Energy operation management method, device and medium of a light storage system

Abstract

The application discloses an energy operation management method and device of a light storage system and a medium. The method comprises the following steps: obtaining operation data of the light storage system and performing feature extraction; constructing a prediction model according to the extracted feature data and by using an improved multi-strategy fusion GTO algorithm; constructing a multi-objective function and solving the constructed multi-objective function by using the improved multi-strategy fusion GTO algorithm to obtain a charging and discharging strategy of the light storage system and a scheduling strategy of a flexible energy utilization device. The device and the medium are used to implement the method. The application has the beneficial effects that the macroscopic system optimization and the microscopic battery mechanism are simultaneously started from two aspects, the low efficiency and the safety problem caused by the lack of battery operation mechanism analysis in the traditional method are effectively avoided, the actual demand of a user and the change of an electricity price are fully considered, the operation management strategy of the light storage system is formulated, and the dual targets of minimizing the operation cost and maximizing the battery life are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of new energy power generation, in particular to an energy operation management method of a light storage system, equipment and medium. BACKGROUND

[0002] The existing energy management method of the energy storage system is divided into two aspects of macro and micro. Macroscopically, the energy can be regulated according to the existing management strategy, which is usually formulated based on inherent experience, engineering knowledge or specific engineering requirements. For example, considering the safety of the battery during use, charging is started when the battery power is lower than the set threshold; considering the user's electricity cost, including construction cost and future benefits, the energy management strategy of charging at the low valley of the power grid and discharging at the peak is formulated. Microscopically, the health state of the battery is the basis for the energy management system to formulate effective strategies, and the life of the battery is an important indicator for measuring the health state of the battery, which can reflect the dynamic parameter changes and operating environment of the battery to a great extent. The methods of battery life prediction mainly include equivalent circuit model, electrochemical model and data-driven model.

[0003] However, in the actual use process, the macro energy management method of the energy storage system analyzes the application scene of the battery from the operation level, and more from the control angle to control the use of the battery, without considering the performance parameter changes caused by various factors during the operation process, and without considering the influence of the specific energy demand or energy mode of the user side on the formulation of the final energy management strategy, so it is only limited to the application in a certain scene, and cannot formulate a changeable energy management strategy from the energy angle.

[0004] For the micro energy management method of the energy storage system, due to the variety of chemical reactions inside the battery, the identification of model parameters is more complex, requiring a large amount of experimental data and unstable parameter estimation, which is difficult to solve and requires a lot of calculation time and effort. For complex actual application scenarios, the matching degree of model parameters with actual parameters is low, resulting in a certain deviation between the prediction result and the actual situation. The data-driven prediction method needs a large amount of data for training, so the data dependency is high, and the influence of different algorithm structures may lead to more complex parameter identification in the training process, resulting in high calculation cost. SUMMARY

[0005] One of the purposes of the present application is to provide an energy operation management method of a light storage system which can solve at least one of the defects in the background art.

[0006] Another purpose of the present application is to provide an electronic device capable of implementing the energy operation management method of the light storage system which can solve at least one of the defects in the background art.

[0007] Still another object of the present application is to provide a computer-readable storage medium capable of implementing an energy operation management method of an optical storage system that solves at least one of the defects in the background art.

[0008] To achieve the above at least one object, the technical solution adopted by the present application is as follows: An energy operation management method of an optical storage system, comprising the following steps: obtaining operation data of the optical storage system, including user-side energy consumption data, photovoltaic power generation data, electricity price data, and energy storage battery operation data, and performing feature extraction on each operation data of the optical storage system based on Pearson correlation coefficients; constructing a prediction model based on a neural network architecture according to the extracted feature data, including a user energy consumption demand prediction model that takes into account flexible energy consumption equipment and rigid energy consumption equipment, and a photovoltaic power generation prediction model and an energy storage battery SOC prediction model; and using an improved multi-strategy fusion GTO algorithm to optimize the network parameters of the prediction model; based on the optimized prediction model, combining system power balance constraints, battery SOC safety constraints, and electricity price mechanisms, constructing a multi-objective function with the objectives of minimizing operation cost and maximizing energy storage battery life; using the above-mentioned improved multi-strategy fusion GTO algorithm to solve the constructed multi-objective function, obtaining the charging and discharging strategy of the optical storage system and the scheduling strategy of the flexible energy consumption equipment.

[0009] Preferably, the process of feature extraction on each operation data of the optical storage system is as follows: calculating a plurality of Pearson correlation coefficients that measure the relationship between each operation data of the optical storage system and user-side energy consumption; setting a threshold coefficient representing strong correlation between the operation data of the optical storage system and user-side energy consumption; and retaining the operation data corresponding to the Pearson correlation coefficients greater than the set threshold coefficient among the calculated plurality of Pearson correlation coefficients as strong correlation feature data.

[0010] Preferably, the prediction model needs to set the parameter weight for different input characteristic parameters, and the specific parameter weight setting process is as follows: selecting an input characteristic parameter from the N input characteristic parameters as a target input characteristic parameter; performing the value judgment process of the target input characteristic parameter: increasing the value of the target input characteristic parameter at t+1 by a set step size Am, and keeping the values of the remaining input characteristic parameters unchanged at t+1; calculating the difference between the output values of the prediction model at t+1 and t; if the calculated output difference is greater than 0, the target input characteristic parameter is defined as + value input; if the calculated output difference is less than 0, the target input characteristic parameter is defined as - value input; if the calculated output difference is 0, the target input characteristic parameter is defined as no value input; reselecting the target input characteristic parameter from the remaining input characteristic parameters and repeating the above value judgment process until all input characteristic parameters complete the value judgment; value sorting according to the output difference corresponding to different input characteristic parameters in the value judgment process, and setting the parameter weight based on the sorting result.

[0011] Preferably, for the value sorting process when the output difference corresponding to two input characteristic parameters is the same: defining the two input characteristic parameters as a first input characteristic parameter and a second input characteristic parameter, and performing the value judgment process of increasing the step size of the two input characteristic parameters by Am- Ai; if the first output difference corresponding to the first input characteristic parameter is greater than the second output difference corresponding to the second input characteristic parameter, the value of the first input characteristic parameter is defined as greater than that of the second input characteristic parameter; if the first output difference corresponding to the first input characteristic parameter is less than the second output difference corresponding to the second input characteristic parameter, the value of the second input characteristic parameter is defined as greater than that of the first input characteristic parameter; if the first output difference corresponding to the first input characteristic parameter is equal to the second output difference corresponding to the second input characteristic parameter, the next round of value judgment process of increasing the step size of the two input characteristic parameters by Am-2Ai is performed until the kth round of value judgment process Am-kAi=0, and the first output difference corresponding to the first input characteristic parameter in the k-1th round of value judgment process is still equal to the second output difference corresponding to the second input characteristic parameter, the values of the two input characteristic parameters are defined as the same; wherein, Ai represents the step size reduction.

[0012] Preferably, the improved multi-strategy fusion GTO algorithm includes the following: introducing sinusoidal disturbance when initializing the population through Tent mapping; changing the linearly decreasing energy parameter to a nonlinearly decreasing one; performing reverse learning on the current individual to generate a reverse solution at a set probability in each iteration process; adding disturbance to the current optimal individual using Cauchy distribution; mapping the chaotic sequence to the solution space to perform GTO position update.

[0013] Preferably, the process of optimizing the network parameters of the prediction model by the improved multi-strategy fusion GTO algorithm is as follows: taking the network parameters of the prediction model as the prey, calculating the population target value and updating the energy parameter of the prey escaping; if the energy parameter of the prey escaping is greater than or equal to 1, the algorithm enters the global search stage and updates the population position; in the global search stage, if the possibility of the prey escaping is greater than or equal to 0.5, the algorithm strengthens the influence on the position of the prey, otherwise, the influence on the position of the prey is weakened; if the energy parameter of the prey escaping is less than 1 or the global search stage is completed, the algorithm enters the local search stage and updates the population position; in the local search stage, the current four optimal solutions are selected as the positions of the attacker, the besieger, the driver and the pursuer to surround the prey; wherein the attacker is the best solution of the population, and the geometric center of the four surrounding roles is the position where the prey is most likely to appear, and is used as the convergence point of the position update of the surrounding roles in the next round of surrounding; if the prey escaping probability is less than the set probability threshold, the four surrounding roles are reversely learned to generate a new solution, otherwise, a new solution is obtained by adding disturbance to the attacker; the new solution and the best solution are judged, if the new solution is less than the best solution, the value of the new solution is assigned to the best solution for updating, otherwise, the value of the best solution remains unchanged; the position of the prey is updated and it is judged whether the maximum number of iterations is reached; if the maximum number of iterations is reached, the best network parameters corresponding to the position of the prey are output, otherwise, the above process is repeatedly executed for one iteration.

[0014] Preferably, the charge and discharge strategy of the optical storage system includes the following contents: when the photovoltaic power generation power is greater than the power of the user side energy consumption equipment, and the SOC of the energy storage battery satisfies 0.1≤SOC<0.8, the photovoltaic power generation is preferentially used to supply power to the user side equipment, and the remaining power is used to charge the energy storage battery and fed into the grid; when the photovoltaic power generation power is greater than the power of the user side energy consumption equipment, and the SOC of the energy storage battery satisfies 0.8≤SOC≤0.9, the photovoltaic power generation is preferentially used to supply power to the user side equipment, and the remaining power is fed into the grid; when the photovoltaic power generation power is less than the power of the user side energy consumption equipment, and the SOC of the energy storage battery satisfies 0.3≤SOC≤0.9, the photovoltaic power generation and the energy storage battery are used to supply power to the user side equipment at the same time; in the valley value electricity price stage and the SOC of the energy storage battery satisfies 0.1≤SOC<0.7, the power is purchased from the grid to supply power to the user side equipment, and the energy storage battery is charged.

[0015] Preferably, the types of flexible energy consumption equipment include interruptible equipment which can adjust the equipment on-off state at any time within the opening time period, and non-interruptible equipment which must be continuously operated until completion once started in the working period; the scheduling strategy for the flexible energy consumption equipment is that the working time of the interruptible equipment is split to the low electricity price or photovoltaic power generation surplus period on the premise that the total working time is unchanged; the working time of the non-interruptible equipment is shifted to the continuous low electricity price period.

[0016] An electronic device comprises a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the energy operation management method of the optical storage system.

[0017] A computer readable storage medium, the storage medium has a computer program stored thereon; when the computer program is executed by a processor, the energy operation management method of the optical storage system is implemented.

[0018] Compared with the prior art, the beneficial effects of the present application are:

[0019] The technical solution of the present application starts from two aspects of macroscopic system optimization and microscopic battery mechanism, effectively avoids the low efficiency and safety problems caused by the lack of battery operation mechanism analysis in traditional methods, fully considers the actual demand of users and the change of electricity price, formulates the operation management strategy of the optical storage system, helps the energy manager to better realize the optimal allocation of energy, realizes the dual goals of minimizing the operation cost and maximizing the battery life. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 It is the overall working procedure diagram of the present application.

[0021] Figure 2 It is the working procedure diagram of the present application for dynamic weight setting.

[0022] Figure 3 It is the working procedure diagram of the present application for value sorting of two input characteristic parameters with the same value.

[0023] Figure 4 It is the Tent value-dimension diagram of the present application based on population initialization through Tent mapping sequence.

[0024] Figure 5 It is the frequency-Tent value diagram of the present application based on population initialization through Tent mapping sequence.

[0025] Figure 6 It is the working procedure diagram of the present application for prediction model optimization through improved multi-strategy fusion GTO algorithm.

[0026] Figure 7 It is the idealized energy flow diagram of the optical storage system in the present application.

[0027] Figure 8 It is the control flow diagram of the present application for executing different strategies at different time periods within 24 hours.

[0028] Figure 9The workflow schematic diagram of the multi-objective function solving of the improved multi-strategy fusion GTO algorithm of the present application.

[0029] Figure 10 The cost comparison schematic diagram of one specific example of the present application before and after optimization.

[0030] Figure 11 The SOC comparison schematic diagram of one specific example of the present application before and after optimization. DETAILED DESCRIPTION

[0031] Hereinafter, the present application will be further described in conjunction with specific embodiments, and it should be noted that, in the description of the present application, the description of the terms “one embodiment”, “some embodiments”, “an example”, “a specific example”, or “some examples” means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms should not be understood as necessarily referring to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, the person skilled in the art can combine and combine different embodiments or examples described in the present specification.

[0032] In the description of the present application, it should be noted that, for the orientation words, such as the terms “center”, “transverse”, “longitudinal”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “clockwise”, “counterclockwise”, etc. The orientation and positional relationship shown in the drawing is based on the orientation or positional relationship shown in the drawing, and is only for the convenience of describing the present application and simplifying the description, and cannot be understood as limiting the specific protection scope of the present application. The device or element indicated must have a specific orientation, be constructed and operated in a specific orientation.

[0033] It should be noted that the terms “first”, “second” and the like in the specification and claims of the present application are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence.

[0034] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0035] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0036] The terms “comprising” and “having”, and any variations thereof, in the specification and claims of this application are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0037] One aspect of this application provides an energy operation management method for a photovoltaic energy storage system, such as... Figure 1 As shown, one preferred embodiment includes the following steps: acquiring operational data of the photovoltaic-storage system, including user-side energy consumption data, photovoltaic power generation data, electricity price data, and energy storage battery operational data, and extracting features from each operational data point based on the Pearson correlation coefficient. Based on the extracted feature data, a prediction model based on a neural network architecture is constructed, including a user energy demand prediction model that considers both flexible and rigid energy-consuming devices, a photovoltaic power generation prediction model, and an energy storage battery SOC prediction model; and an improved multi-strategy fusion GTO algorithm is used to optimize the network parameters of the prediction model. Based on the optimized prediction model, and combining system power balance constraints and system operation mode constraints, a multi-objective function is constructed with the objectives of minimizing operating costs and maximizing energy storage battery lifespan. The improved multi-strategy fusion GTO algorithm is used to solve the constructed multi-objective function to obtain the charging and discharging strategy of the photovoltaic-storage system and the scheduling strategy of the flexible energy-consuming devices.

[0038] It can be understood that the technical scheme of the present application starts from two aspects of macroscopic system optimization and microscopic battery mechanism, effectively avoids the low efficiency and safety problems caused by the lack of battery operation mechanism analysis in traditional methods, fully considers the actual demand of users and the change of electricity price, formulates the operation and management strategy of the light storage system, helps the energy manager to better realize the optimal allocation of energy, realizes the dual goals of minimizing the operation cost and maximizing the battery life. In order to facilitate understanding, each step of the technical scheme of the present application will be described in detail below.

[0039] In the embodiment, the user-side energy consumption data includes historical load, time, equipment operation, indoor population, temperature and humidity, etc.; the photovoltaic power generation data includes solar radiation intensity, cloud amount, temperature, humidity and historical power generation; the electricity price data mainly includes time-of-use electricity price of buying and selling electricity; and the energy storage battery operation data includes battery capacity, charging and discharging efficiency and historical charging and discharging power.

[0040] It can be understood that for the obtained operation data of the light storage system, there may be situations such as non-uniform time dimension, local data anomaly or missing; therefore, after obtaining the operation data of the light storage system, the obtained data can be preprocessed, specifically including data cleaning, normalization and time alignment. The specific process of preprocessing is a known technology to those skilled in the art, and therefore will not be described in detail here.

[0041] In the embodiment, the process of feature extraction of each operation data of the light storage system is as follows: calculating a plurality of Pearson correlation coefficients for measuring the relationship between each operation data of the light storage system and the user-side energy consumption; setting a threshold coefficient representing strong correlation between the operation data of the light storage system and the user-side energy consumption; and retaining the operation data corresponding to the Pearson correlation coefficient greater than the set threshold coefficient in the plurality of calculated Pearson correlation coefficients as strong correlation feature data.

[0042] It can be understood that the specific value of the threshold coefficient can be selected by those skilled in the art according to actual needs; generally, the value of the Pearson correlation coefficient c is greater than 0.8, and it can be considered that the correlation degree of the input operation data and the user-side energy consumption is relatively high; therefore, the value of the threshold coefficient in the embodiment is preferably 0.8. The specific expression of the Pearson correlation coefficient c is:

[0043] .

[0044] Wherein, x j represents the jth input parameter, i.e. the operation data of the light storage system; y j represents the jth output parameter, i.e. the user-side energy consumption data; J represents the total number of parameters, represents the mean value of the input parameters, represents the mean value of the output parameters.

[0045] In this embodiment, since the user actually uses multiple energy-using devices, and the types and energy-using modes of different energy-using devices are different; at the same time, there are many factors affecting the output results of the prediction model; therefore, the prediction model constructed in this embodiment is a complex model based on different types and energy-using modes, which is easily affected by external environment and user energy-using habits. Therefore, when constructing the prediction model, it is necessary to set dynamic parameter weights for different input characteristic parameters; the setting method of the dynamic parameter weights of each prediction model for different input characteristic parameters is the same, so in this embodiment, the user energy-using demand prediction model will be taken as an example to describe the specific setting process of the dynamic parameter weights in detail.

[0046] Specifically, as shown in FIG. 8, one input characteristic parameter is selected from the N input characteristic parameters that have been determined as a target input characteristic parameter n i ; the value judgment process of the target input characteristic parameter n i is performed: the value of the target input characteristic parameter n ni at t+1 is increased by a set step size Δm, and the values of the remaining input characteristic parameters at t+1 remain unchanged; the energy consumption difference ΔP ni between the energy consumption P(t+1) output by the user energy-using demand prediction model at t+1 and the energy consumption P(t) output at t is calculated, i.e., the required output difference. If the calculated energy consumption difference ΔP i is greater than 0, the target input characteristic parameter n ni is defined as a + value input; if the calculated energy consumption difference ΔP i is less than 0, the target input characteristic parameter n ni is defined as a - value input; if the calculated energy consumption difference ΔP i is 0, the target input characteristic parameter n i is defined as a no-value input. The target input characteristic parameter n i is reselected from the remaining input characteristic parameters and the above value judgment process is repeated until all N input characteristic parameters complete the value judgment. The energy consumption difference corresponding to different input characteristic parameters in the value judgment process is sorted according to the value, and the parameter weight is set based on the sorting result.

[0047] It can be understood that the specific value of the step length Δm can be set by the actual needs of those skilled in the art, for example, it can be 10% of the value of the target input characteristic parameter at time t. When setting the parameter weight, a positive weight can be given to the + value input; a negative weight can be given to the - value input; and the value input can be ignored. Therefore, in the technical solution of the present application, when the values are sorted, the + value input is placed in front, and the - value input is placed behind. For the value sorting between multiple + value inputs, the corresponding energy consumption difference can be compared, that is, the greater the energy consumption difference, the greater the degree of positive influence of the input characteristic parameter on the user energy demand prediction model, and therefore the + value input with a larger energy consumption difference can be placed in front of the + value input with a smaller energy consumption difference. Similarly, the value arrangement can also be performed between multiple - value inputs.

[0048] It should be noted that when multiple + value inputs and multiple - value inputs are sorted, there can be two + value inputs and - value inputs with the same energy consumption difference. At this time, for two input characteristic parameters with the same value, there can be two situations, one is that the influence of the two input characteristic parameters on the user energy demand prediction model is indeed the same, and the other is that the influence of the two input characteristic parameters on the user energy demand prediction model is the same under the current step length increment Δm, but can be different under other step length increments. Therefore, in order to ensure the accuracy of the construction of the user energy demand prediction model, the value sorting of the two input characteristic parameters with the same value can be further performed, which will be described in detail below.

[0049] Specifically, as shown in Figure 3 , the value sorting process when the energy consumption differences corresponding to the two input characteristic parameters are the same is as follows: define the two input characteristic parameters as a first input characteristic parameter n i and a second input characteristic parameter n i+1 , and set a certain amount of reduction for the step length Δm applied to the two input characteristic parameters in the above value judgment process, which can be represented as Δi. After the value judgment process of increasing the step length value of the two input characteristic parameters to Δm-Δi, the first energy consumption difference ΔP i corresponding to the first input characteristic parameter n ni and the second energy consumption difference ΔP i+1 corresponding to the second input characteristic parameter n ni+1 can be obtained. The first energy consumption difference ΔP ni and the second energy consumption difference ΔP ni+1 are compared; if the first energy consumption difference ΔP ni is greater than the second energy consumption difference ΔP ni+1 , the first input characteristic parameter n i is defined as a + value input, and the second input characteristic parameter n i+1 is defined as a - value input.i The value is greater than the second input feature parameter n i+1 If the first energy consumption difference ΔP ni Less than the second energy consumption difference ΔP ni+1 Define the second input feature parameter n i+1 The value is greater than the first input feature parameter n i If the first energy consumption difference ΔP ni Equal to the second energy consumption difference ΔP ni+1 The next round of value judgment process involves increasing the step size of Δm-2Δi for both input feature parameters, and repeating the above process until Δm-kΔi=0 in the k-th round of value judgment process, and the first input feature parameter n in the (k-1)-th round of value judgment process is... i The corresponding first energy consumption difference ΔP ni It is still equal to the second input feature parameter n i+1 The corresponding second energy consumption difference ΔP ni+1 We define that the two input feature parameters have the same value.

[0050] It is understandable that the specific value of the step size reduction Δi can be selected according to the actual needs of those skilled in the art. For example, the value of the step size reduction Δi can be 10% to 20% of the step size Δm.

[0051] In this embodiment, the prediction models for user energy demand, photovoltaic power generation, and energy storage battery SOC are generally constructed based on a neural network architecture. However, due to the large amount of operational data in the photovoltaic-energy storage system, directly constructing the prediction models using a neural network architecture may lead to the constructed models easily getting trapped in local optima during use. Therefore, optimization of the neural network architecture is necessary. In this embodiment, an improved multi-strategy fusion GTO algorithm is preferably used to optimize the neural network architecture. For ease of understanding, the acquisition process of the improved multi-strategy fusion GTO algorithm will be described in detail below.

[0052] Specifically, the multi-strategy fusion GTO algorithm is improved through the following aspects:

[0053] (1) Improved Tent initialization population.

[0054] Population initialization includes setting the population size N0, search space dimension D, and maximum number of iterations T. By introducing a sinusoidal perturbation into the traditional Tent map, we avoid getting trapped in small-period loops and fixed points, thus improving population diversity and global search capability. The position formula for each individual in the population is:

[0055] .

[0056] where x n represents the original position of the nth individual, x n+1 represents the updated position of the nth individual after applying the sinusoidal disturbance, a represents a control parameter, and β represents a disturbance factor.

[0057] In one specific example, the population size N0may be taken as 2000, the search space dimension D as 2, and the updated initialization population position after introducing the sinusoidal disturbance as shown in Figure 4 and Figure 5 It can be seen from the figure that the improved population distribution is more uniform and more in line with the actual requirements.

[0058] (2) Nonlinear parameter factor introduction: the linearly decreasing energy parameter E is changed to a nonlinearly decreasing energy parameter E, and the specific expression is as follows:

[0059] E=E max -(E max -E min )(t0 / T) γ .

[0060] where γ represents a nonlinear parameter factor, which is a uniformly distributed random number with a value range of [0, 2], and t0represents the current iteration number.

[0061] (3) Reverse learning strategy: in each iteration process, a reverse solution X c of the current individual X i is generated with a set probability P R to expand the search range and compare the fitness of the reverse solution to complete the selection of the optimal solution; the expression of the reverse solution X R is as follows:

[0062] X R =u b +l b -X i .

[0063] where u b and l b represent the upper and lower bounds of the solution space, respectively.

[0064] (4) Cauchy mutation strategy: the current optimal individual X best is disturbed by using Cauchy distribution Cauchy(0, 1) to prevent falling into local optimum, and the expression of the new solution X new after disturbance is as follows:

[0065] X new =X best + X best ×Cauchy(0, 1).

[0066] (5) GTO position update: Mapping the chaotic sequence to the solution space, specifically represented by the following expression:

[0067] X i =l b +(u b -l b )×x i .

[0068] In the formula, x i This represents the i-th individual in the initial population under the chaotic sequence mapping.

[0069] In this embodiment, optimizing the neural network architecture using the improved multi-strategy fusion GTO algorithm mainly involves treating the network parameters of the neural network architecture as prey, and obtaining the optimal solution through multiple encirclements by the GTO algorithm. For ease of understanding, a detailed description will follow.

[0070] Specifically, such as Figure 6 As shown, after feature extraction and dynamic input parameter weight calculation are completed using the Pearson correlation coefficient, an initial population is generated using an improved Tent chaotic sequence; the target value X of the population is then calculated. best And update the energy parameter E for prey escape.

[0071] If the absolute value of the energy parameter E for prey escape is greater than or equal to 1, the algorithm enters the global search phase and updates the population position. During the global search phase, if the probability r of prey escape is greater than or equal to 0.5, the algorithm strengthens its influence on the prey's position, i.e., it adopts an enhanced trapping strategy; otherwise, the influence on the prey's position weakens, i.e., it adopts a weakened trapping strategy.

[0072] If the absolute value of the prey's escape energy parameter E is less than 1, or if the global search phase is completed, the algorithm enters the local search phase and updates the population position. During the local exploration phase, the four current optimal solutions are sequentially selected as the attacker X. attacker Surrounder X chaser The driver X barrier And the Chaser X driver The location is used to surround and hunt the prey; among them, the attacker X... attacker The optimal solution X for the population best .

[0073] The geometric center of the four hunting characters represents the most likely location of the prey, and serves as the convergence point X for updating the positions of the hunting characters in the next round of hunting. i (t+1); where the convergence point X i The expressions for updating the positions of the capture roles in (t+1) and the next round of capture are as follows:

[0074] X i(t+1)=(X attacker +X chaser +X barrier +X driver ) / 4.

[0075] X attacker =X attacker -a1×|c1×X attacker -m1×X i |

[0076] X chaser =X chaser -a2×|c2×X chaser -m2×X i |

[0077] X barrier =X barrier -a3×|c3×X barrier -m3×X i |

[0078] X driver =X driver -a4×|c4×X driver -m4×X i |

[0079] In the formula, a1, a2, a3, and a4 represent attacker X, respectively. attacker Surrounder X chaser Drive-away X barrier And the Chaser X driver The corresponding random scaling factors; c1, c2, c3, and c4 represent attacker X respectively. attacker Surrounder X chaser The driver X barrier And the Chaser X driver Influence factors during prey pursuit; m1, m2, m3, and m4 represent the attacker X, respectively. attacker Surrounder X chaser The driver X barrier And the Chaser X driver The corresponding Tent-value vector.

[0080] During the capture process, if the probability of the prey escaping, Rand, is less than the set probability threshold P... c Reverse learning is performed on the four encirclement roles to generate a new solution X. new That is, the reverse solution; otherwise, a new solution X is obtained by adding perturbation to the attacker using the Cauchy distribution. new The new solution X is obtained. new With the best solution X best Make a judgment if the new solution X new Less than the optimal solution X best, the value of the new solution X new is assigned to the update of the best solution X best , that is, X best =X new ; otherwise, the value of the best solution X best remains unchanged.

[0081] In the completion of the trapping process, the position of the prey is updated and it is judged whether the current trapping number reaches the maximum trapping number, that is, the single trapping process is an iteration process, and it is judged whether the current iteration number reaches the maximum iteration number T; if the maximum iteration number T is reached, the best network parameter corresponding to the prey position is output, otherwise the above process is repeated to execute a round of iteration process.

[0082] It can be understood that in the process of constructing the initial model, the obtained operation data of the light storage system is divided into training set and test set according to the proportion of 9:1, so as to train the constructed initial model through the training set, and verify the precision of the trained initial model through the test set. When the initial model is optimized by the improved multi-strategy fusion GTO algorithm, the MAE (Mean Absolute Error) of the artificial neural network corresponding to the trained initial model can be set as the fitness of the multi-strategy fusion GTO algorithm. Through the search of the multi-strategy fusion GTO algorithm, the input parameter weight value and the hidden layer threshold value that make the prediction error of the artificial neural network minimum can be obtained. The expression of the fitness function fitness is as follows:

[0083] .

[0084] In the formula, N x represents the number of test set samples, represents the predicted output value of the initial model, represents the actual value of the test set sample.

[0085] In this embodiment, when constructing the user energy demand prediction model, both flexible energy-using equipment and rigid energy-using equipment need to be considered. The rigid energy-using equipment refers to equipment with fixed power demand, which cannot participate in demand response, and cannot change the equipment state frequently according to the user's will, that is, the equipment has zero dispatchable time; for example, refrigerators and freezers. The energy consumption mathematical expression of the rigid energy-using equipment can be defined as:

[0086] .

[0087] In the formula, P base (t) represents the total energy consumption of the L rigid energy-using equipment in the time period t, represents the rated power of the lth rigid energy-using equipment, and I l(t) represents the running state of the lth rigid energy-using equipment at time period t, taking values of {0, 1}, 0 represents that the equipment is closed, and 1 represents that the equipment is opened.

[0088] The flexible energy-using equipment can be divided into interruptible equipment and non-interruptible equipment according to specific application scenarios. The interruptible equipment refers to an energy-using equipment that can adjust the equipment on-off state at any time within a defined opening time period, but the total working time and power are fixed within the time period. For example, electric water heaters and air conditioners. The non-interruptible equipment refers to an energy-using equipment that must be continuously operated until completion once starting within a defined working time period. For example, washing machines and dishwashers.

[0089] The power mathematical expression of the interruptible equipment is defined as follows:

[0090] .

[0091] .

[0092] In the formula, P(t) represents the total power of the interruptible equipment within the time period t, and M represents the total number of the interruptible equipment, Pm represents the rated power of the mth interruptible equipment, (t) represents the running state of the mth interruptible equipment at time period t, taking values of {0, 1}, 0 represents that the equipment is closed, and 1 represents that the equipment is opened; and respectively represent the start time and end time of the mth interruptible equipment allowed to run, Tm represents the total working time of the mth interruptible equipment.

[0093] The power mathematical expression of the non-interruptible equipment is defined as follows:

[0094] .

[0095] .

[0096] .

[0097] .

[0098] In the formula, P(t) represents the total power of the non-interruptible equipment within the time period t, and Q represents the total number of the non-interruptible equipment, Pq represents the rated power of the qth non-interruptible equipment, (t) represents the running state of the qth non-interruptible equipment at time period t, taking values of {0, 1}, 0 represents that the equipment is closed, and 1 represents that the equipment is opened; and respectively represent the start time and the end time of the qth non-interruptible device allowed to run, and respectively represent the start running time and the required continuous running time of the qth non-interruptible device.

[0099] Specifically, the input layer of the user energy demand prediction model optimized by the improved multi-strategy fusion GTO algorithm is set to 9, the output layer is set to 1, and therefore the range of the hidden layer nodes is [4, 20]; the total error of the output prediction value of the same data sample and the number of iterations is calculated one by one to determine that the optimal number of hidden layers is 12. The input feature parameters corresponding to the user energy demand prediction model include the device switching state, the time-of-use price, the historical load, the solar radiation at the previous moment, the solar radiation at the current moment, the time, the temperature, the battery capacity, the number of indoor personnel, and the humidity. Among them, the weight corresponding to the device switching state is 0.883, the weight corresponding to the time-of-use price is 0.872, the weight corresponding to the historical load is 0.87, the weight corresponding to the solar radiation at the previous moment is 0.861, the weight corresponding to the time is 0.857, the weight corresponding to the temperature is 0.842, the weight corresponding to the solar radiation at the current moment is 0.833, the weight corresponding to the battery capacity is 0.814, the weight corresponding to the number of indoor personnel is 0.803, and the weight corresponding to the humidity is 0.783.

[0100] In this embodiment, when constructing the photovoltaic power generation prediction model, the power generation of the photovoltaic system is affected by the solar radiation intensity, the photovoltaic panel area, and the power generation efficiency, and has strong instability and discontinuity; therefore, in actual engineering, the power generation efficiency of the photovoltaic system has a large fluctuation range, and in order to facilitate calculation, the power model of the photovoltaic power generation system can be defined as follows:

[0101] .

[0102] .

[0103] .

[0104] In the formula, P represents the power generation power of the photovoltaic power generation system, E represents the solar radiation intensity, S represents the photovoltaic area, η represents the photovoltaic photoelectric conversion efficiency, T represents the photovoltaic cell temperature, and T represents the environmental temperature. pv pv pv sur

[0105] ​​​​Specifically, the input layer of the photovoltaic power generation prediction model optimized by the improved multi-strategy fusion GTO algorithm is set to 5, the output layer is set to 1, and the optimal number of hidden layers is determined to be 4. In the photovoltaic power generation prediction model, the corresponding input feature parameters include the current solar radiation, historical power generation, cloud cover, temperature, time and season. The obtained operation data of the light storage system is divided into training set and test set according to the proportion of 9:1, so as to realize the construction and test of the photovoltaic power generation prediction model; wherein the weight of the current solar radiation is 0.875, the weight of the historical power generation is 0.863, the weight of the cloud cover is 0.858, the weight of the temperature is 0.842, the weight of the time is 0.814, and the weight of the season is 0.810.

[0106] In this embodiment, when the construction of the energy storage battery SOC prediction model is performed, the SOC (state of charge) of the energy storage battery represents the ratio of the remaining capacity to the rated capacity after the energy storage battery normally operates and completes a single charging or discharging operation, and the mathematical expression is:

[0107] .

[0108] In the formula, SOC(t+1) represents the SOC value of the energy storage battery at t+1 time, SOC(t) represents the SOC value of the energy storage battery at t time, P b (t) represents the charging and discharging power of the energy storage battery in t time period, P b (t) < 0 when charging, and P b (t) > 0 when discharging; δ b is the attenuation coefficient for measuring the capacity of the energy storage battery, which represents the self-discharge rate of the energy storage system in Δt time period; η b represents the charging and discharging efficiency, which is 0.97 when charging and 0.98 when discharging; E r represents the rated capacity of the energy storage battery.

[0109] It should be noted that in order to ensure that the energy storage battery can be normally used, the SOC of the energy storage battery at t time should satisfy: 0.1 ≤ SOC(t) ≤ 0.9.

[0110] Specifically, the input layer of the energy storage battery SOC prediction model optimized by the improved multi-strategy fusion GTO algorithm is set to 5, the output layer is set to 1, and the optimal hidden layer number is determined to be 4. In the energy storage battery SOC prediction model, the corresponding input feature parameters include historical SOC, current charging and discharging power, historical charging and discharging power, temperature and time. The obtained operation data of the light storage system is divided into training set and test set according to the proportion of 9:1, so as to realize the construction and test of the energy storage battery SOC prediction model; wherein the weight of historical SOC is 0.894, the weight of current charging and discharging power is 0.887, the weight of historical charging and discharging power is 0.875, the weight of temperature is 0.87, and the weight of time is 0.862.

[0111] In this embodiment, the system power balance constraint is mainly the power balance constraint between the user side equipment power P load , the power generation power P pv of the photovoltaic system, the charging and discharging power P b of the energy storage battery and the grid power P grid , which is specifically expressed by the following expression:

[0112] P load (t)=P pv (t)+P b (t)+P grid (t)。

[0113] P load (t)=P base (t)+P nc (t)+P nd (t)。

[0114] In this embodiment, the energy flow of the system operation mode in the ideal state is as follows: Figure 7As shown, when the photovoltaic (PV) system generates sufficient power and the electricity price reaches its peak, the PV system's power can supply electricity to the energy storage battery, power user loads (energy-consuming equipment), and sell electricity to the grid. When the PV system generates moderate power and the electricity price reaches its valley, the PV system's power can supply electricity to user loads and charge the energy storage battery. When the PV system generates insufficient power but the energy storage battery's SOC is sufficient, the PV system and the energy storage battery can simultaneously supply electricity to user loads. When the PV system generates insufficient power, the energy storage battery's SOC is not fully charged, and the electricity price is at its valley, electricity can be purchased from the grid to supply electricity to user loads and charge the energy storage battery. Based on the above system operation modes, the operation mode constraints mainly include constraints for PV surplus power grid connection operation mode, PV charging battery operation mode, PV simultaneously charging battery and grid connection operation mode, battery discharging power supply operation mode, grid power purchase operation mode, and grid power purchase and battery charging operation mode. For ease of understanding, each operation mode will be explained in detail below.

[0115] Specifically, the constraints for the photovoltaic surplus power grid connection operation mode are applicable to scenarios such as midday on sunny days, when energy storage batteries are fully charged, and when electricity prices are relatively high; the operating conditions for this mode are: P pv (t)>P load (t), P b (t)=0, P grid (t) < 0 (electricity sales); the power balance equation under this mode is: P pv (t)=P load (t)+|P grid (t)|; The constraint is: SOC(t)≥0.9, P grid (t)≤P grid (t) max P pv (t)≤P pv (t) max Among them, P grid (t) max and P pv (t) max These represent the maximum grid power and the photovoltaic power generation, respectively.

[0116] The constraints for the photovoltaic-to-battery charging operation mode are applicable to scenarios where photovoltaic power generation is sufficient, the battery is not fully charged, and the electricity price is low; the operating conditions for this mode are: P pv (t)>P load (t), P b (t) < 0 (energy storage battery charging), P grid (t)=0 (neither selling nor buying electricity); the power balance equation under this mode is: P pv (t)=P load(t) + |P b (t) + |P b (t) + |P b (t) + |P min (t) + |P pv (t) + |P b (t) + |P max (t) + |P b (t) + |P min (t) + |P b (t) + |P max (t) + |P

[0117] For the photovoltaic charging and grid-connected operation mode constraint, it is suitable for scenarios such as photovoltaic power surplus, partial charging of energy storage battery, moderate electricity selling price, etc. The operation condition of this mode is: P pv (t) > P load (t), P b (t) < 0 (charging of energy storage battery), P grid (t) < 0 (selling electricity); the power balance equation in this mode is: P pv (t) = P load (t) + |P b (t) + |P grid (t) + |P b (t) + |P b (t) + |P min (t) + |P grid (t) + |P grid (t) + |P max .

[0118] For the battery discharging power supply operation mode constraint, it is suitable for scenarios such as night or cloudy days, peak period of electricity price, insufficient photovoltaic power generation, etc. The operation condition of this mode is: P pv (t) < P load (t), P b (t) > 0 (discharging of energy storage battery), P grid (t) = 0 (neither selling electricity nor buying electricity); the power balance equation in this mode is: P pv (t) + P b (t) = P load (t); the constraint condition is: SOC(t) > 0.1, P b (t) ≤ P b (t) max (t) ≥ P b (t) b (t) min .

[0119] For the grid power purchase operation mode constraint, it is applicable to the scenarios of energy storage battery power depletion, insufficient photovoltaic power generation, and power supply guarantee. The operation condition of this mode is: P pv (t) + P b (t) < P load (t), P grid (t) > 0 (power purchase); the power balance equation in this mode is: P load (t) = P pv (t) + P b (t) + P grid (t); the constraint condition is: SOC(t) ≤ 0.1, 0 ≤ P grid (t) ≤ P grid (t) max .

[0120] For the grid power purchase while charging the battery operation mode constraint, it is applicable to the scenarios of low electricity price period, battery charging, and continuous equipment energy demand. The operation condition of this mode is: P pv (t) < P load (t), P b (t) < 0 (energy storage battery charging), P grid (t) > 0 (power purchase); the power balance equation in this mode is: P load (t) + |P b (t)| = P pv (t) + P grid (t); the constraint condition is: SOC(t) < 0.8, M up (t) ≤ M up (t) min , P grid (t) ≤ P grid (t) max . Wherein, M up (t) and M up (t) min represent the power purchase price and the low valley price respectively.

[0121] It should be known that the above operation modes can be automatically selected by means of an optimization algorithm in actual application, the range of each power variable is limited in the constraint condition, and a correct cost model is established, and the operation mode can be selected according to the cost minimization principle and the energy storage battery life optimization, so as to ensure the optimal strategy selection of energy operation. In order to facilitate understanding, the construction process of the multi-objective function with the operation cost minimization and the energy storage battery life maximization as the target will be described in detail below.

[0122] In this embodiment, the operation cost C includes the operation cost C b of the energy storage battery, the operation cost C pv of the photovoltaic, and the power purchase cost Cp i.e. C=C b +C pv +C p The operating cost C of the energy storage battery b The operating cost C of the photovoltaic pv And the electricity purchase cost C of the user p The specific expressions are as follows:

[0123] .

[0124] .

[0125] .

[0126] .

[0127] .

[0128] .

[0129] In the formula, C1(t) represents the electricity purchase cost, C2(t) represents the electricity sale income, C3(t) represents the battery loss cost, P bg And P pvg Respectively represent the on-grid power of the energy storage battery and the photovoltaic system, P gb And P gp Respectively represent the power provided by the power grid to the energy storage battery and the user-side energy consumption equipment, M down Represents the electricity sale price, C4 represents the initial cost, Δt represents the operating time, and C8(t) represents the photovoltaic operation and maintenance cost.

[0130] It can be understood that the lower the SOC value of the energy storage battery, the greater the loss of the energy storage battery; therefore, the maximization of the service life of the energy storage battery can be converted into the maximization of the SOC; and the purpose of the multi-objective function is to achieve the balance between the minimization of the operating cost C and the maximization of the SOC of the energy storage battery.

[0131] Specifically, it can be expressed by the following formula:

[0132] Z(t) = ω1 x min(C) + ω2 x max(SOC).

[0133] In the formula, Z(t) represents the multi-objective function, and ω1 and ω2 respectively represent the weight coefficients of the minimization of the operating cost and the maximization of the SOC of the energy storage battery.

[0134] For the convenience of understanding, the charging and discharging strategy of the photovoltaic energy storage system will be described in detail below: when the photovoltaic power P pv Is greater than the user-side energy consumption equipment power P loadFurthermore, when the SOC of the energy storage battery satisfies the condition that 0.1≤SOC<0.8, photovoltaic power generation prioritizes powering user-side equipment, and the remaining power is used to charge the energy storage battery and connect to the grid. In this case, the weight coefficients ω1 and ω2 in the multi-objective function increase.

[0135] In photovoltaic power generation P pv Greater than the power P of the user-side energy-consuming equipment load Furthermore, when the SOC of the energy storage battery satisfies the condition that 0.8≤SOC≤0.9, photovoltaic power generation prioritizes powering user-side equipment, with the remaining power being fed into the grid; at this time, the weight coefficients ω1 and ω2 in the multi-objective function increase.

[0136] In photovoltaic power generation P pv Less than the power P of the user-side energy-consuming equipment load Furthermore, when the SOC of the energy storage battery satisfies the condition that 0.3≤SOC≤0.9, photovoltaic power generation and energy storage battery simultaneously supply power to user-side equipment; at this time, the weight coefficients ω1 and ω2 in the multi-objective function decrease.

[0137] During off-peak electricity pricing periods and when the SOC of the energy storage battery satisfies 0.1≤SOC<0.7, electricity is purchased from the grid to supply power to user-side equipment and to charge the energy storage battery; at this time, the weight coefficients ω1 and ω2 in the multi-objective function increase.

[0138] Understandably, the scheduling strategy for flexible energy-consuming equipment is as follows: for interruptible equipment, the working time is split into periods of low electricity prices or surplus photovoltaic power generation, while the working time of non-interruptible equipment is shifted to periods of continuous low electricity prices.

[0139] To make it easier to understand, the following will use a 24-hour day as an example to describe in detail the operation process of the entire system.

[0140] like Figure 8 As shown, the system's power variables are first initialized. Given the current time t and the start time Start, the control strategy is executed continuously from the start time Start until 24 hours have passed. The specific strategy execution process is as follows:

[0141] Based on the start time Start, the current time-of-use electricity price M is obtained. up And electricity sales price M down .

[0142] Determine if the current time t falls within the 23:00-7:00 time period, which is the off-peak electricity pricing period at night. If the current time t falls within the 23:00-7:00 time period, electricity needs to be purchased from the grid to supply power to user-side energy-consuming equipment and energy storage batteries. Then, the grid power P at this time is... gridEqual to the power P supplied by the grid to the energy storage battery gb The power P supplied by the power grid to the user-side energy-consuming equipment gp The sum of, i.e., P grid =P gb +P gp .

[0143] If the current time t falls outside the 23:00-7:00 time period, then determine if the current time t falls within the 7:00-8:00 time period. This period represents a time of low sunlight intensity, meaning the photovoltaic system is already operating during this time, but its power generation is low. If the current time t falls within the 7:00-8:00 time period, then determine the photovoltaic system's power generation P. pv and the charging and discharging power P of the energy storage battery b Does it meet the power P requirement of the user-side energy-consuming equipment? load Demand. If satisfied, power can be supplied to user-side energy-consuming equipment through a combination of photovoltaic systems and energy storage batteries, at which point P... load =P pv +P bp Among them, P bp This represents the power supplied by the energy storage battery to the user-side energy-consuming equipment. If this is insufficient, then the photovoltaic system, energy storage battery, and power grid must work together to supply power to the user-side equipment, i.e., P. load =P pv +P bp +P gp .

[0144] If the current time t falls outside the 7:00-8:00 time period, then determine if the current time t falls within the 8:00-19:00 time period. Electricity prices are higher during this period, and it covers both the periods of sufficient and insufficient power generation by the photovoltaic system. If the current time t falls within the 8:00-19:00 time period, then determine the photovoltaic system's power generation P. pv Does it meet the charging and discharging power P of the energy storage battery? b and the power P of user-side energy-consuming equipment load Demand. If satisfied, the photovoltaic system supplies power to user-side energy-consuming equipment, simultaneously charging the energy storage battery, and the remaining electricity is fed into the grid. At this point, P... load =P pv -P pvg -P pvb Among them, P pvg P represents the power that the photovoltaic system sells to the grid, i.e., the power supplied to the grid; pvb This represents the power at which the photovoltaic system charges the energy storage battery. If this condition is not met and the photovoltaic system stops operating, then the charging and discharging power P of the energy storage battery is determined. bwhether the power demand of the user-side energy consuming equipment is met; if yes, the user-side energy consuming equipment is powered by the energy storage battery, at this time, there is P load =P bp ; if not, the user-side energy consuming equipment is powered by the energy storage battery and the power grid, at this time, there is P load =P bp +P gp .

[0145] If the current time t is located outside the time period of 8:00-19:00, at this time the photovoltaic system has stopped working, so for the user-side energy consuming equipment only the energy storage battery and / or the power grid can be used for power supply. Then it can be judged whether the charge-discharge power P b of the energy storage battery meets the power demand of the user-side energy consuming equipment; if yes, the user-side energy consuming equipment is powered by the energy storage battery, at this time, there is P load =P bp ; if not, the user-side energy consuming equipment is powered by the energy storage battery and the power grid, at this time, there is P load =P bp +P gp .

[0146] In the process of completing the above strategy execution process, it can be judged whether the end condition is reached, that is, whether the strategy execution has been continuously for 24 hours; if the end condition is reached, at this time the optimal operation strategy of the system in the next 24 hours can be output, that is, the operation strategy in the next 24 hours is predicted by the prediction model.

[0147] In the embodiment, it can be known from the above charge-discharge strategy and the scheduling strategy of the flexible energy consuming equipment that the solving process of the multi-objective function needs to involve many parameters and has a large amount of calculation, which makes the obtaining process of the optimal solution easily fall into local optimum; therefore, in the technical scheme of the present application, the improved multi-strategy fusion GTO algorithm can be continued to be used to solve the multi-objective function, and the specific solving process is similar to the foregoing content, for the convenience of understanding, which will be simply described below.

[0148] Specifically, as shown in FIG. 8, in the starting stage, the problem is first defined, including determination of the decision variable, establishment of the objective function and construction of the constraint condition and the like. Figure 9 Then population initialization is performed, and an improved Tent mapping with sinusoidal disturbance can be used to generate an initial position for each individual, so that the population can be more uniformly distributed in the solution space.

[0149]

[0150] ​Then the constraint calculation is performed for each individual to check whether the set hard limit is violated; then the adaptive moment estimation method is used to calculate the descent direction of each individual on the objective function for subsequent fine-tuning.

[0151] Then all the current individuals are sorted according to the non-dominated relationship, and the crowding degree between adjacent individuals is calculated, which can measure whether there is enough space around the individual to maintain the diversity of the solution. After the non-dominated sorting and crowding degree calculation are completed, it is judged whether the end condition is met. If not, the non-dominated sorting and crowding degree calculation are continued. If yes, the subsequent steps are continued.

[0152] After the non-dominated sorting and crowding degree calculation are completed, the non-linear energy parameter E is updated and the hunting process is executed. Through the position updating strategy, the reverse learning strategy and the Cauchy mutation strategy, the iteration process is executed to obtain a new population. The objective function value and the constraint violation degree of the new population are calculated, and the offspring individuals and the parent individuals in the new population are merged to obtain a mixed population.

[0153] Then the non-dominated sorting and crowding degree calculation are performed on the mixed population again, and a specified number of excellent individuals are selected from the mixed population to form a global optimal solution, and the Pareto front set in the external archive is updated synchronously.

[0154] The iteration is repeated until the number of iterations reaches the maximum number of iterations, and all optimal individuals in the external archive that are not dominated by any solution are output as the Pareto front solution set.

[0155] It can be understood that each individual in the obtained Pareto front solution set represents a network weight value and system operation strategy combination that achieves the best balance between running cost and energy storage battery life. Those skilled in the art can select a scheme that best meets the actual demand from the Pareto front solution set and directly issue it to the light storage system.

[0156] For the convenience of understanding, the technical scheme of the present application will be described in detail below by taking a comprehensive light storage system of a household photovoltaic- energy storage system-grid-load as an example.

[0157] The initial parameter settings of the devices in the comprehensive light storage system are shown in the following table:

[0158]

[0159] Taking 24 hours of a certain day as an example, the minimum step size Δt=0.5h of the device operation is used to start and stop the device; the specific grid power purchase and power sale prices are shown in the following table:

[0160]

[0161] The operation of the light storage system, the scheduling strategy of the flexible energy-using equipment and the related parameters obtained before and after optimization by means of the multi-strategy fusion GTO algorithm are listed as follows:

[0162] The scheduling strategies of the flexible energy-using equipment before and after optimization are shown in the following table:

[0163]

[0164] The operation strategies of the light storage system before and after optimization are shown in the following table:

[0165]

[0166] According to the above table, the comparison diagrams of the operation efficiency of the light storage system and the service life of the energy storage battery before and after optimization can be drawn as shown in Figure 10 and Figure 11 .

[0167] As can be seen from Figure 10 , the operation efficiency of the light storage system and the service life of the battery are significantly improved by the multi-strategy fusion optimization algorithm: the optimized operation strategy of the light storage system ensures photovoltaic charging during the period of 9:00-15:00, thereby effectively reducing the demand for purchased electricity; charging from the grid during the valley period of 22:00-24:00 reduces the electricity cost; during the electricity peak period of 17:30-18, the battery is preferentially discharged to supply power, avoiding high-priced electricity purchase, and the operation cost is reduced by 92.3%.

[0168] As can be seen from Figure 11 , the SOC of the optimized energy storage battery is always maintained in the healthy range of 0.3-0.85, avoiding damage to the energy storage battery caused by deep charging and discharging; and the fluctuation range of the SOC of the energy storage battery is reduced from ±0.35 before optimization to ±0.28, with a reduction of 20%, thereby helping to prolong the service life of the battery by formulating a reasonable charging and discharging cycle.

[0169] Another aspect of the present application provides an electronic device, wherein a preferred embodiment includes a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program to realize the energy operation management method of the light storage system described above.

[0170] Still another aspect of the present application provides a computer-readable storage medium, wherein a preferred embodiment is that: the storage medium stores a computer program; when the computer program is executed by a processor, the energy operation management method of the light storage system described above is realized.

[0171] The foregoing describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-described embodiments, and the above-described embodiments and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A method for energy operation management of a photovoltaic-storage system, characterized in that, Includes the following steps: Acquire operational data of the photovoltaic-storage system, including user-side energy consumption data, photovoltaic power generation data, electricity price data, and energy storage battery operation data, and extract features from each operational data of the photovoltaic-storage system based on the Pearson correlation coefficient; Based on the extracted feature data, a prediction model based on a neural network architecture is constructed, including a prediction model for user energy demand that takes into account both flexible and rigid energy-consuming equipment, as well as a photovoltaic power generation prediction model and an energy storage battery SOC prediction model; and an improved multi-strategy fusion GTO algorithm is used to optimize the network parameters of the prediction model. Based on the optimized prediction model, and combined with system power balance constraints and system operation mode constraints, a multi-objective function is constructed with the objectives of minimizing operating costs and maximizing the lifespan of energy storage batteries. An improved multi-strategy fusion GTO algorithm is used to solve the constructed multi-objective function, thereby obtaining the charging and discharging strategy of the photovoltaic energy storage system and the scheduling strategy of flexible energy-consuming equipment. The prediction model requires setting parameter weights for different input feature parameters. The specific parameter weight setting process is as follows: Select one input feature parameter from the N known input feature parameters as the target input feature parameter; The process of judging the value of the target input feature parameters: Set the step size for the target input feature parameters at time t+1. The value of m is increased, while the values ​​of the remaining input feature parameters remain unchanged at time t+1; Calculate the difference between the output of the prediction model at time t+1 and time t; If the calculated difference in output values ​​is greater than 0, the target input feature parameter is defined as a +value input; if the calculated difference in output values ​​is less than 0, the target input feature parameter is defined as a -value input; if the calculated difference in output values ​​is 0, the target input feature parameter is defined as a no-value input. From the remaining input feature parameters, select target input feature parameters again and repeat the above value judgment process until all input feature parameters have been valued. The values ​​are ranked according to the difference in output values ​​corresponding to different input feature parameters in the value judgment process, and the parameter weights are set based on the ranking results.

2. The energy operation management method for a photovoltaic-storage system as described in claim 1, characterized in that, The process of feature extraction for various operational data of the photovoltaic-storage system is as follows: Calculate multiple Pearson correlation coefficients to measure the relationship between various operational data of the photovoltaic-storage system and user-side energy consumption; Set a threshold coefficient to characterize the strong correlation between the operating data of the photovoltaic storage system and the energy consumption on the user side; The operational data corresponding to Pearson correlation coefficients that are greater than a set threshold coefficient among the calculated Pearson correlation coefficients are retained as strongly correlated feature data.

3. The energy operation management method for a photovoltaic-storage system as described in claim 1, characterized in that, The value ranking process is as follows when the difference between the output values ​​corresponding to two input feature parameters is the same: Define two input feature parameters as the first input feature parameter and the second input feature parameter, and perform processing on both input feature parameters. m- The value judgment process of increasing the step size of i; If the difference between the first output quantity corresponding to the first input feature parameter is greater than the difference between the second output quantity corresponding to the second input feature parameter, then the value of the first input feature parameter is defined as greater than that of the second input feature parameter. If the difference between the first output quantity corresponding to the first input feature parameter is less than the difference between the second output quantity corresponding to the second input feature parameter, the value of the second input feature parameter is defined as greater than that of the first input feature parameter. If the difference between the first output values ​​corresponding to the first input feature parameter is equal to the difference between the second output values ​​corresponding to the second input feature parameter, then both input feature parameters are processed. m-2 The next round of value judgment process increases the step size of i, until the k-th round of value judgment process has... mk If i=0, and the difference between the first output values ​​corresponding to the first input feature parameter is still equal to the difference between the second output values ​​corresponding to the second input feature parameter during the (k-1)th round of value determination, then the two input feature parameters are defined as having the same value; where... i represents the step size reduction.

4. The energy operation management method for a photovoltaic-storage system as described in any one of claims 1-3, characterized in that, The improved multi-strategy fusion GTO algorithm was obtained by including the following: When initializing the population via Tent mapping, a sinusoidal perturbation is introduced; Change the linearly decreasing energy parameter to a non-linear decreasing one; In each iteration, backward learning is performed on the current individual with a set probability to generate a backward solution; The current best individual is perturbed using the Cauchy distribution; Map the chaotic sequence to the solution space and perform GTO position update.

5. The energy operation management method for a photovoltaic-storage system as described in claim 4, characterized in that, The process of optimizing the network parameters of the prediction model using the improved multi-strategy fusion GTO algorithm is as follows: The network parameters of the prediction model are used as prey to calculate the population target value and update the energy parameters of prey escape. If the energy parameter of the prey escapes is greater than or equal to 1, the algorithm enters the global search phase and updates the population position. During the global search phase, if the probability of the prey escaping is greater than or equal to 0.5, the algorithm's influence on the prey's location is strengthened; otherwise, its influence on the prey's location is weakened. If the energy parameter of the prey's escape is less than 1 or the global search phase is completed, the algorithm enters the local search phase and updates the population position. In the local exploration phase, the current four optimal solutions are selected in turn as the positions of attacker, surroundr, chaser and pursuer to surround and capture the prey. Among them, the attacker is the best solution of the population, and the geometric center of the four surrounding roles is the most likely position of the prey, which is also used as the convergence point for updating the position of the surrounding roles in the next round of encirclement. If the probability of the prey escaping is less than the set probability threshold, reverse learning is performed on the four hunting roles to generate a new solution; otherwise, a perturbation is added to the attacker to obtain a new solution. The new solution is compared with the optimal solution. If the new solution is less than the optimal solution, the value of the new solution is assigned to the optimal solution. Otherwise, the value of the optimal solution remains unchanged. Update the prey's position and determine if the maximum number of iterations has been reached. If the maximum number of iterations has been reached, output the optimal network parameters corresponding to the prey's position; otherwise, continue repeating the above process to execute one round of iteration.

6. The energy operation management method for a photovoltaic-storage system as described in claim 1, characterized in that, The charging and discharging strategies for photovoltaic energy storage systems include the following: When the photovoltaic power generation is greater than the power of the user-side energy-consuming equipment, and the SOC of the energy storage battery satisfies: 0.1≤SOC<0.8, the photovoltaic power generation prioritizes powering the user-side equipment, and the remaining electricity is used to charge the energy storage battery and connect to the grid. When the photovoltaic power generation capacity is greater than the power consumption of the user-side energy-consuming equipment, and the SOC of the energy storage battery satisfies: 0.8≤SOC≤0.9, the photovoltaic power generation prioritizes supplying power to the user-side equipment, and the remaining power is fed into the grid. When the photovoltaic power generation is less than the power of the user-side energy-consuming equipment, and the SOC of the energy storage battery satisfies: 0.3≤SOC≤0.9, the photovoltaic power generation and the energy storage battery can simultaneously supply power to the user-side equipment. During off-peak electricity pricing periods and when the SOC of the energy storage battery satisfies 0.1≤SOC<0.7, electricity is purchased from the grid to supply power to user-side equipment and to charge the energy storage battery.

7. The energy operation management method for a photovoltaic-storage system as described in claim 1, characterized in that, Flexible energy-consuming equipment includes interruptible equipment whose on / off status can be adjusted at any time during the operating period, and non-interruptible equipment that must run continuously until completion once it starts operating during the working period. The scheduling strategy for flexible energy-consuming equipment is as follows: for interruptible equipment, the working time is split into periods of low electricity price or surplus photovoltaic power generation, while the working time of non-interruptible equipment is shifted to periods of continuous low electricity price.

8. An electronic device, characterized in that, It includes a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the energy operation management method of the photovoltaic energy storage system as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program; when the computer program is executed by a processor, it implements the energy operation management method of the photovoltaic energy storage system as described in any one of claims 1-7.

Images