Data-driven power supply anomaly monitoring system

Abstract

The application discloses a power supply abnormality monitoring system based on data driving, comprising a power supply running state acquisition module, an abnormal potential field model construction module, an abnormal interface construction module, a key disturbance sample extraction module, a disturbance state stable projection mapping module, a potential field training target function construction module and an online abnormality monitoring module. The scheme introduces a disturbance state stable projection mapping mechanism, maps the interface neighborhood sample to a stable state manifold, enhances the learning constraint ability of the system to the stable running area structure, constructs a key disturbance sample screening mechanism based on the interface distance, improves the sensitivity to the stable-to-abnormal evolution process of the power supply, reduces the false alarm rate caused by the power supply transient disturbance through a misjudgment constraint mechanism, improves the identification ability of the early fault signs of the power supply through a key disturbance sample reinforcement mechanism, and constrains the rationality of the abnormality judgment boundary by introducing the stable reference sample corrected from the disturbance state, and defines the safe boundary range of the stable running of the power supply.

Description

Technical Field

[0001] This invention relates to the field of data processing, specifically to a data-driven power anomaly monitoring system. Background Technology

[0002] A power supply anomaly monitoring system is a system that collects and analyzes key parameters such as voltage, current, power, and temperature in real time during the operation of power equipment to determine whether it deviates from the normal operating state and to achieve fault early warning and status monitoring. However, conventional power supply anomaly monitoring systems have problems such as insufficient sensitivity in identifying abnormal critical states, and are prone to misjudgment or delayed response in the transition state before a fault occurs; conventional power supply anomaly monitoring systems are also prone to misjudgment or missed detection when the power system experiences complex disturbances, and are not sensitive to early subtle changes, making it difficult to capture fault precursors in a timely manner. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a data-driven power anomaly monitoring system. Addressing the problem that conventional power anomaly monitoring systems lack sensitivity in identifying critical anomaly states and are prone to misjudgment or delayed response in transitional states approaching fault occurrence, this solution constructs a second-order coupled anomaly potential field model. It explicitly characterizes the interaction between multi-dimensional operating parameters through a coupling matrix. A disturbance state stable projection mapping mechanism is introduced to map samples from the interface neighborhood to the stable state manifold, enhancing the system's ability to learn and constrain the structure of the stable operating region. Furthermore, anomaly interfaces are constructed using a zero potential energy surface, and continuous potential function values ​​are used as anomaly exponents, transforming anomaly discrimination into a continuous energy discrimination mechanism. This improves the response accuracy to critical states; it constructs a key disturbance sample screening mechanism based on interface distance to enhance the sensitivity to the evolution of power supply from stable to abnormal; addressing the problems of conventional power supply anomaly monitoring systems, such as the tendency to misjudge or miss when the power system experiences complex disturbances, insensitivity to early subtle changes, and difficulty in timely capturing fault precursors, this solution reduces the false alarm rate caused by power supply transient disturbances through a misjudgment constraint mechanism, reducing the occurrence of misjudging normal disturbances as faults; it enhances the ability to identify early fault symptoms of the power supply through a key disturbance sample reinforcement mechanism; and it clarifies the safe boundary range for stable power supply operation by introducing stable reference samples obtained from disturbance state correction to constrain the rationality of the anomaly judgment boundary.

[0004] The technical solution adopted by the present invention is as follows: The data-driven power supply anomaly monitoring system provided by the present invention includes a power supply operating status acquisition module, an abnormal potential field model construction module, an abnormal interface construction module, a key disturbance sample extraction module, a disturbance state stable projection mapping module, a potential field training objective function construction module, and an online anomaly monitoring module. The power supply operation status acquisition module acquires historical power supply operation data, constructs an operation status vector, and performs rated operating condition normalization processing. The abnormal potential field model construction module constructs an abnormal potential field model based on a normalized state vector, and parametrically expresses the relationship between state variables by introducing a coupling matrix, a linear sensitivity coefficient vector, and a bias parameter in a symmetrical manner. The abnormal interface construction module uses the zero potential energy surface as the boundary condition between stable and abnormal states to form a state discrimination boundary, which serves as the abnormal interface. The key perturbation sample extraction module filters samples based on the potential energy distance from the abnormal interface to construct a key perturbation sample set. The disturbance state stable projection mapping module maps key disturbance samples to stable corrected states that satisfy the potential field stability constraints. The potential field training objective function construction module constructs the potential field training objective function based on the normalized state vector, key perturbation samples, and stable correction state. The online anomaly monitoring module performs online anomaly monitoring on power supply operation data based on the trained abnormal potential field model.

[0005] Furthermore, the abnormal potential field model construction module establishes a mapping relationship between the operating state and the degree of abnormality. Taking the normalized state vector as input, it introduces a second-order coupled potential field, describes the independent influence of each state variable through single-variable terms, describes the coupled influence between different state variables through interaction terms, and outputs the corresponding abnormal potential energy index to characterize the degree of abnormality of the current power supply operating state.

[0006] Furthermore, the key disturbance sample extraction module constructs a sensitive sample screening mechanism based on the output of the abnormal potential field model. By analyzing the distribution position of the samples in the mapping function of the abnormal potential field model, it identifies key disturbance samples in the transition region between the stable state and the abnormal state.

[0007] Furthermore, the disturbance state stability projection mapping module performs stability projection mapping on key disturbance samples, mapping them to the closest stable state that satisfies the potential field stability constraint, which is used to constrain the structure learning of the abnormal potential field model.

[0008] Furthermore, in order to enable the abnormal potential field to have good state discrimination ability, the potential field training objective function construction module defines a potential field training objective function, including a global potential field smoothing constraint term, a misjudgment constraint mechanism term, and a key perturbation sample reinforcement mechanism term.

[0009] Furthermore, the online anomaly monitoring module uses the trained abnormal potential field model to calculate the anomaly index in real time, and performs online anomaly monitoring of power supply operation data.

[0010] The beneficial effects achieved by the present invention using the above solution are as follows: (1) To address the problem that conventional power supply anomaly monitoring systems are not sensitive enough to identify critical abnormal states and are prone to misjudgment or delayed response in the transition state before a fault occurs, this solution constructs a second-order coupled abnormal potential field model and explicitly characterizes the interaction between multi-dimensional operating parameters through the coupling matrix; introduces a disturbance state stable projection mapping mechanism to map the interface neighborhood samples to the stable state manifold, thereby enhancing the system's ability to learn and constrain the structure of the stable operating region; constructs an abnormal interface with a zero potential energy surface and uses the continuous value of the potential function as the anomaly index, so that the anomaly discrimination is transformed into a continuous energy discrimination mechanism, thereby improving the response accuracy to the critical state; and constructs a key disturbance sample screening mechanism based on the interface distance to improve the sensitivity to the evolution process of the power supply from stable to abnormal.

[0011] (2) In view of the problems that conventional power supply anomaly monitoring systems are prone to misjudgment or missed judgment when the power system is subjected to complex disturbances, are not sensitive to early weak changes, and are difficult to capture fault precursors in time, this solution reduces the false alarm rate caused by power supply transient disturbances through a misjudgment constraint mechanism, and reduces the situation of misjudging normal disturbances as faults; enhances the ability to identify early fault signs of power supply through a key disturbance sample enhancement mechanism; and clarifies the safe boundary range of power supply stable operation by introducing a stable reference sample obtained by disturbance state correction to constrain the rationality of the anomaly judgment boundary. Attached Figure Description

[0012] Figure 1 A flowchart illustrating the data-driven power anomaly monitoring system provided by this invention. Figure 2 This is a schematic diagram of the boundary structure of the abnormal potential field.

[0013] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0014] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0015] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0016] Example 1, see Figure 1 The data-driven power supply anomaly monitoring system provided by the present invention includes a power supply operating status acquisition module, an abnormal potential field model construction module, an abnormal interface construction module, a key disturbance sample extraction module, a disturbance state stable projection mapping module, a potential field training objective function construction module, and an online anomaly monitoring module. The power supply operation status acquisition module acquires historical power supply operation data, constructs an operation status vector, and performs rated operating condition normalization processing. The abnormal potential field model construction module constructs an abnormal potential field model based on a normalized state vector, and parametrically expresses the relationship between state variables by introducing a coupling matrix, a linear sensitivity coefficient vector, and a bias parameter in a symmetrical manner. The abnormal interface construction module uses the zero potential energy surface as the boundary condition between stable and abnormal states to form a state discrimination boundary, which serves as the abnormal interface. The key perturbation sample extraction module filters samples based on the potential energy distance from the abnormal interface to construct a key perturbation sample set. The disturbance state stable projection mapping module maps key disturbance samples to stable corrected states that satisfy the potential field stability constraints. The potential field training objective function construction module constructs the potential field training objective function based on the normalized state vector, key perturbation samples, and stable correction state. The online anomaly monitoring module performs online anomaly monitoring on power supply operation data based on the trained abnormal potential field model.

[0017] Example 2, see Figure 1 This embodiment is based on the above embodiment. The power supply operation status acquisition module synchronously collects historical power supply operation data reflecting the power supply energy transmission status, power balance status, and device thermal status, and constructs an operation status vector. The state vector at time t... Represented as: Where, U(·) is the output voltage in V; I(·) is the output current in A; P(·) is the active power in W; Q(·) is the reactive power in Var; PF(·) is the power factor; THD(·) is the total harmonic distortion; and T(·) is the device temperature in °C. For different dimensions, rated operating condition normalization is adopted (actual state variables are divided by rated operating condition quantities) to convert each state variable into the degree of deviation from the rated operating state, thereby obtaining a state representation with unified dimensions; state labeling is performed at each acquisition time to construct a supervision label y(t): y(t)=1 indicates that the power supply is in an abnormal operating state; y(t)=0 indicates that the power supply is in a stable operating state.

[0018] Example 3, see Figure 1 This embodiment is based on the above embodiment. The abnormal potential field model construction module establishes a mapping relationship between the operating state and the degree of abnormality. Taking the normalized state vector as input, a second-order coupled potential field is introduced. The independent influence of each state variable is described by a single variable term, and the coupling influence between different state variables is described by an interaction term. The corresponding abnormal potential energy index is output to characterize the degree of abnormality of the current power supply operating state. The abnormal potential field model mapping function is expressed as: Wherein, the coupling matrix K is used to characterize the contribution of each operating state variable to the abnormal potential field and the coupling influence strength between variables. The diagonal elements represent the independent sensitivity of the operating state variables to the abnormal potential field, and the off-diagonal elements represent the coupled abnormal effect generated by the joint action of two operating state variables. It is initially set as an identity matrix; d is a linear sensitivity coefficient vector used to describe the first-order influence strength of each operating state variable on the abnormal potential field, and is initialized to a zero vector; e is a bias parameter used to characterize the basic abnormal potential energy level of the power system under rated operating conditions, and is initialized to zero. is the abnormal potential energy index of the normalized state vector, and its value represents the degree to which the system deviates from the normal state; z is the normalized state vector; the coupling matrix, the linear sensitivity coefficient vector, and the bias parameter are used as parameters to be optimized. During the training process, they are iteratively updated by minimizing the training objective function based on gradient descent, and symmetry constraints are applied to K after each iteration.

[0019] Example 4, see Figure 1 This embodiment is based on the above embodiment. The anomaly interface construction module uses the zero potential energy surface as the state discrimination boundary between stable and abnormal states to perform operation state discrimination; the anomaly index is defined as: ; It is the anomalous potential energy index of the normalized state vector at time t; anomalous interface ;Judgment conditions: Indicates an abnormal state; It indicates a stable operating state; the interface is derived from the state distribution pattern in historical operating data, indicating the critical position where the system transitions from a stable energy region to an abnormal energy region; The abnormality index is at time t. The larger the value, the more serious the deviation from the stable operating state.

[0020] Example 5, see Figure 1 This embodiment is based on the above embodiment. The key disturbance sample extraction module constructs a sensitive sample screening mechanism based on the output results of the abnormal potential field model. By analyzing the distribution position of the samples in the mapping function of the abnormal potential field model, it identifies key disturbance samples in the transition region between the stable state and the abnormal state, which is used to characterize the critical behavior of the power supply state from stable to abnormal. Samples farther from the anomalous interface correspond to slow changes in the potential field gradient and contribute little to interface learning. Samples closer to the interface, however, correspond to rapid changes in the system state and can reflect the complex disturbance processes of voltage dips, current surges, harmonic abrupt changes, and accelerated temperature rise. The interface distance is defined as... The absolute value of the abnormal potential energy index; construct a key perturbation sample set. , is represented as: ;in, It is the width of the interface neighborhood; It is the distance from the m-th sample to the abnormal boundary.

[0021] Example 6, see Figure 1 This embodiment is based on the above embodiment. The perturbation state stability projection mapping module performs stability projection mapping on key perturbation samples, mapping them to the closest stable state that satisfies the potential field stability constraint. This is used to constrain the structure learning of the abnormal potential field model. The stability projection mapping is expressed as: Where v is the perturbation input state vector; u is the stable correction state vector; It is the non-negative truncated vector of u; It is the number of non-zero elements; It is the state recovery smoothing coefficient, with a value of 10. -4 ~1; It is the result of stability projection mapping, representing the stable corrected state of the key perturbation sample; the stable corrected state label is the stable running state, which is used as the stable reference sample set of the anomalous potential field model to participate in training, and is used to constrain the anomalous potential field model to learn the stable state manifold structure.

[0022] By performing the above operations, this solution addresses the problem that conventional power supply anomaly monitoring systems lack sensitivity in identifying critical anomalies and are prone to misjudgment or delayed response in the transitional state before a fault occurs. It constructs a second-order coupled anomaly potential field model, explicitly characterizing the interaction between multi-dimensional operating parameters through a coupling matrix; introduces a disturbance state stable projection mapping mechanism to map samples from the interface neighborhood to the stable state manifold, enhancing the system's ability to learn and constrain the structure of the stable operating region; constructs anomaly interfaces using a zero potential energy surface and employs continuous potential function values ​​as anomaly exponents, transforming anomaly discrimination into a continuous energy discrimination mechanism, thereby improving the response accuracy to critical states; and constructs a key disturbance sample screening mechanism based on interface distance to enhance sensitivity to the power supply's evolution from stable to anomaly.

[0023] Example 7, see Figure 1 This embodiment is based on the above embodiment. In order to enable the abnormal potential field to have good state discrimination ability, the potential field training objective function construction module defines the potential field training objective function, including the overall potential field smoothing constraint term, the misjudgment constraint mechanism term, and the key perturbation sample reinforcement mechanism term, expressed as: ; Among them, sample misclassification labeling J is the total training objective; N is the number of training samples; C is the error weight coefficient, ranging from 0.1 to 100, representing the system's ability to suppress abnormal states. It is the normalized state vector of the m-th sample; It is the sample weight. It is the anomaly index of the sample; It is the attenuation coefficient, which controls the interface sensitivity, and its value ranges from 0.1 to 1.0; It is the key perturbation enhancement coefficient, which controls the degree of enhancement of the interface sample, and its value ranges from 0.01 to 10. It is the overall potential field smoothing constraint term, the basic potential field, and the learning of the overall potential field structure; This is a misjudgment constraint item to ensure correct judgment; It is a key perturbation sample enhancement mechanism term, which enhances the local curvature learning ability of the potential field function near the zero potential energy surface and improves the ability to distinguish abnormal interfaces; the stable correction state and the original perturbation sample constrain the potential field model in the same perturbation direction, so that the potential field function satisfies the stable convergence characteristics on the perturbation path.

[0024] By performing the above operations, this solution addresses the problems of conventional power supply anomaly monitoring systems, such as the tendency to misjudge or miss anomalies when the power system experiences complex disturbances, insensitivity to early subtle changes, and difficulty in timely detection of fault precursors. It reduces the false alarm rate caused by power supply transient disturbances through a misjudgment constraint mechanism, minimizing the misjudgment of normal disturbances as faults; enhances the ability to identify early fault symptoms through a key disturbance sample reinforcement mechanism; and clarifies the safe boundary range for stable power supply operation by introducing stable reference samples obtained from disturbance state correction to constrain the rationality of anomaly judgment boundaries.

[0025] Example 8, see Figure 1 This embodiment is based on the above embodiment. The online anomaly monitoring module uses the trained abnormal potential field model to calculate the anomaly index in real time, and performs online anomaly monitoring of power supply operation data; it inputs z(t) in real time and calculates... ;determination An alert will be issued for an abnormal state. To maintain a stable operating state; , and These are the final parameters after the abnormal potential field model has been trained.

[0026] Example 9, see Figure 2 This embodiment is based on the above embodiment. The main image on the left is the abnormal potential field boundary structure, which is used to represent the distribution of the normalized operating state of the power supply in the state space. The red area is the high value region of the abnormal potential field, and the blue area is the stable potential field region. The black boundary curve is the abnormal boundary, i.e., the zero potential energy surface. The area below the curve represents the stable operating state, the area above the curve represents the critical state, and the area above the curve represents the abnormal state. z1, z2, and z3 are different types of operating state variables. The blue area represents the stable sample (y=0). The red area represents the abnormal sample (y=1). The green circled dots are the key disturbance samples, which belong to the stable to abnormal transition region. The enlarged view on the right shows the disturbance projection and stable manifold correction. The orange dots are the original disturbance samples, corresponding to the state after being disturbed by voltage drops, current surges, etc. The arrows are the projection paths, representing the process of mapping from the disturbance state to the stable state. The green dots are the stable correction samples, which are the states after being mapped back to the stable manifold. Their function is to pull the samples that are close to instability back to the stable region as constraint learning samples.

[0027] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0028] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A data-driven power anomaly monitoring system, characterized in that: The system includes a power supply operation status acquisition module, an abnormal potential field model construction module, an abnormal interface construction module, a key disturbance sample extraction module, a disturbance state stable projection mapping module, a potential field training objective function construction module, and an online anomaly monitoring module. The power supply operation status acquisition module acquires historical power supply operation data, constructs an operation status vector, and performs rated operating condition normalization processing. The abnormal potential field model construction module constructs an abnormal potential field model based on a normalized state vector, and parametrically expresses the relationship between state variables by introducing a coupling matrix, a linear sensitivity coefficient vector, and a bias parameter in a symmetrical manner. The abnormal interface construction module uses the zero potential energy surface as the boundary condition between stable and abnormal states to form a state discrimination boundary, which serves as the abnormal interface. The key perturbation sample extraction module filters samples based on the potential energy distance from the abnormal interface to construct a key perturbation sample set. The disturbance state stable projection mapping module maps key disturbance samples to stable corrected states that satisfy the potential field stability constraints. The potential field training objective function construction module constructs the potential field training objective function based on the normalized state vector, key perturbation samples, and stable correction state. The online anomaly monitoring module performs online anomaly monitoring on power supply operation data based on the trained abnormal potential field model.

2. The data-driven power anomaly monitoring system according to claim 1, characterized in that: The abnormal potential field model construction module establishes a mapping relationship between the operating state and the degree of abnormality. It takes the normalized state vector as input, introduces a second-order coupled potential field, describes the independent influence of each state variable through single-variable terms, describes the coupling influence between different state variables through interaction terms, and outputs the corresponding abnormal potential energy index to characterize the degree of abnormality of the current power supply operating state.

3. The data-driven power anomaly monitoring system according to claim 1, characterized in that: The key disturbance sample extraction module constructs a sensitive sample screening mechanism based on the output of the abnormal potential field model. By analyzing the distribution position of the samples in the abnormal potential field model mapping function, it identifies key disturbance samples in the transition region between stable and abnormal states.

4. The data-driven power anomaly monitoring system according to claim 3, characterized in that: The disturbance state stability projection mapping module performs stability projection mapping on key disturbance samples, mapping them to the closest stable state that satisfies the potential field stability constraint, which is used to constrain the structure learning of the abnormal potential field model.

5. The data-driven power anomaly monitoring system according to claim 4, characterized in that: In order to enable the abnormal potential field to have good state discrimination ability, the potential field training objective function construction module defines the potential field training objective function, including the global potential field smoothing constraint term, the misjudgment constraint mechanism term, and the key perturbation sample reinforcement mechanism term.

6. The data-driven power anomaly monitoring system according to claim 5, characterized in that: The online anomaly monitoring module uses the trained abnormal potential field model to calculate the anomaly index in real time and perform online anomaly monitoring of power supply operation data.

Images