Intelligent control method and device for photovoltaic energy storage power utilization in industrial park

Abstract

The invention belongs to the technical field of power control, and discloses an intelligent control method and device for photovoltaic energy storage power utilization in an industrial park, which predicts the total reactive power demand of power electronic equipment by combining the historical operation data of the equipment, the real-time environment temperature and the real-time humidity, and performs dynamic reactive power compensation according to the total reactive power demand. Meanwhile, a photovoltaic inverter and an energy storage converter are cooperated to carry out power grid connection point voltage stability control, and a photovoltaic energy storage system cooling system is controlled according to the real-time environment temperature and the real-time humidity; the method can effectively adapt to high-temperature and high-humidity environments and effectively cope with reactive power dynamic changes.

Description

Technical Field

[0001] The present application relates to the technical field of power control, and more particularly, to an intelligent control method and device for photovoltaic energy storage power consumption in industrial parks. Background Art

[0002] Subtropical coastal industrial parks widely deploy photovoltaic energy storage systems to reduce electricity costs. However, in this region, there is a common problem of continuous high temperature and high humidity in summer, which poses a severe challenge to the stable operation of photovoltaic energy storage systems. High temperature environments significantly reduce the power generation efficiency of photovoltaic modules, while high humidity environments exacerbate the energy loss during the charge and discharge processes of energy storage batteries. At the same time, industrial inductive loads deployed within industrial parks and power electronic devices in photovoltaic energy storage systems operate under the combined action of high temperature and high humidity, resulting in a significant increase in their reactive power requirements. To ensure the reliability of power supply in industrial parks, an intelligent control system is urgently needed to achieve effective reactive power compensation and voltage stability control at the grid connection point.

[0003] However, traditional reactive power compensation methods, such as fixed capacitors or traditional reactive compensators, are difficult to effectively adapt to the rapid fluctuations and non-linear characteristics of reactive power demand in subtropical coastal industrial parks under high temperature and high humidity environments. Static compensation strategies cannot dynamically adjust the compensation amount according to the real-time changes in environmental temperature and humidity, easily resulting in under-compensation or over-compensation, which seriously affects the stability of the grid voltage and the overall operating efficiency of the system. Furthermore, high temperature and high humidity environments also accelerate the performance degradation of power electronic devices and energy storage devices, further exacerbating the complexity and uncertainty of reactive power demand. Therefore, there is an urgent need for an intelligent control method that can fully adapt to high temperature and high humidity environments and effectively respond to dynamic changes in reactive power to ensure the stable operation and efficient power consumption of photovoltaic energy storage systems in subtropical coastal industrial parks.

[0004] In view of the above problems, the existing technology urgently needs to be improved. Summary of the Invention

[0005] The purpose of the present application is to provide an intelligent control method and device for photovoltaic energy storage power consumption in industrial parks, which can effectively adapt to high temperature and high humidity environments and effectively respond to dynamic changes in reactive power.

[0006] In a first aspect, the present application provides an intelligent control method for photovoltaic energy storage power consumption in industrial parks, which is used for power consumption control of photovoltaic energy storage systems in subtropical coastal industrial parks. The method includes: S1. Obtain the real-time environmental temperature and real-time humidity of the deployment area of the photovoltaic energy storage system; S2. According to the historical operation data of the equipment, the real-time environmental temperature, and the real-time humidity, use a combined prediction model to predict the total reactive power demand of the industrial park and the power electronic devices of the photovoltaic energy storage system; S3. Dynamically adjust the output of the reactive power compensation device according to the prediction result of the combined prediction model to achieve reactive power compensation; S4. Adjust the operating modes of the photovoltaic inverter and the energy storage converter, and coordinately control the output of active power and reactive power to maintain the voltage stability of the grid connection point; S5. Optimize the operation strategy of the cooling system of the photovoltaic energy storage system according to the real-time ambient temperature and real-time humidity to reduce the operating temperature of the equipment.

[0007] This method combines the historical operation data of the equipment, the real-time ambient temperature and the real-time humidity to predict the total reactive power demand of the power electronic equipment, and accordingly performs dynamic reactive power compensation. At the same time, it coordinately controls the voltage stability of the grid connection point with the photovoltaic inverter and the energy storage converter, and controls the cooling system of the photovoltaic energy storage system according to the real-time ambient temperature and the real-time humidity; it can effectively adapt to high-temperature and high-humidity environments and effectively respond to dynamic changes in reactive power.

[0008] Preferably, the combined prediction model includes a linear regression model and a non-linear neural network model; the input of the linear regression model includes the active power, ambient temperature and humidity in the historical operation data of the equipment, and the output is the first reactive power prediction value; the input of the non-linear neural network model includes the active power, ambient temperature and humidity in the historical operation data, and the output is the second reactive power prediction value; Step S2 includes: S201. Obtain the first reactive power prediction value by using the linear regression model and the second reactive power prediction value by using the non-linear neural network model according to the active power, ambient temperature and humidity in the historical operation data of the equipment; S202. Calculate the first weight of the linear regression model and the second weight of the non-linear neural network model by using the dynamic weight allocation method according to the historical prediction errors of the linear regression model and the non-linear neural network model, and the sum of the first weight and the second weight is 1; S203. Calculate the initial reactive power prediction value through weighted average according to the first reactive power prediction value, the second reactive power prediction value, the first weight and the second weight; S204. Non-linearly correct the initial reactive power prediction value by using the Sigmoid function according to the real-time ambient temperature and the real-time humidity to obtain the final reactive power prediction value.

[0009] In the above way, the prediction accuracy of reactive power can be effectively improved.

[0010] Preferably, step S202 includes: A1. Perform a sliding window process on the historical prediction error to construct an error sequence, and decompose the error sequence by using the wavelet decomposition algorithm to obtain multiple subsequences with different frequencies; A2. For each subsequence, calculate the autocorrelation coefficient. According to the magnitude of the autocorrelation coefficient, use the gradient descent algorithm with an adaptive step size to calculate the first weight of the linear regression model and the second weight of the non - linear neural network model. The sum of the first weight and the second weight is 1.

[0011] Thus, the step size of weight adjustment can be dynamically adjusted according to the stationarity of the error, improving the efficiency and accuracy of weight adjustment.

[0012] Preferably, step A2 includes: B1. For each subsequence, calculate the autocorrelation coefficient. If the autocorrelation coefficient is greater than a preset threshold, determine that the corresponding subsequence is a stationary sequence and execute step B2; otherwise, determine that the corresponding subsequence is a non - stationary sequence and execute step B3; B2. For the stationary sequence, use the gradient descent algorithm with a fixed step size to calculate the first weight of the linear regression model and the second weight of the non - linear neural network model, denoted as the first weight optimal value and the second weight optimal value respectively. The fixed step size is the preset step size; B3. For the non - stationary sequence, use the gradient descent algorithm with a variable step size to calculate the first weight of the linear regression model and the second weight of the non - linear neural network model, denoted as the first weight optimal value and the second weight optimal value respectively. The formula for the variable step size is: st_i = st_max * exp(-|r_i|), where st_i is the step size of the i - th subsequence, st_max is the preset maximum step size, and r_i is the autocorrelation coefficient of the i - th subsequence; B4. Perform weighted averaging on the first weight optimal values and the second weight optimal values calculated for all subsequences to obtain the final first weight of the linear regression model and the final second weight of the non - linear neural network model. The sum of the final first weight and the final second weight is 1.

[0013] Preferably, step S3 includes: S301. Real - time collect the electricity consumption data of various types of loads in the industrial park, and identify the first type of loads with a fast response speed to reactive power demand and the second type of loads with a slow response speed; S302. According to the historical electricity consumption data ratio of the first type of loads and the second type of loads, decompose the total reactive power demand into the predicted value of the reactive power demand of the first type of loads and the predicted value of the reactive power demand of the second type of loads; S303. For the predicted value of the reactive power demand of the first type of loads, adopt a feed - forward control method and directly adjust the output of the reactive power compensation device according to the predicted value to achieve fast compensation; S304. For the predicted reactive power demand value of the second type of load, adopt a feedback control method to monitor the grid voltage in real time. According to the grid voltage deviation and the predicted reactive power demand value of the second type of load, adjust the output of the reactive power compensation device through a PID controller to achieve precise compensation.

[0014] Preferably, step S301 includes: Collect the electricity consumption data of various types of loads in the industrial park in real time to construct an electricity consumption dataset; the electricity consumption data includes active power, reactive power, voltage, and current. Preprocess the electricity consumption dataset to obtain the preprocessed electricity consumption dataset; the preprocessing includes missing value filling, outlier detection and elimination, and moving average filtering. Extract the time-domain features and frequency-domain features of the preprocessed electricity consumption dataset. The time-domain features include mean, variance, peak value, and kurtosis, and the frequency-domain feature includes total harmonic distortion rate to obtain a load feature vector. Use the K-means clustering algorithm to divide the loads into the first type of load and the second type of load according to the load feature vector.

[0015] Preferably, step S4 includes: S401. Monitor the grid connection point voltage in real time, calculate the voltage deviation between the grid connection point voltage and the preset voltage, and obtain the current active power output values of the photovoltaic inverter and the energy storage converter. S402. According to the voltage deviation and the current active power output values, use the droop control algorithm to calculate the reactive power regulation commands for the photovoltaic inverter and the energy storage converter. Among them, the droop coefficient in the droop control algorithm is adjusted according to the real-time ambient temperature and real-time humidity. S403. According to the reactive power regulation commands, adjust the operating modes of the photovoltaic inverter and the energy storage converter to jointly control the output of active power and reactive power. Among them, the photovoltaic inverter preferentially regulates reactive power. When the reactive power of the photovoltaic inverter reaches the upper limit, the energy storage converter continues to regulate reactive power. S404. Monitor the operating states of the photovoltaic inverter and the energy storage converter in real time. If equipment overload or failure is detected, start the standby inverter or energy storage converter and adjust the operating parameters to ensure the stability of the grid connection point voltage.

[0016] Preferably, step S402 includes: Calculate the temperature correction coefficient and humidity correction coefficient through a fuzzy algorithm according to the real-time ambient temperature and real-time humidity. Calculate the comprehensive environment correction coefficient according to the temperature correction coefficient and humidity correction coefficient. Adjust the droop coefficient of the droop control algorithm according to the comprehensive environment correction coefficient. Based on the voltage deviation and the current active power output value, using the adjusted droop coefficient, calculate the reactive power regulation commands for the photovoltaic inverter and the energy storage converter based on the droop control algorithm.

[0017] Preferably, step S5 includes: S501. Monitor the temperatures of the photovoltaic energy storage system inverter, the energy storage battery, and the converter in real time, and construct a device temperature data set; S502. According to the real-time ambient temperature, real-time humidity, and the device temperature data set, calculate the cooling demand level using the fuzzy control algorithm; S503. According to the cooling demand level and the current operating state of the cooling system, adjust the operating parameters of the cooling system using the hierarchical control strategy.

[0018] In a second aspect, the present application provides an intelligent control device for photovoltaic energy storage power consumption in an industrial park, which is used for the power consumption control of the photovoltaic energy storage system in a subtropical coastal industrial park. The device includes: An environmental parameter monitoring module, which is used to obtain the real-time ambient temperature and real-time humidity of the deployment area of the photovoltaic energy storage system; A reactive power demand prediction module, which is used to predict the total reactive power demand of the industrial park and the power electronic devices of the photovoltaic energy storage system using a combined prediction model based on the historical operation data of the devices, real-time ambient temperature, and real-time humidity; A dynamic reactive power compensation control module, which is used to dynamically adjust the output of the reactive power compensation device according to the prediction result of the combined prediction model to achieve reactive power compensation; A voltage stability control module, which is used to adjust the working modes of the photovoltaic inverter and the energy storage converter, and cooperate to control the output of active power and reactive power to maintain the voltage stability at the grid connection point; A device cooling optimization module, which is used to optimize the operation strategy of the cooling system of the photovoltaic energy storage system according to the real-time ambient temperature and real-time humidity to reduce the device operating temperature.

[0019] Beneficial effects: An intelligent control method and device for photovoltaic energy storage power consumption in an industrial park provided by the present application predict the total reactive power demand of power electronic devices by combining the historical operation data of the devices, real-time ambient temperature, and real-time humidity, and perform dynamic reactive power compensation accordingly. At the same time, it cooperates with the photovoltaic inverter and the energy storage converter to control the voltage stability at the grid connection point, and controls the cooling system of the photovoltaic energy storage system according to the real-time ambient temperature and real-time humidity; it can effectively adapt to high-temperature and high-humidity environments and effectively respond to dynamic changes in reactive power. Description of the Drawings

[0020] Figure 1 It is a flowchart of the intelligent control method for photovoltaic energy storage power consumption in an industrial park provided by an embodiment of the present application.

[0021] Figure 2 This is a schematic structural diagram of the intelligent control device for photovoltaic energy storage power consumption in the industrial park provided by the embodiment of the present application.

[0022] Label description: 1. Environmental parameter monitoring module; 2. Reactive power demand prediction module; 3. Dynamic reactive power compensation control module; 4. Voltage stability control module; 5. Equipment cooling optimization module. Specific implementation manners

[0023] Next, the technical solutions in the present application will be clearly and completely described in conjunction with the accompanying drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Usually, the components of the present application described and shown in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application to be protected, but only represents the selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative efforts belong to the scope of protection of the present application.

[0024] It should be noted that: similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present application, terms such as "first" and "second" are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0025] Referring to Figure 1 , the present application proposes an intelligent control method for photovoltaic energy storage power consumption in the industrial park, which is used for the power consumption control of the photovoltaic energy storage system in the subtropical coastal industrial park. The method includes: S1. Obtain the real-time environmental temperature and real-time humidity of the deployment area of the photovoltaic energy storage system; S2. According to the historical operation data of the equipment, the real-time environmental temperature and the real-time humidity, use a combined prediction model to predict the total reactive power demand of the power electronic equipment in the industrial park and the photovoltaic energy storage system; S3. According to the prediction result of the combined prediction model, dynamically adjust the output of the reactive power compensation device to achieve reactive power compensation; S4. Adjust the working modes of the photovoltaic inverter and the energy storage converter, and jointly control the output of the active power and the reactive power to maintain the voltage stability of the grid connection point; S5. According to the real-time environmental temperature and the real-time humidity, optimize the operation strategy of the cooling system of the photovoltaic energy storage system to reduce the equipment operation temperature.

[0026] Among them, in step S1, the ambient temperature and humidity data can be monitored and collected in real time through the ambient temperature sensors and humidity sensors deployed in the photovoltaic energy storage system area, serving as the basic data for subsequent reactive power prediction and control.

[0027] Among them, in step S2, the combined prediction model can fuse the linear regression model and the non - linear neural network model. The linear regression model analyzes the linear relationship between the historical active power data of the load and the ambient temperature and humidity, and preliminarily predicts the reactive power. The non - linear neural network model captures the complex non - linear relationship between the ambient temperature and humidity and the total reactive power demand, further improving the prediction accuracy. The prediction results of the two models can be fused through the dynamic weight allocation method to achieve a more accurate prediction of the total reactive power demand.

[0028] Among them, in step S3, the reactive power compensation device is driven according to the predicted value to dynamically adjust the compensation amount. The output of the reactive power compensation device can quickly respond to the change of the reactive power demand to achieve dynamic compensation.

[0029] Among them, in step S4, the voltage at the grid connection point is monitored in real time. When the voltage deviates, the operating modes of the photovoltaic inverter and the energy storage converter are adjusted simultaneously. The photovoltaic inverter and the energy storage converter work together to maintain the voltage at the grid connection point within the set range by adjusting the output of the active power and reactive power.

[0030] Among them, in step S5, the fuzzy control algorithm can be used to analyze the relationship between the environmental parameters and the equipment temperature, and calculate the cooling demand level. The cooling system, such as a fan or a liquid - cooling system, adjusts the operating parameters, such as the fan speed or the coolant flow rate, according to the cooling demand level to optimize the cooling effect and reduce the equipment operating temperature.

[0031] Specifically, the present application provides an intelligent control method for photovoltaic energy storage power consumption in industrial parks to solve the technical problems of reactive power compensation and voltage stability of photovoltaic energy storage systems in subtropical coastal industrial parks under changing ambient temperature and humidity. This method first obtains the real-time ambient temperature and real-time humidity of the deployment area of the photovoltaic energy storage system, providing a basis for environmental parameters for subsequent control. Then, according to the historical operation data of the equipment, real-time ambient temperature and real-time humidity, a combined prediction model is used to predict the total reactive power demand of the industrial park and the power electronic equipment of the photovoltaic energy storage system. The combined prediction model takes into account the influence of ambient temperature and humidity on the total reactive power demand, improving the prediction accuracy. After that, according to the prediction results of the combined prediction model, the output of the reactive power compensation device is dynamically adjusted to achieve reactive power compensation. The dynamic adjustment compensates according to the predicted demand, avoiding the deficiencies or over-compensation problems of traditional static compensation strategies. Further, the operating modes of the photovoltaic inverter and the energy storage converter are adjusted to jointly control the output of active power and reactive power, maintaining the voltage stability of the grid connection point. The joint control ensures the grid voltage stability and guarantees the reliability of power supply. Finally, according to the real-time ambient temperature and real-time humidity, the operating strategy of the cooling system of the photovoltaic energy storage system is optimized to reduce the operating temperature of the equipment. Optimizing the operating strategy of the cooling system to reduce the equipment temperature improves the system operating efficiency and equipment life. Thus, the method proposed in the present application can dynamically adjust the reactive power compensation and voltage control strategies according to the unique high-temperature and high-humidity environment of subtropical coastal industrial parks, while optimizing equipment cooling, ensuring the stable operation and efficient power consumption of the photovoltaic energy storage system.

[0032] In some possible implementation manners, the combined prediction model includes a linear regression model and a non-linear neural network model; the input of the linear regression model includes the active power, ambient temperature and humidity in the historical operation data of the equipment, and the output is the first reactive power prediction value; the input of the non-linear neural network model includes the active power, ambient temperature and humidity in the historical operation data, and the output is the second reactive power prediction value; Step S2 includes: S201. According to the active power, ambient temperature and humidity in the historical operation data of the equipment, use the linear regression model to obtain the first reactive power prediction value, and use the non-linear neural network model to obtain the second reactive power prediction value; S202. According to the historical prediction errors of the linear regression model and the non-linear neural network model, use the dynamic weight allocation method to calculate the first weight of the linear regression model and the second weight of the non-linear neural network model, and the sum of the first weight and the second weight is 1; S203. According to the first reactive power prediction value, the second reactive power prediction value, the first weight and the second weight, calculate the initial reactive power prediction value through weighted average; S204. According to the real-time ambient temperature and real-time humidity, use the Sigmoid function to perform non-linear correction on the initial reactive power prediction value to obtain the final reactive power prediction value.

[0033] Among them, step S201 refers to using two different types of prediction models, namely the linear regression model and the non-linear neural network model, to perform preliminary predictions on reactive power respectively. Specifically, multiple linear regression models can be used, such as simple linear regression, multiple linear regression, etc., to capture the linear relationship between the total reactive power demand and active power, ambient temperature, and humidity. The non-linear neural network model can use a backpropagation neural network, a convolutional neural network, or a recurrent neural network, etc., to learn and simulate the complex non-linear relationship between the total reactive power demand and input parameters. The device historical operation data includes but is not limited to historical active power data, historical ambient temperature data, and historical humidity data (the active power, ambient temperature, and humidity in the device historical operation data are the historical active power data, historical ambient temperature data, and historical humidity data).

[0034] Among them, step S202 refers to dynamically allocating weights based on the historical prediction errors of the two models. Specifically, the historical prediction error can be obtained by calculating the difference between the model prediction value and the actual value. The dynamic weight allocation method can be implemented using multiple algorithms, such as the gradient descent method, genetic algorithm, or particle swarm optimization algorithm, etc. The principle of weight allocation is that the model with a smaller error is given a higher weight, and the model with a larger error is given a lower weight. The limitation that the sum of the weights is 1 ensures that the weight allocation ratio of the two models is reasonable and easy to calculate.

[0035] Among them, step S203 refers to performing a weighted average on the prediction results of the two models. Specifically, by multiplying the first reactive power prediction value by the first weight, multiplying the second reactive power prediction value by the second weight, and adding the two, an initial reactive power prediction value that comprehensively considers the linear relationship and non-linear relationship is obtained. This way of weighted average can comprehensively utilize the advantages of the two models and reduce the risk of single-model prediction.

[0036] Among them, step S204 refers to using the Sigmoid function to perform non-linear correction on the initial reactive power prediction value. Specifically, due to its S-shaped characteristic, the Sigmoid function can map the input value to the range between 0 and 1, or other preset non-linear ranges, to achieve non-linear adjustment of the initial prediction value. The real-time ambient temperature and real-time humidity are used as the input of the Sigmoid function, so that the correction process can reflect the changes of environmental factors in real time. The specific form of the Sigmoid function can be selected and adjusted according to the actual application scenario, such as the standard Sigmoid function, hyperbolic tangent Sigmoid function, etc.

[0037] Specifically, for the problem of predicting the total reactive power demand in industrial parks, considering that the total reactive power demand is affected by multiple factors and exhibits a mixed characteristic of linearity and non-linearity, the combined prediction model is designed to include a linear regression model and a non-linear neural network model. The linear regression model can effectively capture the linear relationship between the total reactive power demand, active power, and ambient temperature and humidity, while the non-linear neural network model is good at learning and fitting complex non-linear relationships. By combining the two models, information in the data can be more comprehensively extracted, improving the adaptability and accuracy of the prediction model. In terms of the model fusion strategy, a dynamic weight allocation method is introduced. This method adaptively adjusts the weights of each model in the combined prediction according to their performance in historical predictions. Specifically, a sliding window mechanism can be adopted to regularly evaluate the errors of each model in recent predictions. The model with a smaller error will be assigned a higher weight, while the weight will be reduced otherwise. Thus, the combined prediction model can dynamically track the prediction performance of each model and achieve a better combined prediction effect. To further improve the prediction accuracy, especially when the ambient temperature and humidity have a non-linear impact on the total reactive power demand, the Sigmoid function is introduced to correct the prediction results. The Sigmoid function has an S-shaped curve characteristic, which can map the input value to the range between 0 and 1 and introduce a non-linear transformation. By taking the real-time ambient temperature and humidity as the input of the Sigmoid function, the initial prediction value can be non-linearly adjusted, making the final prediction result more in line with the actual situation. For example, in a high-temperature and high-humidity environment, the total reactive power demand may show an accelerating growth trend, and the Sigmoid function can simulate this non-linear growth characteristic and correct the prediction value upward.

[0038] In some preferred embodiments, step S202 includes: A1. Perform a sliding window process on the historical prediction errors to construct an error sequence, and decompose the error sequence using the wavelet decomposition algorithm to obtain multiple subsequences with different frequencies; A2. For each subsequence, calculate the autocorrelation coefficient. According to the magnitude of the autocorrelation coefficient, use the gradient descent algorithm with an adaptive step size to calculate the first weight of the linear regression model and the second weight of the non-linear neural network model. The sum of the first weight and the second weight is 1.

[0039] Among them, steps A1 and A2 are specific limitations on the dynamic weight allocation method. The dynamic weight allocation method aims to adaptively adjust the weights of the linear regression model and the non-linear neural network model in the combined prediction according to the performance of the model in historical predictions, so that the prediction model can better adapt to the changing needs. However, a simple dynamic weight allocation method may not be able to fully cope with the complexity of prediction errors. The errors may contain multiple frequency components. For example, some error components change rapidly and have a high frequency, while some error components change slowly and have a low frequency. To handle these complex errors more precisely, step S202 introduces the wavelet decomposition technique.

[0040] Among them, step A1 preprocesses the historical prediction errors. Specifically, first, a sliding window is used to process the historical prediction errors to construct an error sequence. The size of the sliding window can be adjusted according to actual needs. For example, the error data of the recent period can be selected to capture the changing trend of recent errors. The purpose of the sliding window processing is to organize the discrete historical prediction errors into a continuous sequence for subsequent wavelet decomposition processing. Then, the wavelet decomposition algorithm is used to decompose the error sequence to obtain multiple subsequences with different frequencies. Wavelet decomposition is a signal processing technique that can decompose a complex signal into multiple sub-signals with different frequencies, and each sub-signal represents the component of the original signal within a specific frequency range. For the prediction error sequence, wavelet decomposition can decompose the error into components with different frequencies. For example, the high-frequency components may reflect random noise or sudden disturbances, while the low-frequency components may reflect systematic biases or trend errors. There are various specific implementation algorithms for wavelet decomposition, such as Haar wavelet, Daubechies wavelet, Symlets wavelet, etc. The appropriate wavelet basis function and decomposition level can be selected according to the actual application scenario and the characteristics of the error sequence. Through wavelet decomposition, the complex error signal can be decomposed into multiple relatively simple subsequences, laying a foundation for subsequent differential processing of error components with different frequencies.

[0041] Among them, step A2 calculates weights for each frequency subsequence. Specifically, for each subsequence obtained in step A1, the autocorrelation coefficient is calculated. The autocorrelation coefficient is an index that measures the degree of self-correlation of a sequence and can reflect the correlation between the values at different time points in the sequence. For the prediction error subsequence, the autocorrelation coefficient can reflect the temporal dependence or regularity of the error of this frequency component. If the autocorrelation coefficient of a subsequence is high, it indicates that the error of this frequency has strong regularity and predictability; on the contrary, it indicates that the error of this frequency has strong randomness. There are various specific calculation methods for the autocorrelation coefficient. For example, the Pearson correlation coefficient or the Spearman rank correlation coefficient can be used. According to the magnitude of the calculated autocorrelation coefficient, the gradient descent algorithm with an adaptive step size is used to calculate the first weight of the linear regression model and the second weight of the nonlinear neural network model. The gradient descent algorithm is a commonly used optimization algorithm and can be used to solve the minimum value of a function. In this solution, the goal of the gradient descent algorithm is to minimize the prediction error of the combined prediction model. By continuously iteratively adjusting the weights of the linear regression model and the nonlinear neural network model, the prediction result of the combined prediction model is made closer to the true value. The adaptive step size means that during the iterative process of the gradient descent algorithm, the step size can be dynamically adjusted according to the characteristics of the error. For example, when the error has strong autocorrelation, the step size can be appropriately increased to accelerate the convergence speed; when the autocorrelation of the error is weak, the step size can be appropriately decreased to avoid oscillation or divergence. There are various specific implementation methods for the adaptive step size. For example, the step size can be dynamically adjusted according to factors such as the magnitude of the autocorrelation coefficient, the magnitude of the gradient, or the number of iterations. Through the gradient descent algorithm with an adaptive step size, the weights of the linear regression model and the nonlinear neural network model can be calculated more efficiently and accurately, enabling the weight allocation to better adapt to the characteristics of different frequency error components. The sum of the first weight and the second weight is limited to 1, ensuring that the weight allocation of the two models constitutes a complete weight space and avoiding redundancy or conflict in weight allocation.

[0042] Specifically, before performing dynamic weight allocation in this solution, the historical prediction errors are first decomposed by wavelet transform into multiple subsequences with different frequencies. The advantage of this approach is that the complex error signals can be decomposed into different frequency domains, enabling more refined analysis and processing of error components at different frequencies. For each frequency subsequence, the autocorrelation coefficient is calculated to evaluate the predictability of the error at that frequency. The higher the autocorrelation coefficient, the more regular the error at that frequency, and vice versa, the stronger the randomness. Then, based on the magnitude of the autocorrelation coefficient, the step size of the gradient descent algorithm is adaptively adjusted to calculate the weights of the linear regression model and the non-linear neural network model. For frequency subsequences with a higher autocorrelation coefficient, a larger step size is adopted to accelerate the weight adjustment speed; for frequency subsequences with a lower autocorrelation coefficient, a smaller step size is used to avoid over-adjustment. In this way, the weight allocation can more precisely adapt to the characteristics of errors at different frequencies, thereby improving the prediction accuracy of the combined prediction model.

[0043] In some embodiments, step A2 includes: B1. For each subsequence, calculate the autocorrelation coefficient. If the autocorrelation coefficient is greater than a preset threshold, determine that the corresponding subsequence is a stationary sequence and execute step B2; otherwise, determine that the corresponding subsequence is a non-stationary sequence and execute step B3; B2. For the stationary sequence, use the gradient descent algorithm with a fixed step size to calculate the first weight of the linear regression model and the second weight of the non-linear neural network model, denoted as the first weight optimal value and the second weight optimal value respectively, and the fixed step size is the preset step size; B3. For the non-stationary sequence, use the gradient descent algorithm with a variable step size to calculate the first weight of the linear regression model and the second weight of the non-linear neural network model, denoted as the first weight optimal value and the second weight optimal value respectively. The formula for the variable step size is: st_i = st_max * exp(-|r_i|), where st_i is the step size of the i-th subsequence, st_max is the preset maximum step size, and r_i is the autocorrelation coefficient of the i-th subsequence; B4. Perform weighted averaging on the first weight optimal values and the second weight optimal values calculated for all subsequences to obtain the final first weight of the linear regression model and the final second weight of the non-linear neural network model. The sum of the final first weight and the final second weight is 1.

[0044] Among them, regarding the problem of characteristic differences in different frequency subsequences in the error sequence, a refined weight allocation strategy is adopted. Specifically, first, for each subsequence obtained by wavelet decomposition, the autocorrelation coefficient is calculated, which characterizes the stationarity of the subsequence. A preset threshold is used to distinguish stationary and non-stationary subsequences, which can be set according to actual needs. For example, the preset threshold can be set to 0.5. For subsequences with an autocorrelation coefficient greater than 0.5, they are determined to be stationary sequences; otherwise, they are determined to be non-stationary sequences. Thus, the effective identification of the internal characteristics of the error sequence is achieved.

[0045] Furthermore, for different types of subsequences, a differential weight calculation method is adopted. For stationary subsequences, since their volatility is small and the prediction error is relatively stable, a gradient descent algorithm with a fixed step size is used to ensure the stability of weight update. The preset step size can be adjusted according to the actual application scenario. For example, it can be set to 0.01. For non-stationary subsequences, the data volatility is large and the prediction error is unstable, so a gradient descent algorithm with a variable step size is used. The variable step size calculation formula is: st_i = st_max * exp(-|r_i|). Among them, st_max is the preset maximum step size, which can be set according to actual needs. For example, it can be set to 0.1, and r_i is the autocorrelation coefficient of the i-th subsequence. Thus, the step size is negatively correlated with the autocorrelation coefficient, that is, the lower the autocorrelation of the error sequence (the stronger the non-stationarity), the larger the step size, enabling the weight to quickly adapt to the change of the error and improving the flexibility and tracking ability of the model.

[0046] Finally, by weighted averaging the first weight optimal value and the second weight optimal value calculated for all subsequences, the final first weight of the linear regression model and the final second weight of the non-linear neural network model are obtained. Weighted averaging can comprehensively consider the weight calculation results of each subsequence to obtain an overall optimal weight allocation scheme.

[0047] Specifically, through the above scheme, the weights of the linear regression model and the non-linear neural network model can be adjusted more refinedly, fully considering the complex characteristics inside the error sequence, overcoming the problem of insufficient weight allocation fineness, and improving the prediction accuracy of the combined prediction model. Thus, the accuracy of reactive power compensation and the control effect can be improved, ensuring the stable operation and efficient power consumption of the photovoltaic energy storage system in the subtropical coastal industrial park.

[0048] For example, in a specific embodiment, the preset threshold is set to 0.6, the preset step size is 0.005, and the maximum step size st_max is 0.05. For the 4 error subsequences obtained by wavelet decomposition, the autocorrelation coefficients r_1, r_2, r_3, and r_4 are calculated respectively. Assume r_1 = 0.7, r_2 = 0.8, r_3 = 0.4, and r_4 = 0.5. Since r_1 > 0.6 and r_2 > 0.6, it is determined that the first and second subsequences are stationary sequences, and the gradient descent algorithm with a fixed step size of 0.005 is used to calculate the optimal weight values; since r_3 < 0.6 and r_4 < 0.6, it is determined that the third and fourth subsequences are non-stationary sequences, and the gradient descent algorithm with variable step sizes is used to calculate the weights. The step sizes are st_3 = 0.05 * exp(-|0.4|) ≈ 0.0335 and st_4 = 0.05 * exp(-|0.5|) ≈ 0.0303 respectively. Through iterative calculation using the gradient descent algorithm, the first optimal weight values and the second optimal weight values corresponding to each subsequence are obtained, which are (w1_1, w1_2), (w2_1, w2_2), (w3_1, w3_2), and (w4_1, w4_2) respectively. The final first weight W1 = (w1_1 + w2_1 + w3_1 + w4_1) / 4, and the final second weight W2 = (w1_2 + w2_2 + w3_2 + w4_2) / 4.

[0049] In some preferred embodiments, step S3 includes: S301. Real-time collect the electricity consumption data of various types of loads in the industrial park, and identify the first type of loads with a fast response speed to reactive power demand and the second type of loads with a slow response speed; S302. According to the historical electricity consumption data ratio of the first type of loads and the second type of loads, decompose the total reactive power demand into the predicted value of the reactive power demand of the first type of loads and the predicted value of the reactive power demand of the second type of loads; S303. For the predicted value of the reactive power demand of the first type of loads, adopt a feed-forward control method, and directly adjust the output of the reactive power compensation device according to the predicted value (referring to the predicted value of the reactive power demand of the first type of loads) to achieve fast compensation; S304. For the predicted value of the reactive power demand of the second type of loads, adopt a feedback control method, real-time monitor the grid voltage, and adjust the output of the reactive power compensation device through a PID controller according to the grid voltage deviation and the predicted value of the reactive power demand of the second type of loads to achieve precise compensation.

[0050] Among them, in step S301, the electricity consumption data is collected in real time from various types of loads in the industrial park to comprehensively reflect the electricity consumption characteristics of the park. The load type identification can be achieved through a clustering algorithm, such as the K-means clustering algorithm, which analyzes the historical electricity consumption data characteristics of the loads and divides the loads into the first type of load (fast response) and the second type of load (slow response). Among them, fast and slow response speeds are relative concepts. Here, the first type of load with a fast response speed and the second type of load with a slow response speed mean that the response speed of the first type of load is faster than that of the second type of load.

[0051] Among them, in step S302, the predicted value of the total reactive power demand is provided by the combined prediction model, and this predicted value is then decomposed according to the proportion of the historical electricity consumption data of the two types of loads. The proportion data can be obtained by statistics from the historical operation data of the equipment. For example, if the proportion of the historical electricity consumption data of the first type of load in the historical total electricity consumption data is 1 / 3, and the proportion of the historical electricity consumption data of the second type of load in the historical total electricity consumption data is 2 / 3, then the predicted value of the reactive power demand of the first type of load is 1 / 3 of the total reactive power demand, and the predicted value of the reactive power demand of the second type of load is 2 / 3 of the total reactive power demand.

[0052] Among them, in step S303, for the feedforward control of the first type of load, the output of the reactive power compensation device is directly adjusted according to the predicted value, and the control signal acts quickly on the compensation device.

[0053] Among them, in step S304, for the feedback control of the second type of load, the PID controller is configured to adjust the output of the compensation device according to the grid voltage deviation and the predicted value. The voltage deviation is used as a feedback signal by the PID controller to ensure the compensation accuracy.

[0054] Specifically, aiming at the problem of the difference in load response characteristics in the industrial park, this method proposes a classified dynamic reactive power compensation control strategy. First, in step S301, through real-time power consumption data collection and clustering algorithm, the loads in the industrial park can be effectively classified into two categories with different response speeds to reactive power demand. Thus, the control strategy can be optimized and configured according to the characteristics of different types of loads. Secondly, in step S302, the predicted value of the total reactive power demand is decomposed into the predicted values of the respective demands of the two types of loads, realizing the refined allocation of the total reactive power demand. Further, steps S303 and S304 adopt different control methods. For the first type of load with fast response, the use of feed-forward control realizes fast reactive power compensation and ensures fast response. For the second type of load with slow response, the use of feedback control combined with a PID controller realizes accurate reactive power compensation and ensures compensation accuracy. Through the above steps, this method can take into account the reactive power compensation requirements of both fast-response loads and slow-response loads, improving the overall reactive power compensation effect and the stability of the grid voltage.

[0055] In some possible implementation manners, step S301 includes: Real-time collect the power consumption data of various types of loads in the industrial park to construct a power consumption data set; the power consumption data includes active power, reactive power, voltage, and current; Perform preprocessing on the power consumption data set to obtain a preprocessed power consumption data set; the preprocessing includes missing value filling, outlier detection and removal, and moving average filtering; Extract the time-domain features and frequency-domain features of the preprocessed power consumption data set; the time-domain features include mean, variance, peak value, and kurtosis, and the frequency-domain feature includes total harmonic distortion rate to obtain a load feature vector; Adopt the K-means clustering algorithm to divide the loads into the first type of load and the second type of load according to the load feature vector.

[0056] To ensure classification accuracy. First, the electricity consumption data of various types of loads in the industrial park are collected in real time to form an electricity consumption dataset, laying a data foundation for subsequent analysis. The electricity consumption data includes information such as active power, reactive power, voltage, and current, achieving a comprehensive acquisition of the load operation status. Second, the collected electricity consumption data is preprocessed to obtain the preprocessed electricity consumption dataset. The preprocessing operations include multiple steps. Missing value filling ensures data integrity. For example, the linear interpolation method is used to fill a small amount of missing data that may occur during data collection. Outlier detection and elimination avoid the interference of abnormal data on the analysis results. For example, the outlier detection method based on the 3σ principle is used to identify and eliminate abnormal data that deviates from the normal range. Moving average filtering reduces the impact of data noise and improves data quality. For example, the moving average filtering method with a window size of 5 is used to smooth data fluctuations. Then, time-domain and frequency-domain features are extracted from the preprocessed data. The time-domain features include mean, variance, peak value, and kurtosis, and the frequency-domain feature includes total harmonic distortion rate. These features effectively characterize the electricity consumption characteristics of the load. For example, the mean and variance reflect the average level and fluctuation degree of the load power, the peak value and kurtosis reflect the extreme values and distribution patterns of the load power, and the total harmonic distortion rate reflects the waveform distortion degree of the load current. Finally, the K-means clustering algorithm is used, with the load feature vector as the input, to automatically divide the load into the first category and the second category. The K-means algorithm is an unsupervised learning algorithm that can cluster according to the similarity of the data itself without manual intervention, achieving the automatic classification of the load. For example, the K-means algorithm is set with the number of cluster centers as 2, the distance calculation method as Euclidean distance, and the number of iterations as 100 times. Through the above steps, the accurate classification of the loads in the industrial park is achieved, providing a prerequisite and guarantee for subsequent differential reactive power compensation control strategies for different types of loads. The use of data preprocessing ensures data quality and reduces noise interference. The introduction of feature extraction effectively characterizes the electricity consumption characteristics of the load. The application of the K-means clustering algorithm realizes the automatic classification of the load, improving the classification efficiency and accuracy.

[0057] In some specific embodiments, to further improve the accuracy and adaptability of load classification, more advanced data cleaning methods can be introduced in the preprocessing stage. For example, a denoising method based on wavelet analysis can more effectively remove high-frequency noise in electricity consumption data. In the feature extraction stage, more dimensions and more representative features can be added. For example, the harmonic content of each order can be added to the frequency-domain features, and waveform factor, margin factor, etc. can be added to the time-domain features to more comprehensively describe the electricity consumption characteristics of the load. In terms of clustering algorithms, DBSCAN clustering algorithm or hierarchical clustering algorithm can be tried to meet the load classification requirements of industrial parks with different types and scales. Through the optimization of preprocessing methods, the enhancement of feature extraction, and the improvement of clustering algorithms, the accuracy and robustness of load classification can be further improved, providing a more reliable guarantee for the intelligent control of the photovoltaic energy storage system in industrial parks.

[0058] The present application further proposes that step S4 includes: S401. Monitor the voltage at the grid connection point in real time, calculate the voltage deviation between the voltage at the grid connection point and the preset voltage, and obtain the current active power output values of the photovoltaic inverter and the energy storage converter; S402. According to the voltage deviation and the current active power output values, use the droop control algorithm to calculate the reactive power regulation commands for the photovoltaic inverter and the energy storage converter, where the droop coefficient in the droop control algorithm is adjusted according to the real-time ambient temperature and real-time humidity; S403. According to the reactive power regulation commands, adjust the operating modes of the photovoltaic inverter and the energy storage converter, and coordinately control the output of active power and reactive power. Among them, the photovoltaic inverter preferentially adjusts the reactive power. When the reactive power of the photovoltaic inverter reaches the upper limit, the energy storage converter continues to adjust the reactive power; S404. Monitor the operating states of the photovoltaic inverter and the energy storage converter in real time. If equipment overload or failure is detected, start the standby inverter or energy storage converter and adjust the operating parameters to ensure the stability of the voltage at the grid connection point.

[0059] Among them, in step S401, the voltage at the grid connection point can be obtained by real-time monitoring with a voltage sensor. The voltage deviation is the difference between the voltage at the grid connection point and the preset voltage, and the preset voltage is the target voltage value set according to the grid operation standard. The current active power output values of the photovoltaic inverter and the energy storage converter can be obtained in real time through their respective power sensors.

[0060] Among them, in step S402, the droop control algorithm is used to calculate the reactive power regulation command, and this algorithm determines the required reactive power compensation amount according to the voltage deviation and the current active power output value. The adjustment of the droop coefficient is based on the real-time ambient temperature and real-time humidity, and the droop coefficient can be adjusted based on the fuzzy algorithm. Thus, the droop control algorithm can adapt to environmental changes and improve the control flexibility.

[0061] Among them, in step S403, the working mode adjustment of the photovoltaic inverter and the energy storage converter is carried out according to the reactive power regulation command. The photovoltaic inverter is set to preferentially regulate the reactive power. When the reactive power output of the photovoltaic inverter reaches its rated upper limit, the energy storage converter starts to intervene and continues to regulate the reactive power, realizing the coordinated control of the two to ensure the continuity and effectiveness of reactive power compensation. This hierarchical control strategy can make full use of the reactive power regulation ability of the photovoltaic inverter, and at the same time avoid the frequent operation of the energy storage converter and extend the service life of the energy storage device.

[0062] Among them, in step S404, the operation state monitoring of the photovoltaic inverter and the energy storage converter is realized through the state monitoring module. The monitoring parameters include but are not limited to device temperature, current, voltage, etc. The detection of overload or fault is completed by comparing the monitoring parameters with the preset thresholds. Once an overload or fault is detected, the standby inverter or the energy storage converter is immediately started, and the operation parameters are adjusted to take over the work of the faulty device to ensure the stability of the grid connection point voltage. The timely intervention of the standby device and the adjustment of the operation parameters can effectively improve the reliability and fault tolerance of the system and avoid voltage instability caused by device failures.

[0063] Specifically, in the intelligent control system for photovoltaic energy storage power consumption in a subtropical coastal industrial park, to maintain the voltage stability at the grid connection point, first, the system uses voltage sensors to collect the voltage at the grid connection point in real time, compares the collected value with the preset voltage value, and calculates the voltage deviation. At the same time, the system obtains the current active power output values of the photovoltaic inverter and the energy storage converter. Then, the system adopts a droop control algorithm to calculate the reactive power regulation command based on the voltage deviation and the current active power output value. It should be noted that the droop coefficient in the droop control algorithm is dynamically adjusted, and the adjustment process takes into account the real-time ambient temperature and real-time humidity. The higher the ambient temperature and humidity, the greater the adjustment range of the droop coefficient, and vice versa. In this way, the control system can adaptively adjust the control strategy according to the environmental conditions and more effectively cope with voltage fluctuations in high-temperature and high-humidity environments. Next, the system adjusts the operating modes of the photovoltaic inverter and the energy storage converter according to the calculated reactive power regulation command. Under normal circumstances, the photovoltaic inverter gives priority to undertaking the reactive power regulation task. When the reactive power output of the photovoltaic inverter reaches the upper limit, the energy storage converter starts to work in coordination to jointly regulate the reactive power and ensure that the system has sufficient reactive power compensation capacity. Finally, the system monitors the operating states of the photovoltaic inverter and the energy storage converter in real time. If any device is detected to be overloaded or faulty, the system immediately starts the standby inverter or energy storage converter, adjusts the relevant operating parameters, and seamlessly switches to the standby device to ensure the continuous stability of the voltage at the grid connection point. Through the above steps, this technical solution can effectively maintain the voltage stability at the grid connection point in the high-temperature and high-humidity environment of a subtropical coastal industrial park.

[0064] In some preferred embodiments, step S402 includes: Calculating a temperature correction coefficient and a humidity correction coefficient through a fuzzy algorithm according to the real-time ambient temperature and real-time humidity; Calculating a comprehensive environment correction coefficient according to the temperature correction coefficient and the humidity correction coefficient; Adjusting the droop coefficient of the droop control algorithm according to the comprehensive environment correction coefficient; Calculating the reactive power regulation commands for the photovoltaic inverter and the energy storage converter based on the adjusted droop coefficient and the droop control algorithm according to the voltage deviation and the current active power output value.

[0065] Among them, the temperature correction coefficient and the humidity correction coefficient are calculated by a fuzzy algorithm to handle the non-linear relationship between the ambient temperature and humidity and the droop coefficient, so that the adjustment of the droop coefficient can be more refined. The temperature correction coefficient and the humidity correction coefficient are then combined to obtain a comprehensive environmental correction coefficient, thus comprehensively considering the combined influence of temperature and humidity on the droop coefficient. The comprehensive environmental correction coefficient is then used to adjust the droop coefficient of the droop control algorithm to obtain an adjusted droop coefficient suitable for the current environmental conditions. Finally, based on the adjusted droop coefficient, the droop control algorithm is used to calculate the reactive power regulation commands of the photovoltaic inverter and the energy storage converter, making the calculation of the reactive power regulation commands more accurate, and thus improving the control effect of voltage stability.

[0066] Specifically, to achieve the adaptive adjustment of the droop coefficient, first, the real-time ambient temperature and the real-time humidity are used as the inputs of the fuzzy algorithm. After fuzzyfication, fuzzy inference, and defuzzification processes, the temperature correction coefficient and the humidity correction coefficient are calculated. The fuzzy algorithm uses a pre-set fuzzy rule base, which defines the value ranges of the temperature correction coefficient and the humidity correction coefficient under different temperature and humidity conditions. For example, when the ambient temperature is high and the humidity is high, the fuzzy rule base can set a higher temperature correction coefficient and a higher humidity correction coefficient, and vice versa. The temperature correction coefficient and the humidity correction coefficient are combined by weighted averaging to obtain a comprehensive environmental correction coefficient. As a preferred implementation, the comprehensive environmental correction coefficient can be calculated by the following formula: Comprehensive environmental correction coefficient = w1 * temperature correction coefficient + w2 * humidity correction coefficient, where w1 and w2 are the weights of the temperature correction coefficient and the humidity correction coefficient respectively, and w1 + w2 = 1. The droop coefficient in the droop control algorithm is then adjusted based on the comprehensive environmental correction coefficient. For example, the adjusted droop coefficient can be calculated by the following formula: Adjusted droop coefficient = original droop coefficient * comprehensive environmental correction coefficient. Finally, the voltage deviation and the current active power output value are input into the droop control algorithm with the adjusted droop coefficient, and the reactive power regulation commands of the photovoltaic inverter and the energy storage converter are calculated.

[0067] In some specific embodiments, the fuzzy algorithm can adopt the Mamdani fuzzy algorithm, and its fuzzy rule base can be offline trained and optimized according to the historical environmental data of the subtropical coastal industrial park and the operation data of the photovoltaic energy storage system. For example, the total reactive power demand and voltage fluctuation data of the system under different temperature and humidity conditions can be collected, and through data analysis and expert experience, a fuzzy rule base can be established to determine the fuzzy relationship between temperature, humidity, temperature correction coefficient, and humidity correction coefficient. The input membership function and output membership function of the fuzzy controller can select triangular or Gaussian membership functions to ensure the smoothness and stability of control. The calculation weights w1 and w2 of the comprehensive environmental correction coefficient can be adjusted according to the actual operation situation. For example, in areas where humidity has a greater impact on the system, the weight w2 of the humidity correction coefficient can be appropriately increased. The adjustment method of the droop coefficient can also be selected according to the system characteristics. For example, different adjustment methods such as addition or multiplication can be adopted. Through the above specific embodiments, precise adaptive adjustment of the droop coefficient can be achieved, improving the accuracy and robustness of the voltage control of the photovoltaic energy storage system in the subtropical coastal industrial park.

[0068] In some embodiments, step S5 includes: S501. Monitor the temperature of the inverter of the photovoltaic energy storage system, the temperature of the energy storage battery, and the temperature of the converter in real time, and construct a device temperature data set; S502. According to the real-time environmental temperature, real-time humidity, and the device temperature data set, use the fuzzy control algorithm to calculate the cooling demand level; S503. According to the cooling demand level and the current operating state of the cooling system, use a hierarchical control strategy to adjust the operating parameters of the cooling system.

[0069] Among them, in step S501, monitoring the temperature of the inverter of the photovoltaic energy storage system, the temperature of the energy storage battery, and the temperature of the converter in real time, and constructing a device temperature data set is to comprehensively grasp the operating thermal state of the key devices of the photovoltaic energy storage system and provide accurate basic data for subsequent cooling control. Among them, temperature monitoring can be achieved by using temperature sensors, such as thermocouples, thermistors, or integrated temperature sensors, etc. These sensors can collect the temperature data on the surfaces of the inverter, energy storage battery, and converter in real time. The construction of the device temperature data set can adopt a data acquisition system to integrate, store, and manage the collected temperature data, providing data support for the subsequent fuzzy control algorithm.

[0070] Among them, in step S502, calculating the cooling demand level using a fuzzy control algorithm based on the ambient temperature data, humidity data, and device temperature dataset is the key to implementing the optimized operation strategy of the cooling system in this application. The application of the fuzzy control algorithm enables the system to handle the complex non-linear relationship between the ambient temperature and humidity and the device temperature, and more accurately evaluate the actual cooling demand. The specific implementation process of the fuzzy control algorithm can be as follows: First, determine the fuzzy input variables of the ambient temperature, humidity, and device temperature, and set the fuzzy subsets and membership functions of each variable; then, establish fuzzy control rules. For example, when the ambient temperature is high, the humidity is high, and the device temperature is also high, the cooling demand level is high; when the ambient temperature is moderate, the humidity is low, and the device temperature is also moderate, the cooling demand level is low (the high and low of the temperature and the large and small of the humidity can be divided according to actual needs); next, perform fuzzy processing based on the real-time collected ambient temperature and humidity data and device temperature data to obtain the fuzzy membership degrees of each input variable; finally, perform fuzzy inference according to the fuzzy control rules, and calculate and obtain a clear cooling demand level output value through a defuzzification method. The cooling demand level can be divided into multiple levels, such as low, medium, and high levels, or represented by numerical values. The higher the level, the higher the cooling demand.

[0071] Among them, in step S503, adjusting the operating parameters of the cooling system using a hierarchical control strategy according to the cooling demand level and the current operating state of the cooling system is to minimize the energy consumption of the cooling system and improve the overall operating efficiency and economy of the system on the premise of meeting the cooling demand of the device. The implementation of the hierarchical control strategy can preset different operating parameters of the cooling system according to different cooling demand levels. For example, when the cooling demand level is low, the rotation speed of the cooling fan can be reduced, the number of cooling cycles can be decreased, or the cooling system can be operated intermittently; when the cooling demand level is medium, the medium rotation speed of the cooling fan can be maintained, and the cooling cycle can be operated normally; when the cooling demand level is high, the rotation speed of the cooling fan can be increased, the number of cooling cycles can be increased, or an auxiliary cooling device can be started. The current operating state of the cooling system can be monitored in real time through sensors, such as a fan rotation speed sensor, a coolant flow sensor, etc., so as to perform precise adjustment according to the actual operating state. The specific control parameters of the hierarchical control strategy can be calibrated and optimized according to the actual application scenario and device characteristics.

[0072] Specifically, the optimized operation strategy of the cooling system proposed in this application first monitors the temperature of key equipment in the photovoltaic energy storage system in real time through step S501 to comprehensively grasp the thermal state of equipment operation. Then, in step S502, considering the ambient temperature and humidity and the equipment temperature comprehensively, the fuzzy control algorithm is used to accurately calculate the cooling demand level to achieve a refined evaluation of the cooling demand. Finally, in step S503, based on the cooling demand level and the current operating state of the cooling system, a hierarchical control strategy is adopted to finely adjust the operating parameters of the cooling system to achieve intelligent control of the cooling system. Through the above steps, this application can dynamically adjust the operating strategy of the cooling system according to the actual operating state and environmental conditions of the photovoltaic energy storage system, while ensuring the stable operating temperature of the equipment and reducing the energy consumption of the cooling system. Compared with the traditional fixed-operation mode cooling system, the optimized strategy proposed in this application can more effectively reduce the equipment operating temperature and improve the overall operating efficiency of the system.

[0073] As a preferred embodiment, the solution of this application is specifically implemented as follows: In the photovoltaic energy storage system, PT100 platinum resistance temperature sensors are respectively installed on the inverter, energy storage battery and converter to collect the temperature data of each device in real time for constructing a device temperature data set. At the same time, temperature and humidity sensors are installed in the deployment area of the photovoltaic energy storage system to collect the ambient temperature and humidity data in real time. The controller uses an STM32 microcontroller with a fuzzy control algorithm program preset inside, including three input variables: real-time ambient temperature, real-time humidity and device temperature, and the cooling demand level as the output variable. The fuzzy control rule is set as follows: when the real-time ambient temperature is higher than 35°C, the real-time humidity is higher than 80%, and the device temperature is higher than 50°C, the cooling demand level is high; when the real-time ambient temperature is between 25°C and 35°C, the real-time humidity is between 50% and 80%, and the device temperature is between 40°C and 50°C, the cooling demand level is medium; when the real-time ambient temperature is lower than 25°C, or the real-time humidity is lower than 50%, or the device temperature is lower than 40°C, the cooling demand level is low. The cooling system adopts an air-cooled heat dissipation method, and the fan drive circuit uses PWM speed control. The controller outputs a PWM control signal according to the cooling demand level to adjust the fan speed. When the cooling demand level is low, the PWM duty cycle is 30% and the fan runs at a low speed; when the cooling demand level is medium, the PWM duty cycle is 60% and the fan runs at a medium speed; when the cooling demand level is high, the PWM duty cycle is 90% and the fan runs at a high speed.

[0074] Refer to Figure 2 , this application also provides an intelligent control device for photovoltaic energy storage power consumption in an industrial park, which is used for power consumption control of the photovoltaic energy storage system in a subtropical coastal industrial park. The device includes: The environmental parameter monitoring module 1 is used to obtain the real-time environmental temperature and real-time humidity of the deployment area of the photovoltaic energy storage system (for the specific process, refer to step S1 in the previous text); The reactive power demand prediction module 2 is used to predict the total reactive power demand of the industrial park and the power electronic equipment of the photovoltaic energy storage system according to the historical operation data of the equipment, the real-time environmental temperature and the real-time humidity, by using a combined prediction model (for the specific process, refer to step S2 in the previous text); The dynamic reactive power compensation control module 3 is used to dynamically adjust the output of the reactive power compensation device according to the prediction result of the combined prediction model to achieve reactive power compensation (for the specific process, refer to step S3 in the previous text); The voltage stability control module 4 is used to adjust the working modes of the photovoltaic inverter and the energy storage converter, and cooperate to control the output of active power and reactive power to maintain the voltage stability of the grid connection point (for the specific process, refer to step S4 in the previous text); The equipment cooling optimization module 5 is used to optimize the operation strategy of the cooling system of the photovoltaic energy storage system according to the real-time environmental temperature and the real-time humidity to reduce the equipment operation temperature (for the specific process, refer to step S5 in the previous text).

[0075] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some communication interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical or other forms.

[0076] In addition, the units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units. They can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0077] Furthermore, in each embodiment of the present application, the functional modules can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0078] In this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0079] The above are only the embodiments of the present application and are not intended to limit the protection scope of the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. An intelligent control method for photovoltaic energy storage power consumption in industrial parks, used for power consumption control of photovoltaic energy storage systems in subtropical coastal industrial parks, characterized in that: The method includes: S1. Obtain the real-time ambient temperature and humidity of the photovoltaic energy storage system deployment area; S2. Based on the historical operation data of the equipment, the real-time ambient temperature and the real-time humidity, a combined prediction model is used to predict the total reactive power demand of the power electronic equipment in the industrial park and the photovoltaic energy storage system; S3. According to the prediction results of the combined prediction model, dynamically adjust the output of the reactive power compensation device to achieve reactive power compensation; S4. Adjust the working mode of the photovoltaic inverter and the energy storage converter, coordinately control the output of active power and reactive power, and maintain the voltage stability at the grid connection point; S5. According to the real-time ambient temperature and real-time humidity, optimize the operation strategy of the photovoltaic energy storage system cooling system to reduce the equipment operating temperature.

2. According to claim 1, an intelligent control method for photovoltaic energy storage electricity consumption in an industrial park is characterized in that: The combined prediction model includes a linear regression model and a nonlinear neural network model; the input of the linear regression model includes active power, ambient temperature and humidity in the historical operation data of the equipment, and the output is a first reactive power prediction value; The input of the nonlinear neural network model includes active power, ambient temperature and humidity in historical operation data, and the output is the second reactive power prediction value; Step S2 includes: S201. According to the active power, ambient temperature and humidity in the historical operation data of the device, a linear regression model is used to obtain a first reactive power prediction value, and a nonlinear neural network model is used to obtain a second reactive power prediction value; S202. According to the historical prediction errors of the linear regression model and the nonlinear neural network model, a dynamic weight allocation method is used to calculate the first weight of the linear regression model and the second weight of the nonlinear neural network model, and the sum of the first weight and the second weight is 1; S203. Obtain an initial reactive power prediction value by weighted average calculation based on the first reactive power prediction value, the second reactive power prediction value, the first weight, and the second weight; S204. According to the real-time ambient temperature and real-time humidity, the initial reactive power prediction value is nonlinearly corrected using the Sigmoid function to obtain the final reactive power prediction value.

3. According to claim 2, an intelligent control method for photovoltaic energy storage electricity consumption in an industrial park is characterized in that: Step S202 includes: A1. Perform sliding window processing on historical forecast errors, construct an error sequence, and use wavelet decomposition algorithm to decompose the error sequence to obtain multiple subsequences with different frequencies; A2. For each subsequence, the autocorrelation coefficient is calculated. According to the size of the autocorrelation coefficient, the gradient descent algorithm with adaptive step size is used to calculate the first weight of the linear regression model and the second weight of the nonlinear neural network model. The sum of the first weight and the second weight is 1.

4. According to claim 3, an intelligent control method for photovoltaic energy storage electricity consumption in an industrial park is characterized in that: Step A2 includes: B1. For each subsequence, calculate the autocorrelation coefficient. If the autocorrelation coefficient is greater than a preset threshold, the corresponding subsequence is determined to be a stationary sequence, and step B2 is executed; otherwise, the corresponding subsequence is determined to be a non-stationary sequence, and step B3 is executed; B2. For the stationary sequence, a fixed-step gradient descent algorithm is used to calculate the first weight of the linear regression model and the second weight of the nonlinear neural network model, which are respectively recorded as the first weight preferred value and the second weight preferred value, and the fixed step size is the preset step size; B3. For non-stationary sequences, a variable step-size gradient descent algorithm is used to calculate the first weight of the linear regression model and the second weight of the nonlinear neural network model, which are respectively recorded as the first weight preferred value and the second weight preferred value. The variable step-size calculation formula is: st_i=st_max*exp(-|r_i|), where st_i is the step size of the i-th subsequence, st_max is the preset maximum step size, and r_i is the autocorrelation coefficient of the i-th subsequence; B4. Take a weighted average of the first weight preferred value and the second weight preferred value calculated for all subsequences to obtain the final first weight of the linear regression model and the final second weight of the nonlinear neural network model. The sum of the final first weight and the final second weight is 1.

5. The method for intelligent control of photovoltaic energy storage electricity consumption in an industrial park according to claim 1, characterized in that: Step S3 includes: S301. Real-time collection of power consumption data of various types of loads in the industrial park, identifying the first type of loads that respond quickly to reactive power demand and the second type of loads that respond slowly; S302. According to the historical power consumption data ratio of the first type of load and the second type of load, the total reactive power demand is decomposed into the reactive power demand forecast value of the first type of load and the reactive power demand forecast value of the second type of load; S303. For the reactive power demand prediction value of the first type of load, a feedforward control method is adopted to directly adjust the output of the reactive compensation device according to the prediction value to achieve rapid compensation; S304. For the reactive power demand prediction value of the second type of load, a feedback control method is adopted to monitor the grid voltage in real time. According to the grid voltage deviation and the reactive power demand prediction value of the second type of load, the output of the reactive compensation device is adjusted through the PID controller to achieve precise compensation.

6. The method for intelligent control of photovoltaic energy storage electricity consumption in industrial parks according to claim 5 is characterized in that: Step S301 includes: Collect power consumption data of various types of loads in the industrial park in real time and build a power consumption data set; power consumption data includes active power, reactive power, voltage, and current; Preprocessing the electricity consumption data set to obtain a preprocessed electricity consumption data set; the preprocessing includes missing value filling, outlier detection and elimination, and sliding average filtering; Extract the time domain features and frequency domain features of the preprocessed power consumption data set. The time domain features include mean, variance, peak value and kurtosis, and the frequency domain features include total harmonic distortion rate to obtain the load feature vector. The K-means clustering algorithm is used to divide the load into the first type of load and the second type of load according to the load feature vector.

7. The method for intelligent control of photovoltaic energy storage electricity consumption in an industrial park according to claim 1, characterized in that: Step S4 includes: S401. Real-time monitoring of the grid connection point voltage, calculation of the voltage deviation between the grid connection point voltage and the preset voltage, and acquisition of the current active power output value of the photovoltaic inverter and the energy storage converter; S402. According to the voltage deviation and the current active power output value, a droop control algorithm is used to calculate the reactive power regulation instruction of the photovoltaic inverter and the energy storage converter, wherein the droop coefficient in the droop control algorithm is adjusted according to the real-time ambient temperature and real-time humidity; S403. According to the reactive power regulation instruction, adjust the working mode of the photovoltaic inverter and the energy storage converter, and coordinate the output of active power and reactive power, wherein the photovoltaic inverter prioritizes the adjustment of reactive power. When the reactive power of the photovoltaic inverter reaches the upper limit, the energy storage converter continues to adjust the reactive power; S404. Monitor the operating status of the photovoltaic inverter and energy storage converter in real time. If an equipment overload or failure is detected, start the backup inverter or energy storage converter and adjust the operating parameters to ensure the voltage stability at the grid connection point.

8. The method for intelligent control of photovoltaic energy storage electricity consumption in an industrial park according to claim 7, characterized in that: Step S402 includes: According to the real-time ambient temperature and real-time humidity, the temperature correction coefficient and the humidity correction coefficient are calculated by fuzzy algorithm; Calculate the comprehensive environmental correction factor based on the temperature correction factor and the humidity correction factor; According to the comprehensive environmental correction coefficient, the droop coefficient of the droop control algorithm is adjusted; According to the voltage deviation and the current active power output value, the adjusted droop coefficient is adopted to calculate the reactive power regulation instructions of the photovoltaic inverter and the energy storage converter based on the droop control algorithm.

9. The method for intelligent control of photovoltaic energy storage electricity consumption in an industrial park according to claim 1, characterized in that: Step S5 includes: S501. Real-time monitoring of the inverter temperature, energy storage battery temperature and converter temperature of the photovoltaic energy storage system, and building a device temperature data set; S502. Based on the real-time ambient temperature, real-time humidity and device temperature data set, a fuzzy control algorithm is used to calculate the cooling demand level; S503. According to the cooling demand level and the current operating state of the cooling system, a hierarchical control strategy is adopted to adjust the operating parameters of the cooling system.

10. An intelligent control device for photovoltaic energy storage electricity consumption in industrial parks, used for electricity consumption control of photovoltaic energy storage systems in subtropical coastal industrial parks, characterized in that: The device includes: Environmental parameter monitoring module, used to obtain the real-time ambient temperature and humidity of the photovoltaic energy storage system deployment area; Reactive power demand prediction module, which is used to predict the total reactive power demand of power electronic equipment in industrial parks and photovoltaic energy storage systems using a combined prediction model based on historical equipment operation data, real-time ambient temperature and real-time humidity; A dynamic reactive power compensation control module is used to dynamically adjust the output of the reactive power compensation device according to the prediction results of the combined prediction model to achieve reactive power compensation; Voltage stability control module, used to adjust the working mode of photovoltaic inverter and energy storage converter, coordinate the output of active power and reactive power, and maintain the voltage stability at the grid connection point; The equipment cooling optimization module is used to optimize the operation strategy of the photovoltaic energy storage system cooling system according to the real-time ambient temperature and real-time humidity, and reduce the equipment operating temperature.

Images