A power grid energy storage dynamic regulation method based on real-time monitoring

Abstract

This invention relates to the field of dynamic regulation of grid energy storage, and particularly to a method for dynamic regulation of grid energy storage based on real-time monitoring. The method includes: using real-time monitored transformer operation data and photovoltaic energy storage device operation data from distribution substations to construct a multi-timescale parallel photovoltaic energy storage model based on a time-series convolutional network constrained by grid equations, to generate multi-timescale fused operation characteristics; using these multi-timescale fused operation characteristics to generate photovoltaic energy storage mode operation weights through an operation trend prediction model constructed based on a long short-term memory network and a multilayer perceptron architecture; and using the photovoltaic energy storage mode operation weights and distribution substation operation data to generate the optimal photovoltaic energy storage power of the photovoltaic energy storage device through a rolling time-domain optimization objective function. This invention achieves deep adaptation between the photovoltaic energy storage system and the operation of the distribution substation, ensuring the safe and stable operation of the distribution substation and improving the photovoltaic energy absorption capacity of the distribution substation.

Description

Technical Field

[0001] This invention relates to the field of dynamic regulation of power grid energy storage, and in particular to a method for dynamic regulation of power grid energy storage based on real-time monitoring. Background Technology

[0002] With the rapid popularization of distributed photovoltaic technology and the gradual reduction in the cost of energy storage devices, photovoltaic energy storage devices have been widely used in distribution areas, becoming a core means to realize the consumption of new energy, smooth grid load fluctuations, and improve the reliability and flexibility of power distribution.

[0003] As the final link between the power grid and users, the operation of distribution substations directly affects the power supply quality of the power grid and the user's electricity experience. The proper regulation and control of photovoltaic distribution energy storage devices is the key to giving full play to their efficiency and ensuring the safe and stable operation of distribution substations.

[0004] Currently, the control methods for photovoltaic-energy storage devices in distribution substations are mostly based on analysis and decision-making using operational data at a single time scale. They primarily rely on traditional PID control or fixed threshold control, failing to fully consider the dynamic and complex nature of distribution substation operation. On one hand, transformer operation data and photovoltaic-energy storage device operation data in distribution substations exhibit significant multi-time scale characteristics. Short-term time scales (minutes) show instantaneous fluctuations such as load shocks and sudden changes in photovoltaic output, while long-term time scales (hours and days) show trend changes such as alternating load peaks and valleys and diurnal variations in photovoltaic output. Existing single-time-scale feature extraction methods cannot comprehensively capture these multi-dimensional operational characteristics, resulting in insufficient targeting and accuracy in control decisions.

[0005] In addition, the optimization objectives of existing regulation methods are mostly single-dimensional, such as only pursuing the maximization of photovoltaic absorption rate or the minimization of energy storage loss. This results in the regulation decision not being closely integrated with the typical scenarios of photovoltaic energy storage, making the generated photovoltaic-energy storage active power allocation scheme unreasonable.

[0006] Therefore, how to extract multi-timescale operating characteristics and predict typical operating modes from real-time monitoring data of distribution radio stations and photovoltaic energy storage devices in order to achieve accurate photovoltaic energy consumption and intelligent regulation in distribution radio stations is a technical problem that needs to be solved. Summary of the Invention

[0007] To address this, the present invention provides a grid energy storage dynamic control method based on real-time monitoring. By extracting features at multiple time scales, predicting operating trends and patterns, and dynamically predicting and adjusting, it accurately solves the technical problems existing in the control of photovoltaic energy storage devices in distribution substations, such as feature extraction not being based on multiple time scales, low prediction accuracy, and unreasonable control decisions. This method achieves deep adaptation between the photovoltaic energy storage system and the operation of the distribution substation, ensuring the safe and stable operation of the distribution substation, improving the photovoltaic energy absorption capacity, and optimizing the utilization efficiency of energy storage resources.

[0008] To achieve the above objectives, this invention proposes a dynamic regulation method for grid energy storage based on real-time monitoring, comprising: The real-time monitoring data of transformer operation and photovoltaic energy storage device operation of the distribution substation are obtained and processed through a multi-timescale parallel photovoltaic energy storage model to generate multi-timescale fused operation characteristics. The multi-timescale parallel photovoltaic energy storage model is constructed based on a time-series convolutional network constrained by grid equations. The multi-timescale fused operation characteristics are used to generate photovoltaic energy storage mode operation weights through an operation trend prediction model, wherein the operation trend prediction model is constructed based on a long short-term memory network and a multilayer perceptron architecture. The photovoltaic energy storage mode operation weight and distribution area operation data are used to generate the optimal photovoltaic distribution energy storage power of the photovoltaic distribution energy storage device through a rolling time domain optimization objective function. The rolling time domain optimization objective function includes a voltage cost function, a transformer network loss cost function, and an energy storage operation cost function.

[0009] Furthermore, the process of generating multi-timescale fused operation characteristics through a multi-timescale parallel photovoltaic energy storage model includes: The transformer operation data and photovoltaic energy storage device operation data are processed through a multi-scale parallel temporal convolutional network to generate multi-time-scale features. The multi-timescale features are processed through an attention mechanism to generate timescale weights; The multi-timescale features are weighted based on the timescale weights to generate initial fused running features; The initial fusion operation characteristics are passed through a decoder to generate predicted energy storage operation parameters for the transformer substation, and the physical constraint loss is calculated based on the predicted energy storage operation parameters for the transformer substation. The initial fused operating features are weighted and corrected based on the gradient value of the physical constraint loss to generate the multi-timescale fused operating features; The multi-timescale parallel photovoltaic energy storage model also includes an attention mechanism and a decoder.

[0010] Furthermore, the process of generating the physical constraint loss through the decoder includes: The initial fusion operation features are mapped through a three-level fully connected layer to generate fusion operation mapping features; The fused operation mapping features are passed through multiple output heads to generate the predicted energy storage operation parameters for the transformer area.

[0011] Furthermore, the process of calculating physical constraint losses based on predicted energy storage operation parameters of the transformer substation includes: The dynamic loss term of energy storage is calculated based on the predicted state of charge of photovoltaic energy storage, the predicted charging power of photovoltaic energy storage, the predicted discharging power of photovoltaic energy storage, the rated capacity of photovoltaic energy storage, the charging efficiency of energy storage, and the discharging efficiency of energy storage. The power balance loss term is calculated based on the predicted active power of the distribution area load, the predicted active power output of photovoltaic power, the predicted active power injected into the distribution area by the grid, and the predicted active power of photovoltaic power distribution and energy storage. The energy storage dynamic loss term and the power balance loss term are normalized and weighted summed to generate the physical constraint loss; The predicted operating parameters of the distribution area energy storage include the predicted state of charge of photovoltaic energy storage, the predicted charging power of photovoltaic energy storage, the predicted discharging power of photovoltaic energy storage, the predicted active power of the distribution area load, the predicted active power output of photovoltaic energy storage, the predicted active power injected into the distribution area by the grid, and the predicted active power of photovoltaic distribution and energy storage.

[0012] Furthermore, the process of generating the operating weights of the photovoltaic energy storage mode through the operational trend prediction model includes: The multi-timescale fused operational characteristics are then passed through a long short-term memory network to generate photovoltaic energy storage trend characteristics. The photovoltaic energy storage trend characteristics are normalized by layer to generate normalized photovoltaic energy storage characteristics; The normalized photovoltaic energy storage characteristics are used to generate the photovoltaic energy storage mode operation weights through a multilayer sensor.

[0013] Furthermore, the process of generating the photovoltaic energy storage mode operation weights through a multilayer sensor includes: The normalized photovoltaic energy storage features are used to extract initial features through the first hidden layer to generate initial photovoltaic energy storage features; The initial photovoltaic energy storage characteristics are fused through a second hidden layer to generate fused energy storage characteristics; The fused energy storage features are enhanced in the time domain through a third hidden layer to generate initial mapping features; The initial mapping features are passed through an output layer based on the Softplus function to generate the operating weights of the photovoltaic energy storage mode; The multilayer perceptron includes a first hidden layer, a second hidden layer, and a third hidden layer with successively decreasing convolutional kernel sizes, as well as an output layer.

[0014] Furthermore, the process of calculating the voltage cost function includes: The voltage cost function is calculated based on the difference between the current three-phase effective voltage and the rated voltage of the distribution substation. The process of calculating the cost function for energy storage operation includes: The operating cost function of the photovoltaic energy storage device is calculated based on its current active power and current SOC stress parameters.

[0015] Furthermore, the process of calculating the transformer network loss cost function includes: The line loss value is calculated based on the line resistance of the distribution substation and the current effective value of the line current. The transformer loss value is calculated based on the transformer no-load loss, transformer load loss, current transformer apparent power, and transformer rated capacity of the distribution substation. The line loss value and the transformer loss value are weighted and summed to calculate the transformer network loss cost function.

[0016] Furthermore, the dynamic regulation method for grid energy storage also includes: The scene identification loss term is calculated based on the scene label weight and the photovoltaic energy storage mode operation weight. Based on the optimal photovoltaic power and energy storage power and the sample photovoltaic power, the energy storage command loss term is calculated; Based on the scenario identification loss term and the energy storage command loss term, a combined loss function is calculated, and based on the combined loss function, a multi-timescale parallel photovoltaic energy storage model and an operation trend prediction model are collaboratively optimized and trained.

[0017] Furthermore, the process of calculating the scene recognition loss term includes: The scene recognition loss term is calculated by summing the cross-entropy loss term and the smoothing loss term based on the scene label weight and the photovoltaic energy storage mode operation weight.

[0018] Compared with the prior art, the beneficial effects of the present invention are that, through real-time data-driven, multi-scale feature extraction, accurate trend prediction, and multi-objective optimization and control, the present invention realizes intelligent and precise dynamic control of photovoltaic energy storage systems in distribution substations, thereby ensuring the safe and stable operation of distribution substations, improving the capacity for new energy absorption, and optimizing operational efficiency.

[0019] In particular, this invention constructs a multi-timescale parallel photovoltaic energy storage model by using a temporal convolutional network based on grid equation constraints. This effectively solves the problem that existing single-timescale feature extraction cannot fully capture the dynamic operation of distribution substations, and significantly improves the comprehensiveness, accuracy and reliability of operational feature extraction.

[0020] In particular, the operational trend prediction model built on the architecture of long short-term memory network and multilayer perceptron effectively makes up for the limitations of existing single prediction models, significantly improves the prediction accuracy and response speed of photovoltaic energy storage mode operation trend, provides accurate and reliable basis for dynamic control decision-making, can accurately generate photovoltaic energy storage mode operation weights, and clearly characterize the adaptation relationship between photovoltaic output and energy storage control under different operating conditions. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the dynamic regulation method for power grid energy storage based on real-time monitoring, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the multi-timescale parallel photovoltaic energy storage model of the grid energy storage dynamic control method based on real-time monitoring in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the operation weight of the photovoltaic energy storage mode in the grid energy storage dynamic control method based on real-time monitoring, according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the combined loss function of the grid energy storage dynamic control method based on real-time monitoring, according to an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0023] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0024] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0025] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0026] like Figures 1 to 4 As shown, this invention provides a grid energy storage dynamic control method based on real-time monitoring. Through multi-timescale feature extraction, operation trend pattern prediction, and dynamic prediction and adjustment, it accurately solves the technical problems existing in the control of photovoltaic energy storage devices in distribution substations, such as feature extraction not being based on multiple timescales, low prediction accuracy, and unreasonable control decisions. It achieves deep adaptation between the photovoltaic energy storage system and the operation of the distribution substation, ensuring the safe and stable operation of the distribution substation, improving the photovoltaic energy absorption capacity, and optimizing the utilization efficiency of energy storage resources.

[0027] like Figure 1 As shown in the figure, this embodiment proposes a dynamic control method for grid energy storage based on real-time monitoring, including: The real-time monitoring data of transformer operation and photovoltaic energy storage device operation of the distribution substation are obtained and processed through a multi-timescale parallel photovoltaic energy storage model to generate multi-timescale fused operation characteristics. The multi-timescale parallel photovoltaic energy storage model is constructed based on a time-series convolutional network constrained by grid equations. The multi-timescale fused operation characteristics are used to generate photovoltaic energy storage mode operation weights through an operation trend prediction model, wherein the operation trend prediction model is constructed based on a long short-term memory network and a multilayer perceptron architecture. The photovoltaic energy storage mode operation weight and distribution area operation data are used to generate the optimal photovoltaic distribution energy storage power of the photovoltaic distribution energy storage device through a rolling time domain optimization objective function. The rolling time domain optimization objective function includes a voltage cost function, a transformer network loss cost function, and an energy storage operation cost function.

[0028] Specifically, a distribution substation is a basic power supply unit connecting the high-voltage grid and low-voltage users at the end of the power distribution network. It is also an operational unit for source-grid-load-storage coordination on the distribution network side under the new power system, and its function is to transmit and distribute power on the grid side. Photovoltaic distribution energy storage devices are supporting facilities for power regulation in distribution substations. They are installed on the user side of the distribution substation, next to the transformer, and in the substation's distribution room. Through the charging and discharging, peak shaving and valley filling, and voltage regulation functions of photovoltaic distribution energy storage devices, the power supply quality and operational stability of the distribution substation are improved. In this embodiment, the dilated convolution of a temporal convolutional network (TCN) is used to capture the multi-scale fluctuation characteristics of net load, transformer thermal dynamics, and energy storage state evolution characteristics of the synergistic influence between photovoltaic distribution energy storage devices and distribution substations. Then, long short-term memory networks and multilayer perceptrons are used to extract time-series features and determine the control parameters that match the predicted typical operating mode of the photovoltaic distribution energy storage devices, i.e., the photovoltaic energy storage mode operating weights. The optimal photovoltaic distribution energy storage power of the photovoltaic distribution energy storage devices is determined by using a rolling time-domain optimization objective function.

[0029] like Figure 2 As shown, the process of generating multi-timescale fused operation characteristics through a multi-timescale parallel photovoltaic energy storage model further includes: The transformer operation data and photovoltaic energy storage device operation data are processed through a multi-scale parallel temporal convolutional network to generate multi-time-scale features. The multi-timescale features are processed through an attention mechanism to generate timescale weights; The multi-timescale features are weighted based on the timescale weights to generate initial fused running features; The initial fusion operation characteristics are passed through a decoder to generate predicted energy storage operation parameters for the transformer substation, and the physical constraint loss is calculated based on the predicted energy storage operation parameters for the transformer substation. The initial fused operating features are weighted and corrected based on the gradient value of the physical constraint loss to generate the multi-timescale fused operating features; The multi-timescale parallel photovoltaic energy storage model also includes an attention mechanism and a decoder.

[0030] like Figure 2 As shown, the process of generating the physical constraint loss through the decoder further includes: The initial fusion operation features are mapped through a three-level fully connected layer to generate fusion operation mapping features; The fused operation mapping features are passed through multiple output heads to generate the predicted energy storage operation parameters for the transformer area.

[0031] Furthermore, the process of calculating physical constraint losses based on predicted energy storage operation parameters of the transformer substation includes: The dynamic loss term of energy storage is calculated based on the predicted state of charge of photovoltaic energy storage, the predicted charging power of photovoltaic energy storage, the predicted discharging power of photovoltaic energy storage, the rated capacity of photovoltaic energy storage, the charging efficiency of energy storage, and the discharging efficiency of energy storage. The power balance loss term is calculated based on the predicted active power of the distribution area load, the predicted active power output of photovoltaic power, the predicted active power injected into the distribution area by the grid, and the predicted active power of photovoltaic power distribution and energy storage. The energy storage dynamic loss term and the power balance loss term are weighted and summed to generate the physical constraint loss; The predicted operating parameters of the distribution area energy storage include the predicted state of charge of photovoltaic energy storage, the predicted charging power of photovoltaic energy storage, the predicted discharging power of photovoltaic energy storage, the predicted active power of the distribution area load, the predicted active power output of photovoltaic energy storage, the predicted active power injected into the distribution area by the grid, and the predicted active power of photovoltaic distribution and energy storage.

[0032] Specifically, the process of generating multi-timescale fused operational features can be represented as:

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039] In the formula, The output feature of the m-th branch represents the multi-timescale feature. Represents the ReLU or GELU activation function. , Let represent the weight parameters and convolution bias of the k-th convolution kernel in the m-th branch of a multi-scale parallel temporal convolutional network, respectively. The first input tensor represents the first... Elements of time Let represent the expansion rate of the m-th branch. Preferably, the multi-scale parallel temporal convolutional network has 6 parallel branches with expansion rates of 1, 3, 9, 27, 384, and 1536, respectively. The kernel size is 5 for all branches, and the channel number sequence is 8, 16, 32, 64, 512, and 512, respectively. Therefore, branches 1 and 2 can extract the net load fluctuation characteristics of rapid fluctuations such as photovoltaic cloud shading and load surges on a timescale of 30 seconds to 5 minutes. Branches 3 and 4 can extract the transformer thermal dynamic characteristics of oil temperature rise on a timescale of 10 minutes to 1 hour. Branches 5 and 6 can extract the energy storage state evolution characteristics on an hourly timescale. This represents the time-scale weight of the m-th branch. Represents an exponential function. Let M represent the computation of the m-th attention mechanism, where M represents the total number of branches, preferably 6. Indicates the initial fusion operation characteristics, , These represent the power balance loss term and the energy storage dynamic loss term, respectively. These represent the predicted active power of the distribution transformer area load, the predicted active power output of photovoltaic power, the predicted active power injected into the distribution transformer area by the grid, and the predicted active power of photovoltaic power distribution with energy storage, respectively. These represent the predicted state of charge (SOC) of the photovoltaic (PV) energy storage at times t+1 and t, respectively, reflecting the ratio of the current remaining capacity of the PV energy storage battery to its rated capacity. The time step is indicated, preferably 15 minutes. This indicates the rated capacity of photovoltaic energy storage, which is a fixed parameter determined by the specifications of the energy storage equipment. This indicates the energy storage charging efficiency, with a preferred typical value of 0.95. , These represent the predicted photovoltaic energy storage charging power and the predicted photovoltaic energy storage discharging power, respectively. This represents the energy storage discharge efficiency, with a preferred typical value of 0.92. Represents physical constraint loss. λ represents two weighting coefficients, preferably 1.0 and 0.8, to ensure basic power balance and the importance of energy storage. It is preferable to normalize the power balance loss term and the energy storage dynamic loss term through a reference value to eliminate differences in numerical range before performing a weighted summation calculation. The reference value for normalization can be obtained through experience or initial calculation. λ represents the weighting coefficient, preferably 0.05, to make a moderate correction to the initial integrated operation characteristics based on physical constraints, so as to better cope with the grid changes in the distribution substation area.

[0040] Specifically, the input tensor is a TxF-shaped tensor constructed from transformer operating data and photovoltaic energy storage device operating data. The time series length T is preferably 96, i.e., monitoring data at 24-hour × 15-minute intervals is used. The feature dimension F is the number of features at each time point, and the number of features is preferably 19. That is, the transformer operating data includes the transformer's total active power, total reactive power, transformer load rate, transformer oil temperature, maximum three-phase current, and three-phase current imbalance. The photovoltaic energy storage device operating data includes the photovoltaic's total active power output, total reactive power output, photovoltaic efficiency, total active power of energy storage, total reactive power of energy storage, state of charge, current maximum charge / discharge power, remaining adjustable capacity, and current operating mode (one-hot encoding: charging = 1, 0, 0; discharging = 0, 1, 0; standby = 0, 0, 1). The input tensor also includes weather forecast index (0-1), comprehensive cloud cover, precipitation probability, and seasonal encoding. Therefore, the input tensor comprehensively considers the interaction variables between the distribution transformer area and the photovoltaic energy storage, as well as its own grid operation variables.

[0041] like Figure 3 As shown, the process of generating the operating weights of the photovoltaic energy storage mode through the operating trend prediction model further includes: The multi-timescale fused operational characteristics are then passed through a long short-term memory network to generate photovoltaic energy storage trend characteristics. The photovoltaic energy storage trend characteristics are normalized by layer to generate normalized photovoltaic energy storage characteristics; The normalized photovoltaic energy storage characteristics are used to generate the photovoltaic energy storage mode operation weights through a multilayer sensor.

[0042] Furthermore, the process of generating the photovoltaic energy storage mode operation weights through a multilayer sensor includes: The normalized photovoltaic energy storage features are used to extract initial features through the first hidden layer to generate initial photovoltaic energy storage features; The initial photovoltaic energy storage characteristics are fused through a second hidden layer to generate fused energy storage characteristics; The fused energy storage features are enhanced in the time domain through a third hidden layer to generate initial mapping features; The initial mapping features are passed through an output layer based on the Softplus function to generate the operating weights of the photovoltaic energy storage mode; The multilayer perceptron includes a first hidden layer, a second hidden layer, and a third hidden layer with successively decreasing convolutional kernel sizes, as well as an output layer.

[0043] Specifically, the process of generating the operating weights for the photovoltaic energy storage mode can be expressed as follows:

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] In the formula, This represents the trend characteristics of photovoltaic energy storage, specifically the hidden vector at the final time step T of the Long Short-Term Memory network. Represents the Long Short-Term Memory network. This indicates the characteristics of multi-timescale fusion operation. This represents the normalized photovoltaic energy storage characteristics. Presentation layer normalization operation, , These represent the weight parameters and convolution bias of the normalized convolution kernel, respectively. , , These represent the initial photovoltaic energy storage characteristics, integrated energy storage characteristics, and initial mapping characteristics, respectively. express Activation function , These represent the weight parameters and convolution bias of the convolution kernel in the first hidden layer, respectively. The weight parameters of the first hidden layer are set to 256x64, and the convolution bias is set to 64 for preliminary feature extraction. , These represent the weight parameters and convolution bias of the convolution kernel in the second hidden layer, respectively. The weight parameters of the second hidden layer are set to 128x256, and the convolution bias is set to 128 for feature fusion. , These represent the weight parameters and convolution bias of the convolution kernel in the third hidden layer, respectively. The weight parameters of the third hidden layer are set to 64x128, and the convolution bias is set to 64 for feature fusion. Therefore, the kernel sizes of the first, second, and third hidden layers decrease sequentially. All represent the operating weights of the i-th group of photovoltaic energy storage modes. This indicates a reshaping operation, which adjusts the dimensions and shape of the tensor without changing the total number of elements to adapt to the weight format of the photovoltaic energy storage mode. Softplus represents the Softplus function. , These represent the weight parameters and convolution bias of the output layer's convolution kernel, respectively.

[0050] Furthermore, the process of calculating the voltage cost function includes: The voltage cost function is calculated based on the difference between the current three-phase effective voltage and the rated voltage of the distribution substation. The process of calculating the cost function for energy storage operation includes: The operating cost function of the photovoltaic energy storage device is calculated based on its current active power and current SOC stress parameters.

[0051] Furthermore, the process of calculating the transformer network loss cost function includes: The line loss value is calculated based on the line resistance of the distribution substation and the current effective value of the line current. The transformer loss value is calculated based on the transformer no-load loss, transformer load loss, current transformer apparent power, and transformer rated capacity of the distribution substation. The line loss value and the transformer loss value are weighted and summed to calculate the transformer network loss cost function.

[0052] Specifically, the objective function for rolling time-domain optimization can be expressed as:

[0053]

[0054]

[0055]

[0056]

[0057]

[0058] In the formula, , , , , Let these represent the voltage cost function, line loss value, transformer loss value, transformer network loss cost function, and energy storage operation cost function, respectively. The values ​​are the current three-phase effective voltage values ​​of phase ph at time k+i. This indicates the rated voltage, which is 220V. Represents a set of routes. This represents the line resistance of the l-th line, with a typical value of 0.02. This represents the effective value of the current of the l-th line at time k+i. These represent the transformer no-load loss and the transformer load loss, respectively. The transformer no-load loss is calibrated based on the transformer voltage and rated capacity, while the transformer load loss is taken as 1% of the transformer's rated capacity as a typical value. This indicates the current apparent power of the transformer. This indicates the rated capacity of the transformer. , These are two weighting coefficients, preferably 0.6 and 0.4, which aligns with the rule that line losses are generally greater than transformer losses. , These represent the energy storage cycle cost coefficient and the energy storage degradation cost coefficient, respectively, preferably a typical value of 0.05 yuan / kWh for lithium batteries and a typical value of 0.01-0.05 yuan / unit stress. , These represent the current active power and the current SOC stress parameter, respectively. To optimize the objective function in the rolling time domain, All represent the operating weights of the i-th group of photovoltaic energy storage modes. This represents the operating weight of the i-th group of photovoltaic energy storage modes. Represents the nth group , The optimal photovoltaic power generation for energy storage is obtained by conforming to the rolling time domain (RHO) objective function. This includes active power and SOC stress parameters.

[0059] like Figure 4 As shown, the dynamic regulation method for grid energy storage further includes: The scene identification loss term is calculated based on the scene label weight and the photovoltaic energy storage mode operation weight. Based on the optimal photovoltaic power and energy storage power and the sample photovoltaic power, the energy storage command loss term is calculated; Based on the scenario identification loss term and the energy storage command loss term, a combined loss function is calculated, and based on the combined loss function, a multi-timescale parallel photovoltaic energy storage model and an operation trend prediction model are collaboratively optimized and trained.

[0060] like Figure 4 As shown, the process of calculating the scene recognition loss term further includes: The scene recognition loss term is calculated by summing the cross-entropy loss term and the smoothing loss term based on the scene label weight and the photovoltaic energy storage mode operation weight.

[0061] Specifically, the combination loss function can be expressed as:

[0062]

[0063]

[0064] In the formula, , , These represent the scene recognition loss term, the energy storage command loss term, and the combined loss function, respectively. , Let these represent the scene label weight and the photovoltaic energy storage mode operation weight of the b-th sample at time c, respectively. Let B and C represent the scene label weight and photovoltaic energy storage mode operation weight of the b-th sample at time t / t+1, respectively, where B represents the total number of samples and C represents the total number of time points. This indicates the active power regulation error sub-item and the SOC stress parameter regulation error sub-item. , Let represent the active power of the optimal photovoltaic power storage system and the sample active power, respectively. Let SOC stress parameters and sample SOC stress parameters represent the optimal photovoltaic power combined with energy storage, respectively. , The coefficients are preferably 1.0 and 0.6, which are consistent with the weight configuration of using scene recognition as the main loss term.

[0065] In this embodiment, intelligent and precise dynamic control of the photovoltaic-energy storage system in the distribution substation is achieved through real-time data-driven operation, multi-scale feature extraction, accurate trend prediction, and multi-objective optimization and control. This ensures the safe and stable operation of the distribution substation, enhances the renewable energy absorption capacity, and optimizes operational efficiency. A multi-timescale parallel photovoltaic-energy storage model is constructed using a temporal convolutional network based on grid equation constraints. This effectively solves the problem that existing single-time-scale feature extraction cannot comprehensively capture the dynamic operation of the distribution substation, significantly improving the comprehensiveness, accuracy, and reliability of operational feature extraction. The operational trend prediction model, built based on a long short-term memory network and a multilayer perceptron architecture, effectively compensates for the limitations of existing single prediction models, significantly improving the prediction accuracy and response speed of the photovoltaic-energy storage mode's operational trend. This provides a precise and reliable basis for dynamic control decisions, accurately generating the operational weights of the photovoltaic-energy storage mode and clearly characterizing the adaptation relationship between photovoltaic output and energy storage control under different operating conditions.

[0066] Those skilled in the art will recognize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0067] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for dynamic regulation of power grid energy storage based on real-time monitoring, characterized in that, include: The real-time monitoring data of transformer operation and photovoltaic energy storage device operation of the distribution substation are obtained and processed through a multi-timescale parallel photovoltaic energy storage model to generate multi-timescale fused operation characteristics. The multi-timescale parallel photovoltaic energy storage model is constructed based on a time-series convolutional network constrained by grid equations. The multi-timescale fused operation characteristics are used to generate photovoltaic energy storage mode operation weights through an operation trend prediction model, wherein the operation trend prediction model is constructed based on a long short-term memory network and a multilayer perceptron architecture. The photovoltaic energy storage mode operation weight and distribution area operation data are used to generate the optimal photovoltaic distribution energy storage power of the photovoltaic distribution energy storage device through a rolling time domain optimization objective function. The rolling time domain optimization objective function includes a voltage cost function, a transformer network loss cost function, and an energy storage operation cost function.

2. The method for dynamic regulation of power grid energy storage based on real-time monitoring according to claim 1, characterized in that, The process of generating multi-timescale fused operation characteristics through a multi-timescale parallel photovoltaic energy storage model includes: The transformer operation data and photovoltaic energy storage device operation data are processed through a multi-scale parallel temporal convolutional network to generate multi-time-scale features. The multi-timescale features are processed through an attention mechanism to generate timescale weights; The multi-timescale features are weighted based on the timescale weights to generate initial fused running features; The initial fusion operation characteristics are passed through a decoder to generate predicted energy storage operation parameters for the transformer substation, and the physical constraint loss is calculated based on the predicted energy storage operation parameters for the transformer substation. The initial fused operating features are weighted and corrected based on the gradient value of the physical constraint loss to generate the multi-timescale fused operating features; The multi-timescale parallel photovoltaic energy storage model also includes an attention mechanism and a decoder.

3. The method for dynamic regulation of power grid energy storage based on real-time monitoring according to claim 2, characterized in that, The process of generating physical constraint loss through a decoder includes: The initial fusion operation features are mapped through a three-level fully connected layer to generate fusion operation mapping features; The fused operation mapping features are passed through multiple output heads to generate the predicted energy storage operation parameters for the transformer area.

4. The method for dynamic regulation of power grid energy storage based on real-time monitoring according to claim 2, characterized in that, The process of calculating physical constraint losses based on predicted energy storage operation parameters of the transformer substation includes: The dynamic loss term of energy storage is calculated based on the predicted state of charge of photovoltaic energy storage, the predicted charging power of photovoltaic energy storage, the predicted discharging power of photovoltaic energy storage, the rated capacity of photovoltaic energy storage, the charging efficiency of energy storage, and the discharging efficiency of energy storage. The power balance loss term is calculated based on the predicted active power of the distribution area load, the predicted active power output of photovoltaic power, the predicted active power injected into the distribution area by the grid, and the predicted active power of photovoltaic power distribution and energy storage. The energy storage dynamic loss term and the power balance loss term are normalized and weighted summed to generate the physical constraint loss; The predicted operating parameters of the distribution area energy storage include the predicted state of charge of photovoltaic energy storage, the predicted charging power of photovoltaic energy storage, the predicted discharging power of photovoltaic energy storage, the predicted active power of the distribution area load, the predicted active power output of photovoltaic energy storage, the predicted active power injected into the distribution area by the grid, and the predicted active power of photovoltaic distribution and energy storage.

5. The method for dynamic regulation of power grid energy storage based on real-time monitoring according to claim 1, characterized in that, The process of generating operational weights for photovoltaic energy storage modes through a trend prediction model includes: The multi-timescale fused operational characteristics are then passed through a long short-term memory network to generate photovoltaic energy storage trend characteristics. The photovoltaic energy storage trend characteristics are normalized by layer to generate normalized photovoltaic energy storage characteristics; The normalized photovoltaic energy storage characteristics are used to generate the photovoltaic energy storage mode operation weights through a multilayer sensor.

6. The method for dynamic regulation of power grid energy storage based on real-time monitoring according to claim 5, characterized in that, The process of generating the photovoltaic energy storage mode operation weights through a multilayer sensor includes: The normalized photovoltaic energy storage features are used to extract initial features through the first hidden layer to generate initial photovoltaic energy storage features; The initial photovoltaic energy storage characteristics are fused through a second hidden layer to generate fused energy storage characteristics; The fused energy storage features are enhanced in the time domain through a third hidden layer to generate initial mapping features; The initial mapping features are passed through an output layer based on the Softplus function to generate the operating weights of the photovoltaic energy storage mode; The multilayer perceptron includes a first hidden layer, a second hidden layer, and a third hidden layer with successively decreasing convolutional kernel sizes, as well as an output layer.

7. The method for dynamic regulation of power grid energy storage based on real-time monitoring according to claim 1, characterized in that, The process of calculating the voltage cost function includes: The voltage cost function is calculated based on the difference between the current three-phase effective voltage and the rated voltage of the distribution substation. The process of calculating the cost function for energy storage operation includes: The operating cost function of the photovoltaic energy storage device is calculated based on its current active power and current SOC stress parameters.

8. The method for dynamic regulation of power grid energy storage based on real-time monitoring according to claim 1, characterized in that, The process of calculating the transformer network loss cost function includes: The line loss value is calculated based on the line resistance of the distribution substation and the current effective value of the line current. The transformer loss value is calculated based on the transformer no-load loss, transformer load loss, current transformer apparent power, and transformer rated capacity of the distribution substation. The line loss value and the transformer loss value are weighted and summed to calculate the transformer network loss cost function.

9. The method for dynamic regulation of power grid energy storage based on real-time monitoring according to any one of claims 1 to 8, characterized in that, Also includes: The scene identification loss term is calculated based on the scene label weight and the photovoltaic energy storage mode operation weight. Based on the optimal photovoltaic power and energy storage power and the sample photovoltaic power, the energy storage command loss term is calculated; Based on the scenario identification loss term and the energy storage command loss term, a combined loss function is calculated, and based on the combined loss function, a multi-timescale parallel photovoltaic energy storage model and an operation trend prediction model are collaboratively optimized and trained.

10. The method for dynamic regulation of power grid energy storage based on real-time monitoring according to claim 9, characterized in that, The process of calculating the scene recognition loss term includes: The scene recognition loss term is calculated by summing the cross-entropy loss term and the smoothing loss term based on the scene label weight and the photovoltaic energy storage mode operation weight.

Images