beneficial effects:

[0051] First, the complex system fault location method based on the Ising model of the present invention combines the sensor network with the Ising model and utilizes the characteristics of sensor readings and the Ising model to accurately and efficiently locate possible faults in the complex system. It is particularly suitable for complex systems with multiple sensors, such as air conditioning units.

[0052] Second, the complex system fault location method based on the Ising model of the present invention covers multiple technical links such as data acquisition and processing, model construction, parameter training and fault location. It can be widely used in industrial automation, intelligent equipment maintenance, Internet of Things and other fields, providing strong technical support for ensuring the stable operation and timely maintenance of complex systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a framework diagram of the complex system fault location method based on the Ising model of the present invention.

[0054] Figure 2 A diagram showing the topology of a sensor network described using the two-dimensional Ising model. DETAILED DESCRIPTION

[0055] The following examples may enable those skilled in the art to more fully understand the present invention, but are not intended to limit the present invention in any way. Figure 1 The present invention discloses a complex system fault location method based on the Ising model, comprising the following steps:

[0056] (1) Collect the readings of various sensors in the complex system, including but not limited to flow, temperature, humidity and other data, and record the corresponding timestamps and equipment operating status tags.

[0057] (2) Cleaning the collected sensor data to remove outliers and noise; preferably, the method for cleaning the collected sensor data includes setting a threshold to filter outliers and using statistical methods to remove noise.

[0058] (3) Normalize the data from different sensors so that they have the same scale range. Data normalization can use the min-max normalization method, the formula is:

[0059]

[0060] Where x is the original data, x min and x max are the minimum and maximum values ​​of the data respectively.

[0061] (4) Each sensor in the complex system is mapped to a node in the Ising model, and the state of the sensor is represented by the spin variable, where spin-up represents a normal state and spin-down represents a fault state. Figure 2 shown.

[0062] (5) Determine the interaction parameters in the Ising model based on the correlation between sensors and sensor readings; specifically, the interaction parameters are determined based on the correlation coefficients between different sensor readings and considering the changing trend of the sensor readings.

[0063] (6) Define an energy function, which includes the interaction term, the external field term, and the term related to the sensor reading. The energy function is:

[0064]

[0065] Among them J ij is the interaction parameter, s i is the node spin variable, h i is the external field parameter, d i is the difference between the sensor reading and the expected value, and λ is the adjustment parameter.

[0066] (7) Define the loss function based on the actual device state label, the state predicted by the model, and the energy function.

[0067] The loss function of the present invention should consider both the energy function and the cross-entropy loss function. The energy function plays an important role in the Ising model. It determines the state of the system and the spin state of the node. During the training process, the value of the energy function is also affected by adjusting the model parameters. Normally, when the system reaches a stable state, the energy function will be at a lower value. In fault location, it is hoped that the system will be in a more reasonable state by minimizing the energy function, so as to determine the fault node. On the other hand, minimizing the cross-entropy loss function is to make the model's prediction results as close as possible to the actual device status label, thereby improving the accuracy of fault location. The cross-entropy loss function measures the difference between the probability distribution predicted by the model and the actual label. The model parameters are continuously adjusted through the optimization algorithm so that the loss function value gradually decreases. For multi-classification problems, the formula of the cross-entropy loss function is:

[0068]

[0069] Where C is the total number of categories, y ic is an indicator variable (if the true category of sample i is equal to c, then yic = 1, otherwise 0), is the predicted probability that sample i belongs to category c. Therefore, the total loss function is:

[0070] Ltotal =αL CE +(1-α)E

[0071] The weight parameter α balances the contribution of the cross entropy loss and the energy function.

[0072] (8) Randomly initialize spin variables, interaction parameters, external field parameters, and parameters related to sensor readings.

[0073] (9) Use optimization algorithms to minimize the loss function and update the parameters.

[0074] The optimization algorithm can use the gradient descent method to calculate and update the gradients of the interaction parameters, external field parameters, and parameters related to sensor readings. The process includes:

[0075] (9.1) First calculate the loss function for the interaction parameter J ij Gradient:

[0076] Assuming that the relationship between the loss function and the energy function is known, it can be calculated using the chain rule. The calculation of usually involves taking the partial derivative of the energy function with respect to and then deriving the dependency of the loss function on the energy function.

[0077] Energy function E with respect to J ij Taking partial derivatives we get:

[0078] For the cross entropy loss function L CE About J ij The gradient of J usually needs to be combined with the model prediction probability through the chain rule ij The dependency relationship is calculated, and it is set to Then the total loss function is the interaction parameter J ij The gradient of is:

[0079]

[0080] (9.2) Calculate the loss function for the external field parameter h i Gradient:

[0081] Energy function E with respect to h i Taking partial derivatives we get:

[0082] For the cross entropy loss function L CE About h i The gradient of

[0083] Then the total loss function is the external field parameter h i The gradient of is:

[0084]

[0085] (9.3) Calculate the gradient of the loss function with respect to parameters related to sensor readings (such as the adjustment parameter λ):

[0086] Taking the partial derivative of the energy function E with respect to λ, we can get:

[0087] For the cross entropy loss function L CE The gradient with respect to λ is set to

[0088] Then the gradient of the total loss function with respect to λ is:

[0089]

[0090] (9.4) Parameter update formula:

[0091] For the interaction parameter J ij : where α is the learning rate.

[0092] For the external field parameter h i :

[0093] For parameters related to sensor readings (such as the adjustment parameter λ):

[0094] (10) The Monte Carlo method is used to iteratively update the node state. According to the current energy function and temperature parameters, the node spin state is updated with a certain probability. The specific process of updating the node spin state is:

[0095] (10.1) Calculate the reading x of sensor i i and the expected value μ i The difference d i :d i =x i -μ i

[0096] (10.2) Calculate the terms in the energy function related to the sensor readings:

[0097] The known energy function is Among them J ij is the interaction parameter, s i is the node spin variable, h i is the external field parameter, and λ is the adjustment parameter. The term related to the sensor reading is

[0098] (10.3) Initialize the node spin state: The node spin state can be randomly initialized to up (normal state) or down (fault state), using Indicates the initial state.

[0099] (10.4) Iteratively update the spin state:

[0100] In each iteration, a node is randomly selected and its spin state is updated with a certain probability based on the current energy function and temperature parameter T. The probability of updating the node spin state is calculated using the Monte Carlo method (Metropolis algorithm), and the formula is:

[0101] P(s i →-s i )=min(1,e -ΔE / T )

[0102] Where ΔE is the energy change and T is the temperature parameter.

[0103] After multiple iterations, the system reaches a relatively stable state.

[0104] (11) Repeat the iteration until the system reaches thermal equilibrium or meets certain convergence conditions.

[0105] (12) Determine the fault state of the sensor based on the final node spin state.

[0106] The fault state of the sensor is determined based on the final node spin state. Specifically, if the spin variable s corresponding to a sensor is i =-1, it is considered that the sensor is faulty.

[0107] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A complex system fault location method based on the Ising model, characterized in that: The complex system fault location method comprises the following steps: S1 collects the readings of each sensor in the complex system and records the corresponding timestamps and device operating status tags; cleans the collected sensor data to remove outliers and noise; and normalizes the data from different sensors to make them have the same scale range; S2, each sensor in the complex system corresponds to a node in the Ising model, and the state of the sensor is represented by a spin variable, where spin-up represents a normal state and spin-down represents a fault state; S3, determining the interaction parameters in the Ising model based on the correlation between different sensors and the changing trend of sensor readings; S4, defining an energy function, which includes interaction terms, external field terms, and terms related to sensor readings; S5, defining a loss function based on the actual device operating state label, the state predicted by the Ising model, and the energy function; S6, randomly initializes spin variables, interaction parameters, external field parameters, and parameters related to sensor readings; S7, using an optimization algorithm to minimize the loss function and update the interaction parameters, external field parameters, and parameters related to the sensor readings; S8, using the Monte Carlo method to iteratively update the node state, and update the node's spin state with a preset probability according to the current energy function and temperature parameters; S9, repeat the iteration until the system reaches thermal equilibrium or meets certain convergence conditions; S10, determining the fault state of the sensor according to the final node spin state.

2. The complex system fault location method based on the Ising model according to claim 1, characterized in that: In step S1, the sensor data includes but is not limited to flow, temperature, and humidity data.

3. The complex system fault location method based on the Ising model according to claim 1, characterized in that: In step S1, the process of cleaning the collected sensor data includes: Set a threshold to filter out abnormal values ​​in the sensor data, and then use statistical methods to remove noise in the sensor data.

4. The complex system fault location method based on the Ising model according to claim 1, characterized in that: In step S1, the data from different sensors are normalized using the following formula: Where x is the original data, x min and x max are the minimum and maximum values ​​of the data respectively.

5. The complex system fault location method based on the Ising model according to claim 1, characterized in that: In step S4, the energy function is: Among them J ij is the interaction parameter, s i is the node spin variable, h i is the external field parameter, d i is the difference between the sensor reading and the expected value, and λ is the adjustment parameter.

6. The complex system fault location method based on the Ising model according to claim 1, characterized in that: In step S5, the loss function is: L total =αL CE +(1-a)E The weight parameter α balances the cross entropy loss L CE and the contribution E of the energy function; C is the total number of categories, y ic is an indicator variable, is the predicted probability that sample i belongs to category c.

7. The complex system fault location method based on the Ising model according to claim 1, characterized in that: In step S7, the gradient descent method is used to calculate the gradients of the interaction parameters, the external field parameters, and the parameters related to the sensor readings and update them. The updating process includes the following steps: S71, calculate the loss function for the interaction parameter J ij Gradient: S72, calculate the loss function and the external field parameter h i Gradient: S73, calculate the gradient of the loss function with respect to the parameters related to the sensor readings: S74, updating interaction parameters, external field parameters, and parameters related to sensor readings; For the interaction parameter J ij , the update formula is: Where α is the learning rate; For the external field parameter h i , the update formula is: For the parameter λ related to the sensor reading, the update formula is:

8. The complex system fault location method based on the Ising model according to claim 1, characterized in that: In step S8, the process of updating the spin state of the node includes the following steps: S81, calculate sensor reading x i and the expected value μ i The difference d i :d i =x i -μ i ; S82, calculating the terms in the energy function related to the sensor readings S83, randomly initialize the node spin state to up or down, using Indicates the initial state; S84, using the Monte Carlo method to iteratively update the node state; in each iteration, a node is randomly selected, and the spin state of the node is updated with a certain probability based on the current energy function and temperature parameter T; The probability of updating the node spin state is calculated using the Metropolis algorithm: P(s i →-s i )=min(1,e -ΔE / T ) Where ΔE is the energy change and T is the temperature parameter.

9. The complex system fault location method based on the Ising model according to claim 1, characterized in that: In step S10, determining the fault state of the sensor according to the final node spin state means that if the spin variable s corresponding to a sensor is i =-1, it is considered that the sensor is faulty.

Images