A distributed solar photovoltaic support system fault diagnosis method and system

Abstract

The application provides a kind of distributed solar photovoltaic support system fault diagnosis method and system, comprising: for the influence of environmental change on fault determination, construct digital twin simulation framework based on support structure parameters, material thermal expansion coefficient and wind load model, generate three-dimensional grid environment-motion response surface, and store the theoretical response boundary in edge node;Real-time acquisition of temperature, wind speed, humidity, environment state fingerprint is generated by hash, and the reference parameter is obtained by table lookup, combined with the actual motion of support, drive motor current and temperature data, the absolute deviation from the reference value is calculated;Through multivariate coupling attenuation function weighted fusion deviation, using short window continuous monitoring and long window ratio analysis, trigger preliminary screening alarm;Then call rule tree causal verification module to distinguish binary between card stagnation and pseudo anomaly, the application can improve the environmental adaptability, realize efficient and accurate early screening of mechanical failure.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for mechanical faults in photovoltaic brackets, and in particular to a fault diagnosis method and system for distributed solar photovoltaic bracket systems. Background Technology

[0002] With the continuous expansion of photovoltaic power generation systems, the intelligent operation and maintenance of distributed photovoltaic (PV) supports has become a key focus of the industry. Currently, in the field of health management and fault diagnosis of PV support systems, the mainstream technical solutions are mainly based on data-driven anomaly detection and cloud-based intelligent analysis. One type of technical solution typically transmits a large amount of historical operation and maintenance data to the cloud, relying on deep neural networks (such as convolutional neural networks, temporal autoencoders, or ensemble learning models) for fault prediction and trend analysis, thereby achieving health status assessment and early warning for key components such as drive mechanisms and structural parts. Another type applies edge computing technology, focusing on sinking lightweight fault detection algorithms to the support controller, thereby reducing data transmission latency and improving the system's real-time performance and response efficiency. In recent years, with the rapid development of cloud-edge collaborative architecture, some research has attempted to combine physical models and machine learning methods to achieve dynamic fault diagnosis and adaptive threshold adjustment for support systems, promoting the evolution of intelligent operation and maintenance towards a direction that emphasizes both high accuracy and low latency.

[0003] Existing distributed support system fault diagnosis methods mainly rely on set thresholds or pattern recognition algorithms for support system operating parameters (such as tilt angle change rate, motor current, structural vibration signals, etc.) to determine anomalies. Some technical practices manually set judgment criteria for various environmental conditions, while some solutions attempt to use online trained neural network models for state learning. However, in practical applications, photovoltaic power plants operate outdoors year-round, with significant fluctuations and complex coupling effects of environmental variables such as temperature, wind speed, and humidity. This leads to dynamic changes in many parameters of the support system during normal operation. Traditional diagnostic models struggle to accurately distinguish between normal fluctuations caused by environmental changes and true fault characteristics such as mechanical jamming. On the one hand, they are easily affected by multivariate disturbances, leading to false alarms; on the other hand, they may miss alarms when the initial abnormal manifestations of jamming are insufficient. In particular, purely data-driven models rely too heavily on historical fault samples and long-term labeled data, resulting in limited model generalization ability and online update adaptability. The diagnostic boundary is difficult to adjust in a timely manner with environmental fluctuations, and computational resources are high, making them unsuitable for direct deployment on restricted edge devices.

[0004] Existing technologies are typically suited for the operation and maintenance of photovoltaic (PV) systems in large-scale, stable environments, such as centralized analysis in cloud data centers and periodic batch diagnostics. Their model structures usually require extensive training with labeled samples, heavily rely on historical fault data and continuous data updates, and lack physical interpretability. On the other hand, for edge deployment needs, most lightweight algorithms can only handle single-variable anomalies and cannot cope with the highly nonlinear evolution of dynamic responses under the coupled influence of multiple environmental variables such as temperature, wind speed, and humidity. This makes it difficult to meet the high robustness and high-precision early warning requirements for mechanical jamming in distributed PV systems under extreme weather conditions. Summary of the Invention

[0005] In order to solve the above-mentioned technical problems, the present invention provides a method and system for fault diagnosis of distributed solar photovoltaic support system.

[0006] The technical solution of this invention is implemented as follows: A fault diagnosis method for a distributed solar photovoltaic support system, comprising:

[0007] S1: Based on the structural parameters of the support structure, the thermal expansion coefficient of the material, and the wind load calculation model, the environment-motion response surface is generated through digital twin simulation in the offline stage;

[0008] S2: Real-time acquisition of current environmental variables of photovoltaic support, normalization of the current environmental variables and generation of environmental state fingerprint through hash mapping;

[0009] S3: Based on the environmental state fingerprint, retrieve the corresponding reference parameter group in the environment-motion response surface, and obtain the theoretical tilt angle change rate reference value, drive motor current slope reference value, and backplate temperature gradient reference value under the current environmental state.

[0010] S4: Based on real-time measured bracket tilt angle, drive motor current, and backplate temperature data, calculate the absolute deviation between the measured tilt angle change rate and the theoretical tilt angle change rate reference value, the absolute deviation between the measured drive motor current slope and the drive motor current slope reference value, and the absolute deviation between the measured backplate temperature gradient and the backplate temperature gradient reference value.

[0011] S5: Input the triple absolute deviation into a preset multivariable coupling attenuation function. The multivariable coupling attenuation function dynamically adjusts the attenuation coefficient according to the environmental state fingerprint to generate a weighted fusion deviation index.

[0012] S6: Monitor whether the weighted fusion deviation index continuously exceeds the short window monitoring threshold within a short sliding window of a first preset length, and simultaneously count the percentage of time periods during which the weighted fusion deviation index exceeds the long window statistical threshold within a long sliding window of a second preset length, to obtain the short sliding window monitoring result and the long sliding window statistical result.

[0013] S7: If the monitoring result of the short sliding window and the statistical result of the long sliding window simultaneously meet the preset conditions, a mechanical jamming initial screening alarm signal is triggered.

[0014] S8: Call the preset symbol rule tree form of causal verification module, and perform binary discrimination of jamming fault and environmental induced pseudo-anomaly based on the time correlation characteristics of the target tilt angle achievement state of the support, the current fluctuation mode of the drive motor and the weighted fusion deviation index, and output the final mechanical jamming warning result.

[0015] The present invention also provides a fault diagnosis system for a distributed solar photovoltaic support system, which uses the above-mentioned fault diagnosis method for a distributed solar photovoltaic support system to diagnose faults in the solar photovoltaic support system.

[0016] The present invention provides a fault diagnosis method and system for a distributed solar photovoltaic support system, which has the following beneficial effects:

[0017] (1) This invention effectively overcomes the dependence of traditional data-driven models on a large number of labeled samples and fixed threshold criteria. Under the premise of no historical fault data, no need for cloud collaboration, and no online parameter tuning, it achieves a highly robust early warning for mechanical jamming of distributed photovoltaic brackets. By constructing a kinematic benchmark model that integrates structural mechanics, thermodynamics and aerodynamics, and using digital twin simulation to generate an "environment-motion response surface" covering all working conditions, the dynamic characteristics of normal behavior are solidified in the read-only storage area of ​​edge nodes in a physically interpretable form, so that the system can still establish a reliable reference baseline in the absence of real fault samples. Combined with the environmental state fingerprint coding mechanism, it realizes accurate positioning and context awareness of the current operating conditions, avoids false anomaly misjudgment caused by natural factors such as temperature difference deformation and gust disturbance, and significantly improves the accuracy of early warning and scene adaptability.

[0018] (2) This invention proposes a lightweight, interpretable, and training-free dual-scale dynamic decision architecture that balances real-time performance, sensitivity, and anti-interference capabilities, meeting the stringent engineering requirements of low power consumption, high reliability, and long-term unattended operation for photovoltaic bracket edge controllers. An environmental fingerprint is generated using hash mapping to quickly index baseline parameters. A pre-defined multivariate coupling decay function is used to nonlinearly weight and fuse the joint deviations of tilt angle change rate, current slope, and temperature gradient, forming a comprehensive deviation index with environmental adaptability. A dual-scale sliding window criterion combining short-window transient change detection and long-window trend accumulation analysis is introduced to effectively distinguish between transient disturbances and continuous stagnation, preventing accidental noise from triggering false alarms. Finally, causal verification is performed through a symbolic rule tree, making a deterministic judgment based on the dynamic relationship between physical variables (e.g., the target angle is not reached but the current saturates and oscillates). The entire process requires no model training, parameter updates, or feature recalibration; the inference process is transparent and controllable, and deployment costs and maintenance difficulties are significantly reduced.

[0019] (3) This invention realizes a technological paradigm leap from "black box prediction" to "white box diagnosis", and constructs a closed-loop early warning system based on physical mechanism and environmental perception, which has good scalability and engineering portability. It abandons the mainstream AI architectures such as CNN, Transformer, graph neural network, federated learning, which require a lot of computing power and lack physical interpretation, and instead returns to the essence of electromechanical systems. It transforms the mechanical jamming problem into a pattern recognition task of observable physical quantities deviating from the theoretical trajectory, so that the early warning results have a clear causal chain and debugging and tracing path. The constructed response surface and rule base can be flexibly replaced with different bracket models without the need to re-collect data or adjust the network structure, which is suitable for the rapid deployment of large-scale heterogeneous devices. Attached Figure Description

[0020] Figure 1 This is a flowchart of a fault diagnosis method for a distributed solar photovoltaic support system according to the present invention;

[0021] Figure 2 This is a sub-flowchart of a fault diagnosis method for a distributed solar photovoltaic support system according to the present invention;

[0022] Figure 3 This is another sub-flowchart of a fault diagnosis method for a distributed solar photovoltaic support system according to the present invention. Detailed Implementation

[0023] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0024] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0025] like Figure 1 As shown, the present invention provides a fault diagnosis method for a distributed solar photovoltaic support system, specifically including:

[0026] S1: Based on the structural parameters of the support structure, the coefficient of thermal expansion of the material and the wind load calculation model, an environmental-motion response surface covering the entire working condition range is generated through digital twin simulation in the offline stage. The surface is stored in the read-only storage area of ​​the edge nodes in the form of a three-dimensional mesh. Each mesh point corresponds to the time-series envelope boundary of the theoretical tilt angle change rate under the combination of temperature, wind speed and humidity, the time-series envelope boundary of the drive motor current slope and the time-series envelope boundary of the back plate temperature gradient.

[0027] S2: Real-time acquisition of current ambient temperature, wind speed and humidity data of photovoltaic support, normalization of the environmental variables and generation of a 32-bit binary environmental state fingerprint through hash mapping, which serves as the grid region index identifier of the environment-motion response surface;

[0028] S3: Based on the environmental state fingerprint, retrieve the corresponding reference parameter group in the environmental-motion response surface, and obtain the theoretical tilt angle change rate reference value, drive motor current slope reference value, and backplate temperature gradient reference value under the current environmental state.

[0029] S4: Based on real-time measured bracket tilt angle, drive motor current and back plate temperature data, calculate the absolute deviation between the measured tilt angle change rate and the theoretical reference value, the absolute deviation between the measured drive motor current slope and the theoretical reference value, and the absolute deviation between the measured back plate temperature gradient and the theoretical reference value.

[0030] S5: Input the triple absolute deviation into a preset multivariable coupled attenuation function. This function dynamically adjusts the attenuation coefficient according to the environmental state fingerprint to generate a weighted fusion deviation index. The multivariable coupled attenuation function adopts a piecewise power law attenuation form to suppress the influence of single variable disturbance.

[0031] S6: Monitor whether the weighted fusion deviation index continuously exceeds 1.3 times the benchmark threshold dynamically determined based on environmental state fingerprint within a 15-second short sliding window, and at the same time, count the percentage of time periods in which the weighted fusion deviation index exceeds 1.8 times the benchmark threshold within a 5-minute long sliding window;

[0032] S7: Determine whether the monitoring results of the short sliding window and the statistical results of the long sliding window simultaneously meet the preset conditions. If they do, trigger the mechanical jamming primary screening alarm signal.

[0033] S8: Call the pre-set lightweight causal verification module in the form of a symbol rule tree, and perform binary discrimination of jamming fault and environmentally induced pseudo-anomaly based on the time-series correlation characteristics of the target tilt angle achievement status of the support, the current fluctuation pattern of the drive motor and the weighted fusion deviation index, and output the final mechanical jamming warning result.

[0034] Step S1: Based on the support structure parameters, material thermal expansion coefficient, and wind load calculation model, an environmental-motion response surface covering the entire working condition range is generated offline through digital twin simulation. This surface is stored in the read-only storage area of ​​the edge nodes in the form of a three-dimensional mesh. Each mesh point corresponds to the time-series envelope boundary of the theoretical tilt angle change rate under the combination of temperature, wind speed, and humidity, the time-series envelope boundary of the drive motor current slope, and the time-series envelope boundary of the backplate temperature gradient. Specifically, it includes:

[0035] S1.1: Wind load modeling is performed on the structural parameters and thermal expansion coefficients of the support structure. Based on aerodynamic principles and the geometric features of the support, a wind load calculation model is generated to quantify the dynamic load distribution on the support under different wind speeds. Specifically, based on the structural parameters of the support, geometric features of the support are extracted to obtain the friction coefficient between the support arm length and the shaft; thermal deformation analysis is performed based on the thermal expansion coefficient of the material to generate material thermal deformation compensation parameters; the geometric features of the support and the material thermal deformation compensation parameters are integrated to perform dynamic wind load distribution calculation and generate a wind load calculation model; the wind load calculation model is used as the input for digital twin simulation.

[0036] Based on the structural parameters of the support structure and the thermal expansion coefficient of the materials, an aerodynamic modeling method is adopted (parameters: windward area A, drag coefficient). (Air density ρ, atmospheric pressure P), to achieve quantitative calculation of instantaneous wind pressure on the support surface and key structural nodes under different wind speed conditions;

[0037] Furthermore, through the scaffold geometric feature extraction method (parameter: arm length) Shaft diameter , cross-sectional shape of the support arm This allows for the physical geometric parameterization of the support structure profile, and the acquisition of accurate data on arm length and shaft friction coefficient.

[0038] Furthermore, based on the material thermal expansion coefficient analysis method (parameter: thermal expansion coefficient) Temperature range Material elastic modulus This allows for the prediction of thermal deformation of the support under different temperature conditions, and the generation of material thermal deformation compensation parameters to correct the response error of the kinematic model under high and low temperature environments.

[0039] By coupling the geometric features of the support structure with the material thermal deformation compensation parameters, a dynamic wind load distribution calculation method is adopted (parameters: wind speed v, angle of attack). Local turbulence coefficient This allows for the calculation of instantaneous wind pressure and bending moment distribution at each structural unit node of the support frame, and the generation of a time-series dataset of node loads.

[0040] Furthermore, the finite element method for wind load calculation (parameter: number of nodes) is used. Element stiffness matrix K, load vector The spatial distribution calculation of wind load dynamic response is realized on the grid of the support structure, and a wind load calculation model dataset is generated.

[0041] The wind load calculation model generated by the above method is used as the input basis for the digital twin simulation module to realize the fully coupled simulation preparation of physical environment parameters and support kinematic behavior.

[0042] For example, in a certain distributed photovoltaic power station, the windward area A of the support structure is set as follows: square meters, drag coefficient Set as air density Set as kilograms per cubic meter, atmospheric pressure P is the standard atmosphere. Pascal. Support arm length Measurement is meters, shaft diameter for Meters; the support arm has a rectangular cross-section. The material's coefficient of thermal expansion... for Elastic modulus per degree Celsius for Gipascal, temperature range for Degrees Celsius. At a wind speed v= meters per second, angle of attack = Under the given conditions, the nodal wind pressure is calculated using aerodynamic formulas:

[0043]

[0044] in, For nodal wind pressure, air density, For wind speed, Let A be the drag coefficient and A be the windward area. The calculated nodal wind pressure is: Newton. In thermal deformation analysis, the change in material length. The calculation formula is:

[0045]

[0046] Calculated for The value of meters was used as the material thermal deformation compensation value in the wind load distribution calculation. After finite element analysis, the maximum bending moment at each node was obtained. The newton-meter (N / m) is used to form a complete wind load calculation model, thereby significantly improving the accuracy and stability of the support environment response simulation in digital twin simulation;

[0047] S1.2: Based on the structural parameters of the support structure, the coefficient of thermal expansion of the material, and the generated wind load calculation model, a digital twin simulation framework is constructed to establish a kinematic digital twin of the support. Specifically, this involves: initializing the kinematic model based on the structural parameters of the support to generate a kinematic baseline model of the support body; integrating the coefficient of thermal expansion of the material and the wind load calculation model, performing environmental coupling effect modeling to generate a kinematic-environment coupled response model of the support; applying solar tracking commands to the kinematic-environment coupled response model of the support, performing dynamic behavior simulation to form a kinematic digital twin of the support; and using the kinematic digital twin of the support as the object of full-condition simulation.

[0048] Based on the input support structure parameters, material thermal expansion coefficient, and wind load calculation model generated by step S1.1, a modular digital twin framework construction method (parameters: support geometric feature matrix, material thermal expansion compensation coefficient, dynamic load distribution function) is adopted to realize the structural initialization function of the support kinematic digital twin;

[0049] Furthermore, by using the numerical modeling method of motion equations (parameters: arm length, shaft friction coefficient, gear transmission ratio), the kinematic reference model of the support body is generated, and a theoretical time series dataset containing angular velocity, angular acceleration and dynamic torque is obtained.

[0050] Furthermore, an environmental coupling effect modeling method (parameters: material thermal expansion coefficient vector, wind load calculation model matrix) is adopted to realize the multi-physics field coupling calculation of material thermal deformation and wind load, and generate a kinematic-environmental coupled response model of the support. This model can reflect the transient response characteristics of the support under different temperature and humidity wind field conditions.

[0051] Furthermore, by injecting solar tracking commands (parameters: solar azimuth trajectory function, support target angle sequence), external control input is applied to the kinematic-environment coupled response model of the support, dynamic behavior simulation processing is performed, and a set of theoretical response curves for tilt angle, driving current, and backplate temperature that vary with daytime time are generated.

[0052] By using the data encapsulation interface of the digital twin simulation framework, the dynamic response curve group from the previous step is mapped to the entity object of the kinematic digital twin of the support, realizing a panoramic simulation model that includes structural features, environmental coupling features, and control response features, providing an executable object that can be directly called for subsequent full-condition response surface calculation.

[0053] For example, in a photovoltaic power station support unit, the parameter is arm length. meter, coefficient of friction of shaft Gear transmission ratio Coefficient of thermal expansion of materials The wind load distribution per degree Celsius is determined by the model established in step S1.1 at wind speeds. Maximum torque output at m / s When performing the above construction process, the kinematic baseline model obtains the tilt angle change rate curve through Euler integration, and the environmental coupled response model is superimposed with the temperature gradient. Celsius and humidity The thermal deformation compensation amount under the condition of % is After the solar tracking command is injected, the simulation output shows that the tilt angle remains constant during the midday period. Degree, peak drive current is Ampere, backplate temperature rise rate is Temperatures per minute. This digital twin achieves simultaneous depiction of dynamic characteristics and environmental influences in full-condition simulation, providing a high-resolution theoretical basis for response surface generation and realizing accurate benchmark modeling of mechanical jamming characteristics under different environments;

[0054] S1.3: In the offline phase, the kinematic digital twin of the support is input with a combination of parameters under all operating conditions, and dynamic simulation calculations are performed, covering a temperature range of -20℃ to 60℃, a wind speed of 0 to 15m / s, and a humidity range of 10% to 95%RH. Specifically, the following steps are taken: Dynamic simulation of the support tracking process is performed based on the combination of parameters under all operating conditions to obtain the theoretical time-series curve of the support tilt angle; the theoretical time-series curve of the support tilt angle is differentiated to generate the theoretical tilt angle change rate time-series data; the slope of the drive motor current response is calculated to generate the drive motor current slope time-series data; the gradient of the backplate temperature response is calculated to form the backplate temperature gradient time-series data; and the theoretical tilt angle change rate time-series data, drive motor current slope time-series data, and backplate temperature gradient time-series data are integrated to generate the theoretical tilt angle change rate time-series envelope boundaries, drive motor current slope time-series envelope boundaries, and backplate temperature gradient time-series envelope boundaries.

[0055] S1.4: The generated theoretical tilt rate of change time-series envelope boundaries, drive motor current slope time-series envelope boundaries, and backplate temperature gradient time-series envelope boundaries are subjected to 3D mesh generation. A mesh index system is constructed based on the three variables of temperature, wind speed, and humidity to generate an environment-motion response surface. Specifically, the three variables of temperature, wind speed, and humidity are normalized to generate a standardized environmental variable vector; based on the standardized environmental variable vector, 3D spatial discretization is performed to divide the space into mesh cells with temperatures ranging from -20℃ to 60℃, wind speeds from 0 to 15 m / s, and humidity from 10% to 95% RH; theoretical boundary data mapping is performed on each mesh cell to store the theoretical tilt rate of change time-series envelope boundaries, drive motor current slope time-series envelope boundaries, and backplate temperature gradient time-series envelope boundaries under the corresponding environmental combination; the mapping results of all mesh cells are integrated to generate a 3D mesh-form environment-motion response surface; the environment-motion response surface is used as the input object for storage optimization.

[0056] S1.5: Perform read-only storage optimization on the environment-motion response surface, using a data compression algorithm to reduce storage usage, and store it in a compressed format in the read-only storage area of ​​the edge nodes. Specifically: perform redundant data removal based on the environment-motion response surface to generate a simplified response dataset; perform hash encoding on the simplified response dataset to generate fixed-length data blocks; perform lossless compression on the data blocks to form a compressed data stream; write the compressed data stream into the read-only storage area of ​​the edge nodes to ensure fast retrieval capability of the environment-motion response surface; output the environment-motion response surface stored in the read-only storage area of ​​the edge nodes to provide benchmark parameters for subsequent environmental state fingerprint lookup.

[0057] Step S2: Real-time acquisition of ambient temperature, wind speed, and humidity data for the photovoltaic support; normalization of these environmental variables; and generation of a 32-bit binary environmental state fingerprint via hash mapping. This fingerprint serves as the grid region index identifier for the environment-motion response surface. Specifically, this includes:

[0058] S2.1: Real-time acquisition of raw data signals output by ambient temperature, wind speed and humidity sensors at the photovoltaic bracket installation location to obtain the current raw values ​​of ambient temperature, wind speed and humidity as the initial input source for environmental variable processing;

[0059] The raw data signals output by the ambient temperature sensor, wind speed sensor and humidity sensor at the photovoltaic bracket installation location are acquired using a high-precision analog-to-digital conversion method (parameters: 16-bit sampling resolution, 10Hz sampling frequency) to achieve digital acquisition of environmental variables and ensure that the quantization accuracy of the raw physical quantity signals meets the accuracy requirements of subsequent normalization processing.

[0060] Furthermore, through sensor self-diagnosis and calibration methods (parameters: zero-point drift threshold ±0.05℃, sensitivity drift threshold ±0.5%FS), zero-point correction and sensitivity compensation of the acquired signal are achieved, and the original values ​​of ambient temperature, wind speed and humidity after noise suppression are obtained, ensuring the consistency and reliability of the input data.

[0061] Furthermore, a time-series smoothing method is employed (parameters: sliding window length of 5 points, weighting coefficients). =0.8), short-time mean filtering is applied to the raw value data sequences of the three types of environments to suppress instantaneous fluctuations and generate smoothed instantaneous values ​​of ambient temperature, wind speed and humidity, providing stable input conditions for real-time normalization calculation;

[0062] Furthermore, by using a data consistency verification method (parameter: the criterion is that the simultaneous change rate of temperature, wind speed, and humidity is less than a set threshold), the alignment of multivariate collection timestamps and differential consistency verification are realized, and a three-dimensional environmental original dataset with time synchronization correction is generated.

[0063] Through the above acquisition and preprocessing methods, the analog signal from the previous step is converted into stable digital data of raw ambient temperature, raw wind speed and raw humidity values, thereby realizing real-time acquisition and accurate quantification of multivariable environmental conditions.

[0064] For example, in a photovoltaic power station at an altitude of 1200 meters, the ambient temperature sensor is a PT100 platinum resistance thermometer with a sampling resolution of 16 bits; the wind speed sensor is a three-cup anemometer with a range of 0~30 m / s; and the humidity sensor is a capacitive humidity sensor with a range of 10%~95%RH. The sampling frequency is configured to 10Hz, the zero-point drift threshold is set to ±0.05℃, and the sensitivity drift threshold is set to ±0.5%FS. During sensor initialization, zero-point and sensitivity calibrations are performed. The original output of the temperature channel is corrected from 23.56℃ to 23.50℃, the wind speed channel from 3.12 m / s to 3.10 m / s, and the humidity channel from 45.36%RH to 45.40%RH. During data acquisition, a sliding weighted average filter is applied to every 5 sampling points, with weighting coefficients... With a coefficient of 0.8, taking the temperature channel as an example, the original sequence [23.50, 23.52, 23.49, 23.51, 23.50] outputs a smoothed value of 23.505℃ after processing. After applying timestamp synchronization correction to the three-channel data, the generated three-element environmental original dataset has a temperature of 23.505℃, a wind speed of 3.10m / s, and a humidity of 45.40%RH. After being provided to the normalization module, the output stability was verified, and the data fluctuation was significantly reduced, effectively supporting the accuracy and robustness of the subsequent hash mapping process.

[0065] S2.2: Based on the preset minimum-maximum range threshold of environmental variables, perform linear normalization processing on the collected raw values ​​of ambient temperature, wind speed and humidity, map each physical quantity to the [0,1] standardized interval, generate a normalized environmental variable vector, so as to eliminate the dimensional differences of temperature, wind speed and humidity and build a unified processing basis.

[0066] S2.3: Input the normalized environment variable vector into the preset hash function module, and perform bit-level encoding operation on the normalized environment variables through the deterministic hash mapping algorithm to generate a fixed-length binary hash sequence, so as to achieve a lossless and compact representation of the environment state and ensure the determinism of data processing;

[0067] S2.4: Perform bit truncation on the generated binary hash sequence, extract the first 32 bits of the sequence to form the environment state fingerprint, and generate a 32-bit binary environment state fingerprint through fixed-length truncation operation to meet the storage resource constraints of the edge node and ensure the query efficiency of the index identifier.

[0068] S2.5: The generated 32-bit binary environmental state fingerprint is directly designated as the grid region index identifier of the environment-motion response surface. Through index mapping operation, the corresponding theoretical tilt angle change rate reference value, drive motor current slope reference value and backplate temperature gradient reference value in the surface are quickly located, providing environmental adaptive parameter basis for subsequent dynamic threshold calculation.

[0069] Based on the 32-bit binary environment state fingerprint data generated in step S2.4, a direct index mapping method (parameters: fingerprint value, environment-motion response surface mesh structure definition) is used to establish a one-to-one correspondence between the fingerprint value as a unique index key and the surface storage structure.

[0070] Furthermore, by using the bit-domain parsing method (parameters: high-bit interval, middle-bit interval, and low-bit interval bit length configuration), the 32-bit fingerprint value is divided into temperature, wind speed, and humidity dimension index components according to the preset bit length, and a three-dimensional mesh coordinate index triplet is generated as the input coordinate set for surface positioning;

[0071] Furthermore, through a storage address mapping method (parameters: 3D mesh coordinates, surface data offset reference table), the physical address of the corresponding read-only memory area is calculated using a linear address conversion formula from the ternary coordinate index. The calculation formula is:

[0072]

[0073] in, This is the baseline offset for the temperature dimension. This represents the difference in wind speed coordinates. Store increments for wind speed step size. This represents the difference in humidity coordinates. Store incremental values ​​for humidity step size;

[0074] Furthermore, through a fast lookup process (parameters: physical address, read-only memory access protocol), the theoretical tilt angle change rate reference value, drive motor current slope reference value, and backplane temperature gradient reference value of the corresponding grid point are extracted from the storage area to form a complete set of reference parameters corresponding to the environmental conditions.

[0075] By using direct index mapping technology, fingerprint data and curved mesh are quickly associated as a set of ternary parameters, achieving zero-latency matching of environmental parameters to kinematic reference values, and providing accurate input for environmental adaptation in dynamic threshold calculation;

[0076] For example, when the ambient temperature is 38℃, the wind speed is 8.5m / s, and the humidity is 65%, the normalized value for temperature is 0.783, the normalized value for wind speed is 0.567, and the normalized value for humidity is 0.611. After hash mapping and bit truncation in steps S2.3 and S2.4, the fingerprint value 0x8F3AE41C is obtained. Using a bit-field resolution method, its high 8 bits, middle 12 bits, and low 12 bits are mapped to a temperature index of 200, a wind speed index of 56, and a humidity index of 150, respectively. Using the storage address mapping formula, with a temperature offset... Wind speed step increment Humidity step increment The physical address is obtained by calculating the parameters. The result is Accessing this address via Read-Only Memory Protocol (ROM) retrieves the theoretical tilt rate of change baseline value. Degrees / second, the reference value for the slope of the drive motor current is [value missing]. Ampere / second, backplate temperature gradient reference value ℃ / second, forming a complete set of baseline parameters. In subsequent dynamic threshold calculations, this set of parameters is used as environmental compensation input to achieve stable identification of weak mechanical jamming characteristics under actual environmental conditions, significantly improving the accuracy of early warning.

[0077] like Figure 2 As shown, step S3 involves retrieving the corresponding set of reference parameters from the environment-motion response surface based on the environmental state fingerprint, and obtaining the theoretical tilt angle change rate reference value, drive motor current slope reference value, and backplate temperature gradient reference value under the current environmental state. Specifically, this includes:

[0078] S3.1: Perform grid index parsing on the environmental state fingerprint. Based on the high 8 bits, middle 12 bits and low 12 bits of the 32-bit binary environmental state fingerprint, perform bit mask segmentation operation to generate three-dimensional grid coordinate index values, so as to clarify the discretized query positions of temperature dimension, wind speed dimension and humidity dimension in the environmental-motion response surface.

[0079] The input environmental state fingerprint data is processed by a bitmask segmentation method (parameter: bit width [8,12,12]) to achieve independent parsing of temperature, high wind speed and low humidity information;

[0080] Furthermore, through logical AND operations and bit shift operations (parameters: mask 0xFF000000, mask 0x00FFF000, mask 0x00000FFF), the original index values ​​of the high 8 bits, the middle 12 bits, and the low 12 bits are extracted respectively, and the original indexes of temperature, wind speed, and humidity are obtained.

[0081] Furthermore, by using an integer type conversion method (parameter: unsigned 32-bit integer conversion rule), the original index value is numerically processed, and the extracted high, middle and low bit sequences are converted into temperature index integers, wind speed index integers and humidity index integers, ensuring the physical interpretability of the index value in subsequent spatial mapping operations;

[0082] Furthermore, through a multi-dimensional coordinate combination method (parameter: output is...), This allows for the combination of temperature, wind speed, and humidity indices into a three-dimensional grid coordinate index value, and the generation of a grid discretized query location dataset.

[0083] By using bitmask segmentation and coordinate combination processing, the environmental state fingerprint parsing result from the previous step is transformed into three-dimensional mesh coordinate data that can be discretized in the environment-motion response surface, thus realizing a direct association between the environmental state and the theoretical envelope data storage location.

[0084] For example, for environmental state fingerprints Perform bitmask segmentation, using a mask. Obtain the high-order temperature index bit sequence The temperature index after numerical processing is Use a mask Then shift right by 12 bits to obtain the median wind speed index bit sequence. After numerical processing, the wind speed index is: Use a mask Obtain the low-bit humidity index bit sequence The humidity index after numerical processing is The three indices are combined using a three-dimensional coordinate combination algorithm. The format combination is a three-dimensional grid coordinate (90, 956, 3856), which is used as the input of the discretized query position to the grid coordinate mapping module in step S3.2, which significantly improves the parsing speed and accuracy of indexing to the surface data storage location;

[0085] S3.2: Perform grid coordinate mapping processing based on the three-dimensional grid coordinate index value. Utilize the pre-calibrated temperature range quantization step size, wind speed range quantization step size, and humidity range quantization step size to convert the discretized query position into the three-dimensional grid coordinates of the environment-motion response surface, so as to determine the specific storage areas of the theoretical tilt angle change rate time sequence envelope boundary, the drive motor current slope time sequence envelope boundary, and the backplane temperature gradient time sequence envelope boundary.

[0086] Based on the 3D mesh coordinate index values ​​output from step S3.1, a pre-calibrated interval quantization step-size mapping algorithm is used (parameter: temperature step size). Wind speed step size Humidity step size This transforms the discretized query location into a three-dimensional physical coordinate system of the environment-motion response surface. This is achieved through a step-size mapping algorithm (temperature step). (Depending on the discretization settings in step S1), multiply the temperature dimension index value by Generate physical coordinate values ​​for temperature to achieve precise positioning of the temperature dimension;

[0087] Furthermore, through the step size mapping method (wind speed step size) (Depending on the discretization settings in step S1), multiply the wind speed dimension index value by Generate physical coordinates of wind speed to achieve precise positioning of wind speed dimensions;

[0088] Furthermore, through the step size mapping method (humidity step size) (Depending on the discretization setting in step S1), multiply the humidity dimension index value by Generate physical coordinate values ​​for humidity to achieve precise positioning of the humidity dimension;

[0089] A three-dimensional mapping combination function is used to combine the physical coordinates of temperature, wind speed, and humidity into a three-dimensional grid coordinate system, forming the entry point for the retrieval of the environment-motion response surface.

[0090] Through a three-dimensional cross-indexing mechanism, the above three-dimensional mesh coordinates are mapped to the specific storage area of ​​the curved surface storage structure, thereby realizing the physical location locking of the theoretical tilt angle change rate time sequence envelope boundary, the drive motor current slope time sequence envelope boundary, and the backplane temperature gradient time sequence envelope boundary.

[0091] By combining step-size mapping and cross-indexing, the index value is transformed into three-dimensional grid coordinates that can directly locate the storage area, enabling rapid retrieval of theoretical benchmark data under environmental conditions.

[0092] For example, in a distributed photovoltaic (PV) support system, assuming the discretized temperature range is -20°C to 60°C, the step size is... =5℃; Discretized wind speed range is 0m / s to 15m / s, step size =1.5m / s; Discretized humidity range is 10%RH to 95%RH, step size =5%RH. The temperature index value obtained from step S3.1 is 7, the wind speed index value is 4, and the humidity index value is 6. The physical coordinate values ​​of the temperature are calculated using step-size mapping. get ℃; Wind speed physical coordinates: get m / s; Humidity physical coordinate value: get %RH. ℃ m / s and %RH is combined into three-dimensional mesh coordinates, and the 7th×4th×6th cell of the response surface storage structure is located through the cross-indexing mechanism. In this cell, the theoretical tilt rate change time-series envelope boundary, the drive motor current slope time-series envelope boundary, and the backplate temperature gradient time-series envelope boundary corresponding to this environment combination can be directly read. This ensures that the deviation calculation module can perform parameter compensation and dynamic threshold setting under the current environmental conditions, which significantly improves the accuracy of jamming diagnosis and environmental adaptability.

[0093] S3.3: Perform storage address positioning processing based on the 3D mesh coordinates. Based on the mesh data storage structure of the environment-motion response surface and the starting address offset of the read-only storage area, calculate the linearized storage address of the 3D mesh coordinates to obtain the reference parameter group storage position pointer of the corresponding mesh point in the environment-motion response surface.

[0094] S3.4: Perform parameter reading processing on the reference parameter group storage location pointer. Based on the memory access protocol of the edge node controller, extract the three parameters from the read-only storage area: the theoretical tilt angle change rate reference value, the drive motor current slope reference value, and the backplane temperature gradient reference value, to form a complete reference parameter group dataset under the current environmental conditions.

[0095] S3.5: Perform parameter encapsulation processing on the complete reference parameter set dataset. Based on the preset parameter transfer interface specification, encapsulate the theoretical tilt angle change rate reference value, drive motor current slope reference value and backplate temperature gradient reference value into a standardized parameter structure, and output it to the subsequent deviation calculation module for real-time environmental compensation.

[0096] For the complete baseline parameter set dataset obtained from step S3.4, a structure definition method is used (parameter field: theoretical dip angle change rate baseline value). , Drive motor current slope reference value Backplate temperature gradient reference value This enables hierarchical encapsulation of parameters.

[0097] Furthermore, through field type constraint methods (parameter type: , The company agreed: , , This ensures cross-module data consistency and provides a data structure that can be directly called by the deviation calculation module.

[0098] Furthermore, through memory alignment optimization methods (parameter: byte alignment boundary) Byte, padding strategy: zero padding), to achieve high-speed access in a single read and generate a structure storage block with contiguous addresses;

[0099] Furthermore, through the interface adapter binding method (parameters: deviation calculation module interface ID, calling convention: pass by value), lossless transmission of reference parameters is achieved, and a pointer handle that can be directly mapped on the MCU internal bus is obtained;

[0100] By using parameter encapsulation processing, the results of the previous step are transformed into a unified standardized parameter structure, thereby achieving the expected technical effect of real-time environmental compensation input for the deviation calculation module.

[0101] For example, in an actual photovoltaic power plant edge node, the complete benchmark parameter set dataset includes = , = , = The structure is defined as follows: Field 1 type Field 2 type Field 3 type Total length Byte, using 4-byte alignment, zero padding ensures memory alignment;

[0102] During the packaging process, , and After unit consistency verification, the result is bound to the deviation calculation module interface ID=03 via the interface adapter. The calling convention is pass-by-value, and the transmission delay is less than [specified value]. According to performance testing, this package structure operates at an MCU frequency of [ms]. Under MHz conditions, it can achieve 500 consecutive calls without data errors, which significantly improves the real-time performance and stability of environmental compensation.

[0103] like Figure 3 As shown, step S4 involves calculating the absolute deviations between the measured values ​​of the bracket tilt angle, drive motor current, and backplate temperature based on real-time measured data. Specifically, this includes:

[0104] S4.1: Perform high-precision sampling processing on the original tilt angle signal, drive motor current signal and back sheet temperature signal output by the photovoltaic bracket controller to obtain the real-time bracket tilt angle data sequence, real-time drive motor current data sequence and real-time back sheet temperature data sequence at the current moment, as the basic input data for feature calculation;

[0105] S4.2: Based on the real-time support tilt angle data sequence, the center difference filtering algorithm is applied to perform numerical differentiation processing to eliminate mechanical vibration noise and extract the instantaneous tilt angle change rate feature to obtain the measured value of the tilt angle change rate;

[0106] S4.3: Perform sliding window linear regression algorithm on the real-time drive motor current data sequence to suppress current fluctuation interference and calculate instantaneous change trend, and generate the measured value of drive motor current slope;

[0107] A sliding window linear regression algorithm was used for the real-time drive motor current data sequence (window length: preset number of sampling points). Step interval: This enables local fitting of current trends and instantaneous slope estimation.

[0108] Furthermore, by constructing a two-dimensional dataset of the sampling time series and the current amplitude series within each sliding window, the regression coefficients are calculated using the least squares method to obtain the estimated current slope within that window. The formula is as follows:

[0109]

[0110] in, Let i be the timestamp of the i-th sampling point. This corresponds to the drive motor current value. and These are the average values ​​of the time and current within the window, respectively.

[0111] Furthermore, by performing weighted smoothing on the slope estimates of each sliding window, and using an exponentially weighted moving average algorithm (with the weight parameter α dynamically adjusted according to the intensity of environmental disturbances), the current slope time series curve is denoised and stabilized.

[0112] Furthermore, by detecting the instantaneous peak and valley intervals on the smoothed slope time-series curve, a threshold limitation method is used to remove short-term spikes, ensuring that the generated slope value truly reflects the mechanical load change trend of the drive motor.

[0113] By combining sliding window linear regression and smoothing filtering, the high-frequency fluctuations and low-frequency drifts in the original current signal are effectively separated, generating the measured value of the current slope of the drive motor, thus achieving the expected technical effect of providing a highly reliable feature input for subsequent calculation of absolute deviation.

[0114] For example, in a photovoltaic mounting controller assembled at an edge node, the drive motor current sampling frequency is configured to be 200Hz, and the sliding window length is... Set to 40 points, step interval The slope estimate is set to 20 points (i.e., updated every 0.1 seconds). Within a given sampling period, the average timestamp within the window... For 0.5 seconds, the average current value The initial slope estimate is 3.2A, calculated using the least squares slope formula. A / s, processed by exponentially weighted moving average ( After obtaining the steady-state slope value (=0.3), the steady-state slope value is obtained. A / s. In this scenario, the short-time spike removal module detected and removed two instantaneous peak points. The final measured value of the drive motor current slope was compared with the theoretical benchmark value, which significantly improved the stability and diagnostic accuracy of mechanical jamming trend determination.

[0115] S4.4: Use the real-time backplane temperature data sequence to perform temperature gradient estimation using the first-order forward difference method to compensate for the influence of dynamic changes in ambient temperature and output the measured value of the backplane temperature gradient.

[0116] S4.5: Calculate the differences between the measured value of the tilt angle change rate and the theoretical tilt angle change rate reference value, the measured value of the drive motor current slope and the theoretical drive motor current slope reference value, and the measured value of the backplate temperature gradient and the theoretical backplate temperature gradient reference value, and take the absolute values ​​to generate the absolute deviation of the tilt angle change rate, the absolute deviation of the drive motor current slope, and the absolute deviation of the backplate temperature gradient.

[0117] Step S5: The triple absolute deviation is input into a preset multivariate coupled attenuation function. This function dynamically adjusts the attenuation coefficient based on the environmental state fingerprint to generate a weighted fusion deviation index. The multivariate coupled attenuation function adopts a piecewise power-law attenuation form to suppress the influence of single-variable disturbances. Specifically, it includes:

[0118] S5.1: Based on the three-dimensional mesh parameters stored in the environment-motion response surface, the piecewise power-law attenuation coefficient set of the multivariable coupled attenuation function is calibrated in the offline stage. This coefficient set dynamically maps the attenuation parameters with the environmental state fingerprint as the index, so as to realize the nonlinear attenuation processing of multivariable coupled disturbances of temperature, wind speed and humidity, and output the preset multivariable coupled attenuation function.

[0119] Based on the 3D mesh parameter data stored in the environment-motion response surface, the parameter calibration method (input objects: theoretical tilt angle change rate, drive motor current slope, and back plate temperature gradient under the combination of temperature, wind speed, and humidity for each mesh cell) is used to achieve initial coefficient fitting of the multivariable coupled attenuation function.

[0120] Furthermore, by using a piecewise power law function fitting algorithm (parameter: historical distribution characteristics of triple absolute deviation under various environmental combinations), the power exponent and decay coefficient of each segment interval are calculated, and a set of piecewise coefficients for each grid index is obtained.

[0121] Furthermore, an environment state fingerprint mapping method (parameter: 32-bit fingerprint value) is adopted to achieve a one-to-one correspondence between the segmented power law coefficient set and the environment index, and an index-coefficient mapping table is generated to support fast loading at runtime;

[0122] Furthermore, based on the nonlinear coupled disturbance modeling method (parameters: correlation coefficient matrix of three variables of disturbance characteristics of temperature, wind speed and humidity), the nonlinear response adjustment of the piecewise power law decay function under different environments is realized, ensuring that the exponential response is significantly amplified when multiple variables deviate in the same direction, while the response is suppressed when a single variable is disturbed;

[0123] Using a formulaic approach, the expression for the piecewise power-law decay function is defined through the following MathML formula:

[0124] + +

[0125] in , , It is a piecewise power function, and its value is based on the environmental state fingerprint. Dynamic adjustment;

[0126] By pre-calibrating the multivariable coupled attenuation function, the theoretical parameter relationship of the previous step is transformed into a set of attenuation coefficients that can be directly called at runtime, thereby realizing the environmental adaptive generation of the weighted fusion deviation index.

[0127] For example, in an offline calibration environment with temperatures of 15°C, wind speeds of 5 m / s, and humidity of 50%, the environmental state fingerprint F is encoded as binary "10101100101100101010010110010101". The 3D mesh parameters show the baseline value of the tilt angle change rate. Reference value of drive motor current slope Backplate temperature gradient reference value During the calibration process, piecewise power-law fitting was used to obtain... =1.4、 =1.2、 =1.5. When the measured absolute deviations are respectively , , When the formula calculation steps are as follows: the tilt coefficient term is... ≈0.0636; Current coefficient term is ≈0.1448; the temperature gradient term is The weighted fusion bias index E≈0.2978 is obtained by summing the three factors, with an approximation value of ≈0.0894. This index remains below the environmental dynamic baseline under short-term fluctuations, effectively avoiding misjudgments caused by univariate disturbances and improving the stability and accuracy of diagnosis.

[0128] S5.2: Obtain the absolute deviation of the tilt angle change rate, the absolute deviation of the drive motor current slope, and the absolute deviation of the backplate temperature gradient calculated in step S4 at the current moment as triple absolute deviation input parameters;

[0129] The real-time feature deviation data structure output by step S4 is called in the memory of the edge node controller. The parameter parser module (configured with field mapping table and data type verification rule set) is used to realize field-by-field parsing and numerical validity verification of the absolute deviation of tilt angle change rate, absolute deviation of drive motor current slope and absolute deviation of back plate temperature gradient.

[0130] Furthermore, by using a multi-channel data buffer queue management method (parameters: queue length = 3, timestamp synchronization accuracy ≤ 1ms), time consistency alignment of triple absolute deviation is achieved, and deviation data from different sampling channels are aggregated into a single deviation record according to the current sampling period, ensuring the comparability of input data in the time domain;

[0131] Furthermore, a numerical normalization method is employed (parameters: normalization coefficients are determined based on the same dimension reference range of the environment-motion response surface) to achieve dimensional elimination and unit interval mapping of the triple absolute deviation, generating a normalized absolute deviation vector. ,in , , These are the normalized absolute deviations of the tilt angle change rate, the motor current slope, and the backplate temperature gradient, respectively.

[0132] Furthermore, an outlier removal method is adopted (parameter: the removal threshold is set by the standard deviation multiple of the environmental state fingerprint index) to remove outliers in the deviation vector that may be caused by instantaneous interference, and generate the final triple absolute deviation input parameter set;

[0133] Through the above processing method, the original absolute deviation of the previous step is transformed into structured, normalized and time-aligned three-variable input data, so as to achieve a stable input effect of the multivariable coupled attenuation function and improve the robustness of fault diagnosis.

[0134] For example, at a certain photovoltaic power station, the temperature is ℃, wind speed m / s, humidity is In a %RH operating scenario, the absolute deviation of the tilt angle change rate collected by the edge nodes. ° / s, absolute deviation of motor current slope A / s and absolute deviation of backplate temperature gradient ℃ / min. The resolver normalizes each deviation to the [0,1] interval to obtain the normalized value of the rate of change of inclination angle. Normalized value of motor current slope Normalized value of backplate temperature gradient The data buffer queue aggregates the three data points into a single record according to the same sampling period. The outlier removal module detects no values ​​exceeding the preset standard deviation multiple and retains the complete vector. This vector serves as the input to the S5.4 multivariate coupling attenuation function. After calculation, it generates a weighted fusion deviation index, which significantly improves the early detection capability of lag while maintaining low computational resource consumption.

[0135] S5.3: Obtain the current environment state fingerprint generated by step S2 as the index identifier for dynamic adjustment of the attenuation coefficient;

[0136] S5.4: Input the triple absolute deviation obtained in S5.2 and the environmental state fingerprint obtained in S5.3 into the preset multivariate coupled attenuation function constructed in S5.1, perform piecewise power law attenuation operation, and generate a weighted fusion deviation index to suppress the influence of single variable disturbance on fault diagnosis.

[0137] The absolute deviation of the tilt angle change rate, the absolute deviation of the drive motor current slope, and the absolute deviation of the backplate temperature gradient obtained in S5.2 at the current moment are respectively input into the preset multivariable coupled attenuation function module (parameter index: S5.3 environmental state fingerprint) to realize the matching loading of multidimensional deviation data and environmental adaptive attenuation coefficient;

[0138] Furthermore, by using a lookup table method (index parameter: 32-bit environmental state fingerprint), the offline-calibrated set of piecewise power-law decay coefficients is invoked to retrieve the univariate decay coefficient corresponding to each deviation. Load the deviation into the computation cache to achieve environmental adaptive weight setting of the deviation;

[0139] Furthermore, a piecewise power-law attenuation operator is used to perform nonlinear compression on the triple absolute deviation. The piecewise function is defined as follows: an exponential attenuation segment is used when the deviation value is below the corresponding baseline envelope; a power-law smoothing segment is used when the deviation value is between the baseline and the high threshold segment; and an amplification segment is used when the deviation value exceeds the high threshold segment. The formula is as follows:

[0140]

[0141] in, The weighted fusion bias index. This is the absolute deviation of the rate of change of tilt angle. This is the absolute deviation of the current slope of the drive motor. This represents the absolute deviation of the backplate temperature gradient. These are the attenuation coefficients of each variable mapped from the environmental state fingerprint F. For the exponential coefficients of the piecewise power law;

[0142] Furthermore, by using the threshold discrimination logic of the piecewise function, the dominant effect of large fluctuations of any single variable on the fusion index is suppressed, ensuring that the same deviation of the three variables can trigger the nonlinear leap of the index;

[0143] By using multivariate coupled attenuation calculation, the triple absolute deviation is transformed into a single scalar weighted fusion deviation index, thereby achieving robust quantification of mechanical jamming trends under environmental conditions.

[0144] For example, in the operating environment of a set of support controllers in a distributed photovoltaic power station, the absolute deviation of the tilt angle change rate collected in real time is: Degrees per second, the absolute deviation of the drive motor current slope is Ampere per second, the absolute deviation of the backplate temperature gradient is Temperature per second; environmental state fingerprint F corresponding to The coefficients are respectively , , The power-law exponent p = 2. Substituting into the formula, the calculation process is as follows:

[0145] = + + = Then perform the exponentiation operation. = The weighted fusion bias index was obtained. =0.0379. This index remains low in short-term stable environments, increases significantly when the three variables deviate simultaneously, and achieves sensitive capture of mechanical jamming trends. Moreover, the index does not change significantly under a single environmental disturbance, thus significantly improving the stability and accuracy of alarm judgment.

[0146] S5.5: Output the weighted fusion deviation index generated in S5.4 to the subsequent sliding window monitoring module for judging the mechanical jamming initial screening alarm;

[0147] The input conditions are the temporary data and metadata of the weighted fusion bias index generated in step S5.4;

[0148] A memory writing method (parameters: RAM access address of edge node controller, data length, and write mode) is used to achieve high-speed transfer of the weighted fusion deviation index from the calculation module to the data bus buffer.

[0149] Furthermore, by using an interface mapping method (parameters: sliding window monitoring module input register address, data bus width), the weighted fusion deviation index is aligned with the address of the sliding window monitoring module hardware interface, and a directly accessible data pointer is obtained.

[0150] Furthermore, by using a data encapsulation method (parameters: weighted fusion deviation index numerical precision, timestamp length), the weighted fusion deviation index is bound to the current acquisition time, and a time-stamped weighted fusion deviation index frame structure is generated.

[0151] Furthermore, the integrity of the weighted fusion deviation index frame structure is verified by using a data consistency verification method (parameters: CRC polynomial, initial seed value), and valid data packets that pass the verification are obtained.

[0152] By using data push triggering, the results of the previous step are converted into a weighted fusion deviation index sequence that conforms to the input protocol of the sliding window monitoring module, so as to realize the real-time indicator input for the judgment of mechanical jamming initial screening alarm;

[0153] For example, in the edge controller of a distributed photovoltaic support system, the weighted fusion deviation index output in step S5.4 is 0.452, the timestamp is 1678453200 seconds, and the data length is 8 bytes. This index value is written to RAM address 0x20001000 using a memory write method, with DMA automatic transfer mode and a 32-bit data bus width. In the interface mapping method, the DMA output is mapped to the monitoring module input register address 0x4000A004, achieving zero-copy of the data path. In the data encapsulation method, the index value occupies 4 bytes of floating-point format, the timestamp occupies 4 bytes of unsigned integer, and the frame structure length is 8 bytes. In the data consistency verification method, CRC-16 polynomial is used. The initial seed value is 0xFFFF. The frame structure is verified and passes the verification. Finally, the encapsulated and verified weighted fusion deviation index data packet is pushed to the sliding window monitoring module, enabling the module to simultaneously use the data to perform preliminary screening statistics for mechanical jamming anomalies within a 15-second short window and a 5-minute long window, thereby significantly improving the accuracy of fault identification.

[0154] Step S6: Within a 15-second short sliding window, monitor whether the weighted fusion deviation index continuously exceeds 1.3 times the baseline threshold dynamically determined based on the environmental state fingerprint. Simultaneously, within a 5-minute long sliding window, calculate the percentage of time periods where the weighted fusion deviation index exceeds 1.8 times the baseline threshold. Specifically, this includes:

[0155] S6.1: Based on the current environment state fingerprint, obtain the benchmark threshold corresponding to the environment state fingerprint by looking up the environment-threshold mapping table pre-stored in the read-only storage area of ​​the edge node, and use it as the environment adaptability benchmark input for dynamic threshold calculation;

[0156] S6.2: Perform a scalar multiplication operation on the benchmark threshold to calculate the short window monitoring threshold, which is equal to 1.3 times the benchmark threshold and is used to define the dynamic judgment boundary of short-term abnormal behavior;

[0157] Based on the baseline threshold data output from step S6.1, the scalar multiplication operation method (parameter: coefficient set to 1.3) is used to realize the function of calculating the short window monitoring threshold.

[0158] Furthermore, by using a floating-point scalar multiplication method (parameters: baseline threshold B, coefficient K=1.3), the numerical expansion function of the environmental adaptive short-time decision boundary is realized, and the short-window monitoring threshold value is obtained. ;

[0159] Furthermore, by using a formulaic calculation method, the baseline threshold is multiplied by a constant coefficient to generate the short-window threshold data used for real-time determination, as shown in the following formula:

[0160]

[0161] in, The dynamic baseline threshold given in step S6.1, This is the short-term window anomaly detection coefficient;

[0162] Furthermore, by using a data type conversion method (parameter: floating-point to fixed-point conversion precision 0.001), the data format adaptation of the short window monitoring threshold is achieved, and fixed-point values ​​that can be directly loaded into the deviation index continuity monitoring module are generated.

[0163] By using scalar multiplication and format conversion, the baseline threshold result from the previous step is transformed into a dynamic judgment boundary that can be used within a short-term sliding window, thereby achieving the expected technical effect of rapid identification of transient abnormal states and adaptive adjustment to the environment.

[0164] For example, in a photovoltaic support operation scenario, the baseline threshold B obtained in step S6.1 is 2.457. Using a scalar multiplication coefficient K=1.3 to perform the calculation, we get:

[0165]

[0166] Through calculation, we obtained After floating-point to fixed-point conversion (precision 0.001), the output short-window monitoring threshold is 3.194, which can be directly used by the deviation index continuous over-threshold judgment algorithm in step S6.3. In this scenario, when the real-time weighted fusion deviation index is greater than 3.194 for 15 consecutive seconds, it can be determined that a short-term abnormal continuous state has been established. Combined with long-window statistical conditions, it can achieve the initial screening of mechanical jamming faults and improve the instantaneous fault detection capability in complex environments with a temperature of 41℃, a wind speed of 6m / s, and a humidity of 58%RH.

[0167] S6.3: Within a 15-second short sliding window, continuously monitor the real-time weighted fusion deviation index sequence, determine whether the weighted fusion deviation index continuously exceeds the short window monitoring threshold, and generate a short window monitoring result signal, wherein the short window monitoring result signal represents the instantaneous abnormal continuous state;

[0168] S6.4: Perform a scalar multiplication operation on the baseline threshold to calculate the long window statistical threshold, which is equal to 1.8 times the baseline threshold and is used to define the statistical analysis threshold for the proportion of long-term anomalies.

[0169] The baseline threshold parameter obtained from step S6.1 is processed by scalar multiplication. A fixed-point integer multiplication algorithm (parameters: multiplication coefficient 1.8, bit width 32 bits) is used to expand the baseline threshold to the proportional coefficient required for long-term anomaly statistical judgment.

[0170] Furthermore, the threshold is calculated using a mathematical multiplication formula, as follows:

[0171]

[0172] in, For long window statistics threshold, The baseline threshold;

[0173] Furthermore, an overflow detection method is employed (parameter: overflow limit). This enables numerical security verification of the multiplication process and yields long-window statistical threshold calculation results that conform to the integer operation range constraints of edge nodes.

[0174] Furthermore, by using fixed-point formatting (parameters: scaling factor 1000, rounding mode to zero), the long window statistical threshold is converted into an integer storage format to reduce storage space usage and improve query efficiency.

[0175] Through the above calculations and formatting, the baseline threshold is transformed into a long-window statistical threshold that can be directly used to determine the proportion of long-term anomalies, thereby achieving environmental adaptive quantitative control for anomaly persistence analysis.

[0176] For example, in a photovoltaic support system operation scenario, the edge node retrieves the current environmental state benchmark threshold via S6.1. The long window statistical threshold is generated using multiplication. The calculation formula is as follows: After calculation, we obtained An overflow detection method is used to confirm that the result is within the calculation range, and then a scaling factor is applied. Perform fixed-point conversion to generate integer data. The data is written into the long-window judgment module. Within the subsequent 5-minute statistical period, the system uses this threshold to perform anomaly analysis on the weighted fusion deviation index, under varying ambient temperatures. ℃, wind speed m / s, humidity Under certain conditions, it effectively identified a significant increase in the continuous entrapment behavior of the stent, achieving the expected technical effect of dynamically adapting the long window threshold to complex environments;

[0177] S6.5: Within a 5-minute long sliding window, the percentage of time periods where the weighted fusion deviation index exceeds the long window statistical threshold is calculated to generate long window statistical results. These long window statistical results characterize the relative proportion of abnormal durations, providing a temporal distribution basis for initial fault screening.

[0178] Step S7: Determine whether the monitoring results of the short sliding window and the statistical results of the long sliding window simultaneously meet preset conditions. If they do, trigger a mechanical jamming initial screening alarm signal. Specifically, this includes:

[0179] S7.1: Obtain the short sliding window monitoring result data stream, which contains a Boolean state sequence in which the weighted fusion deviation index continuously exceeds 1.3 times the dynamic benchmark threshold within a 15-second window. Perform continuity verification processing based on bitmask logic operation to generate a short window condition satisfaction flag signal.

[0180] S7.2: Obtain the statistical result data stream of the long sliding window. This data stream contains the percentage of time periods in which the weighted fusion deviation index exceeds 1.8 times the dynamic benchmark threshold within a 5-minute window. Perform the percentage compliance judgment processing based on the preset threshold comparator to generate a long window condition satisfaction flag signal.

[0181] When executing this sub-step, the long sliding window statistical results output by S6.5 are used as the input data object. The result is the percentage of time periods in which the weighted fusion deviation index within a 5-minute window exceeds the long window statistical threshold dynamically determined by the environmental state fingerprint.

[0182] The long window data stream parsing method (parameters: statistical result data stream, timestamp sequence) is adopted to separate the percentage value of each time period from the corresponding time series, so as to ensure the time consistency of the percentage determination process.

[0183] Furthermore, by using a threshold comparison method (parameters: long window statistical threshold, determination percentage threshold), the time period percentage value is compared with the preset percentage determination threshold, and a logical Boolean determination result is generated to ensure the absolute accuracy of the percentage determination.

[0184] Furthermore, an error compensation calculation method (parameter: sampling error compensation coefficient) is adopted to correct the deviation caused by the sampling period in the statistical results of the long sliding window, and to obtain the calculated value of the time period proportion after compensation, so as to eliminate the systematic error in the statistical process;

[0185] Furthermore, through conditional filtering operations (parameters: compensation ratio value, threshold comparison result), the compensation ratio data is judged to meet the standard, and a preliminary state of long window condition satisfaction flag signal is generated;

[0186] By encapsulating and formatting data, the comparison result of the previous step is transformed into a standardized Boolean flag signal, thereby enabling the output of the long window condition satisfaction flag signal and providing stable input conditions for S7.3 logic AND operation;

[0187] For example, in a single-support node of a distributed photovoltaic power station, the statistical result of the long sliding window is a proportion of 0.42. The statistical threshold of the long window is obtained by multiplying the baseline threshold of 4.5 corresponding to the environmental state fingerprint by a multiplication factor of 1.8. The threshold for determining the proportion is set to 0.4 based on experience. A threshold comparison algorithm is used to compare 0.42 with 0.4; if the result is greater than the threshold, a Boolean judgment value of True is obtained. The sampling error compensation coefficient is set to... After compensation calculation, the corrected percentage is... = If the value is still higher than the 0.4 judgment threshold, the conditional filtering operation outputs a "True" judgment. After data encapsulation processing, a long window condition satisfaction flag signal is generated. The signal status is high level, ensuring that it will work with the short window condition satisfaction flag signal to trigger the logic and judgment of the mechanical jamming initial screening alarm in subsequent S7.3.

[0188] S7.3: Perform a logical AND gate operation on the short window condition fulfillment flag signal and the long window condition fulfillment flag signal to generate a dual condition co-fulfillment flag signal, which serves as the necessary triggering basis for the mechanical jamming initial screening alarm;

[0189] S7.4: Based on the dual-condition collaborative satisfaction of the flag signal, the alarm signal generation process is performed. A high-level warning pulse with a duration of 200 milliseconds is generated through the pulse width modulation circuit to form a preliminary screening alarm signal that conforms to the industrial bus protocol.

[0190] S7.5: Perform hardware watchdog verification on the initial screening alarm signal. Use the watchdog timer to monitor whether the signal duration meets the preset pulse width tolerance range, so as to output the mechanical jamming initial screening alarm signal verified by timing compliance to the causality verification module.

[0191] Step S8: A lightweight causal verification module in the form of a pre-set symbolic rule tree is invoked. Based on the temporal correlation characteristics of the support target tilt angle achievement state, drive motor current fluctuation pattern, and weighted fusion deviation index, a binary discrimination between jamming faults and environmentally induced pseudo-anomalies is performed, and the final mechanical jamming warning result is output. Specifically, this includes:

[0192] S8.1: Based on the offline stage kinematic model and fault feature analysis of the support, a causal criterion library in the form of a symbolic rule tree is generated; specifically, based on the support structural parameters, material thermal expansion coefficient and wind load calculation model, the difference features between jamming fault and environmental pseudo-anomaly are extracted through digital twin simulation, and multi-level criterion rules are defined, including physically interpretable conditions such as the continuous increase of the weighted fusion deviation index, the failure to achieve the target tilt angle of the support, and the sawtooth saturation oscillation of the drive motor current. The rule tree structure is solidified and stored in the read-only storage area of ​​the edge node to form the criterion library output of the lightweight causal verification module.

[0193] S8.2: Based on the solidified causal criterion library built in the previous step, the time-series data stream of the support target tilt angle achievement status signal, drive motor current fluctuation morphology feature sequence and weighted fusion deviation index collected by the edge nodes is acquired in real time and used as the feature input data output of the causal verification module.

[0194] S8.3: Based on the solidified causal criterion library and feature input data, the target tilt angle achievement status of the support, the current fluctuation morphology of the drive motor, and the time-series correlation features of the weighted fusion deviation index are input into the symbol rule tree inference engine, and a top-down traversal is performed; specifically, for each rule node, the measured features are compared with the preset threshold conditions, the inference path is determined according to the branch logic, and the preliminary discrimination result is output.

[0195] Based on a solidified causal criterion library and feature input data, a symbolic rule tree inference engine (traversal strategy: top-down depth-first, node determination method: condition matching and branch jump) is used to achieve joint inference of the support target tilt angle achievement status signal, drive motor current fluctuation morphology characteristics and weighted fusion deviation index time-series correlation characteristics.

[0196] Furthermore, by using a node condition matching method (parameters: preset threshold set, feature type index), a quantitative comparison is made between the single physical feature associated with the current node and the preset threshold, and a matching Boolean result data stream is generated.

[0197] Furthermore, through the branch logic judgment module (parameters: rule tree branch table, boolean result data stream), the set of child nodes that meet the conditions can be quickly filtered, and the next traversal of the inference engine can be carried out along the matching path based on the matching results;

[0198] Furthermore, a time-series correlation verification method (parameters: deviation index sequence window length, target tilt angle state change mode, drive motor current morphology mode) is adopted to realize the logical consistency verification of the current node conditions and the predecessor node conditions in the time dimension, and output the time coupling verification flag value.

[0199] Furthermore, the node matching sequence obtained by top-down traversal is transformed into a preliminary discrimination result by using the fault feature pattern aggregation function (parameters: matching path length, time coupling flag value, deviation exponential volatility), and the judgment basis and weight of the result are indicated.

[0200] By using a symbolic rule tree inference engine, the feature input from the previous step is mapped to the causal criterion library as structured preliminary discrimination data, enabling rapid differentiation between stuck faults and environmentally induced pseudo-anomalies.

[0201] For example, in a causal verification module deployed at the edge node of a distributed photovoltaic power station, the sampling period for the target tilt angle achievement status signal of the support is set to 200ms, the length of the characteristic sequence of the current fluctuation pattern of the drive motor is set to 256 points, and the time-series sampling period of the weighted fusion deviation index is set to 100ms. The condition of the first-level node of the rule tree is set as: the mean of the weighted fusion deviation index. If the price continues to rise within the short window, the judgment formula is:

[0202]

[0203] in, This is a sample of the weighted fusion bias index within the current short window. This represents the number of samples.

[0204] The condition for the second-level node is set as follows: the target tilt angle has not been achieved, and the judgment formula is:

[0205]

[0206] in, Angle of inclination for the target This is the current tilt angle.

[0207] The third-level node condition is set as a sawtooth-shaped saturated oscillation in the drive motor current. The judgment criterion is that the sign of the first-order difference of the current alternates frequently and the amplitude is close to the rated value. During the traversal, if the matching result of each node condition is true and the time coupling verification flag is valid, the preliminary judgment result is output as a jamming fault; if any key node condition is not met, the preliminary judgment output is an environmentally induced pseudo-anomaly. In this embodiment, the actual operation results show that the module can complete the judgment within a calculation time of no more than 2ms under the above parameters, and form an immediate response to mechanical jamming faults, effectively avoiding misjudgments caused by environmental factors.

[0208] S8.4: Based on the preliminary judgment result, generate a binary judgment result for stuck fault or environmentally induced pseudo-anomaly; specifically, if the preliminary judgment is a stuck fault, it is marked as a valid alarm, otherwise it is marked as an environmentally induced pseudo-anomaly, and a binary judgment result is output.

[0209] S8.5: Based on the binary discrimination result, if the fault is determined to be mechanical jamming, the final mechanical jamming warning signal is output to the system alarm interface; otherwise, the initial screening alarm is cleared and the cause of false alarm is recorded, and the final mechanical jamming warning result is output.

[0210] The present invention also provides a fault diagnosis system for a distributed solar photovoltaic support system, which uses the above-mentioned fault diagnosis method for a distributed solar photovoltaic support system to diagnose faults in the solar photovoltaic support system.

[0211] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0212] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A fault diagnosis method for a distributed solar photovoltaic support system, characterized in that, Includes the following steps: S1: Based on the structural parameters of the support structure, the thermal expansion coefficient of the material, and the wind load calculation model, the environment-motion response surface is generated through digital twin simulation in the offline stage; S2: Real-time acquisition of current environmental variables of photovoltaic support, normalization of the current environmental variables and generation of environmental state fingerprint through hash mapping; S3: Based on the environmental state fingerprint, retrieve the corresponding reference parameter group in the environment-motion response surface, and obtain the theoretical tilt angle change rate reference value, drive motor current slope reference value, and backplate temperature gradient reference value under the current environmental state. S4: Based on real-time measured bracket tilt angle, drive motor current, and backplate temperature data, calculate the absolute deviation between the measured tilt angle change rate and the theoretical tilt angle change rate reference value, the absolute deviation between the measured drive motor current slope and the drive motor current slope reference value, and the absolute deviation between the measured backplate temperature gradient and the backplate temperature gradient reference value. S5: Input the triple absolute deviation into a preset multivariable coupling attenuation function. The multivariable coupling attenuation function dynamically adjusts the attenuation coefficient according to the environmental state fingerprint to generate a weighted fusion deviation index. S6: Monitor whether the weighted fusion deviation index continuously exceeds the short window monitoring threshold within a short sliding window of a first preset length, and simultaneously count the percentage of time periods during which the weighted fusion deviation index exceeds the long window statistical threshold within a long sliding window of a second preset length, to obtain the short sliding window monitoring result and the long sliding window statistical result. S7: If the monitoring result of the short sliding window and the statistical result of the long sliding window simultaneously meet the preset conditions, a mechanical jamming initial screening alarm signal is triggered. S8: Call the preset symbol rule tree form of causal verification module, and perform binary discrimination of jamming fault and environmental induced pseudo-anomaly based on the time correlation characteristics of the target tilt angle achievement state of the support, the current fluctuation mode of the drive motor and the weighted fusion deviation index, and output the final mechanical jamming warning result.

2. The fault diagnosis method for a distributed solar photovoltaic support system according to claim 1, characterized in that, The wind load calculation model is constructed as follows: Based on the structural parameters of the support structure, geometric feature extraction processing is performed to obtain the geometric features of the support, including the support arm length and the friction coefficient of the rotating shaft; thermal deformation analysis processing is performed based on the thermal expansion coefficient of the material to generate material thermal deformation compensation parameters; the geometric features of the support structure and the material thermal deformation compensation parameters are integrated to perform dynamic wind load distribution calculation to generate a wind load calculation model.

3. The fault diagnosis method for a distributed solar photovoltaic support system according to claim 1, characterized in that, The environment-motion response surface is stored in the edge node read-only storage area in the form of a three-dimensional mesh. Each mesh point corresponds to the time-series envelope boundary of the theoretical tilt angle change rate under the combination of temperature, wind speed, and humidity, the time-series envelope boundary of the drive motor current slope, and the time-series envelope boundary of the backplate temperature gradient.

4. The fault diagnosis method for a distributed solar photovoltaic support system according to claim 1, characterized in that, Step S3 specifically includes: The environmental state fingerprint generated in step S2 is subjected to grid index parsing processing. Based on the high 8 bits, middle 12 bits and low 12 bits of the 32-bit binary environmental state fingerprint, a bit mask segmentation operation is performed to generate three-dimensional grid coordinate index values. Based on the three-dimensional grid coordinate index value, perform grid coordinate mapping processing, and use the pre-calibrated temperature range quantization step size, wind speed range quantization step size and humidity range quantization step size to convert the discretized query position into three-dimensional grid coordinates of the environment-motion response surface. Based on the three-dimensional mesh coordinates, the storage address positioning process is performed. Based on the mesh data storage structure of the environment-motion response surface and the starting address offset of the read-only storage area, the linearized storage address of the three-dimensional mesh coordinates is calculated to obtain the reference parameter group storage position pointer of the corresponding mesh point in the environment-motion response surface. The reference parameter set storage location pointer is processed by performing parameter reading. Based on the memory access protocol of the edge node controller, the theoretical tilt angle change rate reference value, the drive motor current slope reference value, and the backplane temperature gradient reference value are extracted from the read-only storage area to form a complete reference parameter set dataset under the current environmental conditions.

5. The fault diagnosis method for a distributed solar photovoltaic support system according to claim 4, characterized in that, Step S3 further includes: The complete set of reference parameters is encapsulated. Based on the preset parameter transfer interface specification, the theoretical tilt angle change rate reference value, the drive motor current slope reference value, and the backplate temperature gradient reference value are encapsulated into a standardized parameter structure.

6. The fault diagnosis method for a distributed solar photovoltaic support system according to claim 1, characterized in that, Step S4 specifically includes: The original tilt angle signal, drive motor current signal and back sheet temperature signal output by the photovoltaic bracket controller are sampled and processed with high precision to obtain the real-time bracket tilt angle data sequence, real-time drive motor current data sequence and real-time back sheet temperature data sequence at the current moment. Numerical differentiation processing is performed using the central difference filtering algorithm based on the real-time support tilt angle data sequence to obtain the measured value of the tilt angle change rate. The real-time drive motor current data sequence is processed by a sliding window linear regression algorithm to obtain the measured value of the drive motor current slope. The temperature gradient is estimated using the real-time backplane temperature data sequence through a first-order forward difference method to obtain the measured value of the backplane temperature gradient. The measured values ​​of the tilt angle change rate and the theoretical tilt angle change rate reference value, the measured values ​​of the drive motor current slope and the drive motor current slope reference value, and the measured values ​​of the back plate temperature gradient and the back plate temperature gradient reference value are respectively processed by difference calculation, and the absolute values ​​are taken to generate the absolute deviation of the tilt angle change rate, the absolute deviation of the drive motor current slope, and the absolute deviation of the back plate temperature gradient.

7. The fault diagnosis method for a distributed solar photovoltaic support system according to claim 1, characterized in that, Step S5 specifically includes: Based on the three-dimensional mesh parameters stored in the environment-motion response surface, the piecewise power-law attenuation coefficient set of the multivariable coupled attenuation function is calibrated in the offline stage. The piecewise power-law attenuation coefficient set dynamically maps the attenuation parameters with the environmental state fingerprint as the index, performs nonlinear attenuation processing on the multivariable coupled disturbances of temperature, wind speed, and humidity, and outputs the preset multivariable coupled attenuation function. The absolute deviation of the tilt angle change rate, the absolute deviation of the drive motor current slope, and the absolute deviation of the backplate temperature gradient, calculated in step S4, are used as triple absolute deviation input parameters. Obtain the current environment state fingerprint generated in step S2, and use it as an index identifier for dynamic adjustment of the attenuation coefficient; The triple absolute deviation input parameters and the environmental state fingerprint are input into the preset multivariate coupled attenuation function, and a piecewise power-law attenuation operation is performed to generate a weighted fusion deviation index.

8. The method for fault diagnosis of a distributed solar photovoltaic support system according to claim 1, characterized in that, The first preset length is 15 seconds, the second preset length is 5 minutes, the short window monitoring threshold is equal to 1.3 times the baseline threshold, the long window statistical threshold is equal to 1.8 times the baseline threshold, and the baseline threshold is obtained by looking up the environment-threshold mapping table pre-stored in the read-only storage area of ​​the edge node based on the current environment state fingerprint.

9. A fault diagnosis system for a distributed solar photovoltaic support system, characterized in that: The fault diagnosis method for distributed solar photovoltaic support system according to any one of claims 1-8 is used to diagnose faults in the solar photovoltaic support system.

Images