Intelligent management system of energy storage power station

Abstract

The invention discloses an intelligent management system of an energy storage power station. Belongs to the technical field of battery management. The invention discloses an intelligent management system for an energy storage power station, and the system comprises a photovoltaic monitoring module, a wind energy monitoring module, an energy storage monitoring module, a grid-connected control module, a power generation prediction module, an electric quantity prediction module, and a battery function control module. And generating a charging time period and a discharging time period of each battery based on the historical power generation data of the next time period, the time period of accessing the power grid and the output electric power, and controlling each battery to charge and discharge circularly. According to the technical scheme provided by the invention, the historical power generation data of the next time period is accurately obtained, so that the electric energy needing to be charged by the energy storage module in a future time period can be obtained, and meanwhile, the output electric power of the next time period is preset, so that the situation that the battery needs to be charged and discharged urgently is avoided; and the service life of the battery is prolonged.

Description

Technical Field

[0001] This application relates to the field of battery technology, and more specifically, to a smart management system for energy storage power stations. Background Technology

[0002] A microgrid is a small-scale power generation and distribution system consisting of distributed power sources (such as solar, wind, and energy storage systems), electrical loads, power distribution facilities, monitoring and protection devices, etc. It can achieve internal power balance and exchange energy with the external power grid.

[0003] Microgrids are typically built around energy storage power stations, which include wind power modules, solar power modules, energy storage modules, and a grid connection management module. The wind power modules generate wind power, the solar power modules generate solar power, and the energy storage modules store electrical energy. The grid connection management module manages when the microgrid connects to the grid.

[0004] To ensure maximum efficiency, microgrids typically store electricity at night and output it during the day. To guarantee power quality, the daytime output of the microgrid needs to remain stable at a preset value. Therefore, existing solutions calculate the average daily output of the microgrid based on historical wind and solar power generation data, and then input the stored electricity into the grid at rated power during peak electricity price periods each day.

[0005] However, in practice, microgrids typically supply power to the grid between 2 PM and 4 PM, a period when electricity is most scarce and solar energy is most abundant. This is also the time when the photovoltaic modules in the microgrid have the highest photoelectric conversion efficiency. As a result, the energy storage modules need to be charged and discharged simultaneously, which can interfere with the chemical reactions inside the batteries in the energy storage modules, leading to reduced battery efficiency and a shorter battery life. Summary of the Invention

[0006] The summary section of this application is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0007] To address the technical problems mentioned in the background section above, this application provides a smart management system for energy storage power stations, including:

[0008] A photovoltaic monitoring module is used to acquire photoelectric generation data from the photovoltaic power generation module.

[0009] The wind energy monitoring module is used to acquire wind power generation data from the wind power generation module;

[0010] The energy storage monitoring module is used to detect the power consumption data of each battery in the energy storage module;

[0011] The grid connection control module is used to pre-input the time period for the microgrid to connect to the grid and the output power.

[0012] The power generation prediction module calculates historical power generation data for the next time period based on solar power generation data and wind and solar power generation data.

[0013] The power prediction module generates the power of each battery based on power consumption data to obtain the charging and discharging capabilities of each battery.

[0014] The battery function control module generates charging and discharging time periods for each battery based on historical power generation data for the next time period, the time period of grid connection, and the output power, and controls the cyclic charging and discharging of each battery.

[0015] The technical solution provided in this application first accurately acquires the historical power generation data for the next time period, thus obtaining the electrical energy that the energy storage module needs to charge in the future. At the same time, the output power for the next time period is preset, thus determining the electrical energy that needs to be released in the future. Based on this, the battery function control module allows the battery to cycle through charging and discharging. Therefore, for each individual battery, it only needs to perform charging or discharging operations, thus avoiding the need for the battery to both charge and discharge, and increasing the battery's lifespan.

[0016] During the charging and discharging process of a battery, it is necessary to keep the charging current and discharging current of an individual battery within a reasonable range to avoid excessively high or low charging currents. Since the battery pack within an energy storage module needs to be divided into charging and discharging groups, an improper arrangement of the ratio of batteries in the charging and discharging groups will inevitably lead to excessively high charging or discharging currents. Based on this, this application provides the following technical solution:

[0017] The battery function control module calculates the number of batteries that need to be charged in the future based on historical power generation data for the next time period, and calculates the maximum output power for the next time period based on the number of batteries that need to be charged in the future.

[0018] In the technical solution provided in this application, considering the uncontrollability of charging, the number of batteries that need to be charged is first calculated based on historical power generation data. After determining the number of rechargeable batteries, the number of dischargeable batteries is then determined. Based on the number of dischargeable batteries, the maximum output power is finally determined. This allows for setting a reasonable output power when connected to the power grid, thereby avoiding excessive charging or discharging current during battery cycle operation, which could affect the battery's lifespan.

[0019] Generally speaking, wind power is far more random than solar power. Directly using the same model for both wind and solar power will cause the model to tend towards fitting under large datasets, resulting in low model accuracy. To address this issue, this application provides the following technical solution:

[0020] The power generation prediction module includes:

[0021] The wind power forecasting unit is used to predict wind power generation data for the next time period.

[0022] The photoelectric prediction unit is used to predict the photoelectric power generation data for the next time period.

[0023] In the technical solution provided in this application, wind power and solar power are predicted using two different units, thus ensuring the accuracy of the total power generation prediction.

[0024] The battery's capacity is predicted based on its current charge capacity, calculated using the charging or discharging current. While the charge entering and leaving the battery can be monitored via current, the actual amount of charge a battery can take in and output is related to its expected lifespan. A longer battery lifespan (lower internal resistance) results in less energy loss during charging and discharging, and vice versa. Therefore, predicting the battery's maximum charge and discharge capacity requires accurate evaluation of its performance (expected lifespan). However, battery lifespan evaluation is a dynamic process; simply relying on usage frequency and conditions is insufficient for accurate assessment. In traditional battery packs, multiple batteries are managed as a whole, allowing for accurate prediction of the pack's discharge and charge capabilities using large datasets (based on the normal distribution of battery performance). However, this solution requires switching between discharge and charge groups for each battery, necessitating accurate prediction of individual battery performance. Therefore, in order to ensure the accuracy of battery performance prediction, this application provides the following technical solution:

[0025] Furthermore, the power prediction module includes:

[0026] The data collection unit is used to collect the power consumption data of each battery i, where i is the index of the battery;

[0027] The data preprocessing unit is used to preprocess the power consumption data of each battery i to obtain the predicted data H for each battery. i ;

[0028] The data prediction unit inputs the predicted data of each battery into the CNN neural network model to obtain the expected lifespan of each battery.

[0029] The charge / discharge performance calculation unit converts the expected lifespan into a corresponding conversion ratio, calculates the current standard charge / discharge capacity based on the charge / discharge data, and multiplies the conversion ratio by the standard charge / discharge capacity to calculate the charge / discharge capacity of each battery.

[0030] In the technical solution provided in this application, data is collected and preprocessed for each battery, and corresponding prediction data is generated. The prediction is then input into a CNN neural network model, so the expected lifespan of each battery can be accurately predicted. The charging and discharging capacity of each battery is then measured based on the expected lifespan, resulting in higher accuracy. In this way, battery management can reach the lowest level of the battery, which facilitates the implementation of battery charging and discharging management.

[0031] Furthermore, the power consumption data includes: charge / discharge voltage and time curves, charge / discharge current and time curves, and charge / discharge temperature and time curves.

[0032] The data above can accurately describe the changes in battery performance after multiple cycles of use, thereby increasing the accuracy of model predictions.

[0033] Because this solution requires accurate prediction of the performance of each battery, it necessitates increasing model accuracy. However, increasing model accuracy inevitably introduces more input data. Excessive input data leads to too many data items, resulting in significant redundancy and complex relationships between weights within the model, making it difficult to accurately analyze the underlying patterns in the data. To address this issue, this application provides the following solution:

[0034] Furthermore, the data preprocessing unit includes:

[0035] Data dimensionality reducer, used to optimize prediction data H i Perform dimensionality reduction;

[0036] A normalizer is used to normalize data.

[0037] The correlation predictor is used to filter data based on charge-discharge correlation, thus filtering out incorrectly predicted data.

[0038] In this solution, by using data dimensionality reduction, normalization, and filtering, the complexity and redundancy of the data can be reduced, thereby reducing the burden on the model and increasing the model's operational capability and accuracy.

[0039] Furthermore, the dimensionality-reduced prediction data H i ={[t min(v1) [v1], [t]min(v2) [v2], [t] min(a1) [A1], [t] min(a1) [A1], [t] min(a2) [A2], [t] T1 [T1], [t] T2 ,T2]};

[0040] [t min(v1) ,v1]={SV1 s ≥V max}, s=1,2,3…,t min(v1) The time when the battery voltage first reaches its maximum value during charging, v1 is the voltage at which the battery terminal voltage first becomes no less than the maximum voltage value, s is the number of charging cycles, and SV1 is the maximum voltage value. s For the voltage data of the charging voltage versus time curve, V max This is the maximum rated charging voltage.

[0041] [t min(v2) v2]={SV2 d ≥V min}, d = 1, 2, 3, ..., t min(v1) v1 is the time when the battery voltage first reaches its minimum value during discharge; v2 is the voltage at which the battery terminal voltage first does not exceed the minimum voltage value; d is the number of discharge cycles; SV2 s For the voltage data of the discharge voltage versus time curve, V min This is the minimum rated charging voltage.

[0042] [t min(a1) A1] = {SA2] s ≤A max}, s=1,2,3…,t min(a1) A1 is the time when the battery current begins to decrease during discharge, and A2 is the current value when the battery current begins to decrease. max The maximum current value during charging, where s is the number of charging cycles; SA1 s The current data for the charging current versus time curve;

[0043] [t min(a2) A2] = {SA2] d ≥A1}, d=1,2,3…,t min(a2) A1 is the time when the battery current begins to rise during discharge; A2 is the current value when the battery current begins to rise; A1 is the maximum current value during discharge; d is the number of discharges; SA1 d The current data is from the discharge current versus time curve;

[0044] [t T1 ,T1]={ST1 s ≥T1 max}, s=1,2,3…,t min(T1) The time when the battery temperature first reaches its maximum value during charging, T1 is the highest temperature reached by the battery during charging, s is the number of charging cycles, and ST1 is the maximum temperature reached by the battery during charging. s For the temperature data of the charging temperature and time curve, T1 max This represents the maximum charging temperature.

[0045] [t T2 ,T2]={ST1 d ≥T2 max}, s=1,2,3…,t min(T2) T1 is the time when the battery temperature first reaches its maximum value during discharge, T2 is the highest temperature reached during charging, d is the number of charging cycles, and ST2 is the maximum temperature reached during charging. d For the temperature data of the discharge temperature and time curve, T2 max This represents the maximum discharge temperature.

[0046] In the technical solution provided in this application, when performing feature dimensionality reduction, the maximum and minimum values ​​with the most expressive features, as well as obvious abrupt change points, are found in advance from each data curve. These abrupt change points are actually the key points of the battery's power consumption data. This effectively reduces data dimensionality and data redundancy, effectively avoids data loss, and further increases the accuracy of model prediction.

[0047] Predicting the expected lifespan of a battery primarily relies on collected historical battery power consumption data. However, this historical data is affected by sensor accuracy. When sensor accuracy decreases, the collected data becomes inherently flawed, or exhibits drift without clear characteristics due to the reduced sensor measurement precision. In such cases, simple data processing cannot detect changes in sensor accuracy. Therefore, it is foreseeable that using this data to predict the expected battery lifespan will inevitably lead to a significant discrepancy between the predicted results and the actual situation, resulting in reduced accuracy. To address this, this application provides the following technical solution:

[0048] Furthermore, for the predicted data, the correlation coefficient P between each data point is calculated. If the correlation coefficient between each data point is within the preset range, the predicted data is considered normal; otherwise, the battery is deleted.

[0049]

[0050] Among them, P X,Y Represents data items and The correlation coefficient between them and These represent the Xth and Yth data items in the predicted data Hi, respectively.

[0051] The technical solution provided in this application calculates the correlation between data. It can be predicted that if there is a large correlation between the data, it means that the collected data is real, reliable and has reference value. The data input into the neural network model will have greater value for predicting battery life.

[0052] In the process of predicting battery lifespan, the battery undergoes continuous charge-discharge cycles, resulting in long time-series data. This leads to severe gradient decay during neural network propagation. Consequently, the model's internal parameters struggle to accurately adapt to gradient decay characteristics and capture the actual changes in battery lifespan, thus affecting the model's learning ability. Furthermore, in practice, battery lifespan (performance) is not a monotonically decreasing process but can also exhibit an increasing trend. Therefore, this application provides the following technical solution:

[0053] Furthermore, the data prediction unit includes:

[0054] Convolutional layers, in the input prediction data H i Slide the convolution kernel upwards, calculate the dot product between the kernel and the input data, generate feature maps, and then generate the prediction data H. i Local features;

[0055] Pooling layers receive the output from the previous convolutional layer as input and reduce the feature dimension;

[0056] The Attention layer performs a weighted summation of the latent features generated by each convolutional layer to generate the final predicted lifetime score U.

[0057] Furthermore, in the Attention layer,

[0058] R ki =tanh(u k ·q ki ·u i );

[0059] Among them, R ki Let u be an intermediate variable, representing the attention weight between the k-th position and the i-th position. k Represents k hidden states, q ki This represents the weight coefficient matrix to be learned, and tanh represents the bitangent activation function;

[0060]

[0061] pki represents the normalized attention weight, indicating the degree of attention paid by the k-th position to the i-th position;

[0062] R i This represents the hidden variable at the k-th position.

[0063] In the technical solution provided in this application, to address the technical problem that the internal parameters of the model are difficult to accurately adapt to the gradient decay feature and difficult to accurately capture the actual features of changes in the expected battery life, an Attention layer is introduced in the CNN neural network. This layer can filter out feature data with large weight coefficients and give priority to such information. At the same time, the introduced self-attention adaptation mechanism can adaptively allocate weights to each hidden state with a short duration to form a more stable long-term dependency relationship and increase the model's prediction accuracy. Attached Figure Description

[0064] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. Furthermore, throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0065] In the attached diagram:

[0066] Figure 1 This is a schematic diagram of the structure of an intelligent management system for an energy storage power station.

[0067] Figure 2 The curves show the discharge voltage versus time for different discharge cycles.

[0068] Figure 3 This is a schematic diagram of the data prediction unit.

[0069] Figure 4 This is a schematic diagram of the Attention layer. Detailed Implementation

[0070] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0071] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0072] The present application will now be described in detail with reference to the accompanying drawings and embodiments.

[0073] Reference Figure 1 The intelligent management system for energy storage power stations includes a photovoltaic monitoring module, a wind energy monitoring module, an energy storage monitoring module, a grid connection control module, an operation monitoring module, a power generation prediction module, a power generation prediction module, and a battery function control module.

[0074] The photovoltaic (PV) monitoring module acquires photovoltaic (PV) power generation data from the PV power generation modules. This data refers to the historical power generation data of the solar panels at the energy storage station, typically including power generation data from a recent period and historical data for the current time. The same applies to wind power generation data. The wind power monitoring module acquires wind power generation data from the wind power generation modules. The operation monitoring module provides intelligent alarms for operational status, intelligent charge and discharge management, loss analysis, and power statistics analysis. The PV, wind power, and operation monitoring modules are standard functional modules in current energy storage stations; their specific setup methods will not be described here.

[0075] The energy storage monitoring module is used to detect the power consumption data of each battery in the energy storage module; each battery is the smallest energy unit that the energy storage power station can control. Some batteries are fixed into battery packs by circuitry, making it impossible to switch the charge / discharge of individual batteries. Other batteries, however, can be individually charged / discharged, depending on the sensor placement and hardware circuit design. In this solution, to increase accuracy, it is necessary to have an individual charge / discharge switch for each battery. The specific circuit design is not shown here because the current output of a single lithium battery is not large, so the corresponding circuit design is not complex, and the circuit structure is not further provided.

[0076] The grid connection control module is used to pre-input the time period for the microgrid to connect to the grid and the output power. The purpose of building an energy storage power station is actually to input the collected electricity from new energy sources into the grid. However, unlike other energy sources, electricity cannot be stored on the grid. Therefore, the grid needs to adjust its power structure according to the input and load. Solar and wind power, due to their instability, do not meet the conditions for input into the main grid. Therefore, currently, the collected clean energy is stored in batteries. During the peak electricity demand period (afternoon), a stable power transmission request is sent to the grid in advance, and then the power is transmitted to the grid during the corresponding time period. This transforms disordered clean energy into ordered energy. For example, if the energy storage station finds that its own power can output 100 kWh for two hours at night, it sends a request to the national grid to transmit 100 kWh of electricity for two hours from 2 pm to 4 pm.

[0077] Of course, in practice, to ensure effectiveness, requests are generally not sent only after the energy storage station reaches its maximum storage efficiency; instead, adjustments are made dynamically. For example, if it is calculated that the energy storage station can be fully charged by 2 PM tomorrow, a power-on request will be sent to the State Grid ahead of time, thus ensuring that clean energy is not wasted during power transmission.

[0078] This dynamic adjustment, given the known power generation, can achieve maximum benefit based on the State Grid's dynamic electricity price. Specific grid connection management methods will not be described here. The power generation prediction module calculates historical power generation data for the next time period based on solar and wind / solar power generation data. The module includes: a wind power prediction unit for predicting wind power generation data for the next time period; and a solar power prediction unit for predicting solar power generation data for the next time period. Power generation prediction is existing technology, and those skilled in the art can accurately predict it.

[0079] The key to this solution lies in how to avoid simultaneous charging and discharging of batteries. Specifically, the battery function control module generates the charging and discharging time periods for each battery based on historical power generation data for the next time period, the grid connection time period, and the output power, and controls the cyclic charging and discharging of each battery.

[0080] For example, if the historical power generation for the next 24 hours is calculated to be 100, and considering that the electricity price is highest between 2 PM and 4 PM, it is projected that 100 units of power will be generated during that period. This 100 units of power output requires 300 batteries to operate at full power, leaving the remaining 200 batteries idle. Thus, during the charging phase within these 24 hours, priority is given to charging the 300 batteries intended for discharge, while the remaining 200 batteries remain idle. When the discharge phase is reached, the remaining 300 fully charged batteries discharge, and the remaining 200 batteries are used to receive potential wind or solar power.

[0081] Therefore, in this process, it is necessary to accurately predict the electrical energy that each battery can release. In practice, although the internal capacity of a battery can be roughly calculated based on the charging current, considering the losses during charging and battery aging, the estimated capacity needs to be corrected by a correction factor, which is actually the battery lifespan score. Therefore, this solution uses the following method to calculate the lifespan score:

[0082] The power prediction module includes: a data collection unit, a data preprocessing unit, a data prediction unit, and a charge / discharge performance calculation unit.

[0083] The data collection unit and data preprocessing unit are mainly used for data cleaning and preprocessing. The data prediction unit is used to calculate the expected lifespan, or the battery's aging fraction. The charge / discharge performance calculation unit calculates how much electricity the battery can discharge or charge based on its expected lifespan.

[0084] refer to Figure 2 Specifically, the electricity consumption data includes: charge / discharge voltage and time curves, charge / discharge current and time curves, and charge / discharge temperature and time curves.

[0085] The above data is the most critical data for battery life. Although battery internal resistance is also important, there is a certain relationship between battery internal resistance, voltage and current. Introducing it would lead to low data redundancy. After several attempts, this solution directly removes it to reduce the overfitting phenomenon of the model.

[0086] The data collection unit is used to collect the power consumption data of each battery i, where i is the index of the battery;

[0087] The data preprocessing unit is used to preprocess the power consumption data of each battery i to obtain the predicted data H for each battery. i ;

[0088] The data preprocessing unit includes: a data dimensionality reducer, a normalizer, and a correlation predictor.

[0089] Data dimensionality reducer, used to optimize prediction data H i Perform dimensionality reduction;

[0090] The preprocessing in this scheme mainly involves dimensionality reduction of the data to reduce its complexity. The dimensionality-reduced prediction data H i ={[t min(v1) [v1], [t] min(v2) [v2], [t] min(a1) [A1], [t] min(a1) [A1], [t] min(a2) [A2], [t] T1 [T1], [t] T2 ,T2]};

[0091] [t min(v1) ,v1]={SV1 s ≥V max}, s=1,2,3…,t min(v1) The time when the battery voltage first reaches its maximum value during charging, v1 is the voltage at which the battery terminal voltage first becomes no less than the maximum voltage value, s is the number of charging cycles, and SV1 is the maximum voltage value. s For the voltage data of the charging voltage versus time curve, V max This is the maximum rated charging voltage.

[0092] [t min(v2) v2]={SV2 d ≥V min}, d = 1, 2, 3, ..., t min(v1) v1 is the time when the battery voltage first reaches its minimum value during discharge; v2 is the voltage at which the battery terminal voltage first does not exceed the minimum voltage value; d is the number of discharge cycles; SV2 s For the voltage data of the discharge voltage versus time curve, V min This is the minimum rated charging voltage.

[0093] [t min(a1) A1] = {SA2] s ≤A max}, s=1,2,3…,t min(a1) A1 is the time when the battery current begins to decrease during discharge, and A2 is the current value when the battery current begins to decrease. max The maximum current value during charging, where s is the number of charging cycles; SA1 s The current data for the charging current versus time curve;

[0094] [t min(a2) A2] = {SA2] d ≥A1}, d=1,2,3…,t min(a2)A1 is the time when the battery current begins to rise during discharge; A2 is the current value when the battery current begins to rise; A1 is the maximum current value during discharge; d is the number of discharges; SA1 d The current data is from the discharge current versus time curve;

[0095] [t T1 ,T1]={ST1 s ≥T1 max}, s=1,2,3…,t min(T1) The time when the battery temperature first reaches its maximum value during charging, T1 is the highest temperature reached by the battery during charging, s is the number of charging cycles, and ST1 is the maximum temperature reached by the battery during charging. s For the temperature data of the charging temperature and time curve, T1 max This represents the maximum charging temperature.

[0096] [t T2 ,T2]={ST1 d ≥T2 max}, s=1,2,3…,t min(T2) T1 is the time when the battery temperature first reaches its maximum value during discharge, T2 is the highest temperature reached during charging, d is the number of charging cycles, and ST2 is the maximum temperature reached during charging. d For the temperature data of the discharge temperature and time curve, T2 max This represents the maximum discharge temperature.

[0097] The dimensionality reduction in this solution is actually about finding feature points where the information in various data of the battery is offset. Compared with redundant data, this data is more likely to indicate the direction of the battery usage offset.

[0098] A normalizer is used to normalize data; normalization is an existing technology and will not be elaborated on here.

[0099] The correlation predictor is used to filter data based on charge-discharge correlation, eliminating incorrectly predicted data.

[0100] For the predicted data, calculate the correlation coefficient P between each data point. If the correlation coefficient between each data point is within the preset range, the predicted data is normal; otherwise, delete the battery.

[0101]

[0102] Among them, P X,Y Represents data items and The correlation coefficient between them and Let X and Y represent the Xth and Yth data points in the predicted data Hi, respectively. The value of P ranges from -1 to 1, where 1 represents a perfect positive correlation, -1 represents a perfect negative correlation, and 0 represents no linear correlation. Therefore, in practice, only a reasonable threshold needs to be set to control the amount of data introduced. If high prediction requirements are needed, the preset value should be set larger; otherwise, it should be set smaller. For batteries corresponding to inaccurate prediction data, maintenance indicators should be compared, requiring workers to handle them periodically.

[0103] The above describes the preprocessing and filtering methods for the predicted data. After preprocessing and filtering are completed, for each battery, it is only necessary to input its predicted data into the prediction unit.

[0104] refer to Figure 3 The data prediction unit inputs the predicted data of each battery into the CNN neural network model to obtain the expected lifespan of each battery.

[0105] Specifically, the data prediction unit includes: an input layer, convolutional layers, pooling layers, attention layers, and an output layer. The input layer is used to input the prediction data, and the convolutional layers process the input prediction data H... i Slide the convolution kernel upwards, calculate the dot product between the kernel and the input data, generate feature maps, and then generate the prediction data H. i The local features; the pooling layer, which receives the output from the previous convolutional layer as input and reduces the feature dimension; the pooling process is the process of reducing the dimensionality of the data. There are usually multiple convolutional layers and pooling layers. The hidden features obtained from multiple convolutional layers are eventually input to the fully connected layer, which performs feature fusion. In this scheme, the fully connected layer is replaced by the Attention layer.

[0106] refer to Figure 4 The Attention layer performs a weighted summation of the latent features generated by each convolutional layer to generate the final predicted lifetime score U.

[0107] In the Attention layer,

[0108] R ki =tanh(u k ·q ki ·u i );

[0109] Among them, R ki Let u be an intermediate variable, representing the attention weight between the k-th position and the i-th position. k Represents k hidden states, q ki This represents the weight coefficient matrix to be learned, and tanh represents the bitangent activation function;

[0110]

[0111] pki represents the normalized attention weight, indicating the degree of attention paid by the k-th position to the i-th position;

[0112] R i This represents the hidden variable at the k-th position.

[0113] Because the aging levels of the batteries are not uniform, the resulting predicted lifetime scores U are inconsistent. To increase model accuracy, the loss function LS for the data prediction unit is:

[0114]

[0115] Among them, U i It is the actual value of the predicted lifespan score. It is the predicted value of the life expectancy score.

[0116] This approach uses the average percentage error, which can accurately assess the prediction accuracy at different actual values ​​and is independent of the data units. Therefore, it can effectively detect model errors and increase prediction accuracy.

[0117] The above outlines the basic structure of the model and its corresponding loss function, accurately representing the training method and process. However, optimizing the model requires continuous updates to the weight parameters in the convolutional neural network. Existing update methods are slow when dealing with the large sample range required in this approach, significantly increasing the training cost. Therefore, this solution employs the following method for updating the weight parameters:

[0118] Set the initial learning rate and step size α, and the first moment Q. 1 Vector and second moment Q 2 The attenuation coefficients of the vector are set to L1 and L2, and the weight parameter θ is optimized using the loss function LS. The optimization process is as follows:

[0119]

[0120]

[0121] Among them, Q 1 Q is a first-order moment vector used to smooth gradients and reduce the influence of noise on the gradient direction. 2 Let be a second-order moment vector used to adaptively adjust the learning rate for each parameter, t be the time step, represent the current iteration number, b be a non-zero constant, and E be the second-order moment vector. t It is the gradient of the objective function at the current time step with respect to the parameter θ.

[0122] The charge / discharge performance calculation unit converts the expected lifespan into a corresponding conversion ratio, calculates the current standard charge / discharge capacity based on the charge / discharge data, and multiplies the conversion ratio by the standard charge / discharge capacity to calculate the charge / discharge capacity of each battery.

[0123] The conversion ratio can be generated by the charge / discharge performance calculation unit simply by using statistical analysis; the specific method will not be elaborated here.

[0124] The above description is merely a selection of preferred embodiments of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this application.

Claims

1. An intelligent management system for an energy storage power station, comprising: Photovoltaic monitoring module, used to obtain photoelectric generation data of photovoltaic power generation modules; A wind energy monitoring module is used to obtain wind power generation data of the wind power generation module; Energy storage monitoring module, used to detect the power consumption data of each battery in the energy storage module; The grid connection control module is used to pre-enter the time period for the microgrid to connect to the grid and the output power; Operation monitoring module, used for intelligent alarm of operation status, intelligent charge and discharge management, loss analysis, and power statistics analysis; It is characterized by: further comprising: The power generation prediction module calculates the historical power generation data for the next time period based on photovoltaic generation data and wind and solar power generation data; The power prediction module generates the power of each battery based on the power consumption data to obtain the charge and discharge capacity of each battery; The battery function control module generates the charging and discharging time periods for each battery based on the historical power generation data of the next time period, the time period of access to the grid, and the output power, and controls the cyclic charging and discharging of each battery.

2. The energy storage power station intelligent management system according to claim 1, characterized in that: The battery function control module calculates the number of batteries that need to be charged in the future based on the historical power generation data of the next time period, and calculates the maximum output power in the future time period based on the number of batteries that need to be charged in the future.

3. The energy storage power station intelligent management system according to claim 1, characterized in that: The power generation forecast module includes: Wind power prediction unit, used to predict wind power generation data for the next time period; The photovoltaic prediction unit is used to predict the photovoltaic power generation data for the next time period.

4. The energy storage power station intelligent management system according to claim 1, characterized in that: The power forecast module includes: A data collection unit, configured to collect power usage data of each battery i, where i is the battery index; The data preprocessing unit is used to preprocess the power consumption data of each battery i to obtain the predicted data H of each battery i ; A data prediction unit, in which the prediction data of each battery is input into a CNN neural network model to obtain the expected life of each battery; The charge and discharge performance calculation unit converts the expected life into a corresponding conversion ratio, calculates the current standard charge and discharge capacity based on the charge and discharge data, and multiplies the conversion ratio by the standard charge and discharge capacity to calculate the charge and discharge capacity of each battery.

5. The energy storage power station intelligent management system according to claim 4, characterized in that: The electricity consumption data includes: charge and discharge voltage and time curve, charge and discharge current and time curve, charge and discharge temperature and time curve.

6. The energy storage power station intelligent management system according to claim 4, characterized in that: The data preprocessing unit includes: Data dimensionality reducer, used to predict the data H i Perform dimensionality reduction; Normalizer, used to normalize the data; The correlation predictor is used to screen the data for charge and discharge correlation and filter out the data with incorrect prediction.

7. The energy storage power station intelligent management system according to claim 6, characterized in that: The predicted data H after dimensionality reduction i ={[t min(v1) ,v1],[t min(v2) ,v2],[t min(a1) ,A1],[t min(a1) ,A1],[t min(a2) ,A2],[t T1 , T1], [t T2 , T2]}; [t min(v1) ,v1]={SV1 s ≥V max }, s=1,2,3…,t min(v1) The time when the voltage first reaches the maximum value when the battery is charged, v1 is the voltage value at which the battery terminal voltage is first not less than the maximum voltage value, s is the number of charging times, SV1 s is the voltage data of the charging voltage and time curve, V max is the maximum rated charging voltage; [t min(v2) ,v2]={SV2 d ≥V min }, d=1,2,3…, t min(v1) SV2 is the time when the voltage first reaches the minimum value during battery discharge, v2 is the voltage value at which the battery terminal voltage is first not greater than the minimum voltage value, and d is the number of discharges; s is the voltage data of the discharge voltage and time curve, V min is the minimum value of the rated charging voltage; [t min(a1) ,A1]={SA2 s ≤A max }, s=1,2,3…,t min(a1) is the time when the current starts to decrease when the battery is discharged, A1 is the current value when the battery current starts to decrease, A max The maximum current value during charging, s is the number of charging times; SA1 s is the current data of the charging current and time curve; [t min(a2) , A2] = {SA2 d ≥A1}, d=1,2,3…, t min(a2) is the time when the current starts to rise when the battery is discharged, A2 is the current value when the battery current starts to rise, A1 is the maximum current value when discharging, and d is the number of discharges; SA1 d is the current data of the discharge current and time curve; [t T1 , T1] = {ST1 s ≥T1 max }, s=1,2,3…,t min(T1) The time when the temperature first reaches the maximum value when the battery is charging, T1 is the highest temperature reached when the battery is charging, s is the number of charging times, ST1 s The temperature data of the charging temperature and time curve, T1 max is the maximum charging temperature; [t T2 , T2] = {ST1 d ≥T2 max }, s=1,2,3…,t min(T2) The time when the battery temperature first reaches the maximum value when discharging, T2 is the highest temperature reached when charging, d is the number of charging times, ST2 d is the temperature data of the discharge temperature data and time curve, T2 max is the maximum discharge temperature.

8. The energy storage power station intelligent management system according to claim 7, characterized in that: For the predicted data, the correlation coefficient P between each data item is calculated. If the range of the correlation coefficient between each data item is within the preset range, the predicted data is normal. If it is not within the preset range, the battery is deleted. Among them, P X,Y Represents a data item and The correlation coefficient between and They respectively represent the Xth data item and the Yth data item in the predicted data Hi.

9. The energy storage power station intelligent management system according to claim 7, characterized in that: The data prediction unit includes: Convolutional layer, in the input prediction data H i Slide the convolution kernel upward, calculate the dot product of the convolution kernel and the input data, generate a feature map, and generate the predicted data H i local features of The pooling layer receives the output from the previous convolutional layer as input and reduces the feature dimension; The Attention layer performs weighted summation on the implicit features generated by each convolutional layer to generate the final predicted life score U.

10. The energy storage power station intelligent management system according to claim 9, characterized in that: In the Attention layer, R ki =tanh(u k q ki ·u i ); Among them, R ki represents the intermediate variable, indicating the attention weight between the kth position and the ith position, u k Represents k hidden states, q ki Represents the weight coefficient matrix that needs to be learned, and tanh represents the bitangent activation function; pki represents the normalized attention weight, which indicates the degree of attention of the k-th position to the i-th position; R i represents the hidden variable at the kth position.

Images