Intelligent control method for power distribution and efficiency optimization of energy storage power station

Abstract

The invention discloses an intelligent control method for power distribution and efficiency optimization of an energy storage power station, and the method comprises the steps: obtaining real-time operation data and historical load records of all PCS units in the energy storage power station, and obtaining a current load change trend through data fusion processing; according to the current load change trend, a fuzzy logic algorithm is adopted to analyze the performance state and the output proportion of the PCS unit, and the initial load distribution weight of each unit is determined; aiming at the optimized distribution scheme, generating a self-adaptive priority ranking rule by adopting a genetic algorithm, judging whether the rule covers real-time state change or not, and obtaining a dynamic ranking list; extracting a high-priority PCS unit from the dynamic sorting list, obtaining an equipment loss index of the high-priority PCS unit, and determining an output increase proportion of a low-loss unit through index comparison processing; and according to the output increase proportion, fusing the resource utilization rate data, and if the utilization rate is lower than a preset threshold value, redistributing idle unit power to obtain overall efficiency improvement configuration.

Description

Technical Field

[0001] This invention relates to the field of energy storage power stations, specifically to an intelligent control method for power distribution and efficiency optimization in energy storage power stations. Background Technology

[0002] Energy storage systems play a crucial role in modern power systems, ensuring stable grid operation and efficient dispatch through rapid response and flexible adjustment. With the widespread application of renewable energy and complex changes in power load, energy storage systems need to simultaneously handle multiple tasks such as stability control, primary frequency regulation, and automatic generation control (AGC), placing extremely high demands on the system's coordination and responsiveness. The rapid development of energy storage technology has led to an increasing number of application scenarios in the power grid; however, achieving efficient and reliable control in complex and ever-changing grid environments has become a critical issue that urgently needs to be addressed.

[0003] Existing energy storage control methods often face the dual challenges of efficiency and reliability when dealing with large-scale distributed energy storage systems. The core technical difficulty lies in how to achieve batch coordinated management of energy storage units and ensure the accuracy of power allocation and system reliability in dynamic environments. Energy storage units differ in geographical location, state of charge (SBC), and power capacity, making unified scheduling difficult. Furthermore, communication latency and potential fault redundancy further exacerbate the coordination challenges. Communication latency leads to delayed power allocation commands, especially during rapid grid frequency fluctuations, which may prevent timely adjustments to the output power of energy storage units, thus affecting grid stability. Therefore, how to dynamically optimize batch scheduling strategies based on the SBC, power capacity, and response characteristics of energy storage units while meeting grid real-time requirements, to achieve efficient resource utilization and reliable system operation, has become a key issue for energy storage coordinated control systems.

[0004] Solving these problems requires not only overcoming the limitations of managing individual energy storage units, but also optimizing batch dispatch mechanisms at the system level to address the dynamics and uncertainties of power grid operation. This will be a key issue for energy storage systems to efficiently participate in stability control and frequency regulation services. Summary of the Invention

[0005] The purpose of this invention is to solve the above problems and provide an intelligent control method for power allocation and efficiency optimization of energy storage power stations.

[0006] The technical solution adopted by this invention to solve its technical problem is:

[0007] A smart control method for power allocation and efficiency optimization in an energy storage power station includes the following steps:

[0008] S101 acquires real-time operating data and historical load records of each PCS unit in the energy storage power station, and obtains the current load change trend through data fusion processing;

[0009] Based on the current load change trend, S102 uses a fuzzy logic algorithm to analyze the performance status and output ratio of the PCS unit and determine the initial load allocation weight of each unit.

[0010] S103 obtains the power grid demand fluctuation signal through the initial load allocation weight. If the fluctuation signal exceeds the preset threshold, the weight is adjusted to balance the PCS unit load and obtain the optimized allocation scheme.

[0011] S104 uses a genetic algorithm to generate adaptive priority sorting rules for the optimized allocation scheme, determines whether the rules cover real-time state changes, and obtains a dynamic sorting list.

[0012] S105 extracts high-priority PCS units from the dynamic sorting list, obtains their equipment loss indicators, and determines the output increase ratio of low-loss units through indicator comparison processing.

[0013] S106 integrates resource utilization data based on the output increase ratio. If the utilization rate is lower than the preset threshold, the idle unit power is reallocated to obtain an overall efficiency improvement configuration.

[0014] S107 improves overall efficiency by analyzing power plant operating environment variables, using particle swarm optimization algorithm to refine adaptive adjustment parameters, determining whether parameters respond to sudden loads, and obtaining the final power allocation command.

[0015] S108 extracts coordinated operation indicators from the final power allocation command, obtains feedback loop data, determines the command execution effect through data verification processing, and obtains the stable output mode of the power plant.

[0016] Further, step S101 includes:

[0017] Acquire real-time operating data and historical load records of the PCS unit, and process the real-time operating data and historical load records using a time series alignment method to obtain an aligned data sequence;

[0018] By applying the Long Short-Term Memory network algorithm to the aligned data sequence, a preliminary load prediction sequence is obtained;

[0019] If the variance of the preliminary load prediction sequence exceeds a preset threshold, a sliding window filter is used to process the preliminary load prediction sequence to obtain a smoothed load sequence.

[0020] The first-order difference is calculated based on the smoothed load sequence to obtain the differential load sequence;

[0021] By applying the support vector machine algorithm to the differential load sequence, the load change classification label is obtained;

[0022] The trend direction is determined based on the load change classification labels;

[0023] The load change trend is correlated with the real-time operating data of the PCS unit to obtain the fused load trend.

[0024] Further, step S102 includes:

[0025] Obtain the current data sequence, which is determined based on the load trend;

[0026] The performance status of the current data sequence is determined by fuzzy logic to obtain a performance status set. The formula for determining the performance status of the data sequence by fuzzy logic is as follows;

[0027]

[0028] P i Let m represent the fuzzy judgment result of the i-th performance state, and w represent the total number of fuzzy rules. j μ represents the weight of the j-th rule. ij (x) represents the membership value of input x to the i-th performance state in the j-th rule;

[0029] If the performance state set meets the preset conditions, the output ratio is calculated through the membership function to obtain the output ratio set. The formula for calculating the output ratio through the membership function is as follows;

[0030] R k =∫0 1 μ k (y)·ydy / ∫0 1 μ k (y)dyR k μ represents the k-th output ratio value. k (y) represents the membership function of the k-th output state, and y represents the range of values ​​for the output ratio.

[0031] Based on the power output ratio set, a preliminary load allocation weight set is determined using fuzzy inference rules.

[0032] Further, step S103 includes:

[0033] Acquire grid demand fluctuation signals, which are generated through a preset weight allocation method;

[0034] The signal amplitude is determined based on the power grid demand fluctuation signal;

[0035] If the signal amplitude exceeds a preset threshold, the load status of each PCS unit is obtained;

[0036] The current load value of each unit is determined based on the load status of the PCS unit.

[0037] The K-nearest neighbor algorithm is used to perform cluster analysis on the current load value to obtain the load clustering results;

[0038] The weight adjustment factor is determined based on the load distribution characteristics in the load clustering results. The formula for calculating the weight adjustment factor is as follows:

[0039]

[0040] α c N represents the weight adjustment factor for the c-th cluster. c N represents the number of load samples in the c-th cluster. total σ represents the total number of load samples. c This represents the load variance of the c-th cluster. The average loading variance of all clusters is represented by β, which is the adjustment factor.

[0041] The preset weights are adjusted and calculated using the weight adjustment factor to obtain the optimized weights;

[0042] The optimized weights are used to redistribute the load of each PCS unit to obtain an optimized allocation scheme;

[0043] The calculation formula for load redistribution based on optimized weights is as follows:

[0044]

[0045] P opt，i P represents the optimized load allocation value for the i-th PCS unit. preset，i γ represents the preset load value of the i-th PCS unit. i P represents the weight adjustment parameter corresponding to the i-th PCS unit. total This represents the total system load demand, and n represents the total number of PCS units.

[0046] Further, step S104 includes:

[0047] Obtain an initial priority data set, which contains multiple priority parameters;

[0048] The initial priority data set is processed using a genetic algorithm to obtain the first rule set;

[0049] Generate adaptive priority sorting rules based on the first rule set;

[0050] Acquire real-time status change data, which includes current system status information;

[0051] Determine the extent to which the adaptive priority sorting rule covers the real-time state change data;

[0052] If the adaptive priority sorting rule completely covers the real-time state change data, then the first rule set is determined to be the second rule set;

[0053] If the adaptive priority sorting rule does not completely cover the real-time state change data, a genetic algorithm is used to adjust the first rule set to obtain a second rule set.

[0054] Based on the second rule set, the state change process is monitored to obtain the change coverage sequence;

[0055] A dynamic sorted list is obtained by sorting the sequence elements in the changed cover sequence.

[0056] Further, step S105 includes:

[0057] Obtain a set of high-priority PCS units from a dynamic sorting list, the set of high-priority PCS units containing multiple power conversion units to be scheduled;

[0058] For each power conversion unit in the high-priority PCS unit set, obtain the corresponding equipment loss index, which reflects the operating loss status of each power conversion unit.

[0059] The loss ranking value corresponding to each power conversion unit is calculated based on the equipment loss index. The loss ranking value is used to quantify the degree of loss of each unit.

[0060] If the loss ranking value is lower than a preset threshold, the corresponding power conversion unit is judged as a low-loss PCS unit.

[0061] A power output reference value is determined for the low-loss PCS unit, and the power output reference value serves as a reference standard for power adjustment.

[0062] The K-nearest neighbor algorithm is used to calculate the output increase ratio based on the output benchmark value. The output increase ratio is determined based on historical operating data and the current operating status.

[0063] Further, step S106 includes:

[0064] The output ratio data of multiple processing units is obtained, and the resource utilization rate data of each processing unit is merged based on the output ratio data to obtain the merged utilization rate data.

[0065] The fusion utilization rate data is compared with a preset threshold. If the fusion utilization rate data is lower than the preset threshold, the power information of the idle units in each processing unit is obtained.

[0066] Calculate the power parameters to be allocated based on the power information of the idle units, and redistribute the power of each processing unit using the power parameters to be allocated to obtain the power configuration after allocation;

[0067] The overall operating efficiency of each processing unit is calculated based on the allocated power configuration to obtain efficiency improvement configuration data.

[0068] Further, step S107 includes:

[0069] Environment variables are obtained based on efficiency configuration, and these environment variables include system operating status information;

[0070] By performing operational analysis using the aforementioned environmental variables, a preliminary parameter set is obtained;

[0071] The particle swarm optimization algorithm is used to refine the initial parameter set to obtain adaptive parameters.

[0072] Determine the set of optimization parameters based on the adaptive parameters;

[0073] Extract burst load indicators from the set of optimized parameters;

[0074] If the burst load indication exceeds a preset threshold, the parameter response capability is evaluated to obtain a response judgment result;

[0075] A power command is generated based on the response judgment result;

[0076] The power command outputs a final allocation command, which is used to control the system's power allocation.

[0077] Further, step S108 includes:

[0078] Obtain power allocation instructions and extract coordination index data based on the power allocation instructions;

[0079] Feedback data is obtained from the collaborative indicator data, and a cyclic data set is generated from the feedback data;

[0080] The cyclic data set is processed using a data cleaning algorithm to obtain cleaned data;

[0081] If the quality parameters of the cleaned data meet the preset threshold, the execution status of the cleaned data is judged to obtain a preliminary status identifier.

[0082] Based on the preliminary state identifier, a support vector machine classification algorithm is used to classify the state and obtain the classification state result;

[0083] Extract execution effect parameters from the classification status results, and calculate stable parameter values ​​using the execution effect parameters;

[0084] Based on the stable parameter values, the K-nearest neighbor clustering algorithm is used to cluster the data, and a set of clustering results is obtained.

[0085] The output pattern type is determined from the clustering result set to obtain the preliminary output pattern;

[0086] If the deviation between the initial output mode and the standard output mode is less than a preset deviation threshold, then the initial output mode is determined to be a stable output mode.

[0087] The beneficial effects of this invention are:

[0088] 1. This invention addresses the problems of low operating efficiency, uneven load distribution, and insufficient dynamic response in energy storage power stations, employing technologies such as data fusion, fuzzy logic analysis, genetic algorithms, and particle swarm optimization to achieve a comprehensive solution. First, this invention obtains load change trends by fusing real-time and historical data, and uses a fuzzy logic algorithm to evaluate PCS unit performance and determine initial load distribution weights. In response to grid demand fluctuations, the weights are dynamically adjusted to balance the load, generating an optimized distribution scheme. Subsequently, an adaptive priority ranking rule is generated using a genetic algorithm, extracting high-priority, low-loss units and adjusting their output ratios. Resource utilization data is then integrated to redistribute idle power to improve overall efficiency. Finally, a particle swarm optimization algorithm is used to refine the adjustment parameters, respond to sudden load changes, generate final power distribution commands, and verify the execution effect through feedback loops to ensure stable power station output. This invention significantly improves the operating efficiency, dynamic response capability, and system stability of energy storage power stations, achieving intelligent power distribution and optimized resource utilization. Attached Figure Description

[0089] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0090] like Figure 1 As shown, an intelligent control method for power allocation and efficiency optimization in an energy storage power station includes the following steps:

[0091] S101 acquires real-time operating data and historical load records of each PCS unit in the energy storage power station, and obtains the current load change trend through data fusion processing;

[0092] Based on the current load change trend, S102 uses a fuzzy logic algorithm to analyze the performance status and output ratio of the PCS unit and determine the initial load allocation weight of each unit.

[0093] S103 obtains the power grid demand fluctuation signal through the initial load allocation weight. If the fluctuation signal exceeds the preset threshold, the weight is adjusted to balance the PCS unit load and obtain the optimized allocation scheme.

[0094] S104 uses a genetic algorithm to generate adaptive priority sorting rules for the optimized allocation scheme, determines whether the rules cover real-time state changes, and obtains a dynamic sorting list.

[0095] S105 extracts high-priority PCS units from the dynamic sorting list, obtains their equipment loss indicators, and determines the output increase ratio of low-loss units through indicator comparison processing.

[0096] S106 integrates resource utilization data based on the output increase ratio. If the utilization rate is lower than the preset threshold, the idle unit power is reallocated to obtain an overall efficiency improvement configuration.

[0097] S107 improves overall efficiency by analyzing power plant operating environment variables, using particle swarm optimization algorithm to refine adaptive adjustment parameters, determining whether parameters respond to sudden loads, and obtaining the final power allocation command.

[0098] S108 extracts coordinated operation indicators from the final power allocation command, obtains feedback loop data, determines the command execution effect through data verification processing, and obtains the stable output mode of the power plant.

[0099] Step S101 includes: acquiring real-time operating data and historical load records of the PCS unit; processing the real-time operating data and historical load records using a time series alignment method to obtain an aligned data sequence; applying a Long Short-Term Memory (LSTM) network algorithm to the aligned data sequence to obtain a preliminary load prediction sequence; if the variance of the preliminary load prediction sequence exceeds a preset threshold, processing the preliminary load prediction sequence using a sliding window filter to obtain a smoothed load sequence; calculating the first-order difference based on the smoothed load sequence to obtain a differential load sequence; applying a Support Vector Machine (SVM) algorithm to the differential load sequence to obtain a load change classification label; determining the trend direction based on the load change classification label to determine the load change trend; and performing correlation analysis between the load change trend and the real-time operating data of the PCS unit to obtain a fused load trend.

[0100] When acquiring real-time operating data and historical load records of the PCS unit, the data acquisition frequency may be inconsistent; real-time data may be acquired once per second, while historical load records may be recorded once per minute. Time series alignment uses interpolation or resampling techniques to unify data with different time granularities onto the same time base, such as unifying it to one data point every 30 seconds, ensuring that subsequent algorithms can effectively process continuous data sequences.

[0101] Long Short-Term Memory (LSTM) network algorithms are used for deep learning analysis of aligned data sequences. LSTM networks can capture long-term dependencies in PCS cell loads, such as the periodic load patterns formed by daytime charging and nighttime discharging in energy storage systems. By learning the load variation patterns over the past 24 hours, the network predicts load trends for the next 4 hours, generating a preliminary load prediction sequence.

[0102] When the variance of the initial load prediction sequence exceeds the preset threshold, it indicates that the prediction result fluctuates greatly and there may be abnormal peaks. The system then activates the sliding window filtering mechanism. The sliding window filtering uses a 5-minute time window to smooth the prediction sequence, eliminate sudden noise interference, and obtain a more stable and smooth load sequence.

[0103] First-order difference calculations reflect the instantaneous rate of load change by measuring the difference in load values ​​at adjacent time points. If the current load is 80kW and the previous load was 75kW, then the first-order difference is 5kW, indicating that the load is trending upwards. The differential load sequence can highlight the dynamic characteristics of load changes and provide key input for subsequent classification analysis.

[0104] The support vector machine algorithm performs pattern recognition on the differential load sequence, classifying load changes into three categories: rising, falling, and stable.

[0105] SVM can accurately identify different load change patterns by constructing an optimal classification hyperplane. When the difference value is consistently positive and the amplitude is increasing, it is marked as a strong upward trend; when the difference value fluctuates within a range of ±2kW, it is marked as a stable trend.

[0106] The trend judgment of load change classification labels is based on statistical analysis methods. The system statistically analyzes the distribution ratio of various labels within the future prediction time window. When the proportion of rising labels exceeds 60%, it is determined to be an upward trend; when the proportion of falling labels exceeds 60%, it is determined to be a downward trend; otherwise, it is determined to be a stable trend. This classification method can effectively quantify the directional characteristics of load changes.

[0107] The load trend fusion method combines load change trends with real-time operating data from PCS units through correlation analysis. This analysis considers the impact of PCS unit operating parameters such as power factor, temperature, and voltage on the load trend. When PCS unit temperature rises and the load shows an upward trend, the system adjusts the prediction weights to improve its early warning capability for equipment overload risks. This fusion method significantly improves the accuracy and practicality of load forecasting, providing reliable decision support for the intelligent scheduling of energy storage systems.

[0108] Step S102 includes: obtaining the current data sequence, which is determined according to the load trend; using fuzzy logic to determine the performance status of the current data sequence to obtain a performance status set. The formula for determining the performance status of the data sequence using fuzzy logic is as follows;

[0109]

[0110] P i Let m represent the fuzzy judgment result of the i-th performance state, and w represent the total number of fuzzy rules. j μ represents the weight of the j-th rule. ij (x) represents the membership value of input x to the i-th performance state in the j-th rule; if the performance state set meets the preset conditions, the output ratio is calculated through the membership function to obtain the output ratio set. The formula for calculating the output ratio through the membership function is as follows;

[0111] R k =∫0 1 μ k (y)·ydy / ∫0 1 μ k (y)dy

[0112] R k μ represents the k-th output ratio value. k (y) represents the membership function of the k-th output state, and y represents the range of output ratio values; based on the output ratio set, fuzzy inference rules are used to determine the preliminary load allocation weight set.

[0113] The current data sequence is obtained based on the results of previous load trend analysis, including key operating indicators such as power output, voltage and current parameters, and temperature status of each PCS unit.

[0114] If the load trend shows an upward trend, the system will focus on collecting the maximum output capacity data and thermal management status information of each unit to form a current sequence containing power data at 24 time points.

[0115] The preset condition is that at least 70% of the PCS units are in a normal or better performance state. When the performance state set meets this condition, the system initiates the output ratio calculation process. The membership function adopts a trapezoidal distribution pattern, mapping the current output capacity of each unit to the 0-1 interval.

[0116] The fuzzy inference rules are formulated based on expert experience and historical operational data. The rule base includes logical judgments such as "if a unit has excellent performance and a high output ratio, then it should be assigned a higher weight."

[0117] When the performance status of PCS-01 unit is excellent and the output ratio is 0.9, the initial weight is set to 0.25; when the performance of PCS-02 unit is normal and the output ratio is 0.7, the initial weight is 0.18. By traversing all units, a complete set of initial load allocation weights is formed.

[0118] Normalization ensures the rationality and operability of weight allocation. The normalization process adjusts all weight values ​​to a standard form where the sum is 1. This method guarantees the accuracy of load allocation, avoids system operation deviations caused by weight imbalance, and provides a reliable decision-making basis for subsequent real-time scheduling and control.

[0119] Step S103 includes: acquiring a power grid demand fluctuation signal, which is generated through a preset weight allocation method; determining the signal amplitude based on the power grid demand fluctuation signal; if the signal amplitude exceeds a preset threshold, acquiring the load status of each PCS unit; determining the current load value of each unit based on the load status of the PCS unit; performing cluster analysis on the current load value using the K-nearest neighbor algorithm to obtain load clustering results; and determining a weight adjustment factor based on the load distribution characteristics in the load clustering results, wherein the weight adjustment factor is calculated using the following formula.

[0120]

[0121] α c N represents the weight adjustment factor for the c-th cluster. c N represents the number of load samples in the c-th cluster. total σ represents the total number of load samples. c Let σ represent the load variance of the c-th cluster. - Let β represent the average load variance of all clusters, and let β represent the adjustment coefficient. The preset weights are adjusted using the weight adjustment factor to obtain optimized weights. The optimized weights are then used to redistribute the load across each PCS unit to obtain an optimized allocation scheme. The calculation formula for load redistribution based on the optimized weights is as follows:

[0122]

[0123] P opt，i P represents the optimized load allocation value for the i-th PCS unit. preset，i γ represents the preset load value of the i-th PCS unit. i P represents the weight adjustment parameter corresponding to the i-th PCS unit. total This represents the total system load demand, and n represents the total number of PCS units.

[0124] The implementation of optimized load allocation schemes also needs to consider the dynamic response characteristics of the power grid. The dispatch center can ensure that the adjusted load output remains synchronized with the fluctuations in grid demand by monitoring the response speed of each PCS unit in real time. If a PCS unit receives a 210MW output command, it must complete the adjustment within 10 seconds; otherwise, it may lead to local overload of the power grid.

[0125] During the adjustment process, the hot reserve capacity of each unit should also be monitored. If the hot reserve capacity of a unit is less than 20MW, its weight should be appropriately reduced during allocation to avoid overload operation. The hot reserve status of each unit can be predicted through historical data analysis.

[0126] During off-peak hours at night, the amplitude of grid demand fluctuations may drop to 20MW, below the preset threshold of 30MW. In this case, there is no need to trigger load redistribution; instead, the original weights are maintained to reduce the frequency of system adjustments. This dynamic judgment mechanism effectively balances system efficiency and stability. In a real-world scenario, after adjusting load distribution using this method during peak hours, a power grid achieved a more balanced output distribution across PCS units, resulting in a significant improvement in overall operating efficiency.

[0127] The weighting can also be fine-tuned based on the regional characteristics of the power grid. For example, PCS units closer to the load center can be assigned a higher weight to reduce transmission losses.

[0128] Optimized allocation schemes can also be combined with the integration of new energy sources. For example, when wind power output suddenly increases, the weights of PCS units can be adjusted to prioritize the absorption of renewable energy, thereby optimizing the overall energy structure. This comprehensive adjustment can effectively support the green transformation of the power grid.

[0129] Step S104 includes: obtaining an initial priority data set, the initial priority data set containing multiple priority parameters; processing the initial priority data set using a genetic algorithm to obtain a first rule set; generating an adaptive priority sorting rule based on the first rule set; obtaining real-time state change data, the real-time state change data containing current system state information; determining the coverage degree of the adaptive priority sorting rule on the real-time state change data; if the adaptive priority sorting rule completely covers the real-time state change data, then determining the first rule set as a second rule set; if the adaptive priority sorting rule does not completely cover the real-time state change data, then adjusting the first rule set using a genetic algorithm to obtain a second rule set; monitoring the state change process according to the second rule set to obtain a change coverage sequence; and performing sorting calculations on the sequence elements in the change coverage sequence to obtain a dynamic sorting list.

[0130] The construction of the initial priority dataset requires comprehensive consideration of the key operating parameters of each PCS unit in the energy storage system. These priority parameters include dimensions such as unit capacity utilization, response speed, historical reliability indicators, and current health status. For example, an energy storage power station contains 8 PCS units, where unit A has a capacity utilization of 85%, a response time of 50 milliseconds, and a reliability score of 9.2, while unit B has corresponding values ​​of 78%, 45 milliseconds, and 8.8.

[0131] Genetic algorithms use a fitness function to evaluate the merits of different parameter combinations when processing initial priority data. Parameters are iteratively optimized through selection, crossover, and mutation operations, generating a first set of rules after multiple generations of evolution. For example, after multiple generations of evolution, the system determines that the combination with a capacity utilization weight of 0.4, a response speed weight of 0.3, and a reliability weight of 0.3 has the optimal fitness value, thus forming the basic adaptive priority ranking rule.

[0132] Acquiring real-time status change data involves continuous monitoring of the energy storage system's operating environment. This includes key information such as grid frequency fluctuations, load demand changes, temperature status of each PCS unit, and battery state of charge.

[0133] When the power grid frequency fluctuates from 50.0Hz to 49.8Hz, the system needs to respond quickly and adjust the priority order of each unit. At this time, the real-time data display unit C experiences a decrease in response capability due to excessive temperature, and the original rule set may not be able to fully cover this new state combination. The coverage degree is determined by matching degree calculation. When the matching degree is lower than the preset 90% threshold, it indicates that there is a coverage blind spot in the existing rule set.

[0134] The original rule set was designed for operation within the normal temperature range. However, when the ambient temperature exceeds 35 degrees Celsius, the performance characteristics of some PCS units change significantly. In this case, the first rule set needs to be adjusted and optimized.

[0135] The change coverage sequence records the response of each key node during the state change process.

[0136] The sequence elements include information such as timestamps, state change magnitudes, response delays, and processing results. By comprehensively analyzing these sequence elements, the system can identify the actual performance differences of different PCS units under various operating conditions, thereby generating a more accurate dynamic sorting list to ensure the optimal load allocation strategy in complex and ever-changing power grid environments.

[0137] Step S105 includes: obtaining a set of high-priority PCS units from a dynamic sorting list, the set of high-priority PCS units containing multiple power conversion units to be scheduled; obtaining a corresponding equipment loss index for each power conversion unit in the set of high-priority PCS units, the equipment loss index reflecting the operating loss status of each power conversion unit; calculating a loss ranking value for each power conversion unit based on the equipment loss index, the loss ranking value being used to quantify the loss degree of each unit; if the loss ranking value is lower than a preset threshold, then the corresponding power conversion unit is judged as a low-loss PCS unit; determining an output benchmark value for the low-loss PCS unit, the output benchmark value serving as a reference standard for power adjustment; and using the K-nearest neighbor algorithm to calculate the output increase ratio based on the output benchmark value, the output increase ratio being determined based on historical operating data and the current operating status.

[0138] The high-priority PCS unit set is obtained based on the top-ranked units in the aforementioned dynamic sorting list.

[0139] The equipment loss indicators for each power conversion unit cover three dimensions: switching loss, conduction loss, and auxiliary equipment power consumption. The operating loss status of each unit is quantified by real-time monitoring of the current, voltage, and temperature parameters.

[0140] An example of equipment loss indicators is as follows: A 2MW PCS unit, operating at full load, has a switching loss of 15kW, a conduction loss of 8kW, and auxiliary equipment power consumption of 3kW, resulting in a comprehensive loss index of 26kW. The loss ranking value is calculated by normalizing the comprehensive loss index of each unit, obtaining a value between 0 and 1 for horizontal comparison. When the loss ranking value is below a preset threshold of 0.3, the corresponding unit is identified as a low-loss PCS unit, which is capable of undertaking more power output tasks.

[0141] The output baseline value is determined by taking into account the rated power and current load rate of the low-loss PCS unit. For example, the output baseline value of a low-loss unit with a rated power of 2MW and a current load rate of 60% is set to 1.2MW.

[0142] When calculating the output increase ratio, the K-Nearest Neighbors algorithm retrieves historical data samples most similar to the current operating state and determines a reasonable increase ratio by analyzing successful adjustment experiences in these samples. The algorithm selects the five historical operating scenarios that are closest to the current temperature, load, and grid frequency. The average output increase ratio in these scenarios is 15%, so 15% is used as the increase ratio in the current adjustment.

[0143] This decision-making approach based on historical data can avoid the risk of equipment overload caused by excessive adjustments, while ensuring a steady increase in power output.

[0144] The optimized power output configuration, by rationally allocating the output tasks of each unit, improves the overall system efficiency and extends the equipment life, providing strong support for the economical operation of energy storage power stations.

[0145] Step S106 includes: acquiring output ratio data of multiple processing units; fusing resource utilization data of each processing unit with the output ratio data to obtain fused utilization data; comparing the fused utilization data with a preset threshold; if the fused utilization data is lower than the preset threshold, acquiring idle unit power information of each processing unit; calculating power parameters to be allocated based on the idle unit power information; redistributing power to each processing unit using the power parameters to be allocated to obtain the allocated power configuration; and calculating the overall operating efficiency of each processing unit based on the allocated power configuration to obtain efficiency improvement configuration data.

[0146] Output ratio data reflects the percentage of current power output of each PCS unit relative to its rated capacity. A certain energy storage power station contains 10 PCS units, each with a rated power of 500kW. The current actual outputs are 300kW, 450kW, 200kW, etc., corresponding to output ratio data of 60%, 90%, and 40%.

[0147] Resource utilization data takes into account multiple dimensions such as battery capacity usage, inverter load rate, and heat dissipation system operation status of each unit.

[0148] The integrated utilization rate data is calculated using a weighted average method. When the integrated utilization rate of a certain energy storage station is 65%, while the preset threshold is set at 75%, the system determines that there is room for resource allocation optimization. At this time, it is necessary to identify the power information of idle units, that is, the remaining available power capacity of each PCS unit.

[0149] The power parameters to be allocated are determined based on load forecasting and grid dispatching requirements. Assuming the system needs to increase total output by 800kW, this 800kW power is allocated to the three optimally positioned units in a 2:1:3 ratio by analyzing the technical status of each unit, battery state of charge, and temperature conditions. After allocation, the power configuration needs to verify whether each unit exceeds its safe operating range, ensuring that the unit load rate is controlled within 95%.

[0150] Overall operating efficiency calculations involve power conversion efficiency, battery charge / discharge efficiency, and system auxiliary power consumption. A certain PCS unit had an operating efficiency of 92.5% before adjustment. After power reallocation, its operating point was closer to the optimal efficiency range, and its operating efficiency improved to 94.2%. The efficiency improvement configuration data includes the efficiency improvement value for each unit, the power allocation scheme, and expected return metrics.

[0151] Step S107 includes: obtaining environmental variables based on efficiency configuration, the environmental variables containing system operating status information; performing operational analysis using the environmental variables to obtain a preliminary parameter set; refining the preliminary parameter set using a particle swarm optimization algorithm to obtain adaptive parameters; determining an optimized parameter set based on the adaptive parameters; extracting a burst load indication from the optimized parameter set; if the burst load indication exceeds a preset threshold, evaluating the parameter response capability to obtain a response judgment result; generating a power command based on the response judgment result; and outputting a final allocation command through the power command, the final allocation command being used to control system power allocation.

[0152] System operating status information covers key indicators such as processor temperature, memory usage, network latency, and power supply voltage fluctuations. For example, when the processor temperature reaches 75 degrees Celsius, the system will automatically record this temperature value as an important part of the environmental variables. At the same time, it will collect data such as memory usage of 68% and network response latency of 15 milliseconds. These data together constitute a complete set of environmental variables.

[0153] The runtime analysis process identifies the system's current operating mode and performance bottlenecks through in-depth analysis of environmental variables. When the temperature variable shows a continuous upward trend and the memory usage exceeds 70%, the analysis module determines that the system is under high load, thereby generating a preliminary parameter set containing frequency reduction suggestions, load balancing strategies, and other information. This analysis can effectively predict system performance change trends, providing an accurate data foundation for subsequent optimization.

[0154] Particle swarm optimization (PSO) plays a crucial role in parameter refinement by simulating the foraging behavior of bird flocks to find the optimal parameter combination. The algorithm sets the particle swarm size to 50 individuals, with each particle representing a parameter configuration scheme. Through an iterative optimization process, it gradually converges to the optimal parameter values.

[0155] Sudden load indication extraction is a crucial step in optimizing parameter set analysis. It determines whether the system is facing sudden stress by monitoring the load change rate and peak duration. For example, when the system load jumps from 40% to 85% within 5 seconds, the extraction module generates a sudden load indication value of 0.9, which reflects the severity and urgency of the load change.

[0156] The parameter response capability assessment quantifies the system's processing capacity under unexpected conditions. The assessment module checks factors such as currently available processing resources, backup power capacity, and heat dissipation capabilities, and calculates a response capability score based on these factors.

[0157] Power command generation is based on response assessment results, formulating specific power adjustment strategies. When the response capability score is 0.75 and the burst load indication exceeds the threshold, the system generates power commands including "increase the power of core processing units to 120% and reduce the power of non-critical modules to 60%" to ensure that critical tasks receive sufficient power support. The final command allocation output achieves a complete closed loop from analysis to execution. Through precise power control commands, each processing unit can quickly respond to load changes, maintaining the stability and efficiency of the overall system performance.

[0158] Step S108 includes: obtaining a power allocation instruction; extracting coordination index data based on the power allocation instruction; obtaining feedback data from the coordination index data; generating a cyclic data set using the feedback data; processing the cyclic data set using a data cleaning algorithm to obtain cleaned data; if the quality parameters of the cleaned data meet a preset threshold, determining the execution status of the cleaned data to obtain a preliminary status identifier; classifying the state using a support vector machine classification algorithm based on the preliminary status identifier to obtain a classification status result; extracting execution effect parameters from the classification status result and calculating stable parameter values ​​using the execution effect parameters; clustering the data using a K-nearest neighbor clustering algorithm based on the stable parameter values ​​to obtain a clustering result set; determining the output mode type from the clustering result set to obtain a preliminary output mode; if the deviation between the preliminary output mode and the standard output mode is less than a preset deviation threshold, determining the preliminary output mode as a stable output mode.

[0159] When the system receives a power allocation command, it needs to parse out the coordinated operation parameters of each power generation unit, including key indicators such as load sharing ratio, frequency regulation coefficient, and voltage stability margin. These coordinated indicator data reflect the degree of cooperation between the power generation units and the overall operational coordination.

[0160] The application of data cleaning algorithms ensures the reliability of data quality. For abnormal data caused by factors such as sensor noise, communication delays, and equipment failures in the power plant operating environment, the system employs cleaning methods based on statistical principles. For example, when power data fluctuations exceed 15% of the normal range, the system initiates an outlier identification program, correcting the data through neighboring data interpolation or historical trend fitting.

[0161] During the status assessment process, the system evaluates the response capability of each power generation unit based on the cleaned operational data. The initial status identification includes multiple status types such as normal operation, light load operation, heavy load operation, and regulation.

[0162] Support Vector Machine (SVM) classification algorithms establish a multi-dimensional feature space, mapping complex operating states to identifiable classification boundaries to achieve accurate state recognition.

[0163] The calculation of performance parameters involves a comprehensive evaluation across multiple dimensions. The system analyzes key indicators such as power point tracking accuracy, response time, and energy efficiency ratio to calculate comprehensive parameter values ​​reflecting system stability. When stable parameter values ​​remain above 0.85, it indicates that the system is operating well and possesses the fundamental conditions for further optimization.

[0164] The determination of the output mode type directly affects the formulation of subsequent control strategies. The system assesses the stability of the current operating state by comparing the deviation between the initial output mode and the standard output mode. When the deviation is controlled within 3%, the system confirms that a stable output mode has been reached, providing a reliable guarantee for continuous optimization.

Claims

1. A smart control method for power allocation and efficiency optimization in an energy storage power station, characterized in that, Includes the following steps: S101 acquires real-time operating data and historical load records of each PCS unit in the energy storage power station, and obtains the current load change trend through data fusion processing; Based on the current load change trend, S102 uses a fuzzy logic algorithm to analyze the performance status and output ratio of the PCS unit and determine the initial load allocation weight of each unit. S103 obtains the power grid demand fluctuation signal through the initial load allocation weight. If the fluctuation signal exceeds the preset threshold, the weight is adjusted to balance the PCS unit load and obtain the optimized allocation scheme. S104 uses a genetic algorithm to generate adaptive priority sorting rules for the optimized allocation scheme, determines whether the rules cover real-time state changes, and obtains a dynamic sorting list. S105 extracts high-priority PCS units from the dynamic sorting list, obtains their equipment loss indicators, and determines the output increase ratio of low-loss units through indicator comparison processing. S106 integrates resource utilization data based on the output increase ratio. If the utilization rate is lower than the preset threshold, the idle unit power is reallocated to obtain an overall efficiency improvement configuration. S107 improves overall efficiency by analyzing power plant operating environment variables, using particle swarm optimization algorithm to refine adaptive adjustment parameters, determining whether parameters respond to sudden loads, and obtaining the final power allocation command. S108 extracts coordinated operation indicators from the final power allocation command, obtains feedback loop data, determines the command execution effect through data verification processing, and obtains the stable output mode of the power plant.

2. The intelligent control method for power allocation and efficiency optimization of an energy storage power station as described in claim 1, characterized in that, Step S101 includes: Acquire real-time operating data and historical load records of the PCS unit, and process the real-time operating data and historical load records using a time series alignment method to obtain an aligned data sequence; By applying the Long Short-Term Memory network algorithm to the aligned data sequence, a preliminary load prediction sequence is obtained; If the variance of the preliminary load prediction sequence exceeds a preset threshold, a sliding window filter is used to process the preliminary load prediction sequence to obtain a smoothed load sequence. The first-order difference is calculated based on the smoothed load sequence to obtain the differential load sequence; By applying the support vector machine algorithm to the differential load sequence, the load change classification label is obtained; The trend direction is determined based on the load change classification labels; The load change trend is correlated with the real-time operating data of the PCS unit to obtain the fused load trend.

3. The intelligent control method for power allocation and efficiency optimization of an energy storage power station as described in claim 1, characterized in that, Step S102 includes: Obtain the current data sequence, which is determined based on the load trend; The performance status of the current data sequence is determined by fuzzy logic to obtain a performance status set. The formula for determining the performance status of the data sequence by fuzzy logic is as follows; P i Let m represent the fuzzy judgment result of the i-th performance state, and w represent the total number of fuzzy rules. j μ represents the weight of the j-th rule. ij (x) represents the membership value of input x to the i-th performance state in the j-th rule; If the performance state set meets the preset conditions, the output ratio is calculated through the membership function to obtain the output ratio set. The formula for calculating the output ratio through the membership function is as follows; R k =∫1 0 μ k (y)·ydy / ∫0 1 μ k (y)to R k μ represents the k-th output ratio value. k (y) represents the membership function of the k-th output state, and y represents the range of values ​​for the output ratio. Based on the power output ratio set, a preliminary load allocation weight set is determined using fuzzy inference rules.

4. The intelligent control method for power allocation and efficiency optimization of an energy storage power station as described in claim 1, characterized in that, Step S103 includes: Acquire grid demand fluctuation signals, which are generated through a preset weight allocation method; The signal amplitude is determined based on the power grid demand fluctuation signal; If the signal amplitude exceeds a preset threshold, the load status of each PCS unit is obtained; The current load value of each unit is determined based on the load status of the PCS unit. The K-nearest neighbor algorithm is used to perform cluster analysis on the current load value to obtain the load clustering results; The weight adjustment factor is determined based on the load distribution characteristics in the load clustering results. The formula for calculating the weight adjustment factor is as follows: α c N represents the weight adjustment factor for the c-th cluster. c N represents the number of load samples in the c-th cluster. total σ represents the total number of load samples. c This represents the load variance of the c-th cluster. The average loading variance of all clusters is represented by β, which is the adjustment factor. The preset weights are adjusted and calculated using the weight adjustment factor to obtain the optimized weights; The optimized weights are used to redistribute the load of each PCS unit to obtain an optimized allocation scheme; The calculation formula for load redistribution based on optimized weights is as follows: P opt，i P represents the optimized load allocation value for the i-th PCS unit. preset，i γ represents the preset load value of the i-th PCS unit. i P represents the weight adjustment parameter corresponding to the i-th PCS unit. total This represents the total system load demand, and n represents the total number of PCS units.

5. The intelligent control method for power allocation and efficiency optimization of an energy storage power station as described in claim 1, characterized in that, Step S104 includes: Obtain an initial priority data set, which contains multiple priority parameters; The initial priority data set is processed using a genetic algorithm to obtain the first rule set; Generate adaptive priority sorting rules based on the first rule set; Acquire real-time status change data, which includes current system status information; Determine the extent to which the adaptive priority sorting rule covers the real-time state change data; If the adaptive priority sorting rule completely covers the real-time state change data, then the first rule set is determined to be the second rule set; If the adaptive priority sorting rule does not completely cover the real-time state change data, a genetic algorithm is used to adjust the first rule set to obtain a second rule set. Based on the second rule set, the state change process is monitored to obtain the change coverage sequence; A dynamic sorted list is obtained by sorting the sequence elements in the changed cover sequence.

6. The intelligent control method for power allocation and efficiency optimization of an energy storage power station as described in claim 1, characterized in that, Step S105 includes: Obtain a set of high-priority PCS units from a dynamic sorting list, the set of high-priority PCS units containing multiple power conversion units to be scheduled; For each power conversion unit in the high-priority PCS unit set, obtain the corresponding equipment loss index, which reflects the operating loss status of each power conversion unit. The loss ranking value corresponding to each power conversion unit is calculated based on the equipment loss index. The loss ranking value is used to quantify the degree of loss of each unit. If the loss ranking value is lower than a preset threshold, the corresponding power conversion unit is judged as a low-loss PCS unit. A power output reference value is determined for the low-loss PCS unit, and the power output reference value serves as a reference standard for power adjustment. The K-nearest neighbor algorithm is used to calculate the output increase ratio based on the output benchmark value. The output increase ratio is determined based on historical operating data and the current operating status.

7. The intelligent control method for power allocation and efficiency optimization of an energy storage power station as described in claim 1, characterized in that, Step S106 includes: The output ratio data of multiple processing units is obtained, and the resource utilization rate data of each processing unit is fused based on the output ratio data to obtain the fused utilization rate data. The fusion utilization rate data is compared with a preset threshold. If the fusion utilization rate data is lower than the preset threshold, the power information of the idle units in each processing unit is obtained. Calculate the power parameters to be allocated based on the power information of the idle units, and redistribute the power of each processing unit using the power parameters to be allocated to obtain the power configuration after allocation; The overall operating efficiency of each processing unit is calculated based on the allocated power configuration to obtain efficiency improvement configuration data.

8. The intelligent control method for power allocation and efficiency optimization of an energy storage power station as described in claim 1, characterized in that, Step S107 includes: Environment variables are obtained based on efficiency configuration, and these environment variables include system operating status information; By performing operational analysis using the aforementioned environmental variables, a preliminary parameter set is obtained; The particle swarm optimization algorithm is used to refine the initial parameter set to obtain adaptive parameters. Determine the set of optimization parameters based on the adaptive parameters; Extract burst load indicators from the set of optimized parameters; If the burst load indication exceeds a preset threshold, the parameter response capability is evaluated to obtain a response judgment result; A power command is generated based on the response judgment result; The power command outputs a final allocation command, which is used to control the system's power allocation.

9. The intelligent control method for power allocation and efficiency optimization of an energy storage power station as described in claim 1, characterized in that, Step S108 includes: Obtain power allocation instructions and extract coordination index data based on the power allocation instructions; Feedback data is obtained from the collaborative indicator data, and a cyclic data set is generated from the feedback data; The cyclic data set is processed using a data cleaning algorithm to obtain cleaned data; If the quality parameters of the cleaned data meet the preset threshold, the execution status of the cleaned data is judged to obtain a preliminary status identifier. Based on the preliminary state identifier, a support vector machine classification algorithm is used to classify the state and obtain the classification state result; Extract execution effect parameters from the classification status results, and calculate stable parameter values ​​using the execution effect parameters; Based on the stable parameter values, the K-nearest neighbor clustering algorithm is used to cluster the data, and a set of clustering results is obtained. The output pattern type is determined from the clustering result set to obtain the preliminary output pattern; If the deviation between the initial output mode and the standard output mode is less than a preset deviation threshold, then the initial output mode is determined to be a stable output mode.

Images