Data-driven rural photovoltaic energy storage integration device operation evaluation system

Abstract

The present application relates to the technical field of device management, and discloses a data-driven rural photovoltaic energy storage integrated device operation evaluation system, comprising: collecting multi-source telemetry data to construct a reconstructed aligned unified state field, thereby solving the reference and extracting the photovoltaic pollution residual error and energy transfer path; reconstructing the total equivalent heat source of the battery and the power conversion unit, combining the environmental temperature difference to fit the heat balance equation, obtaining the thermal channel impedance and the thermal hysteresis loop area; fusing the meteorological dew point condition to calculate the agricultural moisture aerosol thermal channel hysteresis parameter representing the hysteresis effect; converting the operation pressure index containing the hysteresis parameter into a normalized score, obtaining the comprehensive operation evaluation score by adaptive information entropy, and finally generating the account according to the deviation contribution degree and updating the time series prediction model. The method effectively identifies the device operation degradation risk caused by environmental accumulation and improves the limitations of conventional evaluation methods.

Description

Technical Field

[0001] This invention relates to the field of equipment management technology, and more specifically, to a data-driven operation evaluation system for integrated rural photovoltaic energy storage equipment. Background Technology

[0002] In rural settings, distributed photovoltaic (PV) and energy storage integrated equipment is often deployed in open spaces such as tiled roofs, farm machinery sheds, or outdoor courtyards. During busy farming seasons, aerosols such as dust, straw smoke, or feed powder easily adhere to the equipment casing and component surfaces, forming a wet deposit layer that is difficult to remove by natural wind under the influence of high humidity dew at night or in the early morning. This deposit not only reduces the effective energy absorption of the PV panels but also slowly leads to obstruction of the heat exchange channels and an increase in thermal resistance in outdoor energy storage cabinets. However, existing data-driven evaluation methods mostly focus on conventional electrical parameters (such as battery state of charge, daily power generation, or charge-discharge cycle count), often simply attributing output deviations or abnormal temperature rises under the same irradiance conditions to random weather changes and load fluctuations. Due to the lack of a mechanism to identify the chronic degradation of internal heat dissipation channels, existing evaluation systems often fail to assess the true operational losses and risk accumulation before the equipment triggers a direct high-temperature shutdown alarm, resulting in significant monitoring blind spots. Summary of the Invention

[0003] This invention provides a data-driven operation evaluation system for rural photovoltaic energy storage integrated equipment, which solves the technical problems mentioned in the background art.

[0004] This invention provides a data-driven operation evaluation system for integrated rural photovoltaic energy storage equipment, applicable to an interactive system comprising distributed photovoltaic arrays, outdoor integrated energy storage cabinets, and agricultural loads, configured to execute:

[0005] Collect multi-source operating parameters including meteorological, electrical, thermal, and agricultural conditions, and construct an aligned unified state field;

[0006] Based on the unified state field solution of photovoltaic and load benchmarks, the photovoltaic output loss and equipment energy flow path characterizing the environmental deposition on the component surface are extracted;

[0007] The total equivalent heat source of the energy storage battery and power conversion unit is reconstructed based on the electrothermal characteristic parameters in the unified state field.

[0008] Based on the heat transfer balance model fitted by the total equivalent heat source and the temperature difference inside and outside the equipment, the thermal channel impedance and thermal hysteresis area of ​​the cabinet are obtained.

[0009] By coupling the photovoltaic power loss with the meteorological moisture deposition factor, the degradation parameters of the agricultural moisture aerosol thermal channel characterizing the environmental hysteresis coupling effect are calculated.

[0010] The multi-dimensional operational evaluation index, which includes the degradation parameters of the agricultural wet aerosol thermal channel, is transformed into a normalized score. The index weights are adaptively obtained through the historical distribution variance to obtain the comprehensive operational score.

[0011] An evaluation log is generated based on the deviation contribution of each indicator, and the prediction network is updated using the fitting residual constraint of the heat transfer balance model to output the state prediction result for the next cycle.

[0012] The beneficial effects of this invention are as follows: By incorporating multi-source meteorological and agricultural conditions in rural areas into the evaluation process, and by extracting photovoltaic pollution residuals and reconstructing equivalent heat sources, the slow deterioration state of equipment heat dissipation channels caused by agricultural dust and moisture is effectively separated. This invention no longer relies on preset fixed alarm thresholds, avoiding misjudging the phenomenon of different outputs under the same irradiance caused by aerosol pollution as normal weather or load fluctuations. It achieves quantitative perception of latent aging risks before the equipment triggers a visible high-temperature alarm, effectively improving the accuracy of the evaluation results and their practical operation and maintenance guidance value. Attached Figure Description

[0013] Figure 1 This is a flowchart of the operation evaluation system for the data-driven rural photovoltaic energy storage integrated equipment of the present invention. Detailed Implementation

[0014] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0015] like Figure 1 As shown, the data-driven rural photovoltaic energy storage integrated equipment operation evaluation system is applied to an interactive system that includes distributed photovoltaic arrays, outdoor integrated energy storage cabinets, and agricultural loads, and is configured to execute:

[0016] Collect multi-source operating parameters including meteorological, electrical, thermal, and agricultural conditions, and construct an aligned unified state field;

[0017] Based on the unified state field solution of photovoltaic and load benchmarks, the photovoltaic output loss and equipment energy flow path characterizing the environmental deposition on the component surface are extracted;

[0018] The total equivalent heat source of the energy storage battery and power conversion unit is reconstructed based on the electrothermal characteristic parameters in the unified state field.

[0019] Based on the heat transfer balance model fitted by the total equivalent heat source and the temperature difference inside and outside the equipment, the thermal channel impedance and thermal hysteresis area of ​​the cabinet are obtained.

[0020] By coupling the photovoltaic power loss with the meteorological moisture deposition factor, the degradation parameters of the agricultural moisture aerosol thermal channel characterizing the environmental hysteresis coupling effect are calculated.

[0021] The multi-dimensional operational evaluation index, which includes the degradation parameters of the agricultural wet aerosol thermal channel, is transformed into a normalized score. The index weights are adaptively obtained through the historical distribution variance to obtain the comprehensive operational score.

[0022] An evaluation log is generated based on the deviation contribution of each indicator, and the prediction network is updated using the fitting residual constraint of the heat transfer balance model to output the state prediction result for the next cycle.

[0023] This implementation is applied to an interactive system that includes distributed photovoltaic arrays, outdoor integrated energy storage cabinets, and agricultural loads. The system hardware includes an edge computing terminal, a multi-source data acquisition module, a meteorological monitoring unit, a battery management system, a power conversion control unit, and a substation metering terminal. The edge computing terminal performs the entire process of this solution. The multi-source data acquisition module achieves full terminal clock synchronization through the NTP network time protocol, with a synchronization error not exceeding 10% of the basic sampling time step. Data transmission adopts Modbus RTU, MQTT, or HTTP / HTTPS protocols, adapting to common transmission standards in industrial scenarios.

[0024] Simultaneous acquisition of multi-source operating parameters, including measured photovoltaic power. Irradiance Ambient temperature relative humidity Wind speed AC power DC power Battery terminal voltage Battery terminal current Battery State of Charge (SOC), Battery State of Health (SOH), Cell Temperature Cabinet temperature Surface temperature of the outer casing Auxiliary equipment heat dissipation power Distribution station node voltage Rated voltage of the transformer area Actual agricultural load Agricultural schedule agency parameters Local solar geometry angle and calendar time tags The measured photovoltaic power is in kilowatts (kW), collected by the built-in metering module of the photovoltaic inverter; the irradiance is in watts per square meter (W / m²), collected by the total radiation meter installed next to the photovoltaic array; the ambient temperature is in degrees Celsius (°C), collected by the platinum resistance temperature sensor of the meteorological monitoring unit; the relative humidity is in percentage (%), collected by the humidity sensor of the meteorological monitoring unit; the wind speed is in meters per second (m / s), collected by the wind speed sensor of the meteorological monitoring unit; the AC power is in kilowatts (kW), collected by the AC side metering module of the power conversion unit; and the DC power is in kilowatts (kW), collected by the DC side metering module of the power conversion unit. Block data acquisition; battery terminal voltage is in volts, acquired by the battery management system; battery terminal current is in amperes, acquired by the shunt of the battery management system, with the current sign defined as positive during battery discharge and negative during charging; battery state of charge (SOC) is a dimensionless parameter, ranging from 0 to 1, calculated and output by the battery management system; battery health status is a dimensionless parameter, ranging from 0 to 1, calculated and output by the battery management system; cell temperature is in degrees Celsius, acquired by the cell-level temperature sensor of the battery management system, with the average of all cell temperatures used in the calculation; cabinet temperature is in degrees Celsius. Temperature data is collected by multiple temperature sensors located inside the energy storage cabinet, and the average temperature of all measuring points is used for calculation. The outer surface temperature, in degrees Celsius, is collected by surface temperature sensors installed on the outer wall of the energy storage cabinet. The heat dissipation power of auxiliary equipment, in kilowatts, is collected by the metering module of the power supply circuit for auxiliary equipment such as cooling fans and temperature control units. The voltage at the transformer substation node, in volts, is collected by the metering terminal at the transformer substation outlet. The rated voltage of the transformer substation, in volts, is the nominal rated voltage of the transformer substation and is a fixed value. The measured agricultural load, in kilowatts, is collected by the metering terminal of the agricultural load circuit in the transformer substation. Data collection; the agricultural schedule proxy parameter is a dimensionless parameter, ranging from 0 to 1, and is assigned a value based on the type of agricultural activity. Activities that easily generate aerosols, such as irrigation, agricultural machinery operation, straw processing, and feed processing during busy seasons, are assigned a value of 0.8 to 1; routine field management is assigned a value of 0.3 to 0.7; and the off-season (no agricultural activities) is assigned a value of 0 to 0.2. This parameter is updated synchronously by the agricultural schedule plan. The local solar geometric angle is a dimensionless parameter, calculated from the local latitude and longitude and calendar time stamp using an astronomical formula. The calendar time stamp is a timestamp containing year, month, day, hour, minute, and second, and is output synchronously by all data collection terminals. The above parameters are then used at the i-th original timestamp. The following forms the original telemetry vector, expressed as:

[0025]

[0026] in This is the i-th original timestamp, in seconds. Let be the original telemetry vector under the i-th original timestamp. All acquired parameters correspond to the same original timestamp to ensure the time synchronization of the initial acquisition.

[0027] The median of the differences between adjacent timestamps of multi-source operating parameters is selected as the basic sampling time step to generate a unified evaluation clock. The calculation formula is as follows:

[0028]

[0029] in To assess the total number of sampling points within the assessment window, the assessment window is fixed as a single day, with a time range from 00:00 to 23:59 on that day; This is the median calculation function. The basic sampling time step, measured in seconds, typically ranges from 1 to 300 seconds and can be adaptively adjusted based on the sampling frequency of the acquisition terminal. By selecting the median difference between adjacent timestamps as the basic sampling time step, the problem of uneven time steps caused by packet loss or sampling frequency fluctuations in some acquisition terminals is avoided, adapting to scenarios where multiple source devices have inconsistent sampling frequencies. A unified evaluation clock containing a discrete integer index k is generated based on the basic sampling time step, expressed as follows:

[0030]

[0031] in To ensure a unified evaluation of the k-th discrete moment in time (in seconds, where k is a positive integer), the starting point of the unified evaluation clock is consistent with the start time of the daily evaluation window, and the ending point is consistent with the end time of the daily evaluation window.

[0032] A smooth reconstruction algorithm is used to align multi-source operating parameters to a unified evaluation clock, resulting in the full observation sequence. The smooth reconstruction algorithm employs a Gaussian kernel smoothing interpolation algorithm, adapted to the alignment of non-uniformly sampled multi-source data. The calculation formula is as follows:

[0033]

[0034] in To unify the evaluation of aligned observations at the k-th time point and the j-th parameter dimension under the clock; This represents the original acquired value of the i-th original timestamp and the j-th parameter dimension in the original telemetry vector. Let be the Gaussian kernel function, with the expression: Here, 'u' is the input variable of the kernel function, and the kernel function's cutoff bandwidth is set to 3 times the base sampling time step. Original sampling points exceeding the cutoff bandwidth have their weights reset to zero to reduce computational complexity. For parameter sequences with consecutive packet loss exceeding 3 sampling steps, linear interpolation combined with the historical mean of the same period is used for completion. Completed data are marked as outliers, and a weight decay coefficient of 0.5 is set in subsequent calculations. For outlier data exceeding 3 times the median absolute deviation of the historical normal operating range, the median of adjacent valid sampling points is used for replacement to avoid interference from outliers on the data alignment results. After alignment, the observations of all parameter dimensions constitute the full observation sequence. , This is a column vector containing all parameter dimensions, corresponding to the full observation data at time k under a unified evaluation clock.

[0035] A dimensionless scaling process is applied to the full observation sequence using a standardized function that includes estimates of the statistical median and robust variance, resulting in a unified state field. The calculation formula is as follows:

[0036]

[0037] in To standardize the value of the j-th parameter dimension at the k-th time step in the unified state field; is the statistical median of the j-th parameter dimension over a continuous 90-day historical operating cycle. Median statistics can avoid the impact of extreme outliers on the standardization results. Let be the median absolute deviation of the j-th parameter dimension over a continuous 90-day historical period, and be a robust variance estimator, calculated using the following formula: ,in This represents the full historical sample of the j-th parameter dimension; 1.4826 is the correction coefficient for converting the median absolute deviation to the normal distribution standard deviation, adapting to the distribution characteristics of conventional operating parameters. For cold start scenarios with no historical data for newly commissioned equipment, historical operating data from the same model of equipment in the same region and under the same climatic conditions are used as the benchmark statistical sample. After the equipment has been running for 7 days, this data is gradually replaced with the equipment's own operating data. After standardization, the values ​​of all parameter dimensions are mapped to the same dimension space, eliminating dimensional and numerical differences between different parameter dimensions. The standardized values ​​of all parameter dimensions constitute a unified state field, providing a unified data source for all subsequent calculations.

[0038] The irradiance, ambient temperature, relative humidity, wind speed, local solar geometry, and agricultural schedule proxy parameters in the unified state field are input into the photovoltaic power prediction baseline model, and the ideal photovoltaic power generation baseline is output. The calculation formula is as follows:

[0039]

[0040] in To standardize the evaluation of the ideal photovoltaic power generation baseline at the k-th time point under the clock, the unit is kilowatt, which characterizes the theoretical output power of the photovoltaic array under the condition of no component pollution and no equipment failure; For the baseline model of photovoltaic power prediction, a gradient boosting tree model is adopted. The model input feature dimension is 6, corresponding to the original physical quantities of irradiance, ambient temperature, relative humidity, wind speed, local solar geometry angle, and agricultural schedule proxy parameters. The model is set with 200 trees, a maximum tree depth of 8, a learning rate of 0.05, a subsample sampling ratio of 0.8, and a feature sampling ratio of 0.8. The model training data uses historical operating data since the target photovoltaic array was put into operation. Before training, abnormal data such as component contamination, equipment failure, and abnormal grid periods are removed. The rule for removing abnormal data is that the measured photovoltaic power is lower than 70% of the historical average under the same irradiance conditions, or exceeds 130% of the historical average. The model training adopts five-fold cross-validation, with the goal of minimizing the mean square error. The model convergence condition is that the mean square error of the validation set decreases by less than 1e-6 for 50 consecutive iterations. For newly commissioned arrays without sufficient historical data, a pre-trained model of the same model, installation tilt angle, and region photovoltaic array is used, and fine-tuned by combining it with 7 days of operating data of the new array. As an alternative implementation method, A physical model of the equivalent circuit of a single diode in a photovoltaic module can be used. The model parameters are obtained by combining the factory parameters provided by the photovoltaic module manufacturer with the test results of open-circuit voltage and short-circuit current on site. These are the original physical quantities corresponding to the standardized values ​​of irradiance, ambient temperature, relative humidity, wind speed, local solar geometric angle, and agricultural schedule proxy parameters at the k-th moment in the unified state field.

[0041] Calculate the ratio of the relative deviation between the ideal photovoltaic power generation baseline and the measured photovoltaic power, and extract the positive relative deviation that characterizes the power generation degradation as the photovoltaic output loss. The calculation formula is as follows:

[0042]

[0043] in To uniformly evaluate the photovoltaic power output loss at the k-th moment under the clock, a dimensionless parameter is used, with a value ranging from 0 to 1. The larger the value, the more severe the power generation degradation of the photovoltaic module is due to the deposition of the surface environment. To standardize the evaluation of the measured photovoltaic power at the k-th moment under the clock, the unit is kilowatt; This is a minimum constant, set to 1e-3, used to avoid division by zero errors when the ideal photovoltaic power generation baseline is 0. The max function filters positive relative deviations, retaining only the attenuation portion where the measured photovoltaic power is lower than the ideal photovoltaic power generation baseline, and eliminating negative deviation interference caused by sudden changes in irradiance or measurement errors. For low-irradiance conditions (below 100 watts per square meter) and no-irradiance conditions at night, the photovoltaic output loss is directly set to 0 to avoid calculation errors under low signal-to-noise ratio conditions.

[0044] The ambient temperature, calendar time stamps, agricultural schedule proxy parameters, and transformer node voltage are input into the load prediction baseline model to output the ideal rural load baseline. The calculation formula is as follows:

[0045]

[0046] in To standardize the evaluation of the ideal rural load baseline at time k under the clock, the unit is kilowatt, which represents the theoretical power consumption of rural agricultural load under normal agricultural production conditions; A gated recurrent unit (GRU) time-series forecasting network was used as the baseline model for load forecasting. The network consisted of one encoding layer and 64 hidden neurons, with ReLU activation. The input sequence length was 24 hours, and the output length was one sampling step. Model input features included ambient temperature, hourly features extracted from calendar time stamps, date features, agricultural schedule proxy parameters, and raw physical values ​​of transformer node voltages. Training data consisted of historical agricultural load data from the target transformer area. Abnormal data due to equipment failures or power outages were removed before training. The Adam optimizer was used with an initial learning rate of 1e-3, a batch size of 32, and 100 training epochs. The training objective was to minimize the mean absolute percentage error (MASE), with the learning rate decreasing to 0.5 every 20 epochs. Sub-models were constructed for busy and off-season farming periods. The busy season sub-model was activated when the agricultural schedule proxy parameter was greater than 0.5, and the off-season sub-model was activated when it was less than or equal to 0.5, adapting to the seasonal fluctuations in rural agricultural load. These are the original physical quantity values ​​corresponding to the standardized values ​​of the ambient temperature, calendar time tag, agricultural schedule proxy parameter, and transformer node voltage at the k-th moment in the unified state field.

[0047] Based on the product of battery terminal voltage and battery terminal current, the energy storage charging power and energy storage discharging power are separated according to the power flow direction. The calculation formula is as follows:

[0048]

[0049] in To standardize the evaluation of energy storage charging power at the k-th moment under the clock, the unit is kilowatt; To standardize the evaluation of the energy storage discharge power at the k-th moment under the clock, the unit is kilowatt, and both are non-negative values; To standardize the evaluation of the battery terminal voltage at the k-th moment under the clock, the unit is volts; To standardize the evaluation of the battery terminal current at the k-th moment under the clock, the unit is amperes; The dimensional conversion factor is used to convert volt-amperes to kilowatts. The power values ​​under charging and discharging conditions are extracted separately using the max function to achieve decoupling of charging and discharging power. For the static condition where the absolute value of the battery terminal current is less than 0.01 times the rated current, both charging power and discharging power are set to 0.

[0050] The time-integrated interval between the measured photovoltaic power and the energy storage charging power yields the amount of electricity transferred from photovoltaic to energy storage. The calculation formula is as follows:

[0051]

[0052] in The amount of electricity transferred from the photovoltaic array to the energy storage battery during the d-th daily evaluation period is expressed in kilowatt-hours; D is the set of all discrete moments k corresponding to the unified evaluation clock during the daily evaluation period. The minimum value calculation function is used to accurately match the power overlap range where photovoltaic output directly charges energy storage by taking the minimum value between the measured photovoltaic power and the energy storage charging power, thus avoiding the inclusion of grid charging power in the transfer of electricity from photovoltaic to energy storage. The basic sampling time step, in seconds; The dimensional conversion coefficients from seconds to hours are used to perform trapezoidal time integration through discrete summation.

[0053] By integrating the time interval between the energy storage discharge power and the measured agricultural load, the electricity supplied from the energy storage to the load is obtained. The calculation formula is as follows:

[0054]

[0055] in The energy storage battery supplies electricity to agricultural loads during the d-th daily evaluation period, in kilowatt-hours. To uniformly evaluate the measured agricultural load at the k-th moment under the clock, the unit is kilowatts; by taking the minimum value between the energy storage discharge power and the measured agricultural load, the overlapping range of the power directly supplied to the agricultural load by the energy storage discharge is matched, so as to avoid including the power fed to the grid in the power supplied to the load by the energy storage.

[0056] The unconsumed power integral exceeding the sum of measured photovoltaic power and energy storage charging power from the ideal photovoltaic power generation baseline is calculated, and its proportion relative to the total power generation baseline is taken as the system's potential photovoltaic curtailment rate. The calculation formula is as follows:

[0057]

[0058] in The potential photovoltaic curtailment rate of the system within the d-th daily evaluation period is a dimensionless parameter ranging from 0 to 1. A larger value indicates a weaker photovoltaic absorption capacity of the system. The numerator is the integral of the unabsorbed power within the daily evaluation period. The part exceeding the sum of the measured photovoltaic power and the energy storage charging power of the ideal photovoltaic power generation baseline is filtered by the max function. This represents the photovoltaic power that can be generated theoretically but is not utilized or absorbed by energy storage. The denominator is the total integral of the ideal photovoltaic power generation baseline within the daily evaluation period, representing the theoretical total power generation of the photovoltaic array. This is a minimum constant, with a value of 1e-3, used to avoid the error of dividing by zero when the total integral of the ideal photovoltaic power generation baseline is 0. The ratio of these two values ​​yields the relative curtailment rate, eliminating the impact of differences in the absolute value of power generation under different irradiance conditions.

[0059] Using an open-circuit voltage surface model mapped from the battery's state of charge and cell temperature, combined with measured battery terminal voltage, battery terminal current, and their time-series rate of change, the equivalent ohmic internal resistance and polarization dynamic parameters of the battery are extracted through parameter identification within a local time window. Open-circuit voltage surface model The data was obtained through offline charge-discharge testing. The testing process involved performing constant current charge-discharge tests at a rate of 0.05C on the battery at five temperature gradients: 0℃, 10℃, 25℃, 35℃, and 45℃. The battery was then left to rest for 2 hours at 5% SOC increments, and the open-circuit voltage was recorded after each rest period. A three-dimensional mapping surface between the open-circuit voltage, SOC, and cell temperature was obtained using a third-order polynomial fitting function. The fitting function expression is as follows:

[0060]

[0061] in These are the polynomial fitting coefficients, obtained through least squares fitting; Here, represents the cell temperature, in degrees Celsius. The open-circuit voltage temperature coefficient is the first-order partial derivative of the open-circuit voltage surface model with respect to the cell temperature, expressed as:

[0062]

[0063] The open-circuit voltage temperature coefficient is measured in volts per degree Celsius and typically ranges from -0.5mV / ℃ to 0.5mV / ℃, making it suitable for mainstream energy storage battery systems such as lithium iron phosphate and ternary lithium.

[0064] The formulas for identifying and calculating the battery's equivalent ohmic internal resistance and polarization dynamic parameters are as follows:

[0065]

[0066] in To standardize the evaluation of the battery's equivalent ohmic internal resistance identified at the k-th time under the clock, the unit is ohms; The battery polarization dynamic parameters identified at the same time are expressed in Henry. The optimal values ​​of R and L are obtained by minimizing the sum of squared residuals of the objective function and using the weighted least squares parameter identification function with L2 regularization. is the regularization coefficient, with a value of 1e-6, used to avoid matrix singularity problems during the identification process and improve numerical stability; The local time window corresponding to the k-th time is set to a length of 10 sampling steps, covering the dynamic response process of battery charging and discharging. For transient conditions where the rate of change of charging and discharging current is greater than 0.1C per second, the time window length is adaptively shortened to 5 sampling steps to improve the real-time performance of identification. The weighting coefficients at time r within the local time window are expressed as follows: (The weighting coefficients are given by the expression for time r). ,in The decay time constant is 3 times the sampling step size, and the weighting coefficient is larger for sampling points closer to the k-th time. , These are the battery terminal voltage and battery terminal current at the r-th time within the local time window, respectively. The time-series rate of change of the battery terminal current is expressed in amperes per second. The initial values ​​for the identification process are set to the rated internal resistance and rated polarization parameters provided by the battery manufacturer. The Levenberg-Marquardt algorithm is used for iterative solving. The convergence condition is that the parameter change amplitude is less than 1e-9 for 10 consecutive iterations. The maximum number of iterations is set to 100. If convergence is not achieved after exceeding the maximum number of iterations, the identification result from the previous time step is used to avoid non-convergent values ​​interfering with subsequent calculations. The objective function is constructed based on the first-order equivalent circuit model of the battery. Real-time extraction of the battery's internal resistance and polarization parameters is achieved through weighted least squares identification within a local time window.

[0067] The irreversible Joule heat, composed of the square of the battery terminal current and the battery's equivalent ohmic internal resistance, is calculated and then superimposed with the reversible entropy heat, composed of the battery terminal current, cell temperature, and open-circuit voltage temperature coefficient, to synthesize the heat source of the battery itself. The calculation formula is as follows:

[0068]

[0069] in To standardize the evaluation of the battery's heat generation at time k under a given clock, the unit is watts; the first term on the right-hand side of the formula represents the battery's irreversible Joule heat, generated by the current flowing through the battery's ohmic internal resistance, which is the main source of battery heat generation; the second term represents the battery's reversible entropy heat, generated by the entropy change of the electrochemical reaction, where... The temperature of the battery cell is its thermodynamic temperature, measured in Kelvin. This is the open-circuit voltage temperature coefficient, measured in volts per Kelvin, consistent with the temperature coefficient value in degrees Celsius. The total heat generation power of the battery is fully characterized by the superposition of irreversible Joule heating and reversible entropic heating.

[0070] Calculate the absolute value of the difference between the AC and DC power to extract the heat dissipation sources of the power conversion unit. The calculation formula is as follows:

[0071]

[0072] in To uniformly evaluate the heat dissipation of the power conversion unit at the k-th time under the clock, the unit is watts; This refers to the DC power output, measured in kilowatts. The AC power is expressed in kilowatts (kW); 1000 is the dimensional conversion factor from kW to watt. The operating loss of the power conversion unit is the power difference between the DC and AC sides, which is entirely converted into heat energy, forming the heat source of the power conversion unit. The calculation uses the absolute value of the power to adapt to loss calculations under both charging and discharging conditions. A max function is used to apply a non-negative constraint to the calculation results, eliminating negative loss values ​​caused by measurement errors.

[0073] The total equivalent heat source is obtained by summing the heat generated by the battery itself, the heat dissipation loss of the power conversion unit, and the heat dissipation power of auxiliary equipment. The calculation formula is as follows:

[0074]

[0075] in To standardize the evaluation of the total equivalent heat source at the k-th moment under the clock, the unit is watts; To standardize the evaluation of auxiliary equipment heat dissipation power at the k-th time point under the clock, the unit is kilowatt; 1000 is the dimension conversion factor from kilowatt to watt. The operating power consumption of auxiliary equipment is ultimately converted into heat energy and included in the total heat generation power of the energy storage cabinet. By superimposing these three values, the total heat generation power of all heat sources inside the energy storage cabinet is fully characterized, providing a foundation for the subsequent construction of the heat transfer balance model.

[0076] The average temperature inside the cabinet and the temperature of the battery cells are extracted, and the ambient temperature is subtracted to obtain the overall representative temperature rise of the system. The calculation formula is as follows:

[0077]

[0078] in To standardize the evaluation of the overall representative temperature rise of the system at the k-th time point under the clock, the unit is Celsius; To standardize the evaluation of the cabinet temperature at the k-th moment under the clock, the unit is degrees Celsius; The cell temperature at the same moment is expressed in degrees Celsius. The ambient temperature is taken at the same moment, in degrees Celsius. The average of these two values ​​can comprehensively characterize the overall temperature level inside the energy storage cabinet. By calculating the difference between the internal average temperature and the ambient temperature, the overall temperature rise of the energy storage cabinet relative to the external environment can be obtained, eliminating the influence of ambient temperature fluctuations on heat transfer analysis. Temperature measurement points inside the cabinet should be evenly distributed in the upper, middle, and lower areas of the energy storage cabinet. The cell temperature is taken as the average temperature of all cells in the battery cluster to ensure the representativeness of the temperature rise calculation.

[0079] Calculate the time-series rate of change of the system's overall representative temperature rise at adjacent sampling times. The calculation formula is as follows:

[0080]

[0081] in To uniformly evaluate the time-series rate of change of the overall system temperature at time k under the clock, the unit is degrees Celsius per second; The base sampling time step is in seconds. The rate of temperature change is obtained by dividing the difference in temperature rise between adjacent time points by the base sampling time step. This rate of change characterizes the dynamic temperature change characteristics inside the energy storage cabinet and provides dynamic feature parameters for identifying the heat transfer balance model. For the initial time k=1, the time-series rate of change is set to 0 to avoid calculation errors caused by missing initial values.

[0082] With the goal of minimizing the residuals of the heat transfer balance model equations, the equivalent heat capacity and total thermal conductivity of the cabinet are identified using the total equivalent heat source, the representative temperature rise of the entire system, and its time-series rate of change. The calculation formula is as follows:

[0083]

[0084] in The equivalent heat capacity of the cabinet identified during the d-th single-day evaluation cycle is expressed in joules per degree Celsius. The total thermal conductivity of the cabinet identified within the same cycle is expressed in watts per degree Celsius. It characterizes the convective heat transfer capacity between the energy storage cabinet and the external environment. The higher the value, the stronger the heat dissipation capacity. is the regularization coefficient, with a value of 1e-3, used to improve the numerical stability of the identification process; the objective function is constructed based on a lumped parameter heat transfer balance model, where... The heat power absorbed by the cabinet's heat capacity, measured in watts; The heat power dissipated by the cabinet to the external environment through convection heat transfer, measured in watts; the sum of these two and the total equivalent heat source The difference is the residual of the heat transfer balance equation, expressed in watts, with all addition and subtraction terms having completely uniform dimensions. The identification process employs a constrained least squares algorithm, with the following constraints: , ,in , The upper and lower limits of the equivalent heat capacity of the cabinet are calculated based on the structure, materials, and total mass of the batteries in the energy storage cabinet. , The upper and lower limits of the total thermal conductivity of the cabinet are calculated from the design parameters of the energy storage cabinet's heat dissipation duct and the rated airflow of the cooling fan. The iterative solution adopts the trust region reflection algorithm, and the convergence condition is that the parameter change is less than 1e-6 for 20 consecutive iterations. The maximum number of iterations is set to 200. If the convergence is not achieved after exceeding the maximum number of iterations, the identification result of the previous evaluation cycle is used.

[0085] The reciprocal of the total thermal conductivity is taken as the thermal path impedance, and the ratio of the cabinet's equivalent heat capacity to the total thermal conductivity is taken as the thermal time constant. The calculation formula is as follows:

[0086]

[0087] in The thermal channel impedance of the cabinet is given in the d-th single-day evaluation cycle. The unit is degrees Celsius per watt. The thermal channel impedance characterizes the heat dissipation capacity of the heat dissipation channel of the energy storage cabinet. The larger the impedance value, the weaker the heat dissipation capacity and the more serious the obstruction of the heat dissipation channel. The thermal time constant of the cabinet within the same period is expressed in seconds. The thermal time constant characterizes how fast the temperature of the energy storage cabinet responds. The larger the value, the slower the cabinet temperature responds to changes in heat generation power, and the more obvious the thermal hysteresis effect.

[0088] The integral area of ​​the hysteresis loop representing the overall temperature rise and total equivalent heat source during the charge-discharge cycle is calculated and normalized to represent the thermal hysteresis area characterizing the cabinet's thermal response delay. The charge-discharge cycle is defined as follows: a complete charge-discharge cycle is defined as the battery's state of charge rising from below 20% to above 80% and then falling back below 20%. If multiple partial charge-discharge cycles occur on a single day, the hysteresis area of ​​each cycle is calculated separately and then summed. The formula for calculating the thermal hysteresis area is as follows:

[0089]

[0090] in The thermal hysteresis area in the d-th daily evaluation cycle is a dimensionless parameter ranging from -1 to 1. The larger the absolute value, the more pronounced the hysteresis effect in the cabinet's thermal response. The numerator of the formula uses the trapezoidal integral method to calculate the integral area of ​​the hysteresis loop between the temperature rise and the total equivalent heat source during the charge-discharge cycle. The size of the hysteresis loop directly reflects the degree of hysteresis in the thermal response. A positive value indicates that the temperature rise lags behind the change in heat generation power, while a negative value indicates that the temperature rise leads the change in heat generation power. The denominator is the sum of the absolute values ​​of the integral terms. Through normalization, the influence of different charge-discharge depths and different heat generation power levels on the hysteresis area calculation results is eliminated, resulting in a thermal hysteresis characteristic parameter that can be compared across cycles.

[0091] The meteorological water vapor dew point temperature is calculated based on Magnus's empirical formula, combined with ambient temperature and relative humidity. The calculation formula is as follows:

[0092]

[0093] in To standardize the evaluation of the meteorological water vapor dew point temperature at the k-th moment under the clock, the unit is degrees Celsius. The dew point temperature is the temperature at which air is cooled to saturation under the condition that the water vapor content and air pressure remain constant, and is used to characterize the degree of water vapor saturation in the air. and is the empirical constant of Magnus's empirical formula, where , This constant is suitable for dew point temperature calculations in the range of -45℃ to 60℃; To standardize the evaluation of relative humidity at the k-th moment under the clock, the value is expressed as a percentage, with a range of 1% to 100%, to avoid calculation errors caused by logarithmic input of 0 or negative numbers; It is the natural logarithm function.

[0094] The absolute value of the difference between the outer shell surface temperature and the meteorological water vapor dew point temperature is calculated, and this absolute value is mapped to a wet deposition activity factor characterizing the surface condensation tendency using an exponential decay function. The calculation formula is as follows:

[0095]

[0096] in To uniformly evaluate the wet deposition activity factor at the k-th time under the clock, a dimensionless parameter is used, with a value ranging from 0 to 1. The larger the value, the stronger the condensation tendency on the surface of the energy storage cabinet shell, and the higher the probability of wet deposition of aerosols. To standardize the evaluation of the outer casing surface temperature at the k-th moment under the clock, the unit is degrees Celsius; The temperature variance reference value is 3℃, which is obtained from the historical fluctuation data of local ambient temperature and is used to adjust the decay rate of the exponential decay function. It is a natural exponential function. The closer the surface temperature of the shell is to the dew point temperature, the smaller the absolute value of the difference between the two, the closer the wet deposition activity factor is to 1, and the higher the probability of surface condensation and wet deposition.

[0097] The fluctuation dispersion of wind speed relative to historical stable wind speeds is calculated, and combined with photovoltaic power output loss and wet deposition activity factors, to constitute the proxy for local agricultural wet aerosol deposition intensity. The calculation formula is as follows:

[0098]

[0099] in To standardize the evaluation of the local agricultural moisture aerosol deposition intensity at the k-th time under the clock, a dimensionless parameter is used. The larger the value, the higher the deposition intensity of the local agricultural moisture aerosol, and the more serious the pollution to the heat dissipation channels of photovoltaic modules and energy storage cabinets. This is the median of the historical 30-day wind speed distribution. The median absolute deviation of wind speed over a consecutive 30-day period is calculated from historical wind speed data. The dispersion of wind speed fluctuations is the highest value of dispersion, and the stronger the diffusion and deposition effects of aerosols. This is a local constant with a value of 1e-3, used to avoid calculation errors where the denominator is 0; As a proxy parameter for agricultural activities, it is used to distinguish agricultural wet aerosols from conventional environmental dust. The more frequent the agricultural activities, the larger the proxy parameter value, and the higher the weight of aerosol deposition intensity, thus decoupling agricultural wet aerosols from conventional pollution.

[0100] Extract the time-series increment of the thermal channel impedance and calculate the weighted coupling degree of the surrogate quantity of local agricultural wetland aerosol deposition intensity to the time-series increment. The calculation formula is as follows:

[0101]

[0102] in , is the weighted coupling degree between the local agricultural wet aerosol deposition intensity and the thermal channel impedance increment in the d-th single-day evaluation period. It is a dimensionless parameter. The larger the value, the higher the contribution of agricultural wet aerosol deposition to the increase of thermal channel impedance. The time-series increment of the thermal channel impedance represents the degree of change in the thermal channel impedance. The numerator is the weighted sum of the time-series increment of the thermal channel impedance by the surrogate quantity of agricultural wet aerosol deposition intensity, and the denominator is the sum of the absolute values ​​of the time-series increment of the thermal channel impedance. Through weighted summation and normalization, the degree of coupling between agricultural wet aerosol deposition and thermal channel impedance degradation is quantified. This is a minimum constant, with a value of 1e-3, used to avoid calculation errors where the denominator is 0. For operating conditions where the thermal channel impedance remains unchanged throughout the day, the weighted coupling degree is directly set to 0.

[0103] Multiplying the weighted coupling degree by the thermal hysteresis area yields a parameter characterizing the chronic obstruction and degradation of the agricultural wet aerosol thermal channels. The calculation formula is as follows:

[0104]

[0105] in The parameter for thermal channel degradation caused by agricultural moisture aerosols during the d-th daily evaluation period is a dimensionless parameter. A larger absolute value indicates a more severe degree of chronic degradation of the heat dissipation channel due to agricultural moisture aerosols. By multiplying the weighted coupling degree by the thermal hysteresis area, the influence of agricultural moisture aerosols on the thermal channel impedance and the change in thermal response hysteresis effect are simultaneously incorporated, comprehensively quantifying the degree of chronic degradation of the energy storage cabinet's heat dissipation channel caused by the coupling of agricultural moisture aerosols and moisture. For the increase in thermal resistance caused by non-aerosol factors such as cooling fan failure and mechanical obstruction of the air duct, the degradation parameter is corrected by triggering the equipment fault alarm signal, thus removing the increase in thermal resistance caused by non-aerosol factors from the time-series increment and decoupling the degradation causes.

[0106] The potential photovoltaic curtailment rate of the system is used as the primary operational evaluation indicator, and the calculation formula is as follows:

[0107]

[0108] in , is the first operational evaluation index within the d-th daily evaluation cycle. It is a dimensionless parameter that characterizes the photovoltaic absorption capacity of the system. The larger the value, the weaker the photovoltaic absorption capacity of the system and the worse the operating status.

[0109] The normalized cumulative deviation between the measured agricultural load and the baseline time series of the ideal rural load is calculated as the second operational evaluation index. The calculation formula is as follows:

[0110]

[0111] in The second operational evaluation index within the d-th single-day evaluation cycle is a dimensionless parameter that characterizes the operational stability of agricultural load. The larger the value, the greater the deviation between the actual load operation and the ideal baseline, and the more severe the abnormal load fluctuations. This is a local constant with a value of 1e-3, used to avoid calculation errors where the denominator is 0.

[0112] The cumulative deviation of the node voltage from the rated voltage of the transformer substation is calculated as the third operational evaluation index. The calculation formula is as follows:

[0113]

[0114] in The third operational evaluation index within the d-th daily evaluation cycle is a dimensionless parameter that characterizes the stability of the power supply voltage in the distribution area. The larger the value, the more serious the deviation of the node voltage from the rated voltage, and the worse the power supply quality.

[0115] The equivalent ohmic internal resistance of the coupled battery, the time-series state-of-charge fluctuation amplitude, and the Arrhenius aging acceleration factor based on cell temperature constitute the fourth operational evaluation index, calculated using the following formula:

[0116]

[0117] in The fourth operational evaluation index within the d-th daily evaluation cycle is a dimensionless parameter that characterizes the degree of aging acceleration of the energy storage battery. The larger the value, the faster the battery aging rate. The daily average value of the battery's equivalent ohmic internal resistance during the d-th daily evaluation period, in ohms; The rated charge and discharge current of the battery cluster is expressed in amperes. This is the reference rated power of the battery cluster, measured in watts, used to make the internal resistance heating power dimensionless. is the fluctuation range of the time-series state of charge within the d-th daily evaluation period, is the difference between the maximum and minimum SOC values ​​within the daily period, and is a dimensionless parameter; The Arrhenius aging acceleration factor is based on cell temperature, where The activation energy of the battery aging reaction is expressed in joules per mole. For lithium iron phosphate batteries, the value is 35,000 J / mol, and for ternary lithium batteries, it is 55,000 J / mol. The universal gas constant is 8.314 J / (mol·K); This represents the daily average temperature of the battery cell within the d-th daily evaluation period, in degrees Celsius. Temperature is thermodynamic temperature, and the unit is Kelvin. The baseline temperature is set at 298.15 K, corresponding to 25°C. The Arrhenius factor characterizes the accelerating effect of temperature on battery aging; the higher the temperature, the greater the aging acceleration factor value.

[0118] The product of the agricultural wet aerosol thermal channel degradation parameter and the thermal channel impedance is normalized relative to the reference impedance to obtain the fifth operational evaluation index, calculated using the following formula:

[0119]

[0120] in The fifth operational evaluation index within the d-th single-day evaluation cycle is a dimensionless parameter that characterizes the degree of heat dissipation channel deterioration caused by agricultural wet aerosols. The larger the value, the more severe the heat dissipation channel deterioration. It is the steady-state reference impedance for the thermal path impedance, measured in degrees Celsius per watt. It is obtained by statistically analyzing the average thermal path impedance over 30 consecutive days under healthy operating conditions based on the equipment's factory thermal test data. It is used to provide a reference value for the per-unit processing.

[0121] The severity quantiles of the first to fifth operational assessment indicators were calculated over historical operational cycles and then converted into corresponding normalized scores. The calculation formula is as follows:

[0122]

[0123] in The normalized score is the score corresponding to the j-th operation evaluation indicator within the d-th daily evaluation period. The value of j ranges from 1 to 5, corresponding to the first to fifth operation evaluation indicators. The normalized score ranges from 0 to 1. The larger the value, the worse the operation status of the indicator. This is the rank function of the empirical distribution, used to calculate the current index value within... The ascending ranking of all samples in the historical operating cycle is used, that is, the larger the index value, the higher the ranking and the larger the rank value. In the case of multiple samples with the same value, the average rank method is used. The total number of days in the historical operating cycle is fixed at 90 days. For the cold start scenario of newly commissioned equipment, historical operating data of the same model of equipment in the same region is used as the benchmark sample.

[0124] Calculate the probability of each normalized score in its entire sample, and obtain the distribution information entropy of each indicator according to the information entropy calculation rule. The calculation formula is as follows:

[0125]

[0126] in Let v be the probability of the j-th indicator in the v-th historical sample, representing the proportion of that sample in the entire historical sample set. It is a local constant with a value of 1e-8, used to avoid the problem of meaningless logarithmic operations caused by zero probability values; Let be the distribution information entropy of the j-th indicator, which is a dimensionless parameter with a value range of 0 to 1. The larger the information entropy value, the more uniform the numerical distribution of the indicator, and the lower the discrimination of the operation evaluation. The function is the natural logarithm, and the denominator is... This is the maximum information entropy, used to normalize the information entropy.

[0127] Based on the deviation of the information entropy of each indicator from the maximum information entropy, the objective weight of each indicator is assigned, and the calculation formula is as follows:

[0128]

[0129] in Let be the objective weight of the j-th indicator. The sum of all weights is 1. The smaller the information entropy value of an indicator, the lower its weight. The larger the value, the higher the weight assigned. In other words, the higher the discrimination of an indicator, the greater its weight in the comprehensive evaluation, thus realizing adaptive weight allocation based on the distribution characteristics of indicator data.

[0130] The overall performance score is obtained by weighting and summing the objective weights and their corresponding normalized scores, using the following formula:

[0131]

[0132] in The comprehensive operational score for the d-th daily evaluation period ranges from 0 to 100. The higher the score, the worse the overall operating status of the equipment and the higher the risk of deterioration. By weighted summation and percentage conversion, a comprehensive evaluation result that is intuitive and comparable is obtained.

[0133] Based on the objective weights of each indicator and the degree to which its daily normalized score deviates from the historical mean, the contribution of each indicator to the overall evaluation degradation is calculated, and the indicators are ranked according to their degree of influence. The calculation formula is as follows:

[0134]

[0135] in The deviation contribution of the j-th indicator within the d-th daily evaluation period is represented by the larger value, indicating that the indicator contributes more to the deterioration of the comprehensive operation score on that day. Let j be the historical 90-day average of the normalized score for the j-th indicator. Let be the historical 90-day standard deviation of the normalized score for the j-th indicator; This is a minimum constant, taking a value of 1e-3, used to avoid calculation errors where the denominator is 0. By multiplying the deviation of the normalized score by the objective weight, the contribution of each indicator to the overall evaluation deterioration is quantified. After calculation, the indicators are sorted from largest to smallest according to their contribution values ​​to identify the core influencing factors leading to the deterioration of equipment operating status.

[0136] The overall operational score, parameters of agricultural wet aerosol thermal channel degradation, thermal channel impedance, thermal time constant, thermal hysteresis area, and various operational evaluation indicators and deviation contributions are summarized to generate the daily operational evaluation log sequence. The calculation formula is as follows:

[0137]

[0138] in The operational evaluation ledger vector for the d-th daily evaluation period is a column vector that fully contains all the core parameters of the equipment's operational evaluation for that day, enabling comprehensive recording and backtracking of operational status and providing input features for subsequent status prediction. The ledger data is stored in a time-series database, stored once daily, with a data retention period of no less than 3 years.

[0139] The physical residual of the heat transfer equilibrium model in the current cycle is obtained by calculation using the following formula:

[0140]

[0141] in To uniformly evaluate the physical residuals of the heat transfer balance model fitting at the k-th time under the clock, the unit is watts, which characterizes the fitting accuracy of the heat transfer model. The larger the absolute value of the residual, the worse the model fitting effect and the lower the reliability of the physical constraints. The physical residuals of all times in the current cycle constitute a residual sequence, which is used for the calculation of subsequent physical constraint weights.

[0142] The ratio of the statistical variance of the network's predicted output to the variance of the state-fit physical residuals is calculated to generate adaptive physical law constraint weights. The calculation formula is as follows:

[0143]

[0144] in Let be the adaptive physical law constraint weight corresponding to the d-th single-day evaluation cycle, and be a dimensionless parameter; To calculate the statistical variance of the prediction output of the previous period on the time series prediction network, each parameter in the ledger vector is first normalized to 0-1 to eliminate the difference in units, and then the average variance of the predicted values ​​of all parameters is calculated. The statistical variance of the physical residuals fitted to the heat transfer balance model for the current cycle; This is a local constant, taking a value of 1e-3, used to avoid calculation errors where the denominator is 0. When the variance of the physical residuals is smaller, and the fitting accuracy of the physical model is higher, the constraint weights... The larger the value, the stronger the role of physical constraints in network optimization, thus enabling adaptive constraint weight adjustment based on the fitting accuracy of the physical model.

[0145] The system acquires external input weather forecast data and agricultural scheduling plans. The weather forecast data includes hourly forecasts of irradiance, ambient temperature, relative humidity, and wind speed for the future evaluation period, which are synchronously obtained through the API interface of the public meteorological service platform. The agricultural scheduling plans include the types of agricultural activities, working hours, and expected electricity load for the future evaluation period, which are synchronously obtained through the agricultural production management system and converted into agricultural schedule proxy parameters and agricultural load plan values ​​for the corresponding time period, providing external input characteristics of future operating conditions for the time series forecasting network.

[0146] Using the operational evaluation log sequence, weather forecast data, and agricultural work schedule as input features, the temporal prediction network is optimized with the joint objective function of minimizing the prediction output error and the physical residual of the state fitting after adaptive physical law constraint weight penalty. The output is the predicted state of the system in the next evaluation period. The temporal prediction network adopts a gated recurrent unit encoder-decoder structure. The encoder has two gated recurrent unit layers, each with 128 neurons in its hidden layer. The decoder has one fully connected layer, and the output dimension is consistent with the dimension of the operational evaluation log vector. The network input features include the operational evaluation log sequence within the backtracking time window, weather forecast data for the future evaluation period, and agricultural work schedule data. The backtracking time window q is fixed at 7 days, meaning that the input is the operational evaluation log sequence of the past 7 days, and the output is the predicted state of the system for the next day. The network activation function is the ReLU function, and the output layer uses a linear activation function, adapting to continuous value prediction tasks.

[0147] The formula for calculating the joint objective function of network optimization is as follows:

[0148]

[0149]

[0150] in These are the optimal parameters for the optimized time series prediction network. For time series prediction networks, These are the trainable parameters of the network; Establish a long-term time window sequence for the retrospective ledger of the past 7 days; This refers to the weather forecast data for the (d+1)th evaluation period; The agricultural work schedule plan for the (d+1)th evaluation period; It is a diagonal unit reduction matrix, and the elements on the diagonal are the reciprocals of the historical standard deviations of each parameter in the ledger vector. It is used to strip the heterogeneous features such as temperature, time, and dimensionless coefficients in the evaluation ledger vector of dimensions to ensure the physical legitimacy of the norm space closure. The objective function is an L2 norm. The first term is the sum of squares of the prediction output error, which represents the prediction accuracy of the network. The second term is the sum of squares of the physical residuals after the weight penalty imposed by the adaptive physical law, which is a physical constraint term to ensure that the network prediction results conform to the physical law of heat transfer balance.

[0151] The network training uses the AdamW optimizer with an initial learning rate of 1e-4, a weight decay coefficient of 1e-5, a batch size of 16, and 200 training epochs. The learning rate decays to 0.5 every 50 epochs. An early stopping mechanism is employed: training stops when the loss on the validation set shows no decrease for 20 consecutive epochs to prevent overfitting. The network uses an incremental update method; after the daily evaluation log is generated, the network parameters are updated using the data from that day, with an update frequency of once daily. For the cold start scenario of newly commissioned equipment, a pre-trained model of the same model and region is used for initialization. After the equipment has been running for 14 days, the network is retrained using the equipment's own operational data. After optimization, the corresponding features are input, and the predicted degradation state for the (d+1)th evaluation period is output. .

[0152] For abnormal data conditions, the following processing rules are adopted: For cases where the continuous packet loss of a single parameter does not exceed 3 sampling steps, linear interpolation is used for completion; for cases where the continuous packet loss exceeds 3 sampling steps, the average value of the same historical period is used for completion, and the data is marked as abnormal and its weight is reduced in subsequent calculations; for outlier data that exceeds 3 times the median absolute deviation of the historical normal range, the median of the adjacent valid data is used for replacement; for cases where the data acquisition terminal is offline for more than 1 hour, the evaluation calculation for the day is suspended, and the calculation is recalculated after the data is restored using the completed data.

[0153] For extreme weather conditions, the following processing rules are adopted: For conditions where the irradiance is less than 100 watts per square meter on cloudy or rainy days, the photovoltaic output loss calculation is set to 0, and the photovoltaic power prediction baseline model is switched to the low irradiance adaptation sub-model; For conditions with no irradiance at night, all photovoltaic-related calculation steps are skipped, and historical daytime data is used to calculate the photovoltaic-related indicators for the day; For extreme high temperature and high humidity conditions, the regularization coefficient is increased during the thermal parameter identification process to improve the stability of the identification results; For conditions where meteorological data is missing, the average value of historical meteorological data for the same period is used as a substitute.

[0154] For abnormal equipment operating conditions, the following processing rules are adopted: For equipment alarm conditions such as photovoltaic module failure, battery cell failure, power unit failure, and cooling fan failure, data during the fault period is automatically removed to avoid fault data interfering with the identification of environmental degradation; For battery full charge and full discharge conditions, the local time window for battery parameter identification is adaptively adjusted to avoid the transient process of charge and discharge cutoff; For sudden rises and falls in transformer area voltage and power outages, the load baseline calculation automatically removes data during abnormal periods and uses load data from normal periods to fit the baseline.

[0155] For system cold start conditions, the following processing rules are adopted: For newly commissioned equipment without historical operating data, historical data of equipment of the same model, region and climate conditions are used as the benchmark statistical sample to complete standardization, weight allocation and baseline model initialization; after the equipment has been running for 7 days, it is gradually replaced with the equipment's own operating data; after running for 30 days, all calculation steps are completed entirely using the equipment's own historical data.

[0156] The system hardware deployment adopts an on-site deployment architecture for edge computing terminals. These terminals utilize industrial-grade processors with a minimum computing power requirement of a quad-core ARM Cortex-A53 processor, at least 2GB of memory, and at least 16GB of storage. They are designed to operate in outdoor environments ranging from -40℃ to 70℃. The multi-source data acquisition module communicates with the edge computing terminals via RS485 or Ethernet, with the data transmission cycle consistent with the basic sampling time step. The system interfaces with the photovoltaic inverter, battery management system, power conversion unit, and operation and maintenance management platform using standard industrial communication protocols. Evaluation records and prediction results can be pushed to the operation and maintenance management platform in real time. When the comprehensive operation score exceeds a preset threshold, an operation and maintenance alarm is automatically triggered. The alarm threshold is set at 60 points; a score exceeding 60 points triggers a warning, and a score exceeding 80 points triggers an emergency alarm.

[0157] The regularization coefficient λ is the L2 regularization penalty coefficient in the battery parameter identification process. It is used to constrain the range of values ​​of the identification parameters and avoid the divergence of parameter identification results caused by sampling noise. The smaller the value, the weaker the constraint on the identification parameters, and the larger the value, the stronger the constraint.

[0158] Regularization coefficient L2 regularization penalty coefficient is used in the thermal parameter identification process to constrain the identification range of the cabinet's equivalent heat capacity and total thermal conductivity, avoid the identification results from exceeding the reasonable range of the equipment's physical characteristics, and ensure the physical legality of the identification results.

[0159] decay time constant The exponential decay time constant of the weighting coefficients within the local time window for battery parameter identification is used to control the influence weight of historical sampling points on the current identification result. The larger the value, the slower the weight decay of historical sampling points and the stronger the stability of the identification result; the smaller the value, the faster the weight decay and the stronger the real-time performance of the identification result.

[0160] The empirical distribution rank function rank(·) is a statistical function that ranks a single indicator sample in the entire historical sample in ascending order. It is used to quantify the severity of the current indicator value in the historical operating data. The higher the ranking, the larger the indicator value and the worse the equipment operating status.

[0161] Diagonal unit reduction matrix The inverse of the historical standard deviation of the corresponding parameter in the evaluation ledger vector is used to map ledger parameters of different dimensions and numerical magnitudes to the same dimension space, eliminate the influence of the magnitude difference between different parameters on the calculation of prediction error, and ensure the physical legitimacy of the prediction error term.

[0162] The early stopping mechanism is triggered by the maximum number of iterations during the training of the time-series prediction network where the validation set loss does not decrease continuously. This is used to avoid overfitting and ensure the network's generalization ability.

[0163] The process of partitioning the prediction model dataset is as follows: the entire historical running data is divided into training set, validation set, and test set in a ratio of 7:2:1. The partitioning process adopts time-series ordered partitioning and prohibits random shuffling to ensure the temporal continuity of the data. The training set is used for model parameter fitting, the validation set is used for hyperparameter tuning and early stopping judgment, and the test set is used for final model accuracy verification. The mean absolute percentage error of the test set is required to not exceed 8%; otherwise, the model hyperparameters are readjusted.

[0164] Standardized operating procedures for outlier removal:

[0165] The first step is to identify outliers of a single parameter using the 3-times median absolute deviation criterion and mark sampling points that exceed the threshold.

[0166] The second step is to use the sliding window method to identify continuous packet loss data. The window length is set to 5 sampling steps. The segment with less than 30% of valid data in the window is marked as a continuous packet loss segment.

[0167] The third step is to use the equipment alarm signal linkage method to mark all data during periods of equipment failure and power grid anomalies.

[0168] The fourth step is to process the marked abnormal data by replacing it with the median of adjacent valid data, completing it with linear interpolation, or directly removing it. The proportion of abnormal data removed should not exceed 5% of the total data volume; otherwise, the historical data statistical period should be extended.

[0169] Step-by-step operation process for on-site benchmarking and verification:

[0170] The first step is to detect the surface contamination density of photovoltaic modules. The equal area sampling method is used. Three modules are selected from the upper, middle and lower regions of the photovoltaic array. Surface deposits of 10cm×10cm area are collected from each module. After drying, they are weighed and the contamination density per unit area is calculated.

[0171] The second step is to test the resistance of the heat dissipation duct of the energy storage cabinet. A differential pressure gauge is used to measure the static pressure difference between the inlet and outlet of the duct, and the duct flow resistance is calculated in combination with the air velocity in the duct.

[0172] The third step is to perform a linear correlation test between the accumulated pollution density obtained from on-site testing and the photovoltaic power output loss, with a correlation coefficient of not less than 0.85; and to perform a linear correlation test between the air duct flow resistance and the thermal channel impedance, with a correlation coefficient of not less than 0.8, to verify the accuracy of the evaluation results.

[0173] Screening process for benchmark samples during the commissioning phase of new equipment:

[0174] The first step is to screen existing equipment that is the same model, installation specifications, and climate zone as the target equipment;

[0175] The second step is to screen 90 consecutive days of operating data under healthy operating conditions and remove abnormal data from periods of equipment failure, extreme weather, and abnormal power grid conditions.

[0176] The third step is to standardize the selected sample data to generate a baseline statistical sample for the initial stage of operation, including the historical median, median absolute deviation, and baseline model pre-training weights.

[0177] The fourth step is to gradually replace the baseline sample with newly added data each day at a weight of 10% after the target equipment is put into operation. After 30 days of operation, the data will be completely replaced with the equipment's own data.

[0178] Standardized execution process after an alarm is triggered:

[0179] The first step is to trigger an early warning (comprehensive score of 60-80 points). The system will automatically push the evaluation log, core degradation indicators, and deviation contribution ranking results to the operation and maintenance platform, and generate preventive operation and maintenance suggestions, including component cleaning, air duct inspection, and battery status detection.

[0180] The second step is to trigger an emergency alarm (with a comprehensive score of over 80 points). The system will automatically push emergency alarm information to the operation and maintenance personnel and simultaneously link the energy storage control system to limit the charge and discharge rate to no more than 0.5C to prevent the deterioration caused by high temperature. Normal operation will resume after the operation and maintenance is completed.

[0181] Thermal hysteresis area The positive and negative values ​​are only used to characterize the phase relationship between temperature rise and heat generation power. Positive values ​​indicate that the temperature rise lags behind the heat generation power, which is a typical characteristic of heat dissipation channel deterioration; negative values ​​indicate that the temperature rise leads the heat generation power, which is a normal phenomenon caused by sudden changes in ambient temperature. In the calculation of the deterioration parameters of agricultural wet aerosol heat channels, only the positive values ​​of thermal hysteresis area are used in the calculation, and negative values ​​are directly set to 0.

[0182] Basic sampling time step The legal unit for electricity is standardized to seconds, and it needs to be multiplied by in all electricity integral calculations. The dimensional conversion factor is converted to hourly units to ensure that the unit of the electricity calculation result is kilowatt-hour; in all calculations of thermal dynamic change rate, the second-level unit is directly used. The unit for the rate of temperature change is guaranteed to be degrees Celsius per second.

[0183] The hyperparameters of the photovoltaic power prediction baseline model, such as the number of trees and tree depth, were determined using a Bayesian optimization algorithm, with the optimization objective being to minimize the mean square error of the validation set. The hyperparameters of the time-series prediction network, such as the number of hidden layer neurons and the learning rate, were determined using a grid search method, with a search range of 32-256 hidden layer neurons and a learning rate of 1e-5 to 1e-3, with the optimization objective being to minimize the mean absolute percentage error of the validation set.

[0184] Add a minimum constant to the denominator of all division operations. The value ranges from 1e-8 to 1e-3, and the specific value is determined by the magnitude of the denominator. The smaller the magnitude of the denominator, the higher the value. The smaller the value, the better. Additionally, non-zero constraints are added to all denominator terms. When the calculated value of the denominator is less than... At that time, directly adopt As the denominator, it completely avoids the error of dividing by zero, and there are clear handling rules for all boundary scenarios.

[0185] All iterative algorithms incorporate regularization constraints to prevent matrix singularity issues; all floating-point calculations use double-precision floating-point format to avoid numerical precision loss; all transcendental function inputs are subject to range constraints, with logarithmic function inputs constrained to be greater than 1e-8 and exponential function inputs constrained to be between -20 and 20, to prevent numerical overflow; all iterative algorithms have a maximum number of iterations and a convergence threshold, and if convergence is not achieved after exceeding the maximum number of iterations, the valid result from the previous cycle is automatically adopted to ensure the numerical stability and convergence of the algorithm.

[0186] All acquisition terminals use the NTP network time protocol to synchronize clocks with edge computing terminals, with a synchronization period of 1 minute and a clock synchronization error of no more than 10ms. For high-precision acquisition requirements, the IEEE 1588 PTP precision time protocol can be used, with a synchronization error of no more than 1ms. All acquired data is timestamped after synchronization, with a timestamp accuracy of no less than 1 millisecond, to ensure the time alignment accuracy of multi-source data.

[0187] All thermal parameters are measured in SI units, and the total equivalent heat source... The unit is watts (J / s), equivalent heat capacity. The unit is joules per degree Celsius (J / ℃), and the rate of change of temperature rise. The unit is degrees Celsius per second (°C / s), total thermal conductivity The unit is watts per degree Celsius (W / ℃), and the thermal path impedance is... The unit is degrees Celsius per watt (°C / W), and the thermal time constant is... The unit is seconds (s).

[0188] Adaptive physical constraint weights During the calculation process, the statistical variance of the predicted output is... The calculation must first perform 0-1 min-max normalization on each parameter of the operational evaluation ledger vector. The normalization formula is as follows: ,in , For the historical minimum and maximum values ​​of the corresponding parameters, all parameters are dimensionless after normalization, and the calculated variance is also dimensionless; physical residual variance. Before calculation, the physical residual sequence is first normalized to 0-1 to ensure that the variance is dimensionless. The ratio of the two is... These are dimensionless constraint weights.

[0189] In the calculation of the Arrhenius aging acceleration factor, activation energy The unit is fixed as joules per mole (J / mol), and the universal gas constant is used. The unit is fixed as joules per mole per kelvin (J / (mol·K)), and the cell temperature is converted to thermodynamic temperature in kelvin (K). All parameter units are perfectly matched, with no risk of deviation on the order of 1000 times. The unit of open circuit voltage temperature coefficient is fixed as volts per kelvin (V / K).

[0190] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A data-driven rural photovoltaic energy storage integration device operation evaluation system applied to an interactive system comprising a distributed photovoltaic array, an outdoor integrated energy storage cabinet and an agricultural load, characterized in that, Configured for execution: Multi-source operating parameters, including meteorological, electrical, thermal, and agricultural conditions, are collected to construct an aligned unified state field. This process includes: selecting the median of the differences between adjacent timestamps of the multi-source operating parameters as the basic sampling time step to generate a unified evaluation clock; aligning the multi-source operating parameters to the unified evaluation clock using a smoothing reconstruction algorithm to obtain a full observation sequence; performing dimensionless scaling on the full observation sequence using a standardized function containing statistical median and robust variance estimates to obtain the unified state field; solving for photovoltaic and load benchmarks based on the unified state field to extract photovoltaic output losses and energy flow paths characterizing the environmental deposition on the module surface; reconstructing the total equivalent heat source of the energy storage battery and power conversion unit based on the electrothermal characteristic parameters in the unified state field; and fitting a heat transfer balance model based on the total equivalent heat source and the internal and external temperature difference of the equipment to calculate the thermal channel impedance and thermal hysteresis area of ​​the cabinet. Coupled with photovoltaic power loss and meteorological moisture deposition factor, the degradation parameters of agricultural wet aerosol thermal channels characterizing the environmental hysteresis coupling effect are calculated, including: calculating the meteorological water vapor dew point temperature based on Magnus' empirical formula and combined with ambient temperature and relative humidity; calculating the absolute value of the difference between the shell surface temperature and the meteorological water vapor dew point temperature, and using the exponential decay function to map this absolute value into a wet deposition activity factor characterizing the surface condensation tendency. The fluctuation dispersion of wind speed relative to historical stable wind speed is calculated, and combined with photovoltaic power loss and wet deposition activity factor, a local agricultural wet aerosol deposition intensity surrogate quantity is constructed. The time-series increment of thermal channel impedance is extracted, and the weighted coupling degree of local agricultural wet aerosol deposition intensity surrogate quantity to time-series increment is calculated. The weighted coupling degree is multiplied by the thermal hysteresis area to obtain the agricultural wet aerosol thermal channel degradation parameter characterizing the chronic obstruction and decay of the heat dissipation channel. The multi-dimensional operational evaluation indicators, including parameters related to agricultural aerosol thermal channel degradation, are transformed into normalized scores. The weights of these indicators are adaptively determined using historical distribution variance to derive a comprehensive operational score. This score includes: the system's potential photovoltaic curtailment rate as the first operational evaluation indicator; the normalized cumulative deviation of the measured agricultural load from the ideal rural load baseline as the second operational evaluation indicator; the cumulative deviation of the voltage at transformer substation nodes from the rated voltage as the third operational evaluation indicator; the equivalent ohmic internal resistance of the coupled battery, the amplitude of the time-series state of charge fluctuation, and the Arrhenius aging acceleration factor based on cell temperature as the fourth operational evaluation indicator; the product of agricultural aerosol thermal channel degradation parameters and thermal channel impedance is normalized relative to the reference impedance to obtain the fifth operational evaluation indicator; the severity quantiles of the first to fifth operational evaluation indicators in the historical operational cycle are statistically analyzed and transformed into corresponding normalized scores; the probability of each normalized score in its full sample is calculated, and the distribution information entropy of each indicator is obtained according to the information entropy calculation rule; and the objective weights of each indicator are assigned based on the deviation of its distribution information entropy from the maximum information entropy. The overall performance score is obtained by weighting and summing the objective weights and their corresponding normalized scores. An evaluation log is generated based on the deviation contribution of each indicator, and the prediction network is updated using the fitting residual constraint of the heat transfer balance model. The state prediction result for the next period is output, including: calculating the deviation contribution of each indicator to the overall evaluation degradation based on the objective weight of each indicator and the degree of deviation of the normalized score of the day from the historical mean, and sorting them according to the degree of influence; summarizing the comprehensive operation score, agricultural wet aerosol thermal channel degradation parameters, thermal channel impedance, thermal hysteresis area, and deviation contribution of each indicator to generate the operation evaluation log sequence for the day; obtaining the state fitting physical residual of the heat transfer balance model in the current period; calculating the ratio of the statistical variance of the network prediction output to the variance of the state fitting physical residual to generate adaptive physical law constraint weights; obtaining external input weather forecast data and agricultural work schedule; using the operation evaluation log sequence of the previous evaluation period, weather forecast data, and agricultural work schedule as input features, optimizing the time series prediction network with the joint objective function of minimizing the prediction output error and the state fitting physical residual penalized by the adaptive physical law constraint weights, and outputting the degradation state prediction result of the system in the next evaluation period.

2. The data-driven rural photovoltaic energy storage integrated equipment operation evaluation system according to claim 1, characterized in that, Multi-source operational parameters, including meteorological, electrical, thermal, and agricultural conditions, are collected to construct an aligned unified state field, including: Simultaneously collect multi-source operating parameters, including: measured photovoltaic power, irradiance, ambient temperature, relative humidity, wind speed, AC power, DC power, battery voltage, battery current, battery state of charge, battery health status, cell temperature, cabinet temperature, outer casing surface temperature, auxiliary equipment heat dissipation power, transformer area node voltage, transformer area rated voltage, measured agricultural load, agricultural schedule proxy parameters, local solar geometry angle, and calendar time stamp.

3. The data-driven rural photovoltaic energy storage integrated equipment operation evaluation system according to claim 2, characterized in that, Based on the unified state field solution for photovoltaic and load benchmarks, the photovoltaic output loss and energy flow path of the equipment, which characterize the environmental deposition on the module surface, are extracted, including: The irradiance, ambient temperature, relative humidity, wind speed, local solar geometry angle, and agricultural schedule proxy parameters in the unified state field are input into the photovoltaic power prediction baseline model to output the ideal photovoltaic power generation baseline. Calculate the ratio of the relative deviation between the ideal photovoltaic power generation baseline and the measured photovoltaic power, and extract the positive relative deviation that characterizes the power generation attenuation as the photovoltaic output loss; The ambient temperature, calendar time tag, agricultural schedule proxy parameter and transformer node voltage are input into the load prediction baseline model to output the ideal rural load baseline. Based on the product of the battery terminal voltage and the battery terminal current, the energy storage charging power and energy storage discharging power are separated according to the power flow direction. By integrating the time interval between the measured photovoltaic power and the energy storage charging power, the amount of electricity transferred from photovoltaic to energy storage is obtained; by integrating the time interval between the energy storage discharging power and the measured agricultural load, the amount of electricity supplied from energy storage to the load is obtained. The unconsumed power integral exceeding the sum of the measured photovoltaic power and the energy storage charging power of the ideal photovoltaic power generation baseline is calculated, and its proportion relative to the total integral of the power generation baseline is taken as the potential photovoltaic curtailment rate of the system.

4. The data-driven rural photovoltaic energy storage integrated equipment operation evaluation system according to claim 3, characterized in that, Based on the electrothermal characteristic parameters in the unified state field, the total equivalent heat source of the energy storage battery and power conversion unit is reconstructed, including: Using the open-circuit voltage surface model mapped from the battery state of charge and cell temperature, combined with the measured battery terminal voltage, battery terminal current and their time-series rate of change, the equivalent ohmic internal resistance and polarization dynamic parameters of the battery are extracted by parameter identification within a local time window. The irreversible Joule heat, which is composed of the square of the battery terminal current and the equivalent ohmic internal resistance of the battery, is calculated and superimposed with the reversible entropy heat, which is composed of the battery terminal current, cell temperature and open-circuit voltage temperature coefficient, to synthesize the heat source of the battery body. Calculate the absolute value of the difference between the AC terminal power and the DC terminal power, and extract the heat source of dissipation loss in the power conversion unit; The total equivalent heat source is obtained by summing the heat source generated by the battery body, the heat dissipation loss source of the power conversion unit, and the heat dissipation power of the auxiliary equipment.

5. The data-driven rural photovoltaic energy storage integrated equipment operation evaluation system according to claim 4, characterized in that, Based on the heat transfer balance model fitted to the total equivalent heat source and the temperature difference between the inside and outside of the equipment, the thermal channel impedance and thermal hysteresis area of ​​the cabinet are calculated, including: Extract the average of the cabinet temperature and the cell temperature, and subtract the ambient temperature to obtain the overall representative temperature rise of the system; Calculate the time-series rate of change of the overall representative temperature rise of the system at adjacent sampling times; With the goal of minimizing the residuals of the heat transfer balance model equations, the equivalent heat capacity and equivalent convective heat transfer coefficient of the cabinet are identified by utilizing the total equivalent heat source, the overall representative temperature rise of the system and its time-series change rate. The reciprocal of the equivalent convective heat transfer coefficient is used as the thermal channel impedance, and the ratio of the cabinet's equivalent heat capacity to the equivalent convective heat transfer coefficient is used as the thermal time constant. The calculation system represents the average temperature rise and the integral area of ​​the total equivalent heat source during the charge and discharge cycle. After normalization, it is used as the thermal hysteresis area that characterizes the thermal response delay of the cabinet.

Images