Transformer load rate control method and system based on distributed energy storage

Abstract

The invention relates to the technical field of intelligent power grid operation control, and discloses a transformer load rate control method and system based on distributed energy storage, and the method comprises the steps: obtaining a power grid operation data set; constructing a load rate prediction model, and performing load rate prediction to obtain a complete prediction curve; calculating a load gap and an overload risk identifier according to the complete prediction curve, and carrying out global optimization on a power distribution strategy by adopting a particle swarm algorithm to obtain a power distribution strategy of each energy storage node; establishing an energy storage unit control matrix, and performing dynamic scheduling control to obtain a scheduling scheme which can be continuously executed; actual charging and discharging power is obtained, load dynamic parameter adjustment is carried out, and a cooperative control strategy is obtained; and obtaining an energy storage unit state evaluation report, and carrying out power weight distribution considering the full life cycle of the equipment to obtain a scheduling optimization scheme and a maintenance scheme. According to the method, multi-factor collaborative prospective control of the load rate of the transformer can be realized.

Description

Technical Field

[0001] This invention relates to the field of smart grid operation control technology, and in particular to a transformer load rate control method and system based on distributed energy storage. Background Technology

[0002] Currently, with the expansion of the power system and the growth of electricity demand, intelligent control of transformer load rate and the realization of smart grids have become key links in ensuring the safe and stable operation of the power grid. In one existing technology, transformer load management relies on human experience and simple threshold control. This method suffers from problems such as response lag and difficulty in integrating multi-source heterogeneous data. It is difficult to cope with complex power grid environments and diverse power consumption patterns, and it cannot fully explore the deep correlation between power grid operation data and user behavior patterns, resulting in insufficient accuracy and timeliness of control decisions. At the same time, it cannot effectively handle the time-series coupling relationship between the charging and discharging strategies of energy storage systems and load rate fluctuations, making it difficult to capture dynamic change patterns and solve the time-series matching problem. In summary, existing technologies suffer from insufficient multi-factor coordinated control, which is detrimental to the safe and stable operation of the power grid. Summary of the Invention

[0003] This invention provides a transformer load rate control method and system based on distributed energy storage, so as to achieve forward-looking control of transformer load rate through multi-factor synergy.

[0004] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a transformer load rate control method based on distributed energy storage, comprising: Obtain power grid operation dataset; A load rate prediction model is constructed based on the power grid operation dataset, and a load rate prediction is performed based on the load rate prediction model to obtain a complete prediction curve; The load gap and overload risk indicators are calculated based on the complete prediction curve. Then, the power allocation strategy is globally optimized using the particle swarm optimization algorithm based on the load gap and the overload risk indicators to obtain the power allocation strategy for each energy storage node. A control matrix for energy storage units is established based on the power allocation strategy, and dynamic scheduling control is performed through a real-time communication network to obtain a scheduling scheme that can be executed continuously. The actual charging and discharging power of the energy storage system during the execution process is obtained according to the scheduling scheme, and the load is dynamically adjusted according to the actual charging and discharging power to obtain a collaborative control strategy. Obtain the energy storage unit status assessment report, and based on the energy storage unit status assessment report, perform power weight allocation on the collaborative control strategy considering the entire life cycle of the equipment to obtain the scheduling optimization scheme and maintenance scheme.

[0005] Preferably, obtaining the power grid operation dataset includes: The transformer operating parameters and user electricity consumption data collected by the sensors are obtained to obtain the raw monitoring data stream with time stamps. The original monitoring data stream is standardized and preprocessed to obtain preliminary processed data in a unified format; The time series completeness of the preliminary processed data is calculated, and a threshold judgment is made on the time series completeness. If the time series completeness index is lower than the preset completeness threshold, interpolation is performed to complete the data and obtain the completed power grid monitoring data. The power grid monitoring data is checked for consistency using data quality verification rules. If there is a logical conflict between the operating parameters and the electricity consumption behavior data, abnormal data identification processing is performed to finally obtain the power grid operation dataset.

[0006] Preferably, the step of constructing a load rate prediction model based on the power grid operation dataset and performing load rate prediction based on the load rate prediction model to obtain a complete prediction curve includes: Time series reconstruction is performed based on the transformer load data in the power grid operation dataset to obtain a standardized load change sequence; Extract the periodic features of daily and monthly cycles from the load change sequence, and construct a feature matrix based on the periodic features to obtain a feature matrix containing periodic and trend features; A long short-term memory network prediction model is constructed based on the feature matrix. The long short-term memory network prediction model is trained, and the input layer is set to historical 24-hour load data. The long short-term memory network prediction model is trained until the error converges to within the preset accuracy to obtain the load rate prediction model. The load rate prediction model is used to calculate the predicted load rate for each moment in the next 24 hours, and the basic prediction curve is determined. The confidence distribution of the predicted values ​​at each time point is calculated based on the basic prediction curve to obtain the complete prediction curve.

[0007] Preferably, the step of calculating the load gap and overload risk indicator based on the complete prediction curve, and then using a particle swarm optimization algorithm to globally optimize the power allocation strategy based on the load gap and the overload risk indicator to obtain the power allocation strategy for each energy storage node, includes: The actual electricity demand and the load gap between the rated capacity of the transformer for each time period are calculated based on the predicted curve. The load rate for each time period is determined based on the predicted curve. If the load rate exceeds the safety threshold, an overload risk indicator is generated. A particle swarm optimization solver is constructed based on the load gap and the overload risk indicator, and the charging and discharging parameters of the energy storage unit corresponding to the particle dimension are set. If the fitness function reaches the preset convergence condition, the charging and discharging command sequence is output. The real-time load rate is determined based on the charging and discharging command sequence. If the real-time load rate reaches the preset state transition threshold, a state switching command is sent to the energy storage unit. Based on the state switching command, the discharge power share of each energy storage node is allocated according to the principle of nearby power supply to obtain the power allocation strategy.

[0008] Preferably, the step of establishing an energy storage unit control matrix according to the power allocation strategy and performing dynamic scheduling control through a real-time communication network to obtain a scheduling scheme that can be continuously executed includes: A power allocation matrix is ​​established based on the power allocation strategy, and the power status is judged based on the power allocation matrix. If the output power is lower than the preset power threshold, the corresponding weight value in the power allocation matrix is ​​adjusted to achieve balance, and the energy storage unit control matrix is ​​obtained. Based on the energy storage unit control matrix, control commands are sent to each energy storage unit through a real-time communication network to obtain real-time response status feedback information. A fuzzy control algorithm is used to calculate the matching relationship between the load change rate and the energy storage unit response time in the response state feedback information, and to determine the optimized combination of control parameters. Delay detection is performed based on the combination of control parameters. If the response delay of the energy storage unit exceeds a preset time window, the standby energy storage unit with the best state is selected for replacement, thereby obtaining a scheduling scheme that can be executed continuously.

[0009] Preferably, the step of obtaining the actual charging and discharging power of the energy storage system during execution according to the scheduling scheme, and performing dynamic load parameter adjustment based on the actual charging and discharging power to obtain a cooperative control strategy includes: Based on the scheduling scheme, the power status information of the energy storage unit is obtained, and the power status information is converted from analog to digital and filtered to obtain the actual charging and discharging power. The actual load rate is calculated based on the actual charging and discharging power, and the deviation of the actual load rate is judged. If the deviation between the actual load rate and the predicted load rate is greater than the preset deviation threshold, the dynamic adjustment signal is activated to determine the load deviation judgment result. Based on the load deviation judgment results, power is redistributed to each energy storage unit to obtain a collaborative control strategy.

[0010] Preferably, after obtaining the actual charging and discharging power of the energy storage system during execution according to the scheduling scheme, and performing dynamic load parameter adjustment based on the actual charging and discharging power to obtain a cooperative control strategy, the method further includes: A sliding time window mechanism is used to acquire load change data and energy storage unit response data to obtain a time series training dataset. Based on the time series training dataset, the gradient descent algorithm is used to optimize the network weight parameter configuration of each layer of neurons in the load rate prediction model, and the load rate prediction result is generated based on the network weight parameter configuration to predict the error trend and determine the prediction accuracy evaluation status. The load rate prediction model is retrained based on the prediction accuracy evaluation status to obtain an updated load rate prediction model.

[0011] Preferably, the step of obtaining the energy storage unit status assessment report and, based on the energy storage unit status assessment report, performing power weight allocation on the collaborative control strategy considering the entire life cycle of the equipment to obtain a scheduling optimization scheme and a maintenance scheme includes: Obtain the voltage and current curves of each energy storage unit from the historical charge and discharge data, and calculate the efficiency decay coefficient based on the voltage and current curves to obtain the energy storage unit efficiency decay coefficient. A state assessment report for the energy storage unit is obtained by comparing the state threshold based on the efficiency decay coefficient of the energy storage unit. The power of each energy storage unit in the collaborative control strategy is dynamically weighted based on the status assessment report, and a scheduling instruction sequence is generated based on the adjustment results to obtain the final scheduling optimization scheme. Maintenance instructions are generated based on the status assessment report, resulting in a maintenance plan.

[0012] Secondly, the present invention provides a transformer load rate control system based on distributed energy storage, comprising: The data acquisition module is used to acquire power grid operation datasets. The predictive modeling module is used to construct a load rate prediction model based on the power grid operation dataset, and to predict the load rate based on the load rate prediction model to obtain a complete prediction curve. The strategy generation module is used to calculate the load gap and overload risk indicator based on the complete prediction curve, and to perform global optimization of the power allocation strategy using the particle swarm algorithm based on the load gap and the overload risk indicator to obtain the power allocation strategy for each energy storage node. The control execution module is used to establish an energy storage unit control matrix according to the power allocation strategy, and to perform dynamic scheduling control through a real-time communication network to obtain a scheduling scheme that can be executed continuously. The dynamic parameter adjustment module is used to obtain the actual charging and discharging power of the energy storage system during the execution process according to the scheduling scheme, and to perform dynamic load parameter adjustment based on the actual charging and discharging power to obtain a collaborative control strategy. The lifespan optimization module is used to obtain the energy storage unit status assessment report, and based on the energy storage unit status assessment report, to perform power weight allocation on the collaborative control strategy considering the entire life cycle of the equipment, thereby obtaining a scheduling optimization scheme and a maintenance scheme.

[0013] Thirdly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the transformer load rate control method based on distributed energy storage described in any one of the above.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention constructs a Long Short-Term Memory (LSTM) network prediction model by acquiring power grid operation datasets to predict load rate and construct a complete prediction curve, captures the dynamic change law of load rate, solves the problem of response lag caused by the reliance on human experience in existing technologies, can predict load rate changes in advance, improve the timeliness of control decisions, and realize the principle of effective integration of user behavior patterns and power grid operation data. (2) This invention calculates the load gap and overload risk indicators based on the complete prediction curve, and uses the particle swarm algorithm to perform global optimization of the power allocation strategy to obtain the power allocation strategy of each energy storage node. It comprehensively considers the load gap, overload risk and the timing coupling relationship of energy storage charging and discharging, and realizes the multivariate global optimal solution, which solves the problem that the existing single threshold control cannot handle complex power grid environments. (3) By obtaining the energy storage unit status assessment report, the present invention considers the power weight allocation of the entire life cycle of the equipment in the collaborative control strategy, and obtains the scheduling optimization scheme and maintenance scheme, taking into account the life cycle of the energy storage equipment and the load rate control effect, thus solving the problem of insufficient multi-factor coordination in the existing technology. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the transformer load rate control method based on distributed energy storage provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the transformer load rate control system based on distributed energy storage provided in the second embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Reference Figure 1 The first embodiment of the present invention provides a transformer load rate control method based on distributed energy storage, comprising the following steps: S11, Obtain the power grid operation dataset; S12, construct a load rate prediction model based on the power grid operation dataset, and perform load rate prediction based on the load rate prediction model to obtain a complete prediction curve; S13, calculate the load gap and overload risk indicator based on the complete prediction curve, and use the particle swarm optimization algorithm to globally optimize the power allocation strategy based on the load gap and the overload risk indicator to obtain the power allocation strategy of each energy storage node. S14. Establish an energy storage unit control matrix according to the power allocation strategy, and perform dynamic scheduling control through a real-time communication network to obtain a scheduling scheme that can be executed continuously. S15, obtain the actual charging and discharging power of the energy storage system during the execution process according to the scheduling scheme, and perform dynamic load parameter adjustment according to the actual charging and discharging power to obtain a collaborative control strategy; S16, obtain the energy storage unit status assessment report, and based on the energy storage unit status assessment report, perform power weight allocation on the collaborative control strategy considering the entire life cycle of the equipment to obtain the scheduling optimization scheme and maintenance scheme.

[0018] In step S11, the power grid operation dataset is obtained, including: The transformer operating parameters and user electricity consumption data collected by the sensors are obtained to obtain the raw monitoring data stream with time stamps. The original monitoring data stream is standardized and preprocessed to obtain preliminary processed data in a unified format; The time series completeness of the preliminary processed data is calculated, and a threshold judgment is made on the time series completeness. If the time series completeness index is lower than the preset completeness threshold, interpolation is performed to complete the data and obtain the completed power grid monitoring data. The power grid monitoring data is checked for consistency using data quality verification rules. If there is a logical conflict between the operating parameters and the electricity consumption behavior data, abnormal data identification processing is performed to finally obtain the power grid operation dataset.

[0019] It should be noted that sensors refer to intelligent sensing devices deployed on the transformer itself and its surrounding environment, as well as on the user side, including but not limited to current transformers, voltage sensors, temperature and humidity sensors, and electricity meters. Transformer operating parameters are key physical quantities reflecting the real-time operating status of the transformer, mainly including load current, winding temperature, oil temperature, no-load loss, load loss, temperature, and humidity. User electricity consumption behavior data is time-series data reflecting user electricity consumption patterns, mainly including power consumption, electricity consumption periods, load curves, and electricity consumption. The time stamp is a unified timestamp added to each data point through a synchronized clock mechanism to ensure time alignment of multi-source data.

[0020] For example, at 14:30:00 on January 15, 2024, the system synchronously collected parameters such as transformer load current of 285 amperes, user power consumption of 1850 kilowatts, ambient temperature of -5.2 degrees Celsius, and ambient humidity of 68%, and generated an ISO 8601 format timestamp 2024-01-15T14:30:00Z which was bound to these four parameters to form a raw monitoring data stream carrying a unified time identifier.

[0021] In this embodiment, the raw monitoring data stream undergoes standardized preprocessing. First, format conversion is performed, converting the analog signals output by the sensors into a standard digital format. Next, in the field mapping process, data fields from sensors from different manufacturers are standardized to a standard naming convention, such as mapping fields like temp, temperature, and ambient_temperature to a uniform name. Finally, a Kalman filter algorithm is used to filter out noise from the environmental parameters. Through prediction and update stages, measurement noise is eliminated, smoothing the original temperature measurements of -5.2°C, -4.8°C, and -5.5°C to -5.1°C, -4.9°C, and -5.2°C, effectively improving data quality. After standardization preprocessing including format conversion, field mapping, and noise filtering, the raw monitoring data stream yields preliminary processed data in a unified format.

[0022] Furthermore, the completeness of the preliminary processed data is an indicator of the time series completeness of the preliminary processed data, evaluated by calculating the ratio of the actual number of collected data points to the theoretical number of data points. For example, if the sensor sampling frequency is 5 minutes / sample, then the number of samples in 2 hours is 2*60 / 5=24, and the theoretical number of data points is 24. If the actual number of collected data points in those 2 hours is 20, then the completeness is (20 / 24)*100%=83%. The threshold judgment is that when the data completeness for a certain period is 83%, which is lower than the preset 95% threshold, the system performs interpolation to complete the data. Interpolation completion is based on an interpolation method for similar historical periods, analyzing data patterns from the same time period, similar load conditions, and environmental conditions over the past 30 days to determine the estimated value of missing data.

[0023] Specifically, data quality verification rules refer to verification conditions formulated based on the physical laws or business logic of the power system. For example, the transformer load current and the user's power consumption P must satisfy the relationship P=U*I*φ, where U is voltage, I is current, and φ is the power factor. Logical conflicts refer to data anomalies that violate the verification rules, such as power consumption being zero but the load current not being zero. When the system detects that user power consumption data shows zero power consumption for a certain period, but transformer operating parameters show a load current of 150 amps, it identifies a clear logical conflict and adds an anomaly marker to this type of data, marking it as: Logical Conflict - Pending Manual Verification. More specifically, in the historical load data of a 110kV transformer substation, 15 abnormal peak points were found in the July data. These anomalies are usually caused by sensor failures or communication interruptions. For example, if one of the anomalies is that the load rate suddenly jumps from 0.75 to 1.85 and then returns to 0.75, the system will mark it as an anomaly and replace it with the average value of the preceding and following times. Finally, a complete power grid operation dataset is obtained.

[0024] In step S12, a load rate prediction model is constructed based on the power grid operation dataset, and load rate prediction is performed based on the load rate prediction model to obtain a complete prediction curve, including: Time series reconstruction is performed based on the transformer load data in the power grid operation dataset to obtain a standardized load change sequence; Extract the periodic features of daily and weekly cycles from the load change sequence, and construct a feature matrix based on the periodic features to obtain a feature matrix containing periodic and trend features; A long short-term memory network prediction model is constructed based on the feature matrix. The long short-term memory network prediction model is trained, and the input layer is set to historical 24-hour load data. The long short-term memory network prediction model is trained until the error converges to within the preset accuracy to obtain the load rate prediction model. The load rate prediction model is used to calculate the predicted load rate for each moment in the next 24 hours, and the basic prediction curve is determined. The confidence distribution of the predicted values ​​at each time point is calculated based on the basic prediction curve to obtain the complete prediction curve.

[0025] It should be noted that the transformer load data refers to the daily 24-hour load current sequence, load voltage sequence, and load rate sequence in the power grid operation dataset. The time series reconstruction of the transformer load data uses a sliding window method. This method first removes noise and interference. The original load current values ​​for a certain period are 280A, 295A, 1000A, and 300A (where 1000A is an abnormal sensor value). After processing with a moving average filter (window size 5), these values ​​are corrected to 280A, 295A, 300A, 305A, and 310A. Then, data from different sampling frequencies are unified into an hourly time series. For example, if the sampling frequency is once every 15 minutes, there will be four sampling points per hour. The average of these four sampling points is taken as the data point for the hourly time series, ultimately resulting in a standardized load change sequence.

[0026] In this embodiment, periodic feature extraction refers to the process of identifying periodic patterns in load changes using signal analysis algorithms, such as Fast Fourier Transform (FFT), and quantifying them into feature vectors. The daily periodic feature vector reflects the periodic changes in load rate over 24 hours. For example, in an industrial park, the load rate is highest during the morning peak (8:00-10:00) and second highest during the evening peak (18:00-20:00), formed by extracting the amplitude and phase of the 24-hour frequency components from the Fourier transform. The weekly periodic feature vector reflects the periodic changes in load rate over 7 days. For example, the load rate is higher on weekdays than on weekends, with the weekend peak delayed until 19:00, formed by extracting the amplitude and phase of the 168-hour (7×24) frequency components from the Fourier transform.

[0027] For example, a Fast Fourier Transform (FFT) is performed on the standardized load sequence to obtain the frequency domain component X(f), where f is the frequency (unit: 1 / hour); the frequency corresponding to the daily period (24 hours) is f1 = 1 / 24 = 0.0417 Hz, and the amplitude A1 = 0.5 (load rate) and phase ϕ1 = 0.2 rad of this frequency component are extracted; the frequency corresponding to the weekly period (168 hours) is... =1 / 24=0.00595Hz, extract the amplitude A2=0.2 and phase ϕ2=0.5rad of this frequency component; then calculate the long-term trend characteristics using the moving average method, such as the average daily growth rate of the load in the past month of 0.5%, and obtain the trend characteristic K=0.5%. Finally, take the time step as Δt=1 hour, the time step as t (t=1, 2, ..., N), and calculate the daily cycle characteristic value xday(t), the weekly cycle characteristic value xweek(t), and the trend characteristic value K corresponding to each time step.

[0028] The formula for calculating the daily cycle characteristic value is as follows: ; Formula for calculating the periodic eigenvalue: ; Finally, the daily, weekly, and trend eigenvalues ​​at the same time step t are concatenated into a one-dimensional eigenvector of length D (D is the feature dimension, here D=3). The eigenvectors of all time steps are arranged in chronological order to form the final feature matrix B. 特征 ∈R N*D .

[0029] For example, when constructing a Long Short-Term Memory (LSTM) network prediction model, the input layer of the model structure receives a feature matrix B with 24 hours and 3-dimensional features. 特征 The LSTM layer has 128 hidden units and uses gating mechanisms (input gate, forget gate, output gate) to capture long-term dependencies in the time series. The output layer has one neuron, outputting the predicted load factor for the next 24 hours, with a linear activation function. During model training, the mean squared error (MSE) of the actual load factor and the predicted load factor is used as the loss function. The Adam optimizer learning rate is set to 0.001, with a maximum of 500 training epochs. The batch size of the training hyperparameter is 32, and the training and validation sets are divided in an 8:2 ratio. Training continues until the root mean squared error converges to RMSE < 0.045, at which point the load factor prediction model is obtained. If, by the 300th epoch, the validation set RMSE has converged to 0.030, and there are no significant fluctuations in the subsequent 10 epochs, then the load factor prediction model is considered complete.

[0030] Specifically, the load factor prediction model outputs a basic prediction curve representing the predicted load factor for the next 24 hours. Within the model's preset accuracy, the predicted load factor for an industrial area at 9 AM the following day was 0.81, while the actual value was 0.79, achieving a prediction accuracy of 98%. More importantly, the complete prediction curve combines the basic prediction curve with the confidence interval, forming a comprehensive curve encompassing the predicted value and the confidence range. The confidence distribution is derived based on the probability distribution approximation of Monte Carlo sampling and the statistical properties of the normal distribution. First, a sufficiently large sample size (1000 groups) is generated using Monte Carlo sampling, and the mean μ = 0.76 and variance σ = 0.08 of these samples are calculated. According to the Central Limit Theorem, the distribution of the predicted values ​​can be approximated as a normal distribution N(μ, σ... 2 A 95% confidence interval means that in 100 repeated experiments, the true value will fall within this interval 95 times. For a normal distribution, the 95% confidence interval corresponds to the mean μ plus or minus 1.96*σ. 1.96 is the two-tailed quantile of the standard normal distribution at the 95% confidence level. Therefore, the upper boundary of the confidence interval is μ + 1.96*σ = 0.92, and the lower boundary is μ − 1.96*σ = 0.60, providing a basis for risk assessment for dispatchers.

[0031] In step S13, the load gap and overload risk indicator are calculated based on the complete prediction curve. Then, based on the load gap and overload risk indicator, a particle swarm optimization algorithm is used to globally optimize the power allocation strategy, resulting in the power allocation strategy for each energy storage node, including: The actual electricity demand and the load gap between the rated capacity of the transformer for each time period are calculated based on the predicted curve. The load rate for each time period is determined based on the predicted curve. If the load rate exceeds the safety threshold, an overload risk indicator is generated. A particle swarm optimization solver is constructed based on the load gap and the overload risk indicator, and the charging and discharging parameters of the energy storage unit corresponding to the particle dimension are set. If the fitness function reaches the preset convergence condition, the charging and discharging command sequence is output. The real-time load rate is determined based on the charging and discharging command sequence. If the real-time load rate reaches the preset state transition threshold, a state switching command is sent to the energy storage unit. Based on the state switching command, the discharge power share of each energy storage node is allocated according to the principle of nearby power supply to obtain the power allocation strategy.

[0032] It should be noted that the load gap refers to the difference between the actual electricity demand of the transformer and its rated capacity at a certain future time period. This can be calculated using the predicted load rate. For example, if a transformer has a rated capacity of 2000kW and the predicted load rate at 12:00 the next day is 0.95, then the load gap is 2000 × (1 - 0.95) = 100kW. The safety threshold refers to the maximum allowable load rate for the transformer to operate at; for example, a safety threshold of 0.85 can be set. The overload risk indicator is a quantitative marker of the risk of exceeding the load rate threshold. When the predicted load rate reaches 0.92, exceeding the safety threshold of 0.85, the system immediately generates an overload risk indicator, marking that period as a red warning.

[0033] In this embodiment, during the construction of the particle swarm optimization solver, each particle represents an energy storage scheduling scheme. The particle dimension is set to 12 dimensions (4*3 dimensions), corresponding to the charging and discharging power parameters of the four energy storage units in three time periods (the three key future time periods to be optimized). The fitness function integrates three indicators: load balance F1, energy storage utilization F2, and grid stability F3, with a preset convergence condition of fitness F < 0.02.

[0034] In one example, load balance is the degree of matching between the actual discharge power and the load gap, and is calculated using the following formula: ; Energy storage utilization rate is the energy utilization efficiency of an energy storage system, and the calculation formula is as follows: ; Grid stability measures the degree of fluctuation in discharge power, and the calculation formula is as follows: ; The fitness F is calculated using the following formula: ; in, The dimension particle represents the discharge power of the j-th energy storage unit during time period k. This indicates the load shortfall corresponding to the energy storage unit; Indicates the maximum possible discharge power; This represents the discharge power of the j-th energy storage unit during time period k-1; while These are the weight values ​​for load balance, energy storage utilization, and grid stability, respectively. Based on historical experience, in this embodiment, they are set to 0.4, 0.3, and 0.3, respectively.

[0035] For example, when using a particle swarm optimization solver for iterative search, the velocity and position of the particles need to be continuously updated based on the fitness, as shown in the following update formula: ; ; in, This represents the inertia weight, with a value of 0.7. Indicates the number of iterations; express The current velocity of the particle in the y-th iteration; This represents the velocity of the particle in the j-th energy storage unit during time period k in the (y+1)th iteration; This represents the current position of the particle in the j-th energy storage unit during time period k in the k-th iteration; This represents the current position of the particle in the j-th energy storage unit during time period k in the (y+1)th iteration; A value of 1.5 represents the self-learning factor; A value of 2.0 represents the social learning factor; The value range is [0, 1], representing a random factor; The value range is [0, 1], representing a random factor; Represents the optimal solution for an individual; This represents the global optimal solution, embodying the idea of ​​the full gradient descent method.

[0036] It's worth noting that a maximum number of iterations needs to be set during the iterative search, as this determines the algorithm's convergence speed and the quality of the final solution. Setting it too small may cause the algorithm to converge prematurely, failing to find the globally optimal solution; setting it too large, while potentially finding a better solution, will increase computational cost and time consumption. In this embodiment, the maximum number of iterations is set to 80.

[0037] For example, if the load shortfall is -100kW (requiring 100kW of energy storage discharge) during a certain period, there is an overload risk indicator. By iterating 80 times through a pre-constructed particle swarm optimization solver to reach the preset convergence condition F < 0.02, the final output is a sequence of discharge commands for four energy storage units in three different periods. This sequence scheme satisfies the energy storage capacity constraint while optimizing indicators such as load balance, energy storage utilization, and grid stability.

[0038] Furthermore, the state threshold is the safety critical value for switching the operating state of the energy storage unit. After the charging and discharging command sequence is issued, the real-time load rate monitoring system uses a sliding time window technology to update the load rate value every 5 minutes. When it detects that the load rate of a transformer in an industrial park is rapidly rising from 0.78 to 0.83, and it is expected to exceed the state threshold within 15 minutes, the system sends state switching commands to three energy storage units within 800 meters of the transformer. After receiving the command, the energy storage unit switches its operating mode from constant power charging to discharge preparation state, stabilizing the pre-charge voltage at 95% of the rated value to ensure rapid response capability. The principle of proximity power supply means prioritizing the discharge of the energy storage unit closest to the overloaded transformer to reduce transmission losses and improve response speed. For example, the No. 1 energy storage node closest to the overloaded transformer has a capacity of 5MWh, is allocated a 40% discharge power share, and undertakes a discharge task of 1.52MW. The storage nodes No. 2 and No. 3 have capacities of 3MWh and 4MWh respectively. They are located further apart and are allocated 25% and 35% of the total capacity, corresponding to discharge powers of 0.95MW and 1.33MW, thus yielding the final power allocation strategy.

[0039] In step S14, an energy storage unit control matrix is ​​established according to the power allocation strategy, and dynamic scheduling control is performed through a real-time communication network to obtain a scheduling scheme that can be continuously executed, including: A power allocation matrix is ​​established based on the power allocation strategy, and the power status is judged based on the power allocation matrix. If the output power is lower than the preset power threshold, the corresponding weight value in the power allocation matrix is ​​adjusted to achieve balance, and the energy storage unit control matrix is ​​obtained. Based on the energy storage unit control matrix, control commands are sent to each energy storage unit through a real-time communication network to obtain real-time response status feedback information. A fuzzy control algorithm is used to calculate the matching relationship between the load change rate and the energy storage unit response time in the response state feedback information, and to determine the optimized combination of control parameters. Delay detection is performed based on the combination of control parameters. If the response delay of the energy storage unit exceeds a preset time window, the standby energy storage unit with the best state is selected for replacement, thereby obtaining a scheduling scheme that can be executed continuously.

[0040] It should be noted that the power allocation matrix is ​​established by constructing an n×m dimension power allocation matrix based on the geographical location and capacity specifications of each energy storage node in the power allocation strategy, where n represents the number of energy storage units and m represents the number of time periods. Power status judgment checks whether the actual output power of the energy storage units meets the command requirements (e.g., whether insufficient output is due to equipment failure or excessive temperature). The preset power threshold refers to the maximum or minimum allowed power output value of the energy storage unit, which can ensure equipment safety and dynamic satisfaction of load demands, and is used for weight value adjustment. Weight value adjustment is achieved by adjusting the weight coefficients of matrix elements to balance the power allocation of different energy storage units and ensure that the total output power meets the requirements. Real-time monitoring of the output power of each unit reveals that the actual output of unit A at time step 1 is only 80kW, lower than the preset power threshold of 100kW, possibly due to excessive battery temperature causing power limitation. To balance the total power, the weight value of unit A is reduced to 0.8, while the weight value of unit B is increased to 1.2, resulting in the energy storage unit control matrix.

[0041] In this embodiment, the real-time communication network refers to a dedicated power fiber optic cable or a 5G / 4G wireless public network to achieve bidirectional communication between the master station system and the energy storage unit. Control commands include structured data containing parameters such as target power, execution time period, and priority. The master station sends a command to energy storage unit A via the 5G network: 10:00-10:15, discharge power 80kW; unit A feeds back its status to the master station every 500ms, including actual output power 80kW; remaining capacity 800kWh; battery temperature 35℃; communication delay 50ms; energy storage unit response time 2s.

[0042] Furthermore, the load change rate is the rate of change of the load factor over time; the response time is the time it takes for the energy storage unit to reach the target power after receiving the command. The matching relationship calculation uses a fuzzy control algorithm. The system sets the load change rate as one input variable and the energy storage unit response time as another. First, the load change rate needs to be fuzzified into slow, medium, and fast, and the response time into short, moderate, and long. A trapezoidal function is used as the membership function to calculate the membership degree. For example, if the current load change rate is 2kW / s (fast, membership degree 0.8), the response time of unit 1 is 3s (short, membership degree 0.9), and the response time of unit 2 is 9s (long, membership degree 0.7). According to the preset rules in the fuzzy rule base, if the load change rate is fast and the response time is short, the power weight increases (membership degree 0.7, influence strength 0.2); if the load change rate is fast and the response time is long, the power weight decreases (membership degree 0.5, influence strength 0.2). Through fuzzy reasoning, the original weight of unit 1 was 0.6, and the new weight was 0.6 + 0.7 × 0.2 = 0.74. The original weight of unit 2 was 0.25, and the new weight was 0.25 - 0.5 × 0.2 = 0.15. This yields the optimized combination of control parameters for power allocation weight, response priority, and other parameters, which compensates for the shortcomings of single rules (such as fixed weights) under complex operating conditions and improves the adaptability of scheduling.

[0043] Specifically, during delay detection, a preset time window of 3 seconds is set as the criterion for response timeout. When the response delay of energy storage unit No. 5 reaches 4.2 seconds due to an internal fault, the system immediately selects unit No. 12, which is in the best condition, from the standby unit list as a replacement. During the replacement process, the system updates the values ​​of the corresponding row in the control matrix, transferring the 350kW power task originally allocated to unit No. 5 to unit No. 12, and adjusting the power allocation of adjacent units to achieve a smooth transition, ultimately obtaining a scheduling scheme that can be executed continuously, avoiding scheduling interruptions caused by single-point failures.

[0044] In step S15, the actual charging and discharging power of the energy storage system during execution is obtained according to the scheduling scheme, and the load dynamic parameter adjustment is performed based on the actual charging and discharging power to obtain a cooperative control strategy, including: Based on the scheduling scheme, the power status information of the energy storage unit is obtained, and the power status information is converted from analog to digital and filtered to obtain the actual charging and discharging power. The actual load rate is calculated based on the actual charging and discharging power, and the deviation of the actual load rate is judged. If the deviation between the actual load rate and the predicted load rate is greater than the preset deviation threshold, the dynamic adjustment signal is activated to determine the load deviation judgment result. Based on the load deviation judgment results, power is redistributed to each energy storage unit to obtain a collaborative control strategy.

[0045] It should be noted that step S14 performs communication feedback scheduling optimization, while step S15 collects actual power information for collaborative scheduling. The power status information is power-related data collected in real-time during the operation of the energy storage unit after the scheduling scheme is implemented. During analog-to-digital conversion, the 4-20mA current signal output by the sensor (corresponding to 0-500A current) is converted into a digital quantity (e.g., 0-32767) by the data acquisition card (16-bit ADC), and then converted into an actual current value according to the calibration coefficient (e.g., 1mA=0.1A). During filtering, the original power data may fluctuate due to electromagnetic interference, such as 40kW, 42kW, 38kW, and 45kW. After smoothing using a Kalman filter algorithm (setting the process noise variance to 0.01 and the measurement noise variance to 0.1), the actual charging and discharging power is obtained as 40kW, 41kW, 39kW, and 40kW, effectively suppressing noise.

[0046] In this embodiment, the actual load rate is the ratio of the actual charging / discharging power to the rated power. The preset deviation threshold is the maximum deviation between the actual load rate and the predicted load rate. For example, if a transformer has a rated power of 2000kW and an actual discharge power of 1900kW, the actual load rate is 0.95. The predicted load rate is 0.92, and the preset deviation threshold is ±0.03. An actual load rate of 0.95 equals the upper limit of the threshold and is within the acceptable range, so no adjustment is triggered. If the actual load rate is 0.96 (exceeding the tolerance by 0.01), a dynamic adjustment signal is triggered, indicating that the load rate is too high and an adjustment command to increase the discharge power is needed—this is the load deviation judgment result.

[0047] Furthermore, power redistribution requires the system to recalculate the target output power of each energy storage unit to cover the actual load shortfall, with an overload demand of 2000kW * (0.96 - 0.92) = 80kW. Specifically, this is achieved by adjusting the power weights based on remaining capacity and response time. A relationship curve between remaining capacity, response time, and weight adjustment value is fitted based on historical experience, and then the power redistribution for the target demand is performed according to this curve. The original weight allocation was as follows: Unit 6 (0.8), Unit 7 (0.6), Unit 8 (0.4), and Unit 9 (0.1). Units 6 and 7 have a remaining capacity of 150kW, but are already under heavy load (weights of 0.8 and 0.6 respectively). The weight of unit 6 needs to be appropriately reduced to 0.7, resulting in a 40kW reduction in output power. Unit 9 has a remaining capacity of 300kW, but is not being fully utilized. With a response time of 4 seconds, its weight can be increased to 0.4, resulting in a 120kW increase in output power. Therefore, the adjusted weights of units 6, 7, 8, and 9 are 0.7, 0.6, 0.4, and 0.4, respectively. This coordinated control strategy can cover the 80kW load shortfall.

[0048] In one implementation, after obtaining the actual charging and discharging power of the energy storage system during its execution according to the scheduling scheme, and performing dynamic load parameter adjustment based on the actual charging and discharging power to obtain a collaborative control strategy, the method further includes: A sliding time window mechanism is used to acquire load change data and energy storage unit response data to obtain a time series training dataset. Based on the time series training dataset, the gradient descent algorithm is used to optimize the network weight parameter configuration of each layer of neurons in the load rate prediction model, and the load rate prediction result is generated based on the network weight parameter configuration to predict the error trend and determine the prediction accuracy evaluation status. The load rate prediction model is retrained based on the prediction accuracy evaluation status to obtain an updated load rate prediction model.

[0049] It should be noted that this additional process is a key step in achieving continuous optimization of the load rate prediction model. After the implementation of the collaborative control strategy, the model is updated by sliding time window, the model parameters are optimized by gradient descent, and periodic retraining is performed to ensure that the prediction model can adapt to the dynamic changes in the power grid load and avoid the decrease in prediction accuracy caused by data distribution drift or equipment aging.

[0050] The sliding time window mechanism acquires data over a 12-hour period with a 15-minute step, extracting the latest 12 hours of data from the historical load record table. This data includes fields such as hourly load rate, energy storage unit charging / discharging power, and response time. The extracted window data undergoes imputation (if data for a certain period (within 12 hours) is insufficient or missing, the time period needs to be expanded from 12 hours to 24 hours) and noise filtering (e.g., smoothing the load rate sequence using Kalman filtering) to form the final time-series training dataset. The sliding time window mechanism ensures that the training data always contains the latest load and energy storage response information. In this embodiment, gradient descent is an optimization method that minimizes loss. The direction of parameter updates is determined by calculating the partial derivatives of the loss function with respect to the weight parameters of each layer. When the load rate prediction model contains three hidden layers with 64, 32, and 16 neurons respectively, the backpropagation process requires calculating the gradient values ​​of a total of 2048 weight parameters. The gradient explosion threshold is a key parameter in the algorithm, designed to prevent abnormally large gradient values ​​(gradient explosion) during backpropagation, which could lead to uncontrolled model parameter updates. It needs to be determined comprehensively based on the specific application scenario and training experience. In this embodiment, the system sets the gradient explosion threshold to 5.0. When the absolute value of the gradient of a weight in the second hidden layer reaches 7.2, the gradient pruning mechanism is immediately activated to scale the gradient value to within the threshold range. After the network weight parameters of the model are updated in the reverse direction of the gradient, the historical load data for the previous 24 hours is input, and the predicted load rate value for the next 24 hours is output. During subsequent verification, when the average absolute errors for three consecutive prediction periods (each prediction period is 3 hours) are 2.1%, 2.8%, and 3.4%, respectively, the system identifies an increasing error trend, and the latest error value exceeds the tolerance threshold of 3.0%. The model retraining flag is then set to the active state, and the prediction accuracy evaluation state is obtained.

[0051] Furthermore, since the output of the load rate prediction model continuously optimized using the gradient descent algorithm can no longer meet the error tolerance threshold, the load rate prediction model needs to be retrained. The retraining process involves retraining the model using the latest load data from the past 12 hours (or expanding to 24 hours if the data volume is insufficient) according to step S12, and updating the weight parameters.

[0052] In step S16, an energy storage unit status assessment report is obtained, and based on the energy storage unit status assessment report, the collaborative control strategy is weighted according to the power weight allocation considering the entire life cycle of the equipment to obtain a scheduling optimization scheme and a maintenance scheme, including: Obtain the voltage and current curves of each energy storage unit from the historical charge and discharge data, and calculate the efficiency decay coefficient based on the voltage and current curves to obtain the energy storage unit efficiency decay coefficient. A state assessment report for the energy storage unit is obtained by comparing the state threshold based on the efficiency decay coefficient of the energy storage unit. The power of each energy storage unit in the collaborative control strategy is dynamically weighted based on the status assessment report, and a scheduling instruction sequence is generated based on the adjustment results to obtain the final scheduling optimization scheme. Maintenance instructions are generated based on the status assessment report, resulting in a maintenance plan.

[0053] It should be noted that the voltage-current curve refers to the curve showing the change of voltage (V) and current (A) over time (t) during the charging and discharging process of the energy storage unit. The efficiency degradation coefficient is a quantitative indicator that measures the decrease in energy conversion efficiency of the energy storage unit with the number of cycles. All charge-discharge voltage-current curves of a specific energy storage unit are extracted from historical charge-discharge data, including data from all charge-discharge cycles (e.g., cycle 1: charging voltage 3.8V → discharging voltage 3.2V). Based on the voltage-current curves, the actual capacity of each cycle is calculated using the ampere-hour integration method. ,in (This refers to the discharge current). For example, the actual capacity in the first cycle is considered to be the rated capacity C of the energy storage unit. 额定容量 =2000kWh, the actual capacity in a certain cycle is C 实际容量 =1800kWh, then the attenuation coefficient α = (1-C 实际容量 / C 额定容量 100% = 10%, which means that the capacity of the unit decreases by 10%.

[0054] In this embodiment, a state threshold comparison is performed based on the energy storage unit's efficiency degradation coefficient. The state thresholds are: healthy state (degradation coefficient ≤ 5%), sub-healthy state (degradation coefficient ≤ 15%), and fault state (degradation coefficient > 15%). For example, if the energy storage unit (number E002) has a degradation coefficient of 20%, it is determined to be in a fault state, and a state assessment report for the energy storage unit is generated.

[0055] Furthermore, when making dynamic weight adjustments, it is necessary to adjust the power allocation weight of the energy storage unit in the collaborative control strategy according to the status level (such as healthy, sub-healthy, faulty) in the status assessment report of the energy storage unit (e.g., increase the weight of healthy units by 0.2, decrease the weight of sub-healthy units by 0.2, and set the weight of faulty units to 0); then, based on the allocated weights, a sequence of scheduling instructions containing parameters such as timestamp, target power, and priority is generated to obtain the final scheduling optimization scheme.

[0056] Specifically, energy storage units that are in a faulty state in the status assessment report need to be manually replaced or maintained, thereby generating maintenance instructions containing the energy storage unit number, location, and status, and obtaining the final maintenance plan.

[0057] In summary, this invention discloses a transformer load rate control method based on distributed energy storage, comprising: acquiring a power grid operation dataset; constructing a load rate prediction model based on the power grid operation dataset, and performing load rate prediction based on the load rate prediction model to obtain a complete prediction curve; calculating the load gap and overload risk indicators based on the complete prediction curve, and using a particle swarm optimization algorithm to globally optimize the power allocation strategy based on the load gap and the overload risk indicators to obtain the power allocation strategy for each energy storage node; establishing an energy storage unit control matrix based on the power allocation strategy, and performing dynamic scheduling control through a real-time communication network to obtain a scheduling scheme that can be continuously executed; acquiring the actual charging and discharging power during the execution of the energy storage system based on the scheduling scheme, and performing dynamic load parameter adjustment based on the actual charging and discharging power to obtain a collaborative control strategy; acquiring an energy storage unit status assessment report, and performing power weight allocation considering the entire life cycle of the equipment based on the energy storage unit status assessment report to obtain a scheduling optimization scheme and a maintenance scheme. This invention achieves precise control of transformer load rate and efficient scheduling of energy storage system through multi-dimensional data fusion and intelligent optimization.

[0058] Reference Figure 2 The second embodiment of the present invention provides a transformer load rate control system based on distributed energy storage, comprising: The data acquisition module is used to acquire power grid operation datasets. The predictive modeling module is used to construct a load rate prediction model based on the power grid operation dataset, and to predict the load rate based on the load rate prediction model to obtain a complete prediction curve. The strategy generation module is used to calculate the load gap and overload risk indicator based on the complete prediction curve, and to perform global optimization of the power allocation strategy using the particle swarm algorithm based on the load gap and the overload risk indicator to obtain the power allocation strategy for each energy storage node. The control execution module is used to establish an energy storage unit control matrix according to the power allocation strategy, and to perform dynamic scheduling control through a real-time communication network to obtain a scheduling scheme that can be executed continuously. The dynamic parameter adjustment module is used to obtain the actual charging and discharging power of the energy storage system during the execution process according to the scheduling scheme, and to perform dynamic load parameter adjustment based on the actual charging and discharging power to obtain a collaborative control strategy. The lifespan optimization module is used to obtain the energy storage unit status assessment report, and based on the energy storage unit status assessment report, to perform power weight allocation on the collaborative control strategy considering the entire life cycle of the equipment, thereby obtaining a scheduling optimization scheme and a maintenance scheme.

[0059] It should be noted that the transformer load rate control system based on distributed energy storage provided in this embodiment of the invention is used to execute all the process steps of the transformer load rate control method based on distributed energy storage in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0060] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a data acquisition program. When the processor executes the computer program, it implements the steps in the various embodiments of the transformer load rate control method based on distributed energy storage described above, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the data acquisition module.

[0061] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0062] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0063] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0064] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0065] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0066] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0067] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that 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 for those skilled in the art.

Claims

1. A transformer load rate control method based on distributed energy storage, characterized in that, include: Obtain power grid operation dataset; A load rate prediction model is constructed based on the power grid operation dataset, and a load rate prediction is performed based on the load rate prediction model to obtain a complete prediction curve; The load gap and overload risk indicators are calculated based on the complete prediction curve. Then, the power allocation strategy is globally optimized using the particle swarm optimization algorithm based on the load gap and the overload risk indicators to obtain the power allocation strategy for each energy storage node. A control matrix for energy storage units is established based on the power allocation strategy, and dynamic scheduling control is performed through a real-time communication network to obtain a scheduling scheme that can be executed continuously. The actual charging and discharging power of the energy storage system during the execution process is obtained according to the scheduling scheme, and the load is dynamically adjusted according to the actual charging and discharging power to obtain a collaborative control strategy. Obtain the energy storage unit status assessment report, and based on the energy storage unit status assessment report, perform power weight allocation on the collaborative control strategy considering the entire life cycle of the equipment to obtain the scheduling optimization scheme and maintenance scheme.

2. The transformer load rate control method based on distributed energy storage according to claim 1, characterized in that, The acquisition of the power grid operation dataset includes: The transformer operating parameters and user electricity consumption data collected by the sensors are obtained to obtain the raw monitoring data stream with time stamps. The original monitoring data stream is standardized and preprocessed to obtain preliminary processed data in a unified format; The time series completeness of the preliminary processed data is calculated, and a threshold judgment is made on the time series completeness. If the time series completeness index is lower than the preset completeness threshold, interpolation is performed to complete the data and obtain the completed power grid monitoring data. The power grid monitoring data is checked for consistency using data quality verification rules. If there is a logical conflict between the operating parameters and the electricity consumption behavior data, abnormal data identification processing is performed to finally obtain the power grid operation dataset.

3. The transformer load rate control method based on distributed energy storage according to claim 1, characterized in that, The step of constructing a load rate prediction model based on the power grid operation dataset, and performing load rate prediction based on the load rate prediction model to obtain a complete prediction curve includes: Time series reconstruction is performed based on the transformer load data in the power grid operation dataset to obtain a standardized load change sequence; Extract the periodic features of daily and monthly cycles from the load change sequence, and construct a feature matrix based on the periodic features to obtain a feature matrix containing periodic and trend features; A long short-term memory network prediction model is constructed based on the feature matrix. The long short-term memory network prediction model is trained, and the input layer is set to historical 24-hour load data. The long short-term memory network prediction model is trained until the error converges to within the preset accuracy to obtain the load rate prediction model. The load rate prediction model is used to calculate the predicted load rate for each moment in the next 24 hours, and the basic prediction curve is determined. The confidence distribution of the predicted values ​​at each time point is calculated based on the basic prediction curve to obtain the complete prediction curve.

4. The transformer load rate control method based on distributed energy storage according to claim 1, characterized in that, The process involves calculating the load gap and overload risk indicator based on the complete prediction curve, and then using a particle swarm optimization algorithm to globally optimize the power allocation strategy based on the load gap and overload risk indicator to obtain the power allocation strategy for each energy storage node, including: The actual electricity demand and the load gap between the rated capacity of the transformer for each time period are calculated based on the predicted curve. The load rate for each time period is determined based on the predicted curve. If the load rate exceeds the safety threshold, an overload risk indicator is generated. A particle swarm optimization solver is constructed based on the load gap and the overload risk indicator, and the charging and discharging parameters of the energy storage unit corresponding to the particle dimension are set. If the fitness function reaches the preset convergence condition, the charging and discharging command sequence is output. The real-time load rate is determined based on the charging and discharging command sequence. If the real-time load rate reaches the preset state transition threshold, a state switching command is sent to the energy storage unit. Based on the state switching command, the discharge power share of each energy storage node is allocated according to the principle of nearby power supply to obtain the power allocation strategy.

5. The transformer load rate control method based on distributed energy storage according to claim 1, characterized in that, The step of establishing an energy storage unit control matrix based on the power allocation strategy and performing dynamic scheduling control through a real-time communication network to obtain a continuously executable scheduling scheme includes: A power allocation matrix is ​​established based on the power allocation strategy, and the power status is judged based on the power allocation matrix. If the output power is lower than the preset power threshold, the corresponding weight value in the power allocation matrix is ​​adjusted to achieve balance, and the energy storage unit control matrix is ​​obtained. Based on the energy storage unit control matrix, control commands are sent to each energy storage unit through a real-time communication network to obtain real-time response status feedback information. A fuzzy control algorithm is used to calculate the matching relationship between the load change rate and the energy storage unit response time in the response state feedback information, and to determine the optimized combination of control parameters. Delay detection is performed based on the combination of control parameters. If the response delay of the energy storage unit exceeds a preset time window, the standby energy storage unit with the best state is selected for replacement, thereby obtaining a scheduling scheme that can be executed continuously.

6. The transformer load rate control method based on distributed energy storage according to claim 1, characterized in that, The step of obtaining the actual charging and discharging power of the energy storage system during execution according to the scheduling scheme, and performing dynamic load parameter adjustment based on the actual charging and discharging power to obtain a collaborative control strategy includes: Based on the scheduling scheme, the power status information of the energy storage unit is obtained, and the power status information is converted from analog to digital and filtered to obtain the actual charging and discharging power. The actual load rate is calculated based on the actual charging and discharging power, and the deviation of the actual load rate is judged. If the deviation between the actual load rate and the predicted load rate is greater than the preset deviation threshold, the dynamic adjustment signal is activated to determine the load deviation judgment result. Based on the load deviation judgment results, power is redistributed to each energy storage unit to obtain a collaborative control strategy.

7. The transformer load rate control method based on distributed energy storage according to claim 2 or 6, characterized in that, After obtaining the actual charging and discharging power of the energy storage system during its execution according to the scheduling scheme, and performing dynamic load parameter adjustment based on the actual charging and discharging power to obtain a coordinated control strategy, the method further includes: A sliding time window mechanism is used to acquire load change data and energy storage unit response data to obtain a time series training dataset. Based on the time series training dataset, the gradient descent algorithm is used to optimize the network weight parameter configuration of each layer of neurons in the load rate prediction model, and the load rate prediction result is generated based on the network weight parameter configuration to predict the error trend and determine the prediction accuracy evaluation status. The load rate prediction model is retrained based on the prediction accuracy evaluation status to obtain an updated load rate prediction model.

8. The transformer load rate control method based on distributed energy storage according to claim 1, characterized in that, The process of obtaining the energy storage unit status assessment report and, based on the energy storage unit status assessment report, allocating power weights to the collaborative control strategy considering the entire lifecycle of the equipment to obtain a scheduling optimization scheme and a maintenance scheme includes: Obtain the voltage and current curves of each energy storage unit from the historical charge and discharge data, and calculate the efficiency decay coefficient based on the voltage and current curves to obtain the energy storage unit efficiency decay coefficient. A state assessment report for the energy storage unit is obtained by comparing the state threshold based on the efficiency decay coefficient of the energy storage unit. The power of each energy storage unit in the collaborative control strategy is dynamically weighted based on the status assessment report, and a scheduling instruction sequence is generated based on the adjustment results to obtain the final scheduling optimization scheme. Maintenance instructions are generated based on the status assessment report, resulting in a maintenance plan.

9. A transformer load rate control system based on distributed energy storage, characterized in that, include: The data acquisition module is used to acquire power grid operation datasets. The predictive modeling module is used to construct a load rate prediction model based on the power grid operation dataset, and to predict the load rate based on the load rate prediction model to obtain a complete prediction curve. The strategy generation module is used to calculate the load gap and overload risk indicator based on the complete prediction curve, and to perform global optimization of the power allocation strategy using the particle swarm algorithm based on the load gap and the overload risk indicator to obtain the power allocation strategy for each energy storage node. The control execution module is used to establish an energy storage unit control matrix according to the power allocation strategy, and to perform dynamic scheduling control through a real-time communication network to obtain a scheduling scheme that can be executed continuously. The dynamic parameter adjustment module is used to obtain the actual charging and discharging power of the energy storage system during the execution process according to the scheduling scheme, and to perform dynamic load parameter adjustment based on the actual charging and discharging power to obtain a collaborative control strategy. The lifespan optimization module is used to obtain the energy storage unit status assessment report, and based on the energy storage unit status assessment report, to perform power weight allocation on the collaborative control strategy considering the entire life cycle of the equipment, thereby obtaining a scheduling optimization scheme and a maintenance scheme.

Images