Electrochemical energy storage system efficiency model parameter identification method

Abstract

The invention discloses a parameter identification method for an efficiency model of an electrochemical energy storage system, and the method comprises the steps: collecting multi-working-condition operation data through building a refined loss model covering a PCS, a transformer and a temperature control system, and dividing the multi-working-condition operation data into a training set and a test set; carrying out multi-batch global optimization under a multi-stage closed-loop verification framework by adopting a strategy of combining a least square method and a particle swarm optimization (PSO) algorithm; and a multi-dimensional error index evaluation mechanism is further introduced, an objective function is self-adaptively adjusted and strengthened according to the performance of the test set, and finally, a parameter which is optimal in the test set is selected as a global optimal solution. According to the method, by designing a parameter identification method based on a multi-stage closed-loop verification mechanism and multi-dimensional parameter fitting verification and reinforcement, high-precision parameter identification and decision support are provided for refined efficiency management, state evaluation and loss reduction operation of the electrochemical energy storage system.

Description

Technical Field

[0001] This invention relates to the field of electrochemical energy storage system modeling and parameter identification technology, and more specifically, to a method for identifying parameters of an efficiency model for an electrochemical energy storage system. Background Technology

[0002] Large-scale energy storage power stations involve high investment costs, complex system structures, and diverse operating conditions and control requirements. Key indicators such as the operating efficiency of energy storage equipment, reasonable system control strategies, and operational losses and battery degradation characteristics directly impact the daily operation and overall economic benefits of energy storage power stations. Currently, there is a lack of complete and systematic economic control methods that can comprehensively consider the overall revenue of the power station and battery degradation characteristics. There is a lack of data support for decision-making, as well as a lack of mechanistic models aimed at reducing losses. It is also difficult to directly obtain loss parameters, thus affecting the efficiency management of electrochemical energy storage systems.

[0003] Current parameter identification methods mostly employ least squares, particle swarm optimization (PSO), and gradient descent. However, least squares is prone to getting trapped in local optima, and PSO and gradient descent lack adaptability, making it difficult to further optimize the fitted parameters. Furthermore, existing methods that combine least squares and PSO only provide initial data for PSO, which can easily lead to overfitting or underfitting, resulting in unreliable output loss parameters. To address these issues, a technical solution is proposed. Summary of the Invention

[0004] To overcome the aforementioned deficiencies in the prior art, this invention provides a method for identifying parameters of an efficiency model for an electrochemical energy storage system. By designing a multi-stage closed-loop verification mechanism and introducing enhancement parameters, the fitting parameters are further adaptively optimized to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: Step 1: Establish a power loss prediction model for the energy storage power station, covering the PCS power loss model, the transformer power loss model, and the temperature control system power loss model based on HVDC. During the model establishment process, identify the parameters to be identified.

[0006] The power loss of an energy storage power station also needs to take into account the power loss of the battery module equipment. Therefore, a power loss prediction model for an energy storage power station is established:

[0007] in, This represents the total power loss of the energy storage system. For battery module power loss, For PCS power loss, For transformer power loss, This refers to the power loss of a temperature control system primarily based on HVDC.

[0008] Furthermore, the energy storage PCS includes a DC / AC converter, its control system, and filter circuits. Therefore, the corresponding PCS losses mainly consist of converter losses and filter losses.

[0009] It should be noted that the losses of the converter mainly include static losses and dynamic losses. Static losses are further divided into off-state and on-state losses. Off-state losses account for a small proportion of the total losses and can be ignored. On-state losses are affected by temperature, current, and voltage across the converter. Dynamic losses are switching losses, which are affected by temperature and current. The filter losses are also affected by temperature and current.

[0010] When the PCS topology and modulation method remain unchanged, and the temperature is maintained stable by the temperature control system, the following PCS power loss model can be established:

[0011] in, These are the total PCS loss, converter loss, and filter loss, respectively. This refers to the DC-side current of the PCS. This refers to the DC-side voltage of the PCS. This is a correction factor for the change in current. This is a correction factor for switching losses. This is a correction value for the inductor on-resistance in the converter. This is the correction amount for the inductor's on-resistance in the filter. The iron loss in the filter loss, the , These are the parameters to be identified.

[0012] It should be noted that, This refers to the static loss component of the converter losses. This refers to the dynamic loss component of the converter losses.

[0013] Furthermore, the operating efficiency characteristics of energy storage power station transformers are largely consistent with those of conventional transformers, being significantly affected by the load factor, which in turn is strongly correlated with the power factor. However, since the power transmission of conventional transformers is generally unidirectional, their power factor typically varies within a certain positive range. Energy storage power station transformers, on the other hand, need to meet the bidirectional regulation requirements of the power grid's active and reactive power. Therefore, when calculating their operating efficiency, the power factor cannot be considered a fixed value within a certain range; it should be considered as a variable parameter. In addition, reactive current also generates active power losses within the transformer. Therefore, the project divides the transformer power loss calculation into two parts: active and reactive power. The losses generated in the reactive power part are converted into active power losses using a correction factor.

[0014] In summary, this invention considers transformer load factor and no-load loss as varying parameters, and establishes the following transformer power loss model:

[0015] in, This represents the total power loss of the transformer. , These are the transformer no-load loss and rated load loss, respectively. for The load factor of the transformer at any given time; This refers to the rated capacity of the transformer. Let be the apparent power actually output by the transformer at time t; They are respectively The actual active and reactive power output of the transformer at any given time, the aforementioned , These are the parameters to be identified.

[0016] Furthermore, considering the energy consumption characteristics of each power-consuming component in the temperature control system under both cooling and heating modes, this invention establishes a power loss model for the temperature control system based on HVDC. The temperature control system is a liquid-cooled system, and the model includes cooling demand loss, heating demand loss, total cooling loss, total heating loss, and air-cooled system loss. In cooling mode, the cooling demand loss of the liquid cooling system mainly includes circulation pump loss, refrigeration unit loss, and control loss. The refrigeration unit is a compressor. The system cooling demand loss model is as follows:

[0017] In heating mode, the heating demand loss of the liquid cooling system mainly includes circulation pump loss, heating device loss, and control loss. The heating device is a PTC electric heater. The system heating demand loss model is as follows:

[0018] in, For cooling demand losses, For heating demand losses, For battery heat loss, This refers to the energy loss required to heat the battery to its operating temperature. This refers to the heat power conducted from the external environment through the battery cabinet. The heat transfer coefficient of the battery cabinet enclosure. Let be the surface area of ​​the battery cabinet. The external temperature of the battery cabinet. The temperature set to activate the cooling mode of the liquid cooling system. The quality of the batteries inside the cabinet. For the specific heat capacity of the battery, The temperature set to activate the heating mode of the liquid cooling system. The temperature of the batteries inside the cabinet. Heating time, The external ambient temperature, the These are the parameters to be identified.

[0019] It should be noted that when When the external temperature is higher than the set temperature, Cooling demand ;when When the external temperature is lower than the set temperature, ,and , Since ambient heat dissipation is insufficient to offset electromagnetic heating, a liquid cooling system is still required to meet the cooling demand. .

[0020] In cooling mode, the total cooling loss of the liquid cooling system mainly includes the loss of the circulating pump, the loss of the chiller unit, and the loss of the control system. The total cooling loss model is as follows:

[0021] In heating mode, the total heating loss of the liquid cooling system mainly includes the loss of the circulating pump, the loss of the PTC electric heating system, and the loss of the control system. The total heating loss model is as follows:

[0022] in, This represents the total cooling loss. Total heating loss, For circulation pump losses, For ethylene glycol flow rate, For Yang Cheng, For coolant density, It is the acceleration due to gravity. For compressor wear, To control system losses, For cooling demand losses, The compressor's refrigeration efficiency ratio. For heating demand losses, For pump efficiency, the These are the parameters to be identified.

[0023] It should be noted that the circulating pump is a core loss component of the liquid cooling system, and the influence of the density and viscosity of ethylene glycol needs to be considered. The head is the pipeline resistance that the pump needs to overcome. The compressor loss is related to the cooling capacity and energy efficiency ratio. The control system loss is the loss of auxiliary equipment, which mainly includes controllers, sensors and valves. The control system loss is a fixed value.

[0024] It should be further explained that the thermal efficiency of PTC electric heating is close to 100%, so the PTC electric heating loss is approximately equal to the heating demand loss.

[0025] Furthermore, the PCS cabinet and transformer cabinet are equipped with an air-cooling system to effectively dissipate the heat generated during equipment operation, ensuring normal operation and extending the service life of the equipment.

[0026] The losses in an air-cooled system are primarily determined by the airflow volume based on heat dissipation requirements, and the loss model is derived by considering the operating environment and equipment parameters.

[0027] in, For fan wear, The airflow required for heat dissipation For heat dissipation requirements, air density, The specific heat capacity of air, To allow for the temperature rise inside the cabinet, For airflow resistance, For fan efficiency, For PCS loss, For transformer losses, the These are the parameters to be identified.

[0028] It should be noted that the losses in the air-cooling system refer to the losses of the fans on the PCS or transformer, and the heat dissipation requirements... At that time, the calculation is based on the fan losses on the PCS, and the heat dissipation requirement. At that time, the calculation was performed on the fan losses on the transformer.

[0029] It should be further noted that the parameter to be identified is a quantity that changes in real time or changes with the current, so an algorithm model is required for identification.

[0030] Step 2: Collect N sets of sample data. The sample data includes defined input and output data pairs. The output data includes PCS power loss, transformer power loss, power loss of the temperature control system (mainly HVDC), and corresponding actual losses. To improve the stability of the identification, the sample data undergoes outlier handling, filtering, denoising, and normalization preprocessing to ensure the numerical stability and convergence efficiency of the model during iterative optimization. The sample data is divided into a training set and a test set. The training set is used for optimization, and the test set is used for evaluation.

[0031] It should be noted that the determined input data mentioned above are the determined parameters in the energy storage power station power loss prediction model established in step 1, specifically including: Battery module power loss PCS DC current measurement PCS DC voltage measurement Transformer rated capacity The actual active power output of the transformer at time t The actual reactive power output of the transformer at time t Battery cabinet surface area External temperature of battery cabinet The temperature set when the liquid cooling system starts in cooling mode. Battery quality inside the cabinet Battery specific heat capacity Battery temperature inside the cabinet Required air volume air duct resistance .

[0032] Furthermore, outlier handling involves processing unreasonable values, also known as outliers, present in the dataset. Outliers can be identified as follows: an interval is set for the entire sample data; statistical analysis is performed on the sample data; values ​​outside the interval are outliers. When the data follows a normal distribution... The probability of an exception is less than 0.003, which is an extremely low probability event. This occurs when the sample distance from the average value is greater than... If the sample is an outlier, then the outlier is identified through box plot analysis.

[0033] The filtering and denoising includes amplitude limiting filtering, median filtering, and mean filtering. Amplitude limiting filtering sets a threshold to keep the portion of the sample data that does not exceed the threshold unchanged, preserving most of the features of the sample data while making the overall data sample more stable. Median filtering replaces the value of each sample point with the median value of the corresponding interval, preserving the edge features of the sample data. Mean filtering achieves filtering by calculating the neighborhood average of each sample point, smoothing the signal and reducing noise, while preserving the overall trend of the sample data.

[0034] The normalization process transforms sample data into a standard normal distribution or standardizes and normalizes it, thereby enhancing model training effectiveness and improving prediction performance. At the same time, the standardized data helps to address numerical problems that may arise during the calculation process, thus improving the robustness of the algorithm.

[0035] The division into training and testing sets allows for cross-validation.

[0036] Step 3: Based on N sets of samples, construct an initial objective function using the least squares method, and calculate the loss deviation. The optimization objective of the initial objective function is to minimize the deviation between the model's calculated loss and the actual operating loss, which serves as the optimization criterion for parameter identification. Specifically:

[0037] in, Let be the calculated model loss value for the i-th sample. The actual operating loss corresponds to the sample point, where N is the total number of samples. For the parameters to be identified, For regularization terms, As the regularization factor, in the initial objective function stage, the regularization factor is... If the value is set to 0, the current optimal solution is a parameter vector obtained by minimizing the overall loss deviation of N samples, and the current optimal solution is a local optimal solution.

[0038] Furthermore, the introduction of the regularization term can alleviate the overfitting problem. In the initial stage, in order to obtain the current optimal solution, the regularization factor is set to 0. The parameter vector includes the input data of each loss model and the current group optimal initial value of the parameter to be identified.

[0039] Step 4: Design a multi-stage closed-loop verification mechanism, setting the optimization batch and the number of iterations per batch. Further, the multi-stage closed-loop verification mechanism is divided into an inner loop and an outer loop. The inner loop optimizes parameters on the training set, and the outer loop evaluates error performance on the test set and further adjusts the objective function. This mechanism ensures that the identification process gradually converges to a stable solution in the closed-loop iteration of training-verification-feedback adjustment, alleviating local convergence and overfitting phenomena and improving the generalization ability of the model.

[0040] Furthermore, local convergence refers to the phenomenon that the algorithm can only guarantee that the iteration point converges to the optimal point when the initial point and the optimal point are close. The purpose of avoiding local convergence is to achieve global convergence, that is, the iteration point generated from any initial point converges to the optimal point of the problem.

[0041] Furthermore, the overfitting phenomenon refers to the phenomenon that the model performs well on the training set, but its performance drops significantly on the test set, indicating weak generalization ability. The "good performance" can refer to high accuracy.

[0042] Step 5: In the optimization process of each batch, the Particle Swarm Optimization (PSO) algorithm is introduced to perform global optimization of the objective function. Furthermore, the particle swarm parameters are first initialized, and the initial positions and velocities of the particles are set. In the inner loop iterative optimization process, the positions and velocities of the particles are updated sequentially, the fitness of the particles is calculated, the historical best fitness and position of individual particles are updated, and the historical best fitness and position of the swarm are updated. The global search capability of the PSO algorithm can effectively avoid the risk of getting trapped in local optima, thereby improving the stability and robustness of the optimization results.

[0043] It should be noted that the initial particle swarm parameters also include: particle swarm size, particle dimension, inertia weight, learning factor, and iteration step size range.

[0044] It should be further explained that the inertial weight represents the displacement influence of the previous generation of particles on the current generation of particles, and represents the degree of confidence of the particles in their own motion state. The larger the inertial weight, the stronger the particle's ability to explore new regions and the stronger its global optimization ability, but the corresponding local optimization ability will be weakened. A smaller inertial weight is beneficial for local search, allowing the algorithm to converge to the optimal solution.

[0045] Step 6: Multi-dimensional parameter fitting verification and reinforcement; Step 6 specifically includes: Step 6.1: Evaluate the fitting results using multi-dimensional error indices. Further, the multi-dimensional error indices refer to the mean square error, root mean square error, mean absolute error, maximum absolute error, and minimum absolute error calculated on the test set. These indices reflect the model's performance in terms of overall accuracy, fluctuation range, and extreme errors from multiple perspectives, thereby characterizing the parameter identification effect and determining whether the results are reasonable.

[0046] Furthermore, the mean squared error is specifically as follows:

[0047] in, For the first The true value of each sample For the first The predicted value for n samples, where n is the number of samples.

[0048] The mean squared error measures the average of the squared errors between the predicted and actual values, and is more sensitive to larger errors.

[0049] The root mean square error is specifically:

[0050] Where MSE is the mean squared error.

[0051] The root mean square error has the same dimensions as the original data, which more intuitively reflects the magnitude of the prediction error. The mean absolute error is specifically:

[0052] in, For the first The true value of each sample For the first The predicted value for n samples, where n is the number of samples.

[0053] The mean absolute error reflects the overall average deviation and is not sensitive to outliers.

[0054] The maximum absolute error is specifically:

[0055] in, For the first The true value of each sample For the first The predicted value for each sample.

[0056] The maximum absolute error highlights the worst-case scenario.

[0057] The minimum absolute error is specifically:

[0058] in, For the first The true value of each sample For the first The predicted value for each sample.

[0059] The minimum absolute error can intuitively determine the overall difference of the optimal value.

[0060] Step 6.2: Decision verification is performed based on the error index evaluation results. The decision verification involves setting the current batch as the completed batch if the identification parameters do not show obvious underfitting or overfitting phenomena and meet the actual needs; otherwise, a reinforcement objective function is introduced for further optimization, adaptively adjusted based on test set performance, and the regularization factor is adjusted accordingly. Then, a strengthened objective function is introduced, and the next batch of operations begins.

[0061] Furthermore, the strengthening objective function is:

[0062] in, Let be the calculated model loss value for the i-th sample. The actual operating loss corresponds to the sample point, where N is the total number of samples. For the parameters to be identified, For regularization terms, As a regularization factor, These are the error weighting coefficients.

[0063] It should be noted that, and Adaptive adjustments are made based on the performance on the test set. Specifically, when overfitting occurs, the [adjustment level] is increased. Strengthen regularization; when underfitting occurs, maintain or reduce regularization. Relaxing the fitting ability; when the sample error When it exceeds the threshold, increase This improves the model's fitting accuracy at key points; when samples error When it is less than the threshold, decrease To reduce its interference.

[0064] Furthermore, through and The dual adaptive adjustment ensures that the model maintains a balance between fitting accuracy and complexity, while improving the model's ability to identify key samples and enhancing its practicality.

[0065] It should be noted that the purpose of introducing the error weighting coefficient is to alleviate the underfitting phenomenon, which refers to the inability to fully capture the patterns of the sample data and the insufficient degree of fit to the data.

[0066] Step 7: After completing the set batch, multiple sets of candidate identification parameter solutions are obtained. The candidate solutions are substituted into the test set to calculate the error index. The solutions are sorted according to the overall loss deviation. The identification parameters with the best performance in the test set are selected as the global optimal solution. Furthermore, the global optimal solution can retain one or more sets of optimal solutions to adapt to different application requirements.

[0067] It should be noted that in different batches, there may be two sets of identification parameters that both meet the requirements of the optimal solution. Therefore, it is necessary to retain one or more sets of optimal solutions to prevent the reduction of data accuracy due to the uniformity of the solution during actual operation.

[0068] The technical effects and advantages of the method for identifying efficiency model parameters of an electrochemical energy storage system according to the present invention are as follows: 1. By designing a parameter identification method based on a multi-stage closed-loop verification mechanism and multi-dimensional parameter fitting verification and reinforcement, this mechanism forms a closed loop of "identification-evaluation-decision-re-optimization". Only the model parameters that pass the verification will be output. Otherwise, the objective function will be adaptively adjusted and enter the next batch of iterations, which fundamentally ensures the reliability and accuracy of the final model output. The multi-dimensional parameter fitting verification does not rely on a single index, but uses multiple indexes such as mean square error, mean absolute error, maximum absolute error and minimum absolute error for comprehensive measurement. This provides high-precision identification parameters and decision support for the refined efficiency management, state assessment and loss reduction operation of electrochemical energy storage systems. 2. By integrating the inner and outer loops of the multi-stage closed-loop verification mechanism, the objective function based on least squares and the Particle Swarm Optimization (PSO) algorithm are combined to form a complementary advantage. By utilizing the initial objective function based on least squares, the initial solution of parameters is quickly obtained, which has good approximation in the sense of minimum error. This provides high-quality prior information and search starting point for subsequent optimization, greatly reducing the search space of the PSO algorithm. This allows the PSO algorithm to conduct fine mining in the most promising parameter space, effectively avoiding the inefficiency of blind search in the PSO algorithm. The inner loop uses the PSO algorithm to iterate the optimal solution, while the outer loop iterates and optimizes the objective function based on least squares. This provides higher-precision prior knowledge for the PSO algorithm iteration, and the cooperation between the inner and outer loops ultimately converges to the global optimum with a higher probability. 3. By introducing regularization terms and error weighting coefficients to adaptively adjust the objective function through iterative optimization, penalizing the magnitude of parameters, and incorporating prior knowledge into the optimization process, the model complexity is effectively constrained, overfitting and underfitting phenomena are alleviated, the stability and reliability of parameters are ensured, and the generalization ability of the model is improved. 4. By constructing a global power loss model covering energy storage batteries, PCS, transformers, and HVDC temperature control systems, and collecting sample data covering different equipment, different ambient temperatures, and different operating conditions, we break through the limitations of traditional single-equipment modeling, ensuring the sufficiency and representativeness of training data. Furthermore, we perform outlier processing, filtering and denoising, and normalization preprocessing on the sample data. The model parameters identified based on this sample data avoid over-dependence on specific operating conditions, thereby ensuring the model's excellent generalization ability and robustness under complex operating conditions throughout the power plant's entire life cycle. This provides a foundation for subsequent parameter identification and further ensures the reliability of loss parameters. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0070] Figure 1 This is a flowchart of the method of the present invention; Figure 2 The results are the fitting results for the transformer parameters; Figure 3 The result is the transformer error calculation. Figure 4 Comparison of actual values ​​and model predictions under transformer charging conditions on the test set; Figure 5 To compare the actual values ​​with the model predictions under the transformer discharge state on the test set; Figure 6 Fitting transformer losses under energy storage charging conditions; Figure 7 This is a fitting of transformer losses under energy storage discharge conditions. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] Example 1: Figure 1 A flowchart of the method of the present invention is provided, including: Step 1: Establish a power loss prediction model for the energy storage power station, covering the PCS power loss model, the transformer power loss model, and the temperature control system power loss model based on HVDC. During the model establishment process, identify the parameters to be identified.

[0073] The power loss of energy storage power stations also needs to take into account the power loss of battery module equipment. Therefore, a power loss prediction model for energy storage power stations needs to be established:

[0074] in, This represents the total power loss of the energy storage system. For battery module power loss, For PCS power loss, For transformer power loss, This refers to the power loss of a temperature control system primarily based on HVDC.

[0075] Furthermore, the energy storage PCS includes a DC / AC converter, its control system, and filter circuits. Therefore, the corresponding PCS losses mainly consist of converter losses and filter losses.

[0076] It should be noted that the losses of the converter mainly include static losses and dynamic losses. Static losses are further divided into off-state and on-state losses. Off-state losses account for a small proportion of the total losses and can be ignored. On-state losses are affected by temperature, current, and voltage across the converter. Dynamic losses are switching losses, which are affected by temperature and current. The filter losses are also affected by temperature and current.

[0077] When the PCS topology and modulation method remain unchanged, and the temperature is maintained stable by the temperature control system, the following PCS power loss model can be established:

[0078] in, These are the total PCS loss, converter loss, and filter loss, respectively. This refers to the DC-side current of the PCS. This refers to the DC-side voltage of the PCS. This is a correction factor for the change in current. This is a correction factor for switching losses. This is a correction value for the inductor on-resistance in the converter. This is the correction amount for the inductor's on-resistance in the filter. The iron loss in the filter loss, the , These are the parameters to be identified.

[0079] It should be noted that, This refers to the static loss component of the converter losses. This refers to the dynamic loss component of the converter losses.

[0080] It should be further noted that, depending on the model of the energy storage PCS, the PCS DC current measurement... The value ranges from tens to hundreds of amperes; PCS DC voltage measurement The value can be 400V, 750V, or 1000V; this is the correction value for the inductor on-resistance in the converter. Correction amount of inductor on-resistance in filter The value ranges from tens of milliohms to 1 ohm. , The correction factor takes a value between 0 and 1.

[0081] In the specific implementation process, the DC side voltage of the PCS Typically 750V, PCS DC current measurement The correction amount of the inductor on-resistance in the converter can be obtained through measurement. Correction amount of inductor on-resistance in filter Typically around 500 milliohms.

[0082] Furthermore, the operating efficiency characteristics of energy storage power station transformers are largely consistent with those of conventional transformers, being significantly affected by the load factor, which in turn is strongly correlated with the power factor. However, since the power transmission of conventional transformers is generally unidirectional, their power factor typically varies within a certain positive range. Energy storage power station transformers, on the other hand, need to meet the bidirectional regulation requirements of the power grid's active and reactive power. Therefore, when calculating their operating efficiency, the power factor cannot be considered a fixed value within a certain range; it should be considered as a variable parameter. In addition, reactive current also generates active power losses within the transformer. Therefore, the project divides the transformer power loss calculation into two parts: active and reactive power. The losses generated in the reactive power part are converted into active power losses using a correction factor.

[0083] In summary, this invention considers the transformer load factor and common angle as varying parameters, and establishes the following transformer power loss model:

[0084] in, This represents the total power loss of the transformer. , These are the transformer no-load loss and rated load loss, respectively. for The load factor of the transformer at any given time; This refers to the rated capacity of the transformer. Let be the apparent power actually output by the transformer at time t; They are respectively The actual active and reactive power output of the transformer at any given time, the aforementioned , These are the parameters to be identified.

[0085] It should be noted that the no-load loss of the transformer and transformer rated load loss The no-load loss of the transformer is determined by its capacity. Values ​​range from 1505W to 2590W, rated load loss of transformer. The value ranges from 8190W to 16605W, representing the rated capacity of the transformer. The value ranges from 1250kVA to 2500kVA. The actual active and reactive power output of the transformer at any time The value is determined by the current, voltage, and the phase between the current and voltage.

[0086] In the specific implementation process, when the rated capacity of the transformer used is When the transformer capacity is 1250kVA, the no-load loss is... The rated load loss of the transformer is 1505W. The active and reactive power outputs of the transformer at time t range from 8190W to 9335W, increasing with temperature. The value can be obtained through calculation.

[0087] Furthermore, considering the energy consumption characteristics of each power-consuming component in the temperature control system under both cooling and heating modes, this invention establishes a power loss model for the temperature control system based on HVDC. The temperature control system is a liquid-cooled system, and the model includes cooling demand loss, heating demand loss, total cooling loss, total heating loss, and air-cooled system loss. In cooling mode, the cooling demand loss of the liquid cooling system mainly includes circulation pump loss, refrigeration unit loss, and control loss. The refrigeration unit is a compressor. The system cooling demand loss model is as follows:

[0088] In heating mode, the heating demand loss of the liquid cooling system mainly includes circulation pump loss, heating device loss, and control loss. The heating device is a PTC electric heater. The system heating demand loss model is as follows:

[0089] in, For cooling demand losses, For heating demand losses, For battery heat loss, This refers to the energy loss required to heat the battery to its operating temperature. This refers to the heat power conducted from the external environment through the battery cabinet. The heat transfer coefficient of the battery cabinet enclosure. Let be the surface area of ​​the battery cabinet. The external temperature of the battery cabinet. The temperature set to activate the cooling mode of the liquid cooling system. The quality of the batteries inside the cabinet. For the specific heat capacity of the battery, The temperature set to activate the heating mode of the liquid cooling system. The temperature of the batteries inside the cabinet. Heating time, The external ambient temperature, the These are the parameters to be identified.

[0090] It should be noted that the heat transfer coefficient of the battery cabinet is... The value range is 100 ~1000 Battery cabinet surface area Battery quality inside the cabinet Battery specific heat capacity The value can be obtained based on the equipment model, and the external temperature of the battery cabinet. The temperature set when the liquid cooling system starts its cooling mode. The temperature set when the liquid cooling system starts heating mode Battery temperature inside the cabinet Heating time and external ambient temperature It can be obtained through measurement.

[0091] It needs to be further explained that when When the external temperature is higher than the set temperature, Cooling demand ;when When the external temperature is lower than the set temperature, ,and , Since ambient heat dissipation is insufficient to offset electromagnetic heating, a liquid cooling system is still required to meet the cooling demand. .

[0092] In cooling mode, the total cooling loss of the liquid cooling system mainly includes the loss of the circulating pump, the loss of the chiller unit, and the loss of the control system. The total cooling loss model is as follows:

[0093] In heating mode, the total heating loss of the liquid cooling system mainly includes the loss of the circulating pump, the loss of the PTC electric heating system, and the loss of the control system. The total heating loss model is as follows:

[0094] in, This represents the total cooling loss. Total heating loss, For circulation pump losses, For ethylene glycol flow rate, For Yang Cheng, For coolant density, It is the acceleration due to gravity. For compressor wear, To control system losses, For cooling demand losses, The compressor's refrigeration efficiency ratio. For heating demand losses, For pump efficiency, the These are the parameters to be identified.

[0095] It should be noted that the circulating pump is a core loss component of the liquid cooling system, and the influence of the density and viscosity of ethylene glycol needs to be considered. The head is the pipeline resistance that the pump needs to overcome. The compressor loss is related to the cooling capacity and energy efficiency ratio. The control system loss is the loss of auxiliary equipment, which mainly includes controllers, sensors and valves. The control system loss is a fixed value.

[0096] It should be further explained that the pump efficiency The value is typically 0.5~0.8, for ethylene glycol flow rate. Yangcheng and coolant density It can be obtained through direct or indirect measurement, with indirect measurement referring to calculations based on measurable data. Gravitational acceleration. Take 9.8 compressor refrigeration efficiency ratio The value typically ranges from 2.5 to 4.0, which affects the control system losses. The value is usually between 50w and 200w.

[0097] The coolant can be 50% ethylene glycol. When using 50% ethylene glycol at 25°C, the coolant density is... Approximately 1065 .

[0098] It should be further explained that the thermal efficiency of PTC electric heating is close to 100%, so the PTC electric heating loss is approximately equal to the heating demand loss.

[0099] Furthermore, the PCS cabinet and transformer cabinet are equipped with an air-cooling system to effectively dissipate the heat generated during equipment operation, ensuring normal operation and extending the service life of the equipment.

[0100] The losses in an air-cooled system are primarily determined by the airflow volume based on heat dissipation requirements, and the loss model is derived by considering the operating environment and equipment parameters.

[0101] in, For fan wear, The airflow required for heat dissipation For heat dissipation requirements, air density, The specific heat capacity of air, To allow for the temperature rise inside the cabinet, For airflow resistance, For fan efficiency, For PCS loss, For transformer losses, the These are the parameters to be identified.

[0102] It should be noted that the airflow required for heat dissipation Air density can be calculated or measured using existing data. The specific heat capacity of air is approximately 1.2 kg / m³ at 25°C and standard atmospheric pressure. The value is approximately 1005 J / (kg∙℃), which represents the allowable temperature rise inside the cabinet. The fan efficiency can be set to 10℃ or 15℃. The value is usually between 0.3 and 0.6.

[0103] It should be further clarified that the losses in the air-cooling system refer to the losses of the fans on the PCS or transformer, and the heat dissipation requirements... At that time, the calculation is based on the fan losses on the PCS, and the heat dissipation requirement. At that time, the calculation was performed on the fan losses on the transformer.

[0104] It should be further noted that the parameter to be identified is a quantity that changes in real time or changes with the current, so an algorithm model is required for identification.

[0105] Step 2: Collect N sets of sample data. The sample data includes defined input and output data pairs. The output data includes PCS power loss, transformer power loss, power loss of the temperature control system (mainly HVDC), and corresponding actual losses. To improve the stability of the identification, the sample data undergoes outlier handling, filtering, denoising, and normalization preprocessing to ensure the numerical stability and convergence efficiency of the model during iterative optimization. The sample data is divided into a training set and a test set. The training set is used for optimization, and the test set is used for evaluation.

[0106] It should be noted that the determined input data mentioned above are the determined parameters in the energy storage power station power loss prediction model established in step 1, specifically including: Battery module power loss PCS DC current measurement PCS DC voltage measurement Transformer rated capacity The actual active power output of the transformer at time t The actual reactive power output of the transformer at time t Battery cabinet surface area External temperature of battery cabinet The temperature set when the liquid cooling system starts in cooling mode. Battery quality inside the cabinet Battery specific heat capacity Battery temperature inside the cabinet Required air volume air duct resistance .

[0107] It should be further clarified that the determined input data does not refer to constant data, but rather to known parameters or data that can be easily obtained by those skilled in the art based on actual operating equipment and requirements in practical applications. The output data includes PCS power loss, transformer power loss, and power loss of the temperature control system based on HVDC, which refers to the preliminary estimated data obtained based on the power loss prediction model.

[0108] In addition, the training set contains the defined input data and the corresponding estimated output data, which are used to optimize the objective function. The training process minimizes the error between the model's estimated loss and the actual loss. The test set contains the same input data, but the corresponding output is the actual loss, which evaluates the model's performance and accuracy in real-world scenarios.

[0109] Furthermore, outlier handling involves processing unreasonable values, also known as outliers, present in the dataset. Outliers can be identified as follows: an interval is set for the entire sample data; statistical analysis is performed on the sample data; values ​​outside the interval are outliers. When the data follows a normal distribution... The probability of an exception is less than 0.003, which is an extremely low probability event. This occurs when the sample distance from the average value is greater than... If the sample is an outlier, then the outlier is identified through box plot analysis.

[0110] The filtering and denoising includes amplitude limiting filtering, median filtering, and mean filtering. Amplitude limiting filtering sets a threshold to keep the portion of the sample data that does not exceed the threshold unchanged, preserving most of the features of the sample data while making the overall data sample more stable. Median filtering replaces the value of each sample point with the median value of the corresponding interval, preserving the edge features of the sample data. Mean filtering achieves filtering by calculating the neighborhood average of each sample point, smoothing the signal and reducing noise, while preserving the overall trend of the sample data. The normalization process transforms sample data into a standard normal distribution or standardizes and normalizes it, thereby enhancing model training effectiveness and improving prediction performance. At the same time, the standardized data helps to address numerical problems that may arise during the calculation process, thus improving the robustness of the algorithm.

[0111] The division into training and testing sets can be achieved through cross-validation, a method that can be implemented by those skilled in the art.

[0112] It should also be noted that the input and output data pairs also include the corresponding parameters to be identified for each pair of data. During the data sample collection phase, the corresponding parameters to be identified are manually determined values, usually uniformly sampled within the range of values ​​of the corresponding identification parameters, as the initial values ​​of the optimization model. The optimization model mentioned here refers to the initial objective function and the reinforcement objective function.

[0113] Step 3: Based on N sets of samples, construct an initial objective function using the least squares method and calculate the loss deviation. The optimization objective of the initial objective function is to minimize the deviation between the model's calculated loss and the actual operation loss, which serves as the optimization criterion for parameter identification. Specifically:

[0114] in, Let be the calculated model loss value for the i-th sample. The actual operating loss corresponds to the sample point, where N is the total number of samples. For the parameters to be identified, For regularization terms, As the regularization factor, in the initial objective function stage, the regularization factor is... If the value is set to 0, the current optimal solution is a parameter vector obtained by minimizing the overall loss deviation of N samples, and the current optimal solution is a local optimal solution.

[0115] It should be noted that the total sample size N is determined based on actual needs and can be 1000 or 2000, as well as the regularization factor. The value of is usually between 0 and 1.

[0116] Furthermore, the introduction of the regularization term can alleviate the overfitting problem. In the initial stage, in order to obtain the current optimal solution, the regularization factor is set to 0. The parameter vector includes the input data of each loss model and the current group optimal initial value of the parameter to be identified.

[0117] Step 4: Calculate a multi-stage closed-loop verification mechanism, setting the optimization batch and the number of iterations per batch. Further, the multi-stage closed-loop verification mechanism is divided into an inner loop and an outer loop. The inner loop optimizes parameters on the training set, and the outer loop evaluates error performance on the test set, further adjusting the objective function. This mechanism ensures that the identification process gradually converges to a stable solution in the closed-loop iteration of training-verification-feedback adjustment, alleviating local convergence and overfitting phenomena, and improving the model's generalization ability.

[0118] Furthermore, local convergence refers to the phenomenon that the algorithm can only guarantee that the iteration point converges to the optimal point when the initial point and the optimal point are close. The purpose of avoiding local convergence is to achieve global convergence, that is, the iteration point generated from any initial point converges to the optimal point of the problem.

[0119] Furthermore, the overfitting phenomenon refers to the phenomenon that the model performs well on the training set, but its performance drops significantly on the test set, indicating weak generalization ability. The "good performance" can refer to high accuracy.

[0120] Step 5: In the optimization process of each batch, the Particle Swarm Optimization (PSO) algorithm is introduced to perform global optimization of the objective function. Furthermore, the particle swarm parameters are first initialized, and the initial positions and velocities of the particles are set. In the inner loop iterative optimization process, the positions and velocities of the particles are updated sequentially, the fitness of the particles is calculated, the historical best fitness and position of individual particles are updated, and the historical best fitness and position of the swarm are updated. The global search capability of the PSO algorithm can effectively avoid the risk of getting trapped in local optima, thereby improving the stability and robustness of the optimization results.

[0121] It should be noted that the initial particle swarm parameters also include: particle swarm size, particle dimension, inertia weight, learning factor, and iteration step size range.

[0122] It should be further explained that the inertial weight represents the displacement influence of the previous generation of particles on the current generation of particles, and represents the degree of confidence of the particles in their own motion state. The larger the inertial weight, the stronger the particle's ability to explore new regions and the stronger its global optimization ability, but the corresponding local optimization ability will be weakened. A smaller inertial weight is beneficial for local search, allowing the algorithm to converge to the optimal solution.

[0123] Step 6: Multidimensional parameter fitting verification and reinforcement.

[0124] Step 6 specifically includes: Step 6.1: Evaluate the fitting results using multi-dimensional error indices. Further, the multi-dimensional error indices refer to the mean square error, root mean square error, mean absolute error, maximum absolute error, and minimum absolute error calculated on the test set. These indices reflect the model's performance in terms of overall accuracy, fluctuation range, and extreme errors from multiple perspectives, thereby characterizing the parameter identification effect and determining whether the results are reasonable.

[0125] Furthermore, the mean squared error is specifically as follows:

[0126] in, For the first The true value of each sample For the first The predicted value for n samples, where n is the number of samples.

[0127] The mean squared error measures the average of the squared errors between the predicted and actual values, and is more sensitive to larger errors. The root mean square error is specifically:

[0128] Where MSE is the mean squared error.

[0129] The root mean square error has the same dimensions as the original data, which more intuitively reflects the magnitude of the prediction error. The mean absolute error is specifically:

[0130] in, For the first The true value of each sample For the first The predicted value for n samples, where n is the number of samples.

[0131] The mean absolute error reflects the overall average deviation and is not sensitive to outliers; The maximum absolute error is specifically:

[0132] in, For the first The true value of each sample For the first The predicted value for each sample.

[0133] The maximum absolute error highlights the worst-case scenario. The minimum absolute error is specifically:

[0134] in, For the first The true value of each sample For the first The predicted value for each sample.

[0135] The minimum absolute error can intuitively determine the overall difference of the optimal value.

[0136] In summary, using mean square error, root mean square error, mean absolute error, maximum absolute error, and minimum absolute error simultaneously allows us to observe overall performance, extreme cases, and sensitivity to outliers. In parameter identification tasks, this approach considers both the overall fit and prevents excessive errors under certain conditions.

[0137] Step 6.2: Decision verification is performed based on the error index evaluation results. The decision verification involves setting the current batch as the completed batch if the identification parameters do not show obvious underfitting or overfitting phenomena and meet the actual needs; otherwise, a reinforcement objective function is introduced for further optimization, adaptively adjusted based on test set performance, and the regularization factor is adjusted accordingly. Then, a strengthened objective function is introduced, and the next batch of operations begins.

[0138] Furthermore, the strengthening objective function is:

[0139] in, Let be the calculated model loss value for the i-th sample. The actual operating loss corresponds to the sample point, where N is the total number of samples. For the parameters to be identified, For regularization terms, As a regularization factor, These are the error weighting coefficients.

[0140] It should be noted that, and Adaptive adjustments are made based on the performance on the test set. Specifically, when overfitting occurs, the [adjustment level] is increased. Strengthen regularization; when underfitting occurs, maintain or reduce regularization. Relaxing the fitting ability; when the sample error When it exceeds the threshold, increase This improves the model's fitting accuracy at key points; when samples error When it is less than the threshold, decrease To reduce its interference.

[0141] It should be further explained that the error weighting coefficient The value of the regularization factor is 0 to 1. Sum of error weighting coefficients Manually adjust the values ​​within the range to enhance the fitting effect of the optimized model.

[0142] In the specific implementation process, regularization factor Sum of error weighting coefficients An initial value needs to be assigned, which is usually 0.1, so that it can be adjusted later. The initial value of the threshold is set to 5. During the iterative optimization process, the threshold can be reduced and further adjusted to 3.

[0143] Furthermore, through and The dual adaptive adjustment ensures that the model maintains a balance between fitting accuracy and complexity, while improving the model's ability to identify key samples and enhancing its practicality.

[0144] It should be noted that the purpose of introducing the error weighting coefficient is to alleviate the underfitting phenomenon, which refers to the inability to fully capture the patterns of the sample data and the insufficient degree of fit to the data.

[0145] Step 7: After completing the set batch, multiple sets of candidate identification parameter solutions are obtained. The candidate solutions are substituted into the test set to calculate the error index. The solutions are sorted according to the overall loss deviation. The identification parameters with the best performance in the test set are selected as the global optimal solution. Furthermore, the global optimal solution can retain one or more sets of optimal solutions to adapt to different application requirements.

[0146] It should be noted that in different batches, there may be two sets of identification parameters that both meet the requirements of the optimal solution. Therefore, it is necessary to retain one or more sets of optimal solutions to prevent the reduction of data accuracy due to the uniformity of the solution during actual operation.

[0147] Example 2, parameter identification of the transformer loss model, including: The parameter identification method of this invention yields fitted parameters with small deviations for the transformer loss model, specifically: Select a set of optimal solution pairs of identification parameters to obtain Figure 2 The fitted parameters of the transformer's charging and discharging states, i.e., the identified parameters, are used for multi-dimensional parameter fitting verification based on the obtained identified parameters. The results are as follows: Figure 3 As shown, under charging conditions, the fitting results exhibit a maximum absolute error of 6.156 and a minimum absolute error of only 0.054, indicating excellent fitting of some data. The overall error is small (MSE=5.855, RMSE=2.419), and MAE=1.908, reflecting the model's ability to capture transformer losses during charging. Similarly, under discharging conditions, the maximum absolute error is 26.793, indicating the presence of outliers in the data, suggesting potential overfitting. This can be further optimized by increasing the training batch size. The minimum absolute error is 0.003, with MSE=7.970, RMSE=2.823, and MAE=2.112, indicating a relatively small overall error. The error comparison under charging and discharging conditions verifies the model's correctness. Furthermore, through... Figure 4 Figure 5 It can be seen that the overall error is not large, but there are abnormal data that cause a small number of predictions to have larger deviations.

[0148] Specifically, according to Figure 4 and Figure 5The view shows the actual loss value and the model prediction value. The overall prediction value will change according to the trend of the actual loss value. When the actual loss value changes slowly, the prediction value will stabilize at the same level. When the actual loss value changes rapidly, the prediction value will also drop or rise rapidly. This shows the fitting effect of the present invention on the identification parameters. The phenomenon of prediction value deviation can be alleviated by further optimization or reprocessing the outliers of the sample data.

[0149] Furthermore, according to Figure 4 and Figure 5 The error distribution, besides Figure 5 The results show one value with a large deviation, while the rest are normally distributed with the mean error of 0 as the center. This demonstrates the stability of the accuracy of parameter identification by the method of the present invention.

[0150] from Figure 6 , Figure 7 It can be seen that the transformer loss is positively correlated with the transformer load rate under charging and discharging conditions, that is, the higher the load, the higher the transformer loss, which verifies the rationality of the basis provided by the constructed transformer loss model.

[0151] In summary, the deviation between the fitted parameters and the actual operating data is small, fully verifying the accuracy of the established model and the rationality of the parameter fitting method. This result demonstrates the effectiveness and superiority of the optimization method employed in this invention in parameter identification, providing reliable technical support for efficiency system modeling and performance analysis.

[0152] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0153] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely to distinguish one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0154] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying efficiency model parameters of an electrochemical energy storage system, characterized in that, include: Step 1: Establish a power loss prediction model for the energy storage power station, covering the PCS power loss model, the transformer power loss model, and the temperature control system power loss model based on HVDC. Identify the parameters to be identified during the model building process. Step 2: Collect N sets of sample data, perform data preprocessing, and divide the sample data into training set and test set; Step 3: Based on N sets of samples, construct an initial objective function using the least squares method, calculate the loss deviation as the optimization criterion for parameter identification, and determine the current optimal solution; Step 4: Design a multi-stage closed-loop verification mechanism, and set the optimization batch and the number of iterations per batch; Step 5: In each batch of optimization iterations, the Particle Swarm Optimization (PSO) algorithm is introduced to perform global optimization of the objective function; Step 6: Multi-dimensional parameter fitting verification and reinforcement; Step 6 specifically includes: Step 6.1: Evaluate the fitting results using multi-dimensional error metrics; Step 6.2: Verify the decision based on the error index evaluation results; introduce a reinforcement objective function and adaptively adjust it based on the performance of the test set; Step 7: After completing the set batch, multiple sets of candidate identification parameter solutions are obtained. The candidate solutions are substituted into the test set to calculate the error index. The solutions are sorted according to the overall loss deviation, and the identification parameters with the best performance on the test set are selected as the global optimal solution.

2. The method for identifying efficiency model parameters of an electrochemical energy storage system according to claim 1, characterized in that, Step 1 specifically includes: Step 1.1: Establish a power loss prediction model for energy storage power stations: ， in, This represents the total power loss of the energy storage system. For battery module power loss, For PCS power loss, For transformer power loss, The power loss of the temperature control system, which is mainly based on HVDC; Step 1.2: Establish the PCS power loss model: ， in, These are the total PCS loss, converter loss, and filter loss, respectively. This refers to the DC-side current of the PCS. This refers to the DC-side voltage of the PCS. This is a correction factor for the change in current. This is a correction factor for switching losses. This is a correction value for the inductor on-resistance in the converter. This is the correction amount for the inductor's on-resistance in the filter. The iron loss in the filter loss, the , The parameters to be identified; Step 1.3: Establish a transformer power loss model: ， in, This represents the total power loss of the transformer. , These are the transformer no-load loss and rated load loss, respectively. for The load factor of the transformer at any given time; This refers to the rated capacity of the transformer. Let be the apparent power actually output by the transformer at time t; They are respectively The actual active and reactive power output of the transformer at any given time, the aforementioned , The parameters to be identified; Step 1.4: Establish a power loss model for the temperature control system based on HVDC, including cooling demand loss, heating demand loss, total cooling loss, total heating loss, and air-cooled system loss: Cooling demand loss model: ， Heating demand loss model: ， in, For cooling demand losses, For heating demand losses, For battery heat loss, This refers to the energy loss required to heat the battery to its operating temperature. This refers to the heat power conducted from the external environment through the battery cabinet. The heat transfer coefficient of the battery cabinet enclosure. Let be the surface area of ​​the battery cabinet. The external temperature of the battery cabinet. The temperature set to activate the cooling mode of the liquid cooling system. The quality of the batteries inside the cabinet. For the specific heat capacity of the battery, The temperature set to activate the heating mode of the liquid cooling system. The temperature of the batteries inside the cabinet. Heating time, The external ambient temperature, the The parameters to be identified; Total refrigeration loss model: ， Total heating loss model: ， in, This represents the total cooling loss. Total heating loss, For circulation pump losses, For ethylene glycol flow rate, For Yang Cheng, For coolant density, It is the acceleration due to gravity. For compressor wear, To control system losses, For cooling demand losses, The compressor's refrigeration efficiency ratio. For heating demand losses, For pump efficiency, the The parameters to be identified; Loss model of air-cooled system: ， in, For fan wear, The airflow required for heat dissipation For heat dissipation requirements, air density, The specific heat capacity of air, To allow for the temperature rise inside the cabinet, For airflow resistance, For fan efficiency, For PCS loss, For transformer losses, the These are the parameters to be identified.

3. The method for identifying efficiency model parameters of an electrochemical energy storage system according to claim 1, characterized in that, Step 2 specifically includes: the sample data includes determined input data and output data pairs, the output data includes PCS power loss, transformer power loss, power loss of the temperature control system mainly based on HVDC and the corresponding actual loss, and the data preprocessing includes outlier processing, filtering and noise reduction and normalization.

4. The method for identifying efficiency model parameters of an electrochemical energy storage system according to claim 1, characterized in that, Step 3 specifically includes: the optimization objective of the initial objective function is to minimize the deviation between the model computational loss and the actual operational loss, specifically: ， in, Let be the calculated model loss value for the i-th sample. The actual operating loss corresponds to the sample point, where N is the total number of samples. For the parameters to be identified, For regularization terms, As the regularization factor, in the initial objective function stage, the regularization factor is... If the value is set to 0, the current optimal solution is a parameter vector obtained by minimizing the overall loss deviation of N samples, and the current optimal solution is a local optimal solution.

5. The method for identifying efficiency model parameters of an electrochemical energy storage system according to claim 1, characterized in that: The multi-stage closed-loop verification mechanism consists of an inner loop and an outer loop. The inner loop optimizes parameters on the training set, while the outer loop evaluates error performance on the test set and further adjusts the objective function.

6. The method for identifying efficiency model parameters of an electrochemical energy storage system according to claims 1 and 5, characterized in that: The global optimization process includes: first, initializing the particle swarm parameters, setting the initial position and velocity of the particles, and during the inner loop iterative optimization process, sequentially updating the position and velocity of the particles, calculating the fitness of the particles, updating the historical best fitness and position of individual particles, and updating the historical best fitness and position of the swarm.

7. The method for identifying efficiency model parameters of an electrochemical energy storage system according to claim 1, characterized in that, Step 6.1 specifically includes: the multi-dimensional error index refers to the mean square error, root mean square error, mean absolute error, maximum absolute error and minimum absolute error calculated on the test set, which reflects the model’s performance in terms of overall accuracy, fluctuation range and extreme error, and realizes the characterization of parameter identification effect and judges whether the result is reasonable.

8. The method for identifying efficiency model parameters of an electrochemical energy storage system according to claim 1, characterized in that, Step 6.2 specifically includes: the decision verification is as follows: when the identification parameters do not show obvious underfitting or overfitting phenomena and can meet the actual needs, the current batch is set as the completed batch; otherwise, a reinforcement objective function is introduced for further optimization, and adaptive adjustment is made based on the performance on the test set, by adjusting the regularization factor. Then, a strengthened objective function is introduced, and the next batch of operations begins.

9. A method for identifying efficiency model parameters of an electrochemical energy storage system according to claims 4 and 8, characterized in that, The enhancement objective function: ， in, Let be the calculated model loss value for the i-th sample. The actual operating loss corresponds to the sample point, where N is the total number of samples. For the parameters to be identified, For regularization terms, As a regularization factor, These are the error weighting coefficients. and Adaptive adjustments are made based on the performance on the test set. Specifically, when overfitting occurs, the system is increased... When underfitting occurs, maintain or reduce... When the sample error When it exceeds the threshold, increase When the sample error When it is less than the threshold, decrease .

10. The method for identifying efficiency model parameters of an electrochemical energy storage system according to claim 1, characterized in that, Step 7 specifically includes: the global optimal solution can retain one or more sets of optimal solutions to adapt to different application requirements.

Images