New energy equipment performance index construction method and system based on multi-modal fusion

Abstract

The application relates to the technical field of new energy equipment data analysis and operation and maintenance management, and discloses a new energy equipment performance index construction method and system based on multi-modal fusion. The method collects multi-modal data of a target equipment and maps the multi-modal data into standard equivalent unit values; an initial dynamic capability boundary is calculated based on a physical constraint model; the dynamic capability boundary is updated by checking and correcting equipment environment parameters by using spatial neighborhood environment data; a state characteristic vector is constructed, a performance loss attribution category is determined based on projection of a difference vector in different characteristic dimensions, and a performance compensation coefficient is generated; and a normalized performance deviation index is generated based on the corrected dynamic capability boundary, the performance compensation coefficient and actual output data. The application eliminates equipment specification differences, removes the influence of an external environment on performance through a physical model and an attribution mechanism, and realizes objective evaluation of real performance of heterogeneous new energy equipment.

Description

Technical Field

[0001] This invention relates to the field of data analysis and operation and maintenance management technology for new energy equipment, and in particular to a method and system for constructing performance indicators for new energy equipment based on multimodal fusion. Background Technology

[0002] With the large-scale deployment of new energy infrastructure such as electric vehicle charging stations, distributed photovoltaic inverters, and energy storage converters, the refined operation and management of equipment assets is becoming increasingly important. In order to improve the return on assets and ensure grid security, operators need to quantitatively evaluate the operating performance of massive distributed equipment to identify inefficient or abnormal equipment.

[0003] Traditional equipment performance evaluation methods primarily rely on simple output statistics, such as calculating equipment time utilization, energy conversion efficiency, or downtime rate. However, in practical applications, this statistically-based evaluation system has significant limitations. First, new energy equipment typically has multiple power specifications, and the absolute output data of equipment with different rated power are not directly comparable, lacking a unified normalized benchmark. Second, the actual output capability of power electronic equipment is significantly constrained by the physical environment. For example, in high-temperature environments, equipment needs to implement over-temperature derating strategies, or when the grid voltage is low, input current needs to be limited. Existing evaluation methods often only focus on the final output result, failing to analyze the dynamic physical boundaries of equipment operation. This can easily lead to misjudging objective power limitations caused by excessively high ambient temperatures or poor grid quality as performance defects of the equipment itself, resulting in distorted evaluation results.

[0004] Furthermore, existing technologies largely rely on local sensor data from individual devices for calculations. When the device's own temperature or electrical sensors drift or fail, performance indicators calculated based on erroneous data will be severely biased. Simultaneously, limited device performance is often the result of the coupling effect between the external power grid environment and the state of internal components. Existing technologies lack effective feature separation and attribution methods, making it difficult to mathematically eliminate the influence of uncontrollable external factors and accurately pinpoint the responsible party for performance loss, thus affecting the accuracy of operation and maintenance decisions. Summary of the Invention

[0005] This invention primarily addresses the technical problem that existing performance evaluation methods for new energy equipment struggle to distinguish between objective constraints of the external environment and internal performance degradation of the equipment, resulting in a lack of comparability and fairness in evaluation results for equipment with different power specifications and under different geographical environments.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] The first aspect of this invention provides a method for constructing performance indicators for new energy equipment based on multimodal fusion, comprising the following steps:

[0008] Multimodal data of the target device is collected and mapped to standard equivalent unit values. The standard equivalent unit values ​​are used to eliminate the dimensional influence caused by differences in device hardware specifications.

[0009] A physical constraint model is constructed based on the multimodal data, and the initial dynamic capability boundary of the target device under the current operating conditions is calculated based on the standard equivalent unit values.

[0010] Acquire neighborhood environment data within the target device's spatial neighborhood, perform confidence verification and correction on the environmental parameters in the multimodal data based on the neighborhood environment data, update the physical constraint model using the corrected parameters, and obtain the corrected dynamic capability boundary;

[0011] Construct the state feature vector of the target device, calculate the difference vector between the state feature vector and the reference vector, and determine the attribution category of performance loss based on the projection components of the difference vector in different feature dimensions, and generate the corresponding performance compensation coefficient.

[0012] Based on the actual output data in the corrected dynamic capability boundary, the performance compensation coefficient, and the standard equivalent unit values, a normalized performance deviation index is generated.

[0013] In one possible implementation, the process of acquiring multimodal data of the target device and mapping it to standard equivalent unit values ​​includes: parsing the multimodal data to obtain rated power, real-time operating data, and static parameters; calculating the ratio of the rated power to a preset global reference power to determine the hardware multiplier; dividing the output power data and output current data in the real-time operating data by the hardware multiplier to convert them into the standard equivalent unit values, while maintaining the original physical units of the voltage and temperature data.

[0014] In one possible implementation, the process of constructing a physical constraint model based on the multimodal data and calculating the initial dynamic capability boundary includes: extracting the measured internal key point temperature and measured input voltage from the multimodal data; determining the current carrying capacity attenuation ratio coefficient of the power device in different internal temperature ranges using a temperature derating function; determining the limiting ratio coefficient of output power being limited as the voltage decreases when the input voltage is lower than the rated voltage lower limit using a grid voltage constraint function; substituting the measured values ​​into the above functions to calculate the temperature derating coefficient and voltage constraint coefficient, selecting the minimum value of the two as the boundary constraint factor, and thus determining the initial dynamic capability boundary.

[0015] In one possible implementation, the process of confidence verification and correction of environmental parameters in the multimodal data includes: calculating the arithmetic mean of the external ambient temperatures of all devices in the spatial neighborhood as the neighborhood average; calculating the absolute value of the deviation between the measured external ambient temperature and the neighborhood average; when the absolute value of the deviation exceeds a preset temperature deviation threshold, replacing the measured external ambient temperature with the neighborhood average; subsequently, calculating the current power loss based on real-time operating data, deriving the corrected internal key point temperature in combination with preset thermal resistance parameters, and substituting it into the physical constraint model to recalculate the dynamic capability boundary.

[0016] In one possible implementation, the spatial neighborhood is determined based on the geographical coordinates and device type labels of the device. This is achieved by delineating a circular area with the geographical coordinates as the center and a preset distance as the radius, and then filtering a set of compatible devices within the area.

[0017] In one possible implementation, the process of constructing the state feature vector of the target device involves calculating external grid-side features and internal device-side features, and performing standardization processing to generate dimensionless standard scores for each. The external grid-side features include input voltage deviation and input voltage ripple, while the internal device-side features include temperature rise rate, fan speed deviation, and efficiency degradation. The reference vector is set as a zero vector, representing an ideal fault-free operating state.

[0018] In one possible implementation, the process of determining the performance loss attribution category and generating the performance compensation coefficient includes: calculating the difference vector between the state feature vector and the reference vector; defining the external power grid weight vector and the internal equipment weight vector, extracting the components of the difference vector in the feature dimensions of the external power grid side and the feature dimensions of the internal equipment side, respectively, and calculating the corresponding external influence magnitude and internal influence magnitude; determining the dominant factor by comparing the magnitude relationship between the two types of magnitudes: when the external influence magnitude is significantly greater than the product of the internal influence magnitude and the sensitivity coefficient, it is determined that the performance limitation is dominated by external factors, and the performance compensation coefficient is set to a value of one; otherwise, it is determined that the performance is dominated by internal equipment factors, and the performance compensation coefficient is set to a value of zero.

[0019] In one possible implementation, the process of generating the normalized performance deviation index includes: converting the control power command value of the target device into a standard equivalent unit value, and taking the smaller value between the value and the corrected dynamic capability boundary as the theoretical target power; calculating the difference between the theoretical target power and the actual output data to obtain the instantaneous physical performance gap; using the difference between the value and the performance compensation coefficient as a weight to weight the instantaneous physical performance gap to obtain the effective performance loss value; performing time-cumulative calculations on the effective performance loss value and the theoretical target power within the observation period, and finally calculating the ratio of the cumulative effective performance loss value to the cumulative theoretical target power, and subtracting the ratio from the value to obtain the index.

[0020] A second aspect of this invention provides a system for constructing performance indicators for new energy equipment based on multimodal fusion, comprising:

[0021] The data standardization module is configured to collect multimodal data from the target device and map it to standard equivalent unit values.

[0022] The boundary calculation module is configured to construct a physical constraint model based on the multimodal data and calculate the initial dynamic capability boundary of the target device based on the standard equivalent unit values.

[0023] The verification and analysis module is configured to acquire neighborhood environment data within the spatial neighborhood of the target device, and based on this, perform confidence verification and correction on the environmental parameters in the multimodal data to obtain the corrected dynamic capability boundary; the verification and analysis module is also configured to construct the state feature vector of the target device, determine the performance loss attribution category based on the projection of the difference between it and the reference vector in different feature dimensions, and generate performance compensation coefficients.

[0024] The indicator generation module is configured to generate a normalized performance deviation indicator based on the actual output data in the modified dynamic capability boundary, the performance compensation coefficient, and the standard equivalent unit values.

[0025] In summary, the present invention has at least one of the following beneficial technical effects:

[0026] 1. This invention eliminates the differences in rated power specifications between heterogeneous devices by introducing standard equivalent unit values ​​and hardware multiplier coefficients, enabling performance comparison of new energy devices of different capacity levels under a unified dimension.

[0027] 2. This invention constructs a dynamic capability boundary based on a physical constraint model, which can accurately identify objective output capability limitations caused by grid voltage fluctuations or excessively high ambient temperatures, thus avoiding misjudging performance degradation caused by external environment as equipment failure.

[0028] 3. This invention introduces a confidence verification mechanism for spatial neighborhood data, uses group statistical characteristics to correct sensor drift data of individual devices, and combines thermal resistance models to infer internal temperature, effectively preventing boundary calculation deviations caused by environmental sensor failures and improving the accuracy of evaluation benchmarks.

[0029] 4. This invention utilizes a vector projection method in a high-dimensional feature space to achieve automatic attribution of performance losses. By separating external power grid characteristics from internal equipment characteristics, it can intelligently determine the source of performance bottlenecks and generate dynamic performance compensation coefficients accordingly. This enables automatic exemption from performance losses not caused by equipment responsibility, ensuring the objectivity and fairness of performance evaluation results. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the architecture of the new energy equipment performance index construction system based on multimodal fusion according to the present invention;

[0031] Figure 2 This is a flowchart illustrating the method for constructing performance indicators for new energy equipment based on multimodal fusion according to the present invention.

[0032] Figure 3 This is a schematic diagram of the multi-source heterogeneous data standardization preprocessing workflow according to the present invention;

[0033] Figure 4 This is a schematic diagram of the initial dynamic capability boundary construction process based on physical constraints according to the present invention;

[0034] Figure 5 This is a schematic diagram of the boundary confidence verification process based on spatiotemporal neighborhood features according to the present invention;

[0035] Figure 6 This is a schematic diagram of the performance attribution and compensation process based on state feature difference according to the present invention;

[0036] Figure 7 This is a schematic diagram of the normalized performance deviation generation process according to the present invention;

[0037] Figure 8 This is a comparison chart of the power curves of a device according to Embodiment 2 of the present invention during power grid fluctuations;

[0038] Figure 9 This is a scatter plot of fault clustering in the multidimensional feature space according to Embodiment 2 of the present invention;

[0039] Figure 10 This is a bar chart comparing the performance scores of the three methods according to Embodiment 2 of the present invention in different scenarios.

[0040] Among them, 100, New Energy Equipment Performance Indicator Construction System; 101, Data Standardization Module; 102, Boundary Calculation Module; 103, Verification and Analysis Module; 104, Indicator Generation Module; 200, New Energy Equipment. Detailed Implementation

[0041] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0042] See attached document Figure 1 This invention provides a new energy equipment performance index construction system 100, which is applied to perform unified performance evaluation of distributed heterogeneous new energy equipment. The new energy equipment performance index construction system 100 includes: a data standardization module 101, a boundary calculation module 102, a verification and analysis module 103, and an index generation module 104.

[0043] The new energy equipment performance indicator construction system 100 is connected to multiple new energy devices 200 via a communication network. The new energy devices 200 include DC charging piles, AC charging piles, or energy storage inverters. Each new energy device 200 is equipped with sensor components to collect internal operating status data and external environmental data. These sensor components include voltage sensors, current sensors, temperature sensors, and speed sensors. The new energy equipment performance indicator construction system 100 is deployed on a cloud server or edge computing cluster to execute data processing and indicator calculation logic.

[0044] The data standardization module 101 is configured to receive raw operating data, environmental data, and static parameters of the new energy equipment 200. The data standardization module 101 performs data cleaning and time alignment operations, mapping the operating data of equipment with different power levels to standard equivalent unit values. These standard equivalent unit values ​​are used to eliminate the dimensional effects caused by differences in the rated specifications of different equipment.

[0045] Boundary calculation module 102 is connected to data standardization module 101. Boundary calculation module 102 is configured to calculate the initial dynamic capability boundary of the device under current operating conditions based on the device's physical characteristic curve function, combined with the current internal temperature, external ambient temperature, and grid input voltage. The initial dynamic capability boundary characterizes the theoretical maximum output capability of the device without external data verification.

[0046] The verification analysis module 103 is connected to the boundary calculation module 102. The verification analysis module 103 is configured to perform spatiotemporal collaborative verification and state feature analysis. The verification analysis module 103 acquires the operating data of other devices within the geographical neighborhood of the target device, calculates the mean value of the neighborhood environment, and performs confidence verification and correction on the environmental parameters of the target device based on the mean value of the neighborhood environment, thereby updating the dynamic capability boundary. The verification analysis module 103 is also configured to construct the device's state feature vector, determine the attribution category of performance loss by calculating the difference between the measured feature vector and the reference feature vector, and generate corresponding performance compensation coefficients. The state feature vector contains physical parameters reflecting the underlying operating mechanism of the device.

[0047] The indicator generation module 104 is connected to the verification and analysis module 103. The indicator generation module 104 is configured to receive actual output data, the corrected dynamic capability boundary, and performance compensation coefficients, and generate a normalized performance deviation indicator through a preset calculation model.

[0048] See attached document Figure 2 This invention provides a method for constructing performance indicators for new energy equipment, comprising the following steps:

[0049] S100, the data standardization module 101 collects multimodal data of the target device within a specific time window. The multimodal data includes operational data, environmental data, and hardware static parameters. The data standardization module 101 calculates the hardware scaling factor based on the ratio of the target device's rated power to a preset reference power, and converts the operational data into values ​​based on standard equivalent units.

[0050] S200, Boundary calculation module 102 constructs an initial dynamic capability boundary based on physical constraints. Boundary calculation module 102 calls a preset temperature derating function and grid voltage constraint function. The temperature derating function describes the current-carrying capacity decay characteristics of power devices at different temperatures, and the grid voltage constraint function describes the limiting characteristics of input voltage fluctuations on output power. Boundary calculation module 102 substitutes the collected temperature and voltage data into the temperature derating function and grid voltage constraint function to calculate the initial dynamic capability boundary.

[0051] S300, the verification analysis module 103 performs boundary confidence verification based on spatiotemporal neighborhood features. The verification analysis module 103 determines the spatial neighborhood set of the target device and calculates the average external ambient temperature of all devices within this set. The verification analysis module 103 calculates the absolute value of the deviation between the measured external ambient temperature of the target device and the average neighborhood ambient temperature. When the absolute value of the deviation exceeds a preset threshold, the verification analysis module 103 determines that the sensor data of the target device has drifted, and replaces the measured external ambient temperature with the average neighborhood ambient temperature, recalculates and outputs the corrected dynamic capability boundary.

[0052] S400, the verification analysis module 103 performs performance attribution and compensation based on state characteristic difference. The verification analysis module 103 constructs a state feature vector containing underlying feature dimensions, including input voltage ripple, fan speed, and junction temperature change rate. The verification analysis module 103 calculates the difference vector between the current measured feature vector and the standard reference feature vector. The verification analysis module 103 projects the difference vector onto the external power grid side feature dimension and the internal equipment side feature dimension, respectively. When the projected value of the external power grid side feature dimension is significantly greater than the projected value of the internal equipment side feature dimension, the verification analysis module 103 determines that the current performance limitation is dominated by external factors and sets the performance compensation coefficient to the compensation value; otherwise, it sets the performance compensation coefficient to the non-compensation value.

[0053] S500, the indicator generation module 104 generates the normalized performance deviation. The indicator generation module 104 calculates the normalized performance deviation using the corrected dynamic capability boundary as the denominator and the sum of the actual output data and the compensation term as the numerator. The compensation term is the product of the performance compensation coefficient and the performance loss value, where the performance loss value is the difference between the corrected dynamic capability boundary and the actual output data. The indicator generation module 104 outputs the normalized performance deviation as a quantitative indicator for evaluating the true performance of the target equipment.

[0054] The following provides a detailed description of each step of the method of the present invention.

[0055] See attached document Figure 3 In this embodiment, step S100 eliminates the differences in sampling frequency, hardware specifications, and communication protocols among different new energy devices, and establishes a unified evaluation benchmark.

[0056] Step S100 specifically includes the following steps:

[0057] Step S101: Construct multimodal data acquisition vectors.

[0058] Data standardization module 101 establishes a connection with the sensor network of the new energy equipment through a communication interface, targeting the set of equipment. Any device in Collect its continuous time series Multimodal data on the platform.

[0059] Multimodal data is encapsulated into raw state vectors Specifically, the original state vector It includes a data subset with three dimensions: high-frequency operational data, low-frequency environmental data, and static configuration parameters.

[0060] Among them, high-frequency operating data includes grid-side input voltage. Equipment-side output current and real-time output power Its sampling frequency is set to the second or hundreds of milliseconds level (e.g., 1Hz to 10Hz) to capture power grid fluctuations and load change characteristics;

[0061] Low-frequency environmental data includes the critical temperature of power devices inside the equipment. (such as IGBT junction temperature or heatsink temperature) and the ambient temperature outside the equipment. Considering thermal inertia, its sampling frequency is set to the minute level (e.g., 0.01Hz to 0.1Hz).

[0062] Static configuration parameters include the device's nameplate rated power. And device topology type identifier.

[0063] The parsing of communication protocols (such as Modbus, CAN, etc.) and message unpacking during the data acquisition process are well-known technologies to those skilled in the art, and will not be elaborated here.

[0064] Step S102: Perform time window alignment for multi-frequency data.

[0065] For the original state vector To address the issue of inconsistent sampling frequencies across different dimensions of data, this embodiment employs a time synchronization strategy based on high-frequency data. The fundamental principle is that although the rate of change of parameters such as ambient temperature is much lower than that of electrical parameters, constructing a unified physical model requires complete state input at the same moment. Therefore, by using interpolation algorithms to fill the gaps in low-frequency data along the time axis, the complete physical state of the device at any high-frequency sampling point can be reconstructed.

[0066] The data standardization module 101 uses the timestamps of high-frequency operational data as a reference axis to interpolate and align low-frequency environmental data. Specifically, when the timestamps of high-frequency data... Located at two consecutive sampling times of low-frequency data and When the interval is between, a linear interpolation algorithm is used to calculate the time. The corresponding environmental parameter values. Let... For the temperature variable to be aligned, at time... The calculation formula is as follows:

[0067] ;

[0068] in, For a moment interpolated temperature, and They are time points and The measured temperature value, , , All are Unix timestamps or relative time values.

[0069] Furthermore, for discrete state variables (such as fault alarm codes and power-on / off status bits) in the original state vector, since their state abrupt changes are not continuously differentiable, the zero-order hold method is used for processing, i.e., taking the time step... The value at the most recent sampling time is used as the value at the current time to maintain the continuity of the state.

[0070] Step S103: Perform numerical mapping based on standard equivalent unit (SEU).

[0071] To address the issue of incomparable dimensions in performance evaluation of devices with different power ratings (e.g., 7kW AC charging piles and 120kW DC charging piles), this embodiment introduces a hardware scaling factor. Normalization is performed. The physical significance of this process is to convert absolute physical quantities into relative "capacity utilization rates," thereby enabling the performance of equipment of different specifications to be compared within the same coordinate system.

[0072] The system presets a global reference power. This reference power serves as a normalized reference for mapping all devices. The value can be set based on the average power of the cluster of devices to be evaluated, for example, set to 10kW, or set to 1kW per unit power. Hardware scaling factor. Defined as a device Rated power Compared with reference power The ratio is calculated using the following formula:

[0073] ;

[0074] Based on this coefficient, the data normalization module 101 maps all physical quantities related to power capacity to standard equivalent unit values. Let... For equipment At any moment The measured power, then its corresponding standard equivalent power The calculation is as follows:

[0075] ;

[0076] Through the mapping described above, the power data of all devices is converted into values ​​relative to the reference unit. Similarly, the same factoring is used for other capacity-related parameters such as output current. Alternatively, normalization can be performed based on the corresponding coefficient of the rated current.

[0077] It should be noted that for non-capacity-related physical quantities, including voltage, temperature, and time parameters, the original physical units are retained and they are not included in the SEU mapping in order to preserve the direct readability of their physical properties.

[0078] See attached document Figure 4 In this embodiment, step S200 establishes a physical model of equipment operation, converting the static rated power on the equipment nameplate into a "dynamic capability boundary" that changes in real time with operating conditions.

[0079] The core principle is that the power output capability of new energy equipment is not fixed, but is limited by the current heat dissipation conditions and the quality of the power grid. By quantifying these physical constraints, it is possible to isolate objective performance degradation caused by environmental or grid factors, thereby providing an accurate reference system for subsequently identifying hardware and software faults in the equipment itself.

[0080] Step S200 specifically includes the following steps:

[0081] Step S201: Construct a temperature derating function model.

[0082] The boundary calculation module 102 establishes a nonlinear mapping relationship describing the output power capability as a function of internal temperature, based on the thermal characteristics of power semiconductor devices (such as IGBTs or MOSFETs) and the design parameters of the device's heat dissipation system. The physical basis of this relationship is that the junction temperature of the power device must be controlled within the safe operating area (SOA). When the heat sink temperature rises, the current flowing through the device must be reduced in order to maintain a constant junction temperature, which in turn leads to a decrease in the upper limit of output power.

[0083] In this embodiment, the temperature derating function It is constructed as a piecewise linear function. Specifically, a first temperature threshold is set. Second temperature threshold First temperature threshold (For example, 70℃) is usually determined according to the equipment's thermal design specifications, corresponding to the thermal equilibrium point when the heat dissipation system is running at full load; the second temperature threshold (For example, 90°C) is determined according to the power device's datasheet, corresponding to the device's highest permissible safe junction temperature or over-temperature protection trigger point.

[0084] When the temperature of key points inside the equipment Below At that time, the equipment has full-load output capability; when In and When the output capability decreases linearly, the output capability decreases linearly; when Exceed When this occurs, the equipment enters over-temperature protection mode, and its output capacity drops to zero. Temperature derating factor. The calculation formula is as follows:

[0085] ;

[0086] in, The temperature of key internal points of the device after being collected and aligned in step S100. It is a dimensionless temperature derating factor, with a value range of [0,1].

[0087] Step S202: Construct the grid voltage constraint function model.

[0088] The boundary calculation module 102 establishes constraint logic describing the variation of output power capability with grid input voltage based on the device's input stage circuit topology (such as a three-phase rectifier circuit) and input current limitations. This logic follows the law of power conservation. There is a maximum physical limit to the current on the input side of the device. Under the premise that the input voltage The maximum power that the equipment can draw during descent It will inevitably decrease proportionally.

[0089] In this embodiment, the grid voltage constraint function It is also constructed as a piecewise function. A lower limit for the rated voltage is set. and cutoff voltage Lower limit of rated voltage (For example, 85% of the rated voltage) is usually determined according to power grid operation standards (such as GB / T 12325), representing the minimum voltage at which the equipment is allowed to operate at full power; cutoff voltage (For example, 70% of the rated voltage) is determined based on the equipment's undervoltage protection threshold.

[0090] When the input voltage Higher than At that time, the device is allowed to output full power; when Below But higher When the voltage decreases, the output power is limited; when Below When this occurs, the equipment stops outputting. Grid voltage constraint coefficient. The calculation formula is as follows:

[0091] ;

[0092] in, The effective value of the grid-side input voltage acquired in step S100. This is a dimensionless voltage constraint coefficient, ranging from [0,1]. This function ensures that, when calculating evaluation indicators, objective power reduction caused by grid voltage fluctuations will not be misjudged as a performance failure of the equipment itself.

[0093] Step S203: Calculate the initial dynamic capability boundary.

[0094] Boundary calculation module 102 will calculate the current time. Measured internal temperature and input voltage Substituting the values ​​into the temperature derating function and the grid voltage constraint function respectively, we can obtain the current temperature derating factor. and voltage constraint coefficient Based on the principle of system constraints, the final output capability of a device depends on the minimum value among all constraints.

[0095] Boundary calculation module 102 combined with the rated power of the equipment Calculate the initial dynamic capability boundary at the current moment. The calculation formula is as follows:

[0096] ;

[0097] in, Indicates at time Under current temperature and power grid conditions, the theoretically maximum allowable power output of the equipment. If the standard equivalent unit (SEU) from step S103 is used for calculation, then the formula... Replace it with the value 1 (i.e., the unit base), and the calculated result will be... This is the normalized dynamic capability boundary coefficient.

[0098] Through step S203, the system completes the transformation from static rated parameters to dynamic real-time capability boundaries, and these boundary values... This constitutes the theoretical upper limit for subsequent judgments on whether the equipment has "unpowered" power loss.

[0099] See attached document Figure 5 In this embodiment, step S300 addresses the data distortion problem caused by drift, damage, or local heat source interference of individual device sensors by introducing statistical characteristics of group data to perform reverse verification and correction of individual data.

[0100] This step is based on the principle of spatial correlation of microclimate environment, which states that within a certain geographical range, the external ambient temperature of different devices should tend to be consistent. If the measured ambient temperature of a device deviates significantly from the population mean of that area, the sensor data of that device is deemed unreliable. In this case, the system uses the population consensus data to reconstruct its physical capability boundary to prevent performance evaluation bias caused by sensor failure. For example, it avoids misjudging downtime caused by falsely reported high temperatures due to sensor damage as reasonable over-temperature protection derating.

[0101] Step S300 specifically includes the following steps:

[0102] Step S301: Construct a spatial neighborhood device set.

[0103] The verification and analysis module 103, based on Geographic Information System (GIS) data, filters out the target equipment to be evaluated. A cluster of neighboring devices operating within the same microclimate environment. Specifically, a target device is defined. geographic coordinates Centered on, at a preset distance A circular spatial region with a radius of [radius value]. Preset distance. The range of values ​​is set to 500 meters to 2000 meters. This range can effectively cover equipment in the same charging station, the same parking lot, or adjacent blocks, ensuring that meteorological conditions such as atmospheric temperature and solar radiation intensity in the area are highly similar.

[0104] Simultaneously, to ensure the consistency of the comparison benchmark, it is also necessary to filter device type labels to ensure that neighboring devices and the target device belong to the same or compatible device types. Spatial Neighborhood Set The definition is as follows:

[0105] ;

[0106] in, For the collection of all online devices, For other devices in the neighborhood, Indicates device With equipment The geographical distance between them can be calculated using the semi-versus formula.

[0107] Step S302: Extract the true values ​​of neighborhood environment features.

[0108] Verification analysis module 103 obtains the set All devices in the same time The external ambient temperature data is collected, and its statistical characteristic value is calculated as a reference for the true ambient temperature of the area at the current moment. In order to eliminate the influence of outliers caused by the failure of some neighboring devices, this embodiment can use the extreme value averaging method or the median method.

[0109] In this embodiment, the arithmetic mean is used. As a reference, the calculation formula is as follows:

[0110] ;

[0111] in, This represents the total number of devices in the neighborhood set. Indicates neighboring devices At any moment Measured ambient temperature. This is the average value. This reflects the time-varying microclimate region. The objective environmental heat level.

[0112] Step S303: Perform anomaly detection and data correction.

[0113] Verification analysis module 103 calculates the target device Measured external temperature and neighborhood mean The absolute deviation between them, and the preset temperature deviation threshold. Compare. Temperature deviation threshold. The settings take into account both the sensor's measurement accuracy error (typically ±1℃) and the allowable fluctuation range of the local microenvironment, with a value range of 5℃ to 10℃. If the deviation exceeds this threshold, the device is deemed faulty. If environmental sensor data is distorted (low confidence level), the measured value will be forcibly replaced with the neighborhood mean; otherwise, the measured value will be retained. Corrected ambient temperature. The determination logic is as follows:

[0114] ;

[0115] Step S304: Reconstruct the dynamic capability boundary.

[0116] When step S303 determines that data correction has occurred, it indicates that the initial boundary calculated based on the original measured data in step S200 may have a deviation. At this time, the verification analysis module 103 uses the corrected ambient temperature. Re-derive the estimated values ​​of the temperature at key points inside the equipment. This updates the dynamic capability boundary. The derivation process is based on the device's thermal resistance model, and the calculation formula is as follows:

[0117] ;

[0118] in, The equivalent thermal resistance parameter (unit: ℃ / W) preset for the equipment is determined by the equipment's thermal design data; For a moment The power loss of the equipment can be calculated based on the difference between the real-time input power and the output power, or obtained by looking up the efficiency curve based on the current load rate.

[0119] Subsequently, the reconstructed internal temperature Substitute into the temperature derating function in step S201 In the process, the corrected temperature derating factor is calculated, and then the final capability boundary after confidence verification is obtained. This step ensures that even in the event of sensor failure, the system can still accurately assess the output capabilities that the equipment "should" possess based on objective environmental conditions, providing an accurate theoretical denominator for subsequent calculations of the equipment's actual performance.

[0120] See attached document Figure 6 In this embodiment, step S400 solves the technical problem that traditional evaluation methods cannot distinguish between performance degradation caused by objective environment and performance degradation caused by equipment failure.

[0121] Its core principle lies in constructing a multi-dimensional state space, mapping the real-time operating state of the equipment to a vector point in the space. Since grid quality deterioration and internal equipment faults manifest in different directions in the multi-dimensional feature space (e.g., grid problems mainly cause changes in the voltage dimension, while equipment faults mainly cause changes in the temperature or speed dimension), by calculating the projected components of the state vector in different feature directions, the contribution of different factors to performance deviations can be quantified. This mechanism achieves fair evaluation based on physical attribution, ensuring that performance deductions are only applied when the equipment itself is defective.

[0122] Step S400 specifically includes the following steps:

[0123] Step S401: Construct a multidimensional state feature vector.

[0124] The verification and analysis module 103 extracts key features from the raw data collected in step S100 and constructs a feature vector representing the current operating state. To eliminate the impact of differences in the magnitude of different physical units (such as voltage unit V and rotational speed unit RPM) on distance calculation, it is necessary to normalize each feature component. In this embodiment, the Z-Score normalization method is used, utilizing the mean of historical operating data. and standard deviation Transform the raw data into dimensionless standardized scores. Feature vectors. Defined as Dimensional vector:

[0125] ;

[0126] Among them, the characteristic components of the external power grid include:

[0127] Normalized voltage deviation ( ): The standardized value of the deviation between the measured input voltage and the rated voltage, reflecting the steady-state fluctuation of the grid voltage;

[0128] Normalized voltage ripple ( : The standardized value of the input voltage ripple factor or total harmonic distortion (THD), reflecting the purity of the power grid.

[0129] The internal characteristic components of the device include:

[0130] Normalized temperature rise rate ( ): Standardized value of temperature change at key points inside the equipment per unit time, reflecting the dynamic response of the heat dissipation system;

[0131] Normalized fan speed deviation ( ): The standardized value of the deviation between the measured speed and the commanded speed of the cooling fan, reflecting the health of the active cooling components;

[0132] Normalization efficiency decay ( ): The standardized value of the difference between the theoretical optimal efficiency and the current measured efficiency, reflecting the loss status of the power device.

[0133] Step S402: Calculate the Euclidean distance between the difference vector and the reference datum.

[0134] The system pre-sets a standard reference vector. This vector represents the feature value of the device under ideal power grid conditions and without its own faults. Since Z-score normalization has been used, the feature value corresponding to the ideal state approaches 0; therefore, it is set... The zero vector is 0. The verification analysis module 103 calculates the measured feature vector at the current time. With reference vector Difference vector between :

[0135] ;

[0136] The difference vector Length of the module It represents the total degree to which the current operating state deviates from the ideal state, and its orientation in space implies the physical cause of the deviation.

[0137] Step S403: Perform projection calculation based on weighted direction.

[0138] To separate the influence of external and internal factors, this embodiment defines two feature weight vectors: the external power grid weight vector. and internal device weight vector These two vectors are binary mask vectors used to filter features in a specific dimension.

[0139] exist In this context, the dimensions corresponding to power grid characteristics such as voltage deviation and ripple are set to 1, while the other dimensions are set to 0.

[0140] exist In this context, dimensions corresponding to equipment characteristics such as temperature rise, efficiency, and fan are set to 1, while the remaining dimensions are set to 0.

[0141] Verification analysis module 103 calculates the difference vector The projection modulus in the two directional subspaces mentioned above yields the external influence modulus. and internal influence modulus :

[0142] ;

[0143] ;

[0144] in, This represents the Hadamard product of vectors (i.e., element-wise multiplication). This represents the L2 norm of a vector. The larger the value, the more likely the current performance deviation is caused by power grid quality issues; The higher the value, the more likely it is caused by a problem with the device itself.

[0145] Step S404: Attribution determination and compensation coefficient generation.

[0146] The verification analysis module 103 compares the magnitudes of the two projection moduli to determine the dominant factors causing performance limitations, and generates binary performance compensation coefficients accordingly. The decision logic is as follows:

[0147] Set sensitivity coefficient (The value range is usually from 0.8 to 1.2, and 1.0 is used in this embodiment) and dead zone threshold. (The value ranges from 0.1 to 0.5, used to filter out small fluctuations caused by measurement noise).

[0148] like These fluctuations are considered normal operating fluctuations and are not attributed to any cause; they are assumed to be normal. (That is, no punitive record will be made).

[0149] like Then the following comparison is performed:

[0150] like The current performance limitation was determined to be primarily due to poor external power grid quality, which is not the equipment's fault, triggering a liability exemption claim. ;

[0151] like The current performance limitation is determined to be primarily due to equipment malfunction or aging, which is considered equipment responsibility and therefore no compensation will be provided. .

[0152] Based on the above logic, the compensation coefficient The calculation formula is as follows:

[0153] ;

[0154] Through step S404, the system outputs... As a dynamic gating signal, it will participate in subsequent performance score calculations. When When this occurs, it indicates that the current performance degradation is exempted and not included in the equipment failure downtime or performance loss integral; when When this occurs, it indicates that the current performance degradation has been identified as equipment responsibility and will be accurately included in performance evaluation.

[0155] See attached document Figure 7 In this embodiment, step S500 performs a comprehensive calculation on the dynamic capability boundary, real-time output data and attribution compensation coefficient obtained from the previous steps to generate a quantitative indicator that can objectively reflect the true performance level of the equipment. In this embodiment, it is defined as Normalized Performance Deviation (UPDI).

[0156] This step is based on the principle of normalization evaluation, which maps the actual performance of the equipment to its theoretical limits under specific environmental and operating conditions. By using a dynamic boundary as a benchmark in the denominator, the rigid influence of ambient temperature on capacity assessment is eliminated; by introducing attribution compensation in the numerator, passive performance losses caused by external power grid factors are removed. This approach makes the equipment performance mathematically comparable across different geographical locations and seasons.

[0157] Step S500 specifically includes the following steps:

[0158] Step S501: Calculate the instantaneous physical performance gap.

[0159] The indicator generation module 104 first obtains the current time. After confidence level verification, dynamic capability boundary (Obtained from step S300), the measured output power of the device and the power command value issued by the control system .in, The target power determined by the device controller (such as the main control unit of a charging pile) based on load demand (such as BMS requests) and scheduling instructions, if the original unit of the instruction value is watts, also needs to be divided by the hardware multiplier factor according to the method described in step S103. This is converted to standard equivalent unit values ​​to maintain dimensional consistency; specifically, The value is the smaller of the current load demand power and the grid dispatch limit power, in order to exclude power reduction caused by non-equipment reasons due to insufficient demand at the load end (such as a vehicle about to be fully charged).

[0160] The system calculates the theoretical target power at the current moment. This value is the smaller of the dynamic capability boundary and the power command value.

[0161] This logic ensures the rationality of the evaluation benchmark: when the load demand is less than the equipment capacity boundary, the load demand shall prevail; when the load demand exceeds the equipment capacity boundary, the physical boundary of the equipment shall prevail.

[0162] Subsequently, the instantaneous physical performance gap was calculated. That is, the difference between the theoretical target power and the measured output power:

[0163] ;

[0164] Introduced here The function is designed to prevent negative values ​​caused by measurement noise from interfering with the cumulative results.

[0165] Step S502: Synthesize the effective performance loss after compensation.

[0166] The indicator generation module 104 utilizes the performance compensation coefficient generated in step S400. For instantaneous physical performance gap After weighted correction, the effective performance loss value is obtained. The calculation logic is as follows:

[0167] ;

[0168] According to the definition of step S400, when it is determined that the performance limitation is caused by an external power grid, ,at this time , making Zeroing out means that the performance deficit at that moment is exempted and not attributed to equipment responsibility; when the cause is determined to be the equipment itself... ,at this time This means that the performance shortfall at that moment is fully accounted for as equipment responsibility. This step enables automated data cleaning based on physical attribution.

[0169] Step S503: Time-domain integration and index aggregation.

[0170] To evaluate the overall performance of the equipment over a period of observation TT, the index generation module 104 discretizes and accumulates the effective performance loss value and the theoretical target power in the time domain. Observation period The duration of a single charging session can be set according to business needs, such as 24 hours or one month.

[0171] Step S504: Generate Normalized Performance Deviation (UPDI).

[0172] The final evaluation formula for the system is based on the UPDI metric, defined as the effective energy delivery rate of the equipment during the evaluation period. The calculation formula is as follows:

[0173] ;

[0174] in:

[0175] This represents the total number of sampling points within the observation period.

[0176] The sampling time interval (e.g., 1 second or 1 minute);

[0177] Numerator This represents the undelivered energy (unit: kWh) that was clearly attributable to equipment malfunction or performance defects during the evaluation period.

[0178] denominator This represents the total energy (in kWh) that the equipment should theoretically deliver during the evaluation period, based on current environmental conditions and load requirements.

[0179] To prevent tiny positive numbers with a denominator of zero (e.g., 10) -6 ), used to handle the computational stability of the device when it is in standby or no-load state.

[0180] The value range of UPDI is [0,1].

[0181] When UPDI≈1, it means that after eliminating external power grid interference, the actual output of the device closely matches its physical capacity boundary or load requirements, and the device is in a healthy operating state.

[0182] When UPDI is significantly less than 1 (e.g., below 0.9), it indicates that the device has an intrinsic performance bottleneck, such as premature derating due to decreased heat dissipation efficiency, partial damage to the power module, or sluggish control response.

[0183] This indicator allows the operation and maintenance system to directly identify inefficient equipment with good environmental conditions but weak output, without falsely reporting normal equipment that is forced to be derated due to low grid voltage or extremely high ambient temperature.

[0184] Example 1:

[0185] To more intuitively illustrate the operating logic of the present invention under actual working conditions, the following explanation is based on a specific charging station scenario.

[0186] Scene background:

[0187] The No. 5 DC charging pile (target device i) in a charging station located in the city center area was selected as the evaluation object.

[0188] Equipment parameters: Rated power =120kW, rated voltage =380V.

[0189] Time period: 14:00-14:15 on July 15th (high temperature and high load period).

[0190] Neighborhood environment: There are 10 charging piles of the same model in this site, and the other 9 are operating normally.

[0191] The sequence of events:

[0192] At 14:05, due to a surge in regional power grid load, the grid input voltage dropped to 340V (approximately 89% of the rated value). Simultaneously, charging pile number 5 was charging an electric bus with a battery SOC of 20% (BMS power requirement 120kW). At this time, the cooling fan at pile number 5, due to dust accumulation, slowed from the commanded 3000RPM to 2000RPM, causing an abnormal increase in the internal IGBT junction temperature.

[0193] The processing procedure of this invention is as follows:

[0194] S100, Data Standardization:

[0195] The system collected =340V, internal temperature =85℃ (abnormally high) Actual output power =90kW. The system maps it to a standard equivalent unit.

[0196] S200, Boundary Calculation:

[0197] Based on voltage constraint function Due to the reduced voltage, the input current is limited, and the theoretical capacity under voltage limitation is calculated to be 107kW.

[0198] Based on temperature derating function If we only look at the measured temperature, the equipment is already in the high-temperature derating range.

[0199] Initial dynamic capability boundary The value is temporarily calculated as a lower value.

[0200] S300, Verification Analysis:

[0201] The system obtained an average external ambient temperature of 38℃ from the other 9 devices in the vicinity. The actual measured external temperature of pile No. 5 was 39℃, which is within the threshold range, indicating that the external temperature sensor is functioning normally. At this time, It is still subject to physical constraints.

[0202] S400, Attribution and Compensation:

[0203] The system constructs feature vectors and calculates difference vectors. .

[0204] External feature projection Voltage drop detected, projection value is large.

[0205] Internal feature projection Anomalies were detected in both "fan speed deviation" and "temperature rise rate," with similarly large projected values. Furthermore, in this case, the internal feature projection value... The calculated value is significantly greater than the external characteristic projection value (because although the voltage drop exists, it is not enough to cause such a severe temperature surge; the dominant factor is fan failure).

[0206] determination: The responsibility was determined to be with the equipment itself.

[0207] Output: Performance Compensation Coefficient (No exemption).

[0208] S500, Indicator Generation:

[0209] Theoretical target: Although the voltage limits the capacity to 107kW, the actual output is only 90kW due to fan failure.

[0210] Performance gap: =107kW−90kW=17kW.

[0211] Effective loss: due to Effective loss =17kW.

[0212] Result: The UPDI index dropped significantly during this period, accurately identifying a fault in the equipment that was "not performing as it should" rather than simply attributing its performance degradation to grid fluctuations.

[0213] Example 2:

[0214] To verify the effectiveness of the present invention, a backtest was conducted using a real historical dataset provided by a charging operator.

[0215] Experimental setup:

[0216] Data set: Contains full operational data of 50 120kW DC charging piles over 30 days, with a sampling frequency of 1Hz.

[0217] Fault injection: Two typical fault labels are manually injected into the dataset:

[0218] Type A (External Interference): Grid voltage fluctuations, extreme high temperature weather (not equipment responsibility).

[0219] Type B (Internal Fault): Heat sink blockage, power module aging, control board communication delay (equipment responsibility).

[0220] Comparison method:

[0221] Method 1 (Traditional Utilization Rate Method): Only calculate the actual power / rated power.

[0222] Method 2 (Static Threshold Method): Data is removed only when the temperature exceeds a fixed threshold, without performing dynamic boundary calculations.

[0223] Method 3 (the method of this invention): UPDI index based on dynamic boundary and multimodal attribution.

[0224] Analysis of experimental results:

[0225] (1) Fitting effect of dynamic boundary:

[0226] Appendix Figure 8 The figure shows a comparison of the power curves of a certain device during grid fluctuations. The horizontal dashed line represents the rated power reference of the device (120kW); the solid line represents the dynamic capability boundary calculated by the present invention based on the physical model; and the dotted line represents the actual output power curve of the device.

[0227] from Figure 8 As can be seen, the grid voltage drops between T=100s and T=300s. The traditional rated power benchmark (dashed line) remains unchanged, causing the "utilization rate" to appear to decrease during this period. However, the dynamic capability boundary (solid line) calculated by this invention automatically adjusts downwards following physical constraints and closely matches the actual output power (dotted line). The calculated UPDI index remains around 1.0, indicating that this invention successfully avoids misjudging power drops caused by grid issues as equipment failure.

[0228] (2) Accuracy of fault attribution:

[0229] Appendix Figure 9 The diagram shows a scatter plot of fault clustering in a multidimensional feature space. In the figure, the X-axis represents the normalized voltage ripple feature; the Y-axis represents the normalized temperature rise rate feature. Circular markers represent injected type A fault samples (external interference); triangular markers represent injected type B fault samples (internal faults); the solid line in the middle represents the attribution decision boundary determined by the algorithm of this invention.

[0230] Experimental results show that external interference samples (circles) mainly cluster in the high-value region of the X-axis, while internal fault samples (triangles) mainly cluster in the high-value region of the Y-axis. The projection determination algorithm of this invention can clearly separate the two types of labels on both sides of the decision boundary, achieving an attribution accuracy of 98.5%.

[0231] (3) Comparison of the effectiveness of evaluation indicators:

[0232] Appendix Figure 10 The bar charts show the performance ratings of three methods in different scenarios, distinguished by different fill patterns. White-filled bars represent Method 1 (traditional utilization method); diagonally filled bars represent Method 2 (static threshold method); and black-filled bars represent Method 3 (UPDI of this invention).

[0233] The data comparison is shown in the table below:

[0234] Scene Method 1 (white bar) scoring Method 2 (Diagonal Column) Scoring This invention (black bar) rating Evaluation conclusions Scenario 1: Low grid voltage, equipment operating normally 0.75 (Low, misjudgment) 0.75 (Low, misjudgment) 0.99 (High, accurate) This invention provides for exemption from liability. Scenario 2: Suitable environment, equipment fan malfunction 0.60 (Low) 0.85 (High, missed detection) 0.60 (Low, accurate) This invention has error detection capability.

[0235] As shown in the figure, in scenario 1, the present invention (black bar) can correctly identify objective limitations and is given a high score; in scenario 2, the present invention can keenly capture the power gap of "should have been issued but not issued" and is given a low score.

[0236] Simulation results show that the UPDI index proposed in this invention reduces the false alarm rate by 45% and improves the fault detection rate by 30% under complex operating conditions compared to traditional methods. This system effectively eliminates the evaluation blind spots of a single data source through the fusion analysis of multimodal data, providing a reliable basis for the refined operation and maintenance of new energy equipment.

[0237] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing performance indicators for new energy equipment based on multimodal fusion, characterized in that, Includes the following steps: Collect multimodal data from the target device and map it to standard equivalent unit values; A physical constraint model is constructed based on the multimodal data, and the initial dynamic capability boundary of the target device based on the standard equivalent unit values ​​is calculated. Obtain neighborhood environment data within the target device's spatial neighborhood, and based on this, perform confidence verification and correction on the environmental parameters in the multimodal data to obtain the corrected dynamic capability boundary; Construct the state feature vector of the target device, determine the performance loss attribution category based on the projection of the difference between the target device and the reference vector on different feature dimensions, and generate performance compensation coefficients. Based on the actual output data in the corrected dynamic capability boundary, the performance compensation coefficient, and the standard equivalent unit values, a normalized performance deviation index is generated.

2. The method for constructing performance indicators for new energy equipment based on multimodal fusion according to claim 1, characterized in that, The acquisition of multimodal data from the target device, mapped to standard equivalent unit values, includes: The multimodal data is analyzed to obtain the rated power, real-time operating data, and static parameters. Calculate the ratio of the rated power to the preset global reference power to obtain the hardware multiplier coefficient; Using the hardware scaling factor as a normalization factor, the output power data and output current data in the real-time operating data are divided by the hardware scaling factor to convert them into the standard equivalent unit values; for the voltage data and temperature data in the multimodal data, the original physical units are maintained.

3. The method for constructing performance indicators for new energy equipment based on multimodal fusion according to claim 2, characterized in that, The step of constructing a physical constraint model based on the multimodal data and calculating the initial dynamic capability boundary of the target device based on the standard equivalent unit values ​​includes: The measured internal key point temperature and measured input voltage are obtained from the multimodal data; Construct a temperature derating function, which is configured to determine the rate of decrease in the current carrying capacity of the power device under different internal temperature ranges. Construct a grid voltage constraint function, wherein the grid voltage constraint function is configured to determine a limiting proportional coefficient that restricts the output power as the voltage decreases when the input voltage is lower than the lower limit of the rated voltage. Substitute the measured internal key point temperature and the measured input voltage into the temperature derating function and the power grid voltage constraint function, respectively, to calculate the corresponding temperature derating coefficient and voltage constraint coefficient. The minimum value between the temperature derating factor and the voltage constraint factor is selected and used as the initial dynamic capability boundary, or the minimum value is multiplied by a preset unit reference value to obtain the initial dynamic capability boundary.

4. The method for constructing performance indicators for new energy equipment based on multimodal fusion according to claim 3, characterized in that, The process of performing confidence verification and correction on the environmental parameters in the multimodal data based on this, to obtain the corrected dynamic capability boundary, includes: Calculate the arithmetic mean of the external ambient temperature of all devices within the spatial neighborhood, and use it as the neighborhood ambient mean. The measured external ambient temperature is obtained from the multimodal data, the absolute value of its deviation from the mean value of the neighborhood environment is calculated, and it is determined whether the absolute value of the deviation exceeds a preset temperature deviation threshold. If the temperature deviation threshold is exceeded, the measured external ambient temperature is replaced by the mean value of the neighborhood environment to obtain the corrected ambient temperature. The current power loss of the target device is calculated based on the real-time operating data, and the corrected internal key point temperature is derived by superimposing the product of the corrected ambient temperature, the preset thermal resistance parameter, and the current power loss. Substitute the corrected internal key point temperature into the physical constraint model to recalculate the corrected dynamic capability boundary.

5. The method for constructing performance indicators for new energy equipment based on multimodal fusion according to claim 1, characterized in that, The process of determining the spatial neighborhood includes: Obtain the geographic coordinates and device type label of the target device from the multimodal data; A circular area is defined with the geographic coordinates as the center and a preset distance as the radius; Traverse all devices within the circular spatial region, and filter out devices that are compatible with the device type label to form the spatial neighborhood.

6. The method for constructing performance indicators for new energy equipment based on multimodal fusion according to claim 1, characterized in that, The constructed state feature vector of the target device includes: Based on the multimodal data, external grid-side characteristics and internal equipment-side characteristics are calculated; the external grid-side characteristics include at least input voltage deviation and input voltage ripple; the internal equipment-side characteristics include at least temperature rise rate, fan speed deviation, and efficiency degradation. The extracted external power grid side features and internal equipment side features are standardized to generate dimensionless standard scores. The state feature vector is constructed by combining the standard components of all features. The reference vector is set to a vector with all elements being zero, representing an ideal, fault-free operating state.

7. The method for constructing performance indicators for new energy equipment based on multimodal fusion according to claim 6, characterized in that, The process of determining the performance loss attribution category based on the projection of its difference with the reference vector across different feature dimensions and generating performance compensation coefficients includes: Calculate the difference vector between the state feature vector and the reference vector; Define an external power grid weight vector and an internal equipment weight vector. Use the external power grid weight vector to extract the components of the differential vector on the feature dimension of the external power grid side, and calculate the external influence magnitude. Use the internal equipment weight vector to extract the components of the differential vector on the feature dimension of the internal equipment side, and calculate the internal influence magnitude. Compare the magnitudes of the external influence modulus and the internal influence modulus; When the external influence magnitude is greater than the product of the internal influence magnitude and the preset sensitivity coefficient, it is determined that the performance limitation is dominated by external factors, and the performance compensation coefficient is set to a value of one. Otherwise, if the performance limitation is determined to be mainly due to internal factors of the equipment, the performance compensation coefficient will be set to zero.

8. The method for constructing performance indicators for new energy equipment based on multimodal fusion according to claim 1, characterized in that, The process of generating a normalized performance deviation index based on the actual output data from the corrected dynamic capability boundary, the performance compensation coefficient, and the standard equivalent unit values ​​includes: Obtain the control power command value of the target device and convert it into the standard equivalent unit value; The smaller of the corrected dynamic capability boundary and the converted control power command value is selected as the theoretical target power; Calculate the difference between the theoretical target power and the actual output data. If the difference is less than zero, set it to zero to obtain the instantaneous physical performance gap. Calculate the difference between the first value and the performance compensation coefficient, and multiply the difference as a weighting factor with the instantaneous physical performance gap to obtain the effective performance loss value; The effective performance loss value and the theoretical target power during the observation period are calculated by time accumulation.

9. The method for constructing performance indicators for new energy equipment based on multimodal fusion according to claim 8, characterized in that, The normalized performance deviation index is calculated in the following manner: Calculate the cumulative total of the effective performance loss value during the observation period; Calculate the cumulative total of the theoretical target power within the observation period, and add a small positive number to prevent division by zero error to obtain the denominator term; Calculate the ratio of the cumulative total to the denominator term; The normalized performance deviation index is obtained by subtracting the ratio from the numerical value.

10. A system for constructing performance indicators for new energy equipment based on multimodal fusion, used to execute the method for constructing performance indicators for new energy equipment based on multimodal fusion as described in any one of claims 1-9, characterized in that, include: The data standardization module is configured to collect multimodal data from the target device and map it to standard equivalent unit values. The boundary calculation module is configured to construct a physical constraint model based on the multimodal data and calculate the initial dynamic capability boundary of the target device based on the standard equivalent unit values. The verification and analysis module is configured to acquire neighborhood environment data within the spatial neighborhood of the target device, and based on this, perform confidence verification and correction on the environmental parameters in the multimodal data to obtain the corrected dynamic capability boundary; the verification and analysis module is also configured to construct the state feature vector of the target device, determine the performance loss attribution category based on the projection of the difference between it and the reference vector in different feature dimensions, and generate performance compensation coefficients. The indicator generation module is configured to generate a normalized performance deviation indicator based on the actual output data in the modified dynamic capability boundary, the performance compensation coefficient, and the standard equivalent unit values.

Images