Method and system for evaluating health condition of new energy equipment

Abstract

The invention discloses a health condition evaluation method and system for new energy equipment, and the main points of the technical scheme are that the method comprises the following steps: multi-source data collection and preprocessing: deploying a sensor array to collect equipment operation data, fault feature data, environment data and operation and maintenance historical data, employing an EMD-SVD algorithm to carry out the denoising of a vibration signal, and carrying out the preprocessing of the data; processing the multi-dimensional data through Z-score standardization; constructing a dynamic weight calculation model; health state grading evaluation; and health trend prediction and active early warning: constructing an LSTM-Attention prediction model, inputting a historical CHI sequence, operation data and environmental parameters, and outputting health state prediction values in future 1-3 months. A complete closed loop is formed from data acquisition, evaluation and prediction to operation and maintenance decision, full-life-cycle health archive management of equipment is supported, and data support is provided for equipment type selection optimization and life cycle cost analysis.

Description

Technical Field

[0001] The present invention relates to the field of new energy equipment operation and maintenance technology, and specifically to a new energy equipment health status assessment method and system based on multi-dimensional data fusion and dynamic weight algorithm, which is suitable for the full life cycle health management of new energy core equipment such as wind turbines, photovoltaic inverters, and energy storage battery packs. Background Art

[0002] With the advancement of the "dual carbon" goals, my country's installed capacity of renewable energy has exceeded 1.3 billion kilowatts, accounting for over 50% of the total installed capacity. However, new energy equipment operates in complex environments (such as high altitude, salt spray, and low temperatures), and its failure rate is 30%-50% higher than that of traditional power generation equipment. Traditional equipment condition monitoring systems have the following technical bottlenecks:

[0003] Limitations of single-parameter monitoring: Existing systems are mostly based on single-physical quantity threshold alarms, such as vibration and temperature. For example, the IEC61400-11 standard only stipulates an early warning when the effective value of the vibration velocity exceeds 7.1 mm / s. However, actual equipment failures often manifest as multi-parameter coupling anomalies (such as an increase in gearbox oil temperature accompanied by a simultaneous increase in vibration amplitude and bearing temperature). The single-parameter monitoring missed alarm rate is as high as 40%.

[0004] Static weighting assessment flaws: Traditional analytic hierarchy process (AHP) methods rely on subjective expert experience to determine fixed weights for evaluation indicators, making them incapable of adapting to changing equipment operating conditions. For example, when operating a photovoltaic inverter under high-temperature and high-irradiation conditions, the weight of the cooling system should be significantly higher than when operating at room temperature. This static weighting results in a 25%-30% decrease in assessment accuracy.

[0005] Inadequate health trend prediction: Existing prediction models often use linear regression or simple neural networks, which have poor adaptability to the nonlinear and time-varying characteristics of equipment degradation. For example, for lithium-ion batteries, traditional models have a root mean square error (RMSE) of over 15% in cycle life prediction, failing to meet the operational and maintenance requirements of accurate predictions three months in advance.

[0006] Inefficient multi-source data fusion: Equipment operating data, historical fault data, environmental data, and operation and maintenance data were not deeply integrated, resulting in data utilization rates of less than 30%. For example, wind turbine fault diagnosis failed to incorporate abnormal wind speed-power curve data, resulting in a 2-3 cycle delay in early fault identification.

[0007] To address these issues, existing technologies have attempted to introduce data fusion algorithms and intelligent models. For example, CN112363215A discloses a device status assessment method based on DS evidence theory, but does not address the issue of dynamic weight adjustment. CN113156243B proposes a device health management system integrated with the Internet of Things (IoT), but lacks a description of a real-time update mechanism for the prediction model. Therefore, there is an urgent need for an assessment method and system that can dynamically integrate multi-dimensional data, adaptively adjust weights, and accurately predict health trends. Summary of the Invention

[0008] The purpose of the present invention is to provide a method and system for evaluating the health status of new energy equipment to solve the problems raised in the above background technology.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A method for evaluating the health status of new energy equipment, comprising the following steps:

[0011] Multi-source data acquisition and preprocessing: Deploy a sensor array to collect equipment operation data, fault characteristic data, environmental data, and operation and maintenance history data. Use the EMD-SVD algorithm to denoise vibration signals, and perform Z-score standardization to process multi-dimensional data.

[0012] Construction of dynamic weight calculation model: A four-level evaluation index system is established, and the AHP-entropy weight method is used to calculate the dynamic weight. The subjective weight and the objective weight are integrated through the real-time working condition adjustment factor α. The calculation formula is W = αW + (1-α)W, where W is the dynamic fusion subjective weight; W is the dynamic fusion objective weight; α is calculated in real time based on the equipment load rate and ambient temperature.

[0013] Health status grading assessment: A health status assessment cloud generator is constructed using cloud model theory. The cloud membership of each indicator is calculated, and the equipment comprehensive health index (CHI) is calculated using the weighted average method. The formula is CHI = \sum_{i=1}^nW_i\timesC_i, where W_i is the dynamic weight, representing the weight coefficient of the i-th evaluation indicator in the comprehensive assessment, reflecting the relative importance of the indicator to the equipment health status. C_i is the cloud membership value, indicating the fuzzy membership degree of the i-th evaluation indicator to a certain health status level. The value range is [0,1]. The closer the value is to 1, the higher the probability that the indicator belongs to the corresponding health level.

[0014] Health trend prediction and proactive warning: Build an LSTM-Attention prediction model, input historical CHI sequences, operating data, and environmental parameters, and output health status predictions for the next 1-3 months. Set up a three-level warning mechanism to trigger different levels of warnings and operation and maintenance strategies based on the prediction results.

[0015] Preferably, a four-level evaluation index system is used for evaluation, and the four-level evaluation index system includes:

[0016] First-level indicators: equipment operating parameters, fault characteristic parameters, environmental impact parameters, and operation and maintenance history parameters;

[0017] Secondary indicators: electrical parameters, mechanical parameters, vibration characteristics, thermal performance parameters;

[0018] Level 3 indicators: voltage harmonic distortion rate, gearbox vibration RMS value, bearing fault frequency amplitude, and heat dissipation efficiency.

[0019] Preferably, the multi-type sensors include:

[0020] Vibration acceleration sensors are deployed in vibration-sensitive areas of equipment bearing seats and gearbox housings;

[0021] Infrared thermal imager, used to obtain the surface temperature distribution of the equipment;

[0022] Rogowski coil current sensor is used to collect high-frequency current waveform data.

[0023] Preferably, the EMD-SVD denoising algorithm specifically includes:

[0024] Step 1: Perform empirical mode decomposition (EMD) on the original vibration signal to obtain 10-15 intrinsic mode functions (IMFs);

[0025] Step 2: Perform singular value decomposition (SVD) on each IMF component, retaining the singular values ​​with the first 30% energy share to reconstruct the signal;

[0026] Step 3: Filter the effective IMF components through the kurtosis criterion and reconstruct the denoised vibration signal.

[0027] Preferably, in the cloud model theory:

[0028] The health state cloud generator parameters are: expectation E_c = 0.8\pm0.1, entropy En = 0.1\pm0.05, excess entropy He = 0.05\pm0.02;

[0029] The sub-health cloud generator parameters are: expectation E_c=0.65\pm0.15, entropy En=0.15\pm0.08, excess entropy He=0.08\pm0.03.

[0030] Preferably, when calculating the comprehensive health index CHI, a weighting coefficient \beta=1.2-1.5 is set for the fault characteristic parameter, and a weighting coefficient \gamma=0.8-1.0 is set for the environmental impact parameter. The calculation formula is: CHI=\sum_{i=1}^n(\beta_iW_i)\timesC_i; wherein \beta_i is dynamically adjusted according to the indicator category.

[0031] The present invention also discloses a new energy equipment health status assessment system for implementing the above method, comprising:

[0032] Edge acquisition layer: Contains a high-precision sensor array and edge computing units to achieve real-time data acquisition, local processing and feature extraction, and supports Modbus / TCP and MQTT protocol data upload;

[0033] Platform management layer: includes the data center and algorithm engine. The data center builds a digital twin model of the device and integrates full lifecycle data. The algorithm engine deploys dynamic weight calculation, health assessment, and trend prediction modules, and supports online tuning of algorithm parameters.

[0034] Application service layer: includes health status visualization interface, intelligent early warning module, operation and maintenance decision support system, realizes assessment result visualization, multi-channel early warning and differentiated operation and maintenance strategy generation, and connects to the enterprise ERP system to realize operation and maintenance process automation.

[0035] Preferably, the edge computing unit ECU is configured as follows:

[0036] The processor is ARM Cortex-A72 dual-core or above, with a main frequency of ≥1.5GHz;

[0037] Integrated 4GB DDR4 memory and 32GB eMMC storage, supporting local data caching for more than 72 hours;

[0038] It has multiple communication interfaces such as RS485, CAN, Ethernet, and supports ModbusRTU and ProfibusDP protocols.

[0039] Preferably, the digital twin model constructed by the data center platform includes:

[0040] Equipment geometric model, built based on 3D laser scanning data;

[0041] Physical simulation model, integrating multi-body dynamics and heat transfer equations, supports performance simulation under real-time working conditions;

[0042] The fault knowledge base contains more than 1,000 historical fault cases and corresponding characteristic parameter thresholds.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention can improve evaluation accuracy through multi-dimensional integration: by integrating four-dimensional data of equipment operation, fault characteristics, environment, and operation and maintenance, 18 core evaluation indicators are constructed (such as the envelope spectrum energy of the gearbox vibration signal, the SOC-SOH coupling coefficient of the battery pack, etc.), which improves the evaluation accuracy by more than 40% compared with traditional single-dimensional evaluation.

[0045] The dynamic weights of the present invention adapt to complex working conditions: the improved AHP-entropy weight method combined with the real-time working condition adjustment factor reduces the weight calculation error from 20% of the traditional method to below 8%, and the evaluation reliability is significantly improved, especially in extreme environments (such as low-temperature starting of wind turbines and overload operation of photovoltaic inverters).

[0046] The present invention can accurately predict and realize proactive operation and maintenance: the LSTM-Attention model reduces the RMSE to 8% in battery remaining life prediction, and the lead time for wind turbine gearbox failure prediction reaches 45 days, which is a 60% improvement in predictive ability compared with traditional models, effectively reducing unplanned downtime by more than 30%.

[0047] This invention can achieve a closed-loop management of the entire life cycle: forming a complete closed loop from data collection, evaluation and prediction to operation and maintenance decision-making, supporting the health record management of the equipment throughout its life cycle, providing data support for equipment selection optimization and life cycle cost analysis, and is expected to reduce operation and maintenance costs by 25%-30%. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flowchart of the present invention; DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] Example 1

[0051] See also Figure 1 The present invention provides a method for evaluating the health status of new energy equipment, comprising the following steps:

[0052] Multi-source data acquisition and preprocessing: Deploy a sensor array to collect equipment operation data, fault characteristic data, environmental data, and operation and maintenance history data. Use the EMD-SVD algorithm to denoise vibration signals, and perform Z-score standardization to process multi-dimensional data.

[0053] Construction of dynamic weight calculation model: A four-level evaluation index system is established, and the AHP-entropy weight method is used to calculate the dynamic weight. The subjective weight and the objective weight are integrated through the real-time working condition adjustment factor α. The calculation formula is W = αW + (1-α)W, where W is the dynamic fusion subjective weight; W is the dynamic fusion objective weight; α is calculated in real time based on the equipment load rate and ambient temperature.

[0054] Health status grading assessment: A health status assessment cloud generator is constructed using cloud model theory. The cloud membership of each indicator is calculated, and the equipment comprehensive health index (CHI) is calculated using the weighted average method. The formula is CHI = \sum_{i=1}^nW_i\timesC_i, where W_i is the dynamic weight, representing the weight coefficient of the i-th evaluation indicator in the comprehensive assessment, reflecting the relative importance of the indicator to the equipment health status. C_i is the cloud membership value, indicating the fuzzy membership degree of the i-th evaluation indicator to a certain health status level. The value range is [0,1]. The closer the value is to 1, the higher the probability that the indicator belongs to the corresponding health level.

[0055] Health trend prediction and proactive warning: Build an LSTM-Attention prediction model, input historical CHI sequences, operating data, and environmental parameters, and output health status predictions for the next 1-3 months. Set up a three-level warning mechanism to trigger different levels of warnings and operation and maintenance strategies based on the prediction results.

[0056] The evaluation is conducted using a four-level evaluation index system, which includes:

[0057] First-level indicators: equipment operating parameters, fault characteristic parameters, environmental impact parameters, and operation and maintenance history parameters;

[0058] Secondary indicators: electrical parameters, mechanical parameters, vibration characteristics, thermal performance parameters;

[0059] Level 3 indicators: voltage harmonic distortion rate, gearbox vibration RMS value, bearing fault frequency amplitude, and heat dissipation efficiency.

[0060] The multi-type sensors include:

[0061] Vibration acceleration sensors are deployed in vibration-sensitive areas of equipment bearing seats and gearbox housings;

[0062] Infrared thermal imager, used to obtain the surface temperature distribution of the equipment;

[0063] Rogowski coil current sensor is used to collect high-frequency current waveform data.

[0064] The EMD-SVD denoising algorithm specifically includes:

[0065] Step 1: Perform empirical mode decomposition (EMD) on the original vibration signal to obtain 10-15 intrinsic mode functions (IMFs);

[0066] Step 2: Perform singular value decomposition (SVD) on each IMF component, retaining the singular values ​​with the first 30% energy share to reconstruct the signal;

[0067] Step 3: Filter the effective IMF components through the kurtosis criterion and reconstruct the denoised vibration signal.

[0068] Among them, in the cloud model theory:

[0069] The health state cloud generator parameters are: expectation E_c = 0.8\pm0.1, entropy En = 0.1\pm0.05, excess entropy He = 0.05\pm0.02;

[0070] The sub-health cloud generator parameters are: expectation E_c=0.65\pm0.15, entropy En=0.15\pm0.08, excess entropy He=0.08\pm0.03.

[0071] When calculating the comprehensive health index CHI, a weighting coefficient of \beta=1.2-1.5 is set for the fault characteristic parameter, and a weighting coefficient of \gamma=0.8-1.0 is set for the environmental impact parameter. The calculation formula is: CHI=\sum_{i=1}^n(\beta_iW_i)\timesC_i; where \beta_i is dynamically adjusted according to the indicator category.

[0072] This embodiment also discloses a new energy equipment health status assessment system for implementing the above method, including:

[0073] Edge acquisition layer: Contains a high-precision sensor array and edge computing units to achieve real-time data acquisition, local processing and feature extraction, and supports Modbus / TCP and MQTT protocol data upload;

[0074] Platform management layer: includes the data center and algorithm engine. The data center builds a digital twin model of the device and integrates full lifecycle data. The algorithm engine deploys dynamic weight calculation, health assessment, and trend prediction modules, and supports online tuning of algorithm parameters.

[0075] Application service layer: includes health status visualization interface, intelligent early warning module, operation and maintenance decision support system, realizes assessment result visualization, multi-channel early warning and differentiated operation and maintenance strategy generation, and connects to the enterprise ERP system to realize operation and maintenance process automation.

[0076] The edge computing unit ECU is configured as follows:

[0077] The processor is ARM Cortex-A72 dual-core or above, with a main frequency of ≥1.5GHz;

[0078] Integrated 4GB DDR4 memory and 32GB eMMC storage, supporting local data caching for more than 72 hours;

[0079] It has multiple communication interfaces such as RS485, CAN, Ethernet, and supports ModbusRTU and Profib usDP protocols.

[0080] Among them, the digital twin model constructed by the data center includes:

[0081] Equipment geometric model, built based on 3D laser scanning data;

[0082] Physical simulation model, integrating multi-body dynamics and heat transfer equations, supports performance simulation under real-time working conditions;

[0083] The fault knowledge base contains more than 1,000 historical fault cases and corresponding characteristic parameter thresholds.

[0084] Among them, the edge acquisition layer is composed of a high-precision sensor array (vibration, temperature, electrical parameters, etc.) and an edge computing unit (ECU), which realizes real-time data acquisition, local denoising processing and preliminary feature extraction (such as the RMS value and kurtosis value of the vibration signal, the hot spot temperature difference of the temperature field, etc.), and uploads the data to the platform layer through Modbus / TCP and MQTT protocols.

[0085] The platform management layer includes the data center and algorithm engine:

[0086] Data center: Builds digital twin models of new energy equipment, integrates data from the entire equipment life cycle (design parameters, operating data, fault records, operation and maintenance logs, etc.), and supports data cleaning, storage (time series database InfluxDB + relational database MySQL) and visual modeling;

[0087] Algorithm engine: Deploys dynamic weight calculation module, health assessment module, and trend prediction module, supports online tuning of algorithm parameters (such as the number of LSTM network layers, attention mechanism weight matrix update cycle, etc.), and provides API interface for application layer calls.

[0088] The application service layer includes a health status visualization interface, an intelligent early warning module, and an operation and maintenance decision support system:

[0089] Visual interface: real-time display of equipment 3D model, health index curve, and dynamic change trend of each indicator weight;

[0090] Intelligent early warning: Early warning information is released through multiple channels such as SMS, email, and platform pop-up windows, along with possible causes of failure and handling suggestions;

[0091] Operation and maintenance decision-making: Generate differentiated operation and maintenance strategies based on health assessment results (such as extended inspection cycles for healthy equipment, customized maintenance plans for sub-healthy equipment, and quick replacement work orders for faulty equipment), and connect to the enterprise ERP system to automate the operation and maintenance process.

[0092] Among them, the present invention can improve the evaluation accuracy through multi-dimensional integration: by integrating four-dimensional data of equipment operation, fault characteristics, environment, and operation and maintenance, 18 core evaluation indicators are constructed (such as the envelope spectrum energy of the gearbox vibration signal, the SOC-SOH coupling coefficient of the battery pack, etc.), which is more than 40% higher than the traditional single-dimensional evaluation accuracy.

[0093] The dynamic weights of the present invention adapt to complex working conditions: the improved AHP-entropy weight method combined with the real-time working condition adjustment factor reduces the weight calculation error from 20% of the traditional method to below 8%, and the evaluation reliability is significantly improved, especially in extreme environments (such as low-temperature starting of wind turbines and overload operation of photovoltaic inverters).

[0094] The present invention can accurately predict and realize proactive operation and maintenance: the LSTM-Attention model reduces the RMSE to 8% in battery remaining life prediction, and the lead time for wind turbine gearbox failure prediction reaches 45 days, which is a 60% improvement in predictive ability compared with traditional models, effectively reducing unplanned downtime by more than 30%.

[0095] This invention can achieve a closed-loop management of the entire life cycle: forming a complete closed loop from data collection, evaluation and prediction to operation and maintenance decision-making, supporting the health record management of the equipment throughout its life cycle, providing data support for equipment selection optimization and life cycle cost analysis, and is expected to reduce operation and maintenance costs by 25%-30%.

[0096] The following is a detailed explanation of the four-level evaluation indicator system:

[0097] 1. Four-level evaluation indicator system

[0098] Core definition: Divide the evaluation indicators of the health status of new energy equipment into target layer, criterion layer, indicator layer, and sub-indicator layer, forming a structured evaluation framework to achieve progressive analysis from macro to micro.

[0099] Parameter description of each level:

[0100] Target layer (top layer)

[0101] Parameter meaning: The ultimate goal of the assessment, that is, the comprehensive health status score of new energy equipment (represented by H).

[0102] Function: As the guide of the entire evaluation system, it integrates the calculation results of lower-level indicators to form a unified quantitative conclusion on health status.

[0103] Criteria layer (first-level indicators):

[0104] Parameter meaning: reflects the core dimensions of device health, usually including:

[0105] C_1: Equipment operating status (such as load rate, speed, vibration, etc.);

[0106] C_2: Environmental adaptability (such as temperature, humidity, dust concentration, etc.);

[0107] C_3: Historical maintenance data (such as failure frequency, maintenance time, parts replacement records, etc.);

[0108] C_4: Energy efficiency performance (such as energy consumption efficiency, energy conversion rate, etc.).

[0109] Function: Decompose the target layer into quantifiable core dimensions to facilitate layered weight calculation.

[0110] Indicator layer (secondary indicators):

[0111] Parameter meaning: Specific detailed indicators of the criterion layer, for example:

[0112] If the criterion layer is "equipment operating status C_1", the indicator layer may include:

[0113] D_{11}: real-time load rate;

[0114] D_{12}: vibration amplitude;

[0115] D_{13}: Motor temperature.

[0116] Function: Connects the criterion layer and the sub-indicator layer to provide a more specific evaluation dimension.

[0117] Sub-indicator layer (third-level indicators):

[0118] Parameter meaning: The underlying data unit of the indicator layer, usually real-time sensor data or statistical data that can be directly collected, for example:

[0119] If the indicator layer is "real-time load rate D_{11}", the sub-indicator layer may be:

[0120] X_{111}: current hourly average load rate;

[0121] X_{112}: Daily maximum load rate peak.

[0122] Function: Serves as the raw data input for evaluation and is used to calculate the weights and scores of underlying indicators.

[0123] 2. Dynamic weight calculation of AHP-entropy weight method

[0124] 1. Subjective weight (AHP)

[0125] Definition: A subjective weighting method based on expert experience or industry standards that determines the relative importance of indicators by constructing a judgment matrix.

[0126] Parameter Description:

[0127] Judgment matrix: For indicators at the same level (such as criterion levels C_1-C_4), the importance of each pair of indicators is compared through expert scoring (1-9 scaling method), forming a matrix \mathbf{A}=[a_{ij}], where a_{ij} represents the importance ratio of indicator i to indicator j (a_{ij}=1 / a_{ji}).

[0128] Weight calculation: The subjective weight of each indicator W_s=[w_{s1},w_{s2},\dots,w_{sn}] is calculated by the eigenvalue method (such as the eigenvector corresponding to the maximum eigenvalue \lambda_{\text{max}}), which must meet the consistency test (through the consistency index CI and random consistency ratio CR<0.1).

[0129] Function: Incorporate expert knowledge and industry standards to reflect prior experience in assessment.

[0130] 2. Objective weight (entropy weight method)

[0131] Definition: An objective weighting method based on data volatility, using entropy to measure the information disorder of indicators. The greater the data variation, the higher the weight.

[0132] Parameter Description:

[0133] Data matrix: After standardizing the underlying sub-indicator layer data, a matrix \mathbf{X}=[x_{ij}] (i is the number of samples, j is the number of indicators) is formed. The standardization method is, for example, range standardization:

[0134] x'_{ij}=\frac{x_{ij}-\min(x_j)}{\max(x_j)-\min(x_j)}

[0135] Entropy calculation: The entropy of indicator j is E_j = -\frac{1}{\lnm}\sum_{i=1}^m\left(\fr ac{x'_{ij}}{\sum_{i=1}^mx'_{ij}}\right)\ln\left(\frac{x'_{ij}}{\su m_{i=1}^mx'_{ij}}\right), where m is the sample size.

[0136] Weight calculation: The objective weight of indicator j is w_{oj}=\frac{1-E_j}{\sum_{j=1}^n(1-E_j)}, forming a weight vector W_o=[w_{o1},w_{o2},\dots,w_{on}].

[0137] Function: Avoid subjective bias and objectively reflect the importance of indicators based on data.

[0138] 3. Real-time working condition adjustment factor

[0139] 1. Definition

[0140] The adjustment parameter used to dynamically fuse the subjective weight W_s and the objective weight W_o, with a value range of [0,1].

[0141] When \alpha=1, the weight depends entirely on subjective judgment;

[0142] When \alpha=0, the weight depends entirely on objective data;

[0143] In actual applications, \alpha is dynamically adjusted according to the real-time operating status of the device and environmental parameters to achieve adaptive optimization of weights.

[0144] Example 2

[0145] Wind turbine health assessment:

[0146] Data acquisition and preprocessing: Deployed on a 1.5MW doubly-fed wind turbine: vibration sensors at the gearbox input (X / Y / Z directions) and the generator bearing seat (radial direction), with a sampling frequency of 10240Hz.

[0147] Temperature sensors include gearbox oil temperature (PT100) and generator stator temperature (embedded thermocouple). Electrical sensors include inverter output current (Hall sensor) and grid voltage (voltage transformer). Operating data from January to December 2023 was collected, and EMD-SVD denoising was performed on the gearbox vibration signal to extract 12 fault characteristic parameters, including kurtosis, margin factor, and energy operator.

[0148] Dynamic weight calculation:

[0149] Establish a wind turbine evaluation index system (Table 1):

[0150]

[0151]

[0152] The triangular fuzzy number method is used to construct the judgment matrix, and the subjective weight is calculated after consistency test. The entropy weight method is used to calculate the objective weight, and the weight is dynamically adjusted by \alpha=0.3+0.4\times(v / 15) (v is the real-time wind speed, m / s), so that the weight of the mechanical vibration index under high wind speed conditions can be automatically increased by 15%-20%.

[0153] Health status assessment:

[0154] A wind turbine health assessment cloud model was constructed, and the health status cloud parameters were set as follows: E_c=0.8, En=0.1, He=0.05; the sub-health status was E_c=0.65, En=0.15, He=0.08. The monitoring data on August 15, 2023 showed that the gearbox oil temperature was 85°C (historical threshold 75°C) and the vibration RMS value was 6.8mm / s (warning value 7.1mm / s). However, through dynamic weight calculation, due to the current wind speed of 14m / s (high load condition), the oil temperature index weight was increased from 0.10 to 0.13, and the comprehensive CHI value was calculated to be 0.72, which was determined to be a sub-healthy state, triggering a targeted lubrication system inspection work order, and discovering the gearbox oil level sensor failure in advance, thereby avoiding the expansion of gear wear.

[0155] Trend forecast and early warning:

[0156] An LSTM-Attention model was trained using 500 sets of historical CHI data, with an input sequence length of 30, 128 hidden layer neurons, and an attention weight matrix updated every two hours. A November 2023 forecast indicated that the CHI value of a particular wind turbine would continue to drop to 0.55 over the next 30 days, and bearing temperature fluctuations would increase. This triggered an orange alert, prompting the operations team to replace the bearings in advance, avoiding downtime losses of approximately 500,000 yuan.

[0157] Example 3: Photovoltaic Inverter Health Assessment

[0158] For a 100kW string photovoltaic inverter, we collected data such as input voltage, output current, IGBT module temperature, and fan speed to build an evaluation index system that includes electrical parameters (voltage imbalance, total harmonic distortion THD) and thermal performance parameters (module temperature difference, heat dissipation efficiency). 2 ), the weight of cooling system-related indicators automatically increased by 30%, effectively identifying the problem of reduced cooling efficiency due to fan dust accumulation, providing an early warning and arranging cleaning and maintenance two weeks in advance, avoiding IGBT overheating and damage.

[0159] System deployment and application:

[0160] After deploying this system, a new energy power generation group has achieved centralized health management of 300 wind turbines and 500 inverters:

[0161] Fault diagnosis accuracy increased from 75% to 92%;

[0162] Unplanned downtime was reduced by 35%, and annual power generation increased by 2.5%;

[0163] Operation and maintenance costs decreased by 28%, with manual inspection costs reduced by 40% and spare parts inventory turnover increased by 30%.

[0164] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the health status of new energy equipment, characterized in that: The following steps are involved: Multi-source data acquisition and preprocessing: Deploy a sensor array to collect equipment operation data, fault characteristic data, environmental data, and operation and maintenance history data. Use the EMD-SVD algorithm to denoise vibration signals, and perform Z-score standardization to process multi-dimensional data. Construction of dynamic weight calculation model: A four-level evaluation index system is established, and the AHP-entropy weight method is used to calculate the dynamic weight. The subjective weight and objective weight are integrated through the real-time working condition adjustment factor α. The calculation formula is W = αW + (1-α)W, where W is the dynamic fusion subjective weight; W is the dynamic fusion objective weight; α is calculated in real time based on the equipment load rate and ambient temperature; Health status grading assessment: A health status assessment cloud generator is constructed using cloud model theory. The cloud membership of each indicator is calculated, and the equipment comprehensive health index (CHI) is calculated using the weighted average method. The formula is CHI = \sum_{i=1}^nW_i\timesC_i, where W_i is the dynamic weight, representing the weight coefficient of the i-th evaluation indicator in the comprehensive assessment, reflecting the relative importance of the indicator to the equipment health status. C_i is the cloud membership value, indicating the fuzzy membership degree of the i-th evaluation indicator to a certain health status level. The value range is [0,1]. The closer the value is to 1, the higher the probability that the indicator belongs to the corresponding health level. Health trend prediction and proactive warning: Build an LSTM-Attention prediction model, input historical CHI sequences, operating data, and environmental parameters, and output health status predictions for the next 1-3 months. Set up a three-level warning mechanism to trigger different levels of warnings and operation and maintenance strategies based on the prediction results.

2. The health status assessment method of a new energy device according to claim 1, characterized in that: The evaluation is conducted using a four-level evaluation index system, which includes: First-level indicators: equipment operating parameters, fault characteristic parameters, environmental impact parameters, and operation and maintenance history parameters; Secondary indicators: electrical parameters, mechanical parameters, vibration characteristics, thermal performance parameters; Level 3 indicators: voltage harmonic distortion rate, gearbox vibration RMS value, bearing fault frequency amplitude, and heat dissipation efficiency.

3. The health status assessment method of a new energy device according to claim 1, characterized in that: The multi-type sensors include: Vibration acceleration sensors are deployed in vibration-sensitive areas of equipment bearing seats and gearbox housings; Infrared thermal imager, used to obtain the surface temperature distribution of the equipment; Rogowski coil current sensor is used to collect high-frequency current waveform data.

4. The health status assessment method of a new energy device according to claim 1, characterized in that: The EMD-SVD denoising algorithm specifically includes: Step 1: Perform empirical mode decomposition (EMD) on the original vibration signal to obtain 10-15 intrinsic mode functions (IMFs); Step 2: Perform singular value decomposition (SVD) on each IMF component, retaining the singular values ​​with the first 30% energy share to reconstruct the signal; Step 3: Filter the effective IMF components through the kurtosis criterion and reconstruct the denoised vibration signal.

5. The health status assessment method of a new energy device according to claim 1, characterized in that: In the cloud model theory: The health state cloud generator parameters are: expectation E_c = 0.8\pm0.1, entropy En = 0.1\pm0.05, excess entropy He = 0.05\pm0.02; The sub-health cloud generator parameters are: expectation E_c=0.65\pm0.15, entropy En=0.15\pm0.08, excess entropy He=0.08\pm0.

03.

6. The health status assessment method of a new energy device according to claim 1, characterized in that: When calculating the comprehensive health index CHI, a weighting coefficient of \beta = 1.2-1.5 is set for the fault characteristic parameter, and a weighting coefficient of \gamma = 0.8-1.0 is set for the environmental impact parameter. The calculation formula is: CHI = \sum_{i=1}^n(\beta_iW_i)\timesC_i; where \beta_i is dynamically adjusted according to the indicator category.

7. A new energy equipment health status assessment system implementing the method of claims 1-6, characterized in that: include: Edge acquisition layer: Contains a high-precision sensor array and edge computing units to achieve real-time data acquisition, local processing and feature extraction, and supports Modbus / TCP and MQTT protocol data upload; Platform management layer: includes the data center and algorithm engine. The data center builds a digital twin model of the device and integrates full lifecycle data. The algorithm engine deploys dynamic weight calculation, health assessment, and trend prediction modules, and supports online tuning of algorithm parameters. Application service layer: includes health status visualization interface, intelligent early warning module, operation and maintenance decision support system, realizes assessment result visualization, multi-channel early warning and differentiated operation and maintenance strategy generation, and connects to the enterprise ERP system to realize operation and maintenance process automation.

8. The method and system for evaluating the health status of new energy equipment according to claim 7, characterized in that: The edge computing unit ECU is configured as follows: The processor is ARM Cortex-A72 dual-core or above, with a main frequency of ≥1.5GHz; Integrated 4GB DDR4 memory and 32GB eMMC storage, supporting local data caching for more than 72 hours; It has multiple communication interfaces such as RS485, CAN, Ethernet, and supports ModbusRTU and ProfibusDP protocols.

9. The evaluation system according to claim 7, characterized in that: The digital twin model constructed by the data center platform includes: Equipment geometric model, built based on 3D laser scanning data; Physical simulation model, integrating multi-body dynamics and heat transfer equations, supports performance simulation under real-time working conditions; The fault knowledge base contains more than 1,000 historical fault cases and corresponding characteristic parameter thresholds.

Images