A warehouse environment dynamic control optimization method based on artificial intelligence

Abstract

The application discloses a warehouse environment dynamic control optimization method based on artificial intelligence, mainly relates to the technical field of control optimization, and aims to solve the problems that the feature representation of the existing scheme is difficult to reflect the local fluctuation mode and phase difference of different micro-environments in the warehouse, and the feature representation is difficult to reflect the local fluctuation mode and phase difference of different micro-environments in the warehouse. It includes: introducing population diversity index to adjust the convergence coefficient, adaptive adjustment factor and spiral shape parameter involved in whale optimization algorithm; adopting the fitness function of fusing total energy consumption target and constraint violation penalty, and adopting the dynamic penalty coefficient adjustment strategy based on historical violation total, combining fuzzy logic to make the penalty change smoothly with the iteration process; after each iteration is completed, the next generation of whale population is constructed through elite reservation, hybrid operation and diversity maintenance; when the preset termination condition is met, the current global optimal whale individual position is output.

Description

Technical Field

[0001] This application relates to the field of control optimization technology, and in particular to a dynamic control optimization method for a warehouse environment based on artificial intelligence. Background Technology

[0002] With the rapid development of e-commerce and new retail, the scale of fast-moving consumer goods (FMCG) warehousing is constantly expanding, and storage density and turnover frequency are increasing significantly. Temperature and humidity, as core elements for ensuring the quality and safety of FMCG and extending its shelf life, directly affect the energy consumption of environmental control equipment. However, current related technical solutions still have many problems that need to be solved, making it difficult to meet the needs of precise control and energy-saving optimization of the warehousing environment.

[0003] The core problems of existing technologies are mainly reflected in four aspects: First, conventional normalization methods cannot capture the spatiotemporal dependence of temperature and humidity data, making it difficult to reflect the local fluctuations and phase differences of the warehouse microenvironment, resulting in incomplete modeling input information; Second, swarm intelligence optimization algorithms randomly initialize the population, ignoring historical high-quality solution information, resulting in low-quality and scattered initial solutions, slow convergence, and easy getting trapped in local optima; Third, the algorithm control parameters are fixed or simply linearly decreasing, making it impossible to dynamically balance the exploration and development intensity, and difficult to adapt to multi-peak and time-varying search spaces; Fourth, the penalty intensity of static constraint processing methods is unreasonable, which can easily fragment the search space and make it difficult to guarantee the feasibility of the solution, failing to achieve a balance between constraint satisfaction and energy consumption optimization. Summary of the Invention

[0004] This application provides an artificial intelligence-based dynamic control optimization method for warehouse environments to solve the above-mentioned problems.

[0005] This application provides a method for dynamic control and optimization of warehouse environment based on artificial intelligence, the method including:

[0006] Perform wavelet decomposition on the original warehousing environment data and calculate the normalized wavelet coefficients;

[0007] Based on normalized wavelet coefficients, historical control strategy, and random weights, the initial position of each individual whale involved in the whale optimization algorithm is calculated.

[0008] Based on the randomly generated switching probability, the search method is selected, and chaotic perturbation and local enhancement factor are introduced to iteratively update the position of individual whales;

[0009] The convergence coefficient, adaptive adjustment factor, and spiral shape parameters involved in the whale optimization algorithm are adjusted by introducing a population diversity index; a fitness function that integrates the total energy consumption target and the constraint violation penalty is adopted, and a dynamic penalty coefficient adjustment strategy based on the total historical violation is used, combined with fuzzy logic to make the penalty change smoothly with the iteration process.

[0010] After each iteration, the next generation of whale population is constructed through elite preservation, crossbreeding operations, and diversity maintenance; when the preset termination condition is met, the current global best individual whale position is output.

[0011] Based on the current location of the best individual whale in the global system, the optimal storage environment data is obtained.

[0012] In one implementation of this application, before performing wavelet decomposition on the original warehousing environment data and calculating the normalized wavelet coefficients, the method further includes:

[0013] In the current warehousing environment, deploy a network of temperature and humidity sensors covering all storage areas;

[0014] Furthermore, the temperature and humidity sensor network involves sensors that automatically collect temperature and humidity data at fixed time intervals;

[0015] Simultaneously, environmental data and environmental control equipment data corresponding to the warehouse's external envelope structure are collected as boundary conditions.

[0016] In one implementation of this application, wavelet decomposition is performed on the original warehousing environment data, and normalized wavelet coefficients are calculated, specifically including:

[0017] Wavelet decomposition is performed on the original warehousing environment data to obtain multi-scale wavelet coefficients;

[0018] For each scale and spatial location, based on historical wavelet coefficient data, the mean and standard deviation of the wavelet coefficients at that scale and location are calculated respectively.

[0019] Using a dynamic time warping algorithm, wavelet coefficient data sequences at different spatial locations are time-phase aligned, and the time offset parameters corresponding to each scale and each location are calculated and determined.

[0020] K-means clustering analysis was performed on historical wavelet coefficient data. The number of clusters was determined by the elbow method, and the cluster centers of wavelet coefficients at each scale and location were obtained.

[0021] Based on the mean and standard deviation, the wavelet coefficients are scaled. Then, a Gaussian window function is used to weight the time offset, improving the smoothness and consistency of the time features. Finally, the deviation of the current wavelet coefficients from the typical environmental pattern is calculated by using the natural exponential function, combined with the normalized decay coefficient and the cluster center, and the normalized wavelet coefficients are obtained.

[0022] In one implementation of this application, the initial position of each individual whale involved in the whale optimization algorithm is calculated based on normalized wavelet coefficients, historical control strategy, and random weights, specifically including:

[0023] Extract the normalized wavelet coefficients that have been calculated and the historical control strategy data of the warehousing environment, and generate the random weight parameters required by the algorithm.

[0024] Based on the number of target parameters for warehouse environment control optimization, the total number of individuals in the whale optimization algorithm and the dimension size corresponding to each whale individual are set.

[0025] For each individual whale, the normalized wavelet coefficients, the matched historical control strategy parameters, and the generated random weight values ​​for each dimension parameter of the individual are extracted.

[0026] According to the allocated basic weight ratio, the three types of data are weighted and fused. Through numerical superposition and constraint adjustment, the initial position parameter value of the current dimension of the individual whale is calculated.

[0027] By comparing the actual range of values ​​for the warehouse environment control parameters, boundary constraints are applied to the calculated initial position parameters, and values ​​that exceed the reasonable range are eliminated.

[0028] This continues until all dimensional parameters of all individual whales have been calculated, forming a complete initial position set for the whale optimization algorithm.

[0029] In one implementation of this application, a search method is selected based on randomly generated switching probabilities, and chaotic perturbations and local enhancement factors are introduced to iteratively update the position of individual whales, specifically including:

[0030] Before updating the position in each iteration, a random number uniformly distributed between 0 and 1 is generated. If the generated switching probability is less than the preset probability threshold, an encirclement search is performed; otherwise, a spiral search is performed.

[0031] When performing a bounding search, a whale individual's position is randomly selected from the current population as a reference position; the distance between the current individual and the reference position is calculated, and the basic search step size is determined by combining the adaptively adjusted convergence coefficient and the adaptive adjustment factor; a logistic chaotic perturbation factor is introduced to perturb and adjust the basic search results; after completing the calculation, the updated whale individual positions are obtained.

[0032] When performing a spiral search, the distance between the current whale individual and the current global best individual is calculated; a basic spiral movement trajectory is generated according to the spiral update rule; a normal distribution local enhancement factor is superimposed to perturb the optimal solution; after the calculation is completed, the position of the whale individual after this spiral search update is obtained.

[0033] Complete a single iteration of position update, and determine the final position of the current whale individual after this iteration based on the selected search method and corresponding perturbation enhancement processing, thus completing the position iteration update of a single individual;

[0034] The entire population is traversed and updated. The above steps are repeated for all individual whales in the population to complete the current round of position updates for all individuals.

[0035] In one implementation of this application, the convergence coefficient, adaptive adjustment factor, and spiral shape parameters involved in the whale optimization algorithm are adjusted by introducing a population diversity index, specifically including:

[0036] Calculate the mean vector of the current positions of all individual whales, and then calculate the standard deviation of the positions of all individuals in the population based on the mean vector;

[0037] The convergence coefficient is calculated using the nonlinearly decaying convergence factor and the first random number; wherein the convergence factor gradually decays from the initial value to the preset minimum value according to a nonlinear exponential law.

[0038] The adaptive adjustment factor is obtained by multiplying the second random number and the population diversity adaptive adjustment coefficient; wherein, the population diversity adaptive adjustment coefficient is calculated in real time based on the population diversity index.

[0039] An oscillating decay model is adopted, in which the initial spiral shape parameter value decays exponentially with the number of iterations, while sinusoidal periodic fluctuations are superimposed to calculate the spiral shape parameter.

[0040] In one implementation of this application, a fitness function that integrates the total energy consumption target and constraint violation penalties is adopted, and a dynamic penalty coefficient adjustment strategy based on the total historical violations is employed, specifically including:

[0041] Thermodynamic and mass transfer models were used to simulate the temperature and humidity distribution at different locations and times in the warehouse under the current control strategy.

[0042] Based on the equipment operating parameters and operating time, the power consumption of the equipment in the environment is added to obtain the total energy consumption value of the control strategy;

[0043] Determine whether the temperature exceeds the preset upper and lower limits at each location and at each time point, calculate the square value of the temperature deviation exceeding the limit and sum them to obtain the total amount of temperature constraint violation;

[0044] The sum of squares of deviations from the preset humidity range is calculated synchronously to obtain the total amount of humidity constraint violation;

[0045] Based on the basic penalty coefficients for temperature and humidity, and combined with the total number of constraint violations in the current iteration and the historical maximum total number of violations, the basic penalty magnitude is calculated.

[0046] The sigmoid function is used to make the penalty coefficient increase smoothly with the number of iterations;

[0047] Generate dynamic penalty coefficients for temperature and humidity respectively;

[0048] Calculate the absolute deviation between the simulated temperature and the optimal temperature setpoint at each location and time, and generate a reward value through an exponential decay function;

[0049] The total reward of the control strategy is obtained by summing the reward values ​​at all spatiotemporal points.

[0050] The core optimization objective is to use the negative value of total energy consumption, plus the dynamic penalty negative value for violations of temperature and humidity constraints, and the reward term for the optimal temperature.

[0051] Each iteration recalculates the total number of constraint violations, dynamic penalty coefficient, and reward term for the current population.

[0052] In one implementation of this application, after each iteration, the next generation of whale population is constructed through elite preservation, crossbreeding operations, and diversity maintenance, specifically including:

[0053] Calculate the fitness value of all current whale individuals and sort them from highest to lowest fitness.

[0054] According to the set elite ratio, select the top few individuals with the highest fitness and directly retain them to enter the next generation of the population;

[0055] From the non-elite individuals that were never retained, two different individuals were randomly selected as parents.

[0056] The control strategy vectors of the two parents are convexly combined and fused according to random weights to generate a basic new individual;

[0057] Injecting adaptively normally distributed noise into the base new individuals generates new hybrid individuals;

[0058] Repeat this process to generate a sufficient number of new hybrid individuals to supplement the next generation of the population;

[0059] Calculate the overall distribution standard deviation of the current population to obtain population diversity;

[0060] If the diversity is higher than the preset diversity threshold, no adjustment is made and the population construction is completed directly; if the diversity is lower than the preset diversity threshold, the population reset mechanism is triggered.

[0061] Select several individuals with the worst fitness in the current population, and add uniform random perturbation to the global optimal control strategy to generate new replacement individuals.

[0062] Replace the worst individual with the newly generated replacement individual;

[0063] By integrating elite individuals, hybridizing to generate new individuals, and maintaining diversity among the reset individuals, a complete next-generation whale population is formed.

[0064] As can be seen from the above technical solutions, this application has the following advantages:

[0065] 1. A multi-scale normalization method based on wavelet transform and dynamic time warping is adopted. Multi-scale features are extracted by wavelet decomposition of the original environmental data, and data sequences at different spatial locations are aligned by dynamic time warping. Combined with cluster analysis, the normalization parameters are adaptively adjusted so that the normalized wavelet coefficients carry both spatiotemporal phase information and the degree of deviation of local patterns, thereby enhancing the spatiotemporal consistency of feature representation and the ability to capture local fluctuations.

[0066] 2. Based on the historical optimal control strategy, the initial position of the population is constructed by the cluster centers. By calculating the offset of each cluster center from the reference vector and introducing random weights and random perturbations, a differentiated initial solution with data distribution information is generated, so that the initial population is distributed in the high-potential area in the decision space, thereby improving the quality of the initial solution and the diversity of the population.

[0067] 3. In the encirclement search of the whale optimization algorithm, a logistic chaotic perturbation factor is introduced, and a local enhancement factor based on normal distribution is superimposed in the spiral search. At the same time, the convergence coefficient, adaptive adjustment factor and spiral shape parameter are dynamically adjusted by nonlinear decay, oscillation decay and diversity feedback according to the population diversity index to achieve an adaptive balance between global exploration and local development.

[0068] 4. A dynamic penalty coefficient adjustment strategy based on the total historical violations is adopted to make the penalty coefficient rise smoothly with the iteration process. An exponential decay reward term is added to the fitness function to encourage the temperature to approach the optimal set value. Combined with adaptive noise injection in the hybridization operation and a local reset mechanism based on the diversity threshold, an effective solution to the high-dimensional nonlinear strongly constrained optimization problem is achieved. Attached Figure Description

[0069] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0070] Figure 1 This is a flowchart of a dynamic control and optimization method for a warehouse environment based on artificial intelligence, provided in an embodiment of this application.

[0071] Figure 2 This is a biaxial bar chart comparing temperature and humidity constraint violations, provided in an embodiment of this application.

[0072] Figure 3 This is a line graph comparing the convergence curves of different optimization algorithms provided in an embodiment of this application. Detailed Implementation

[0073] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0074] Those skilled in the art should understand that the embodiments described below are merely preferred embodiments of this disclosure and do not imply that this disclosure can only be implemented through these preferred embodiments. These preferred embodiments are merely used to explain the technical principles of this disclosure and are not intended to limit the scope of protection of this disclosure. Based on the preferred embodiments provided by this disclosure, all other embodiments obtained by those skilled in the art without creative effort should still fall within the scope of protection of this disclosure.

[0075] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0076] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0077] The embodiment provides an artificial intelligence-based method for dynamic control and optimization of the warehouse environment, such as... Figure 1 As shown in the embodiments of this application, the method mainly includes the following steps:

[0078] Step 110: Perform wavelet decomposition on the original warehousing environment data and calculate the normalized wavelet coefficients.

[0079] In some embodiments, before performing wavelet decomposition on the original warehousing environment data and calculating the normalized wavelet coefficients, the method further includes:

[0080] In the current warehousing environment, deploy a network of temperature and humidity sensors covering all storage areas;

[0081] Furthermore, the temperature and humidity sensor network involves sensors that automatically collect temperature and humidity data at fixed time intervals;

[0082] Simultaneously, environmental data and environmental control equipment data corresponding to the warehouse's external envelope structure are collected as boundary conditions.

[0083] This step can be specifically described as deploying a temperature and humidity sensor network covering all storage areas in a fast-moving consumer goods (FMCG) warehousing environment. The sensor installation locations need to take into account factors such as shelf height, aisle layout, and air conditioning supply paths to ensure that they can comprehensively reflect the real state of the microenvironment in different spaces within the warehouse.

[0084] All sensors automatically collect temperature and humidity data at fixed time intervals (e.g., every hour) and upload them in real time to a central data server via industrial bus or wireless network, forming a raw time-series database containing timestamps, location numbers, and environmental parameter values.

[0085] To ensure the effectiveness of subsequent multi-scale analysis and optimization strategies, the data collection process needs to last for at least four weeks to accumulate sufficient historical data for extracting statistical characteristics and fluctuation patterns at different time granularities at each location.

[0086] At the same time, meteorological parameters such as outdoor temperature and humidity, solar radiation intensity, wind speed and direction near the warehouse's outer perimeter are collected simultaneously, as well as equipment characteristic data such as the operating status, power consumption and energy efficiency ratio of various environmental control equipment (air conditioning units, dehumidifiers, humidifiers, ventilation fans), to provide input boundary conditions for the energy consumption simulation model.

[0087] Based on data collection, the dynamic control optimization problem of the storage environment is defined as follows: Under the premise that the temperature and humidity of all storage areas are strictly within the preset allowable range at all times, the total cumulative energy consumption of all participating control equipment is minimized by reasonably setting the temperature and humidity setpoint sequence of each independent control zone over a future period. The decision variables are the hourly temperature and humidity setpoints of each zone, the dimensionality of which is determined by the number of zones, the number of variables to be optimized in each zone, and the length of the optimization time window. Constraints include that the simulated temperature and humidity values ​​at each location at each time point must not exceed the allowable range, and the control setpoints themselves should also conform to the physical adjustability range of the equipment. The optimization objective is the sum of the products of the power and operating time of various equipment such as refrigeration, heating, dehumidification, and humidification, obtained through thermodynamic and mass transfer model simulations. This optimization problem is characterized by high dimensionality, nonlinearity, strong constraints, and dynamic time-varying nature; therefore, it requires an artificial intelligence optimization algorithm that can fully mine historical data distribution information and adaptively balance exploration and development during iteration.

[0088] Warehouse environment data (temperature and humidity) exhibits spatiotemporal fluctuations and multi-scale correlations. Data at different time granularities and spatial locations show scale differences. Conventional methods such as min-max normalization cannot effectively capture the spatiotemporal dependencies between data, resulting in insufficient feature representation and difficulty in adapting to the precise control requirements of dynamic environments.

[0089] This application employs a multi-scale normalization method based on wavelet transform and dynamic time warping, and combines cluster analysis to adaptively adjust the normalization parameters to enhance the spatiotemporal consistency of features and the ability to capture local fluctuations.

[0090] In practical implementation, wavelet decomposition is performed on the original environmental data to extract multi-scale features, and dynamic time warping is used to align data sequences at different spatial locations. Local data distribution characteristics are obtained through cluster analysis, and then normalized wavelet coefficients are calculated, expressed as:

[0091]

[0092] In the formula, Indicates the first Scale, First Position, number Normalized wavelet coefficients at any given time are suitable for temperature or humidity data. They characterize the normalized wavelet coefficients of environmental data at a specific scale, location, and time, and have the characteristics of enhancing spatiotemporal consistency and capturing local fluctuations.

[0093] Indicates the first Scale, First Position, number The wavelet coefficients at each time point are obtained by performing discrete wavelet decomposition on the original environmental data (temperature or humidity time series). Specifically, the sequence is decomposed into multiple layers using wavelet basis functions (such as Daubechies wavelet, commonly db4), and detail coefficients and approximation coefficients at different time scales are extracted. This allows for the extraction of multi-scale features from the original data and the capture of spatiotemporal fluctuations.

[0094] This represents a time index, indicating a specific monitoring time, with a value range of [value range missing]. ;

[0095] This represents a spatial location index, indicating a specific sensor number within the warehouse, with a value range of [value range missing]. ;

[0096] The scale index represents different time granularities obtained through wavelet decomposition, with values ​​ranging from [value range missing]. For example, if the original data is in hourly granularity, then This likely corresponds to an hourly timescale. Corresponding to a daily scale (aggregated 24 hours). Corresponding to a weekly scale (aggregated 168 hours);

[0097] Indicates the first Position at The historical wavelet coefficient mean at a given scale represents the long-term environmental fluctuation level at that location, and is calculated as follows: ;

[0098] Indicates the first Position at The standard deviation of historical wavelet coefficients at a given scale characterizes the intensity of environmental fluctuations at that location, and is calculated as follows: ;

[0099] This represents the total number of time points, with an example value of 672, corresponding to hourly data over 4 weeks.

[0100] This represents the total number of scales, with an example value of 3.

[0101] This represents the total number of spatial locations, i.e., the number of sensors; an example value is 20.

[0102] This represents a Gaussian window function used to weight time offsets, enhancing the matching degree of local time features and improving the smoothness and consistency of normalized wavelet coefficients on the time axis.

[0103] Indicates the first Position at The time offset parameter at the scale is used to align the time phase of data sequences at different locations. It is obtained by optimizing the dynamic time warping algorithm. Specifically, the time offset parameter is extracted by aligning data sequences at different locations and calculating the minimum cumulative distance path.

[0104] This represents the time scale parameter, which controls the window width of the wavelet function; an example value is 5.

[0105] Represents the natural exponential function;

[0106] This represents the normalization attenuation coefficient, which adjusts the degree of influence of the difference between the current wavelet coefficient and the typical value on the normalization result. The example value is 0.1.

[0107] Indicates the first Position at The wavelet coefficient cluster centers at a given scale reflect typical environmental patterns at that location. This is achieved by analyzing historical wavelet coefficients. -means clustering is obtained by collecting historical wavelet coefficient data and applying... - The means clustering algorithm obtains the centers of the clusters.

[0108] Step 120: Based on normalized wavelet coefficients, historical control strategy, and random weights, calculate the initial position of each individual whale involved in the whale optimization algorithm.

[0109] Specifically, it includes:

[0110] Extract the normalized wavelet coefficients that have been calculated and the historical control strategy data of the warehousing environment, and generate the random weight parameters required by the algorithm.

[0111] Based on the number of target parameters for warehouse environment control optimization, the total number of individuals in the whale optimization algorithm and the dimension size corresponding to each whale individual are set.

[0112] For each individual whale, the normalized wavelet coefficients, the matched historical control strategy parameters, and the generated random weight values ​​for each dimension parameter of the individual are extracted.

[0113] According to the allocated basic weight ratio, the three types of data are weighted and fused. Through numerical superposition and constraint adjustment, the initial position parameter value of the current dimension of the individual whale is calculated.

[0114] By comparing the actual range of values ​​for the warehouse environment control parameters, boundary constraints are applied to the calculated initial position parameters, and values ​​that exceed the reasonable range are eliminated.

[0115] This continues until all dimensional parameters of all individual whales have been calculated, forming a complete initial position set for the whale optimization algorithm.

[0116] More specifically, this step can be:

[0117] Improved population initialization for the whale optimization algorithm:

[0118] The standard whale optimization algorithm typically initializes the population position in a random manner, ignoring the clustering structure information contained in the normalized data, resulting in poor quality of the initial solution. When faced with high-dimensional and dynamic warehouse environment data, this can easily lead to slow convergence and getting stuck in local optima.

[0119] This application generates the initial position of the population based on the global statistical characteristics and clustering structure information of normalized data, thereby improving the quality of the initial solution and the population diversity.

[0120] In practical implementation, a baseline vector is determined based on historical control strategies. This vector, combined with individual offsets calculated using cluster centers and random weights, and random perturbations, generates the initial position for each individual whale, represented as:

[0121]

[0122] In the formula, Indicates the first The initial position of each individual whale, with dimensions of This represents an initial control strategy;

[0123] This indicates the number of control variables, such as temperature setpoint and humidity setpoint. Each position vector represents a complete control strategy. ;

[0124] The mean vector representing the historical optimal control strategy is obtained by taking the arithmetic mean of the global optimal control strategy vectors from each historical optimization. Its dimension is... ;

[0125] It should be noted that the control strategy vector is the decision variable directly manipulated by the improved whale optimization algorithm. It is a mathematical vector that fully describes the detailed control scheme of each area of ​​the warehouse in the future. For the algorithm, the position of each individual whale is a control strategy vector. The algorithm iteratively optimizes and finally outputs the globally optimal individual position, which is the optimal control strategy vector. For actual control, this vector is sent to the controllers of equipment such as air conditioners and dehumidifiers, telling them what values ​​the temperature and humidity should be set at each hour in the future.

[0126] Indicates the first The offset of each individual whale, calculated based on random weights and cluster centers, is used to introduce variability based on data distribution. The calculation method is expressed as follows: ;

[0127] This represents the total number of clusters, determined by the elbow method, with an example value of 5.

[0128] This represents the clustering index, with a value range of 100. ;

[0129] Indicates the first The cluster centers, obtained by k-means clustering of normalized wavelet coefficients, represent a typical warehouse environment pattern, that is, a pattern containing... The normalized wavelet coefficients at all positions, including those mentioned above, are clustered using k-means to obtain... Cluster centers, Indicates the first There are 10 cluster centers, which represent typical local environmental patterns in the storage environment (such as temperature and humidity fluctuations in different seasons, time periods, and regions).

[0130] Indicates the first The individual relative to the first The random weighting coefficients of each cluster center are used to balance the contribution of different cluster centers to the individual offsets. This incorporates differences based on data distribution and is calculated as follows: ;

[0131] Indicates the first The individual relative to the first The initial random weights of the cluster centers follow a uniform distribution. ;

[0132] Indicates the interval Uniform distribution on;

[0133] This represents the cluster attenuation coefficient, used to adjust the degree of attenuation of the contribution of cluster centers to the offset. An example value is 0.1.

[0134] Represents the L2 norm;

[0135] This indicates the initial scaling factor, which controls the overall magnitude of the offset. An example value is 0.5.

[0136] Indicates the first The random perturbation factor of each individual whale follows a uniform distribution. This is used to increase the randomness of the initial population;

[0137] This represents the index of an individual whale, with a value range of [value range missing]. ;

[0138] This indicates the size of the whale population, i.e., the total number of individuals, and is preferably set as follows: .

[0139] It's important to note that the control strategy vector is the decision variable directly manipulated by the improved whale optimization algorithm. It's a mathematical vector that comprehensively describes the detailed control scheme for each area of ​​the warehouse over a future period. For the algorithm, the position of each individual whale is a control strategy vector. Through iterative optimization, the algorithm outputs the globally optimal individual position, which is the optimal control strategy vector. In practical control, this vector is sent to the controllers of equipment such as air conditioners and dehumidifiers, informing them of the appropriate temperature and humidity settings for each hour in the future.

[0140] In practical implementation, each whale's position vector corresponds to a complete control strategy, and its dimension... This equals the total number of all controllable variables. If the warehouse uses independent zone control, then the number of control variables... For example, if the area is divided into three zones, and each zone has its own set temperature and humidity, then... If we optimize the future The hourly setting value is then... .

[0141] Step 130: Based on the randomly generated switching probability, select the execution search method, and introduce chaotic perturbation and local enhancement factor to iteratively update the position of individual whales.

[0142] Specifically, it includes:

[0143] Before updating the position in each iteration, a random number uniformly distributed between 0 and 1 is generated. If the generated switching probability is less than a preset probability threshold, an encirclement search is performed; otherwise, a spiral search is performed.

[0144] When performing a bounding search, a whale individual's position is randomly selected from the current population as a reference position; the distance between the current individual and the reference position is calculated, and the basic search step size is determined by combining the adaptively adjusted convergence coefficient and the adaptive adjustment factor; a logistic chaotic perturbation factor is introduced to perturb and adjust the basic search results; after completing the calculation, the updated whale individual positions are obtained.

[0145] When performing a spiral search, the distance between the current whale individual and the current global best individual is calculated; a basic spiral movement trajectory is generated according to the spiral update rule; a normal distribution local enhancement factor is superimposed to perturb the optimal solution; after the calculation is completed, the position of the whale individual after this spiral search update is obtained.

[0146] Complete a single iteration of position update, and determine the final position of the current whale individual after this iteration based on the selected search method and corresponding perturbation enhancement processing, thus completing the position iteration update of a single individual;

[0147] The entire population is traversed and updated. The above steps are repeated for all individual whales in the population to complete the current round of position updates for all individuals.

[0148] This step can be specifically described as follows:

[0149] The standard whale optimization algorithm uses fixed parameters to control the surrounding prey and spiral update behavior. In the dynamic optimization of the storage environment control problem, the search efficiency is low and it is difficult to balance the algorithm's global exploration and local development capabilities.

[0150] This application employs an adaptive parameter adjustment mechanism based on population diversity and convergence state to dynamically switch search behavior, and incorporates chaotic mapping and local enhancement operators to improve the algorithm's global optimization capability in complex spaces. The specific steps are as follows:

[0151] 1) Location update:

[0152] Based on the randomly generated switching probability, either encirclement search or spiral search is selected, and chaotic perturbation and local enhancement factors are introduced to achieve a dynamic balance between the algorithm's global exploration and local development capabilities, enhance its adaptability to dynamic changes in the storage environment, improve convergence accuracy, and avoid getting trapped in local optima.

[0153] In specific implementation, the first The whale in the The position update method for the next iteration is represented as follows:

[0154] In the formula, Indicates the first The whale in the The position at the nth iteration represents the... The position of each whale in the next iteration, i.e., the updated control strategy;

[0155] This indicates a randomly selected whale location within the current population, used to guide exploration of unknown areas;

[0156] This represents the convergence coefficient, used to control the encirclement step size, and its value decreases adaptively with iteration;

[0157] This represents an adaptive adjustment factor, which is used to adjust the search step size based on the dynamic changes in population diversity.

[0158] Indicates the first The whale in the The position at the nth iteration represents the... The position of each whale in the current iteration, i.e., the current control strategy;

[0159] The chaotic perturbation factor is obtained through iterative logistic mapping, i.e. The initial value is randomly set at Intervals are used to increase the traversal of the search trajectory;

[0160] This indicates an assignment operation;

[0161] The probability of behavior switching is represented in A random number that is uniformly distributed within an interval;

[0162] This represents the distance vector between the current individual and the current global best individual, calculated as follows: ;

[0163] This represents the position of the whale with the best global fitness in the current iteration, and represents the best control strategy found so far, which is obtained by comparing the fitness values ​​of all individuals.

[0164] This represents the helical shape parameter, which controls the pitch of the helical update, i.e. the tightness of the helical line;

[0165] Indicates a A random number that is uniformly distributed within an interval;

[0166] Represents the natural constant. Characterizes the magnitude of spiral renewal;

[0167] This represents the cosine function, used to generate periodic motion in spiral updates;

[0168] This represents the local enhancement factor, used to introduce random perturbations in the spiral search to enhance local exploitation capabilities. Its calculation method is expressed as follows: ;

[0169] This represents the local enhancement coefficient, used to control the perturbation intensity of the local enhancement factor; an example value is 0.01.

[0170] This represents a standard normal distribution with a mean of 1 / 2. The standard deviation is .

[0171] It should be noted that the standard whale optimization algorithm's position update formula is: bounding search Spiral search However, warehouse environment data is highly time-varying, and the random exploration of the standard whale optimization algorithm is prone to losing traversal. This application introduces a chaotic perturbation factor to address the specific characteristics of dynamic control tasks in warehouse environments. With local enhancement factors In the encirclement search, a chaotic perturbation factor is multiplied to enhance the global traversal capability by utilizing the pseudo-randomness of the logistic mapping. In the spiral search, a local enhancement factor based on normal perturbation is added to avoid insufficient oscillation near the optimal solution, thereby improving the algorithm's ability to escape local optima and its precision mining accuracy in complex search spaces.

[0172] It should also be noted that the control strategy is a vector of decision variables output by the optimization algorithm. It defines the setpoints or operating parameters of each environmental control device (such as air conditioners and dehumidifiers) within a future time period. Each whale position represents a control strategy vector. For example, assuming the warehouse is divided into three independent control zones, optimizing the temperature and humidity setpoints for each hour over the next six hours (one set per hour), then... Dimensions , represented as ,in Indicates the first District No. The temperature set point for the hour. Indicates the first District No. Humidity setpoint for the hour.

[0173] It should also be noted that the chaotic perturbation factor Introducing logistic chaotic mapping into the encirclement search, leveraging its pseudo-randomness and ergodicity, enhances global search capabilities and avoids premature convergence. This is particularly suitable for problems with complex data distributions and multiple local extrema in warehouse environments. Furthermore, it provides a local enhancement factor. By superimposing normal random perturbations into the spiral search, the fine search capability of the optimal solution neighborhood is improved, the temperature and humidity setpoints are precisely adjusted, and energy consumption is effectively reduced.

[0174] 2) Adaptive parameter and chaotic perturbation coordinated adjustment mechanism:

[0175] By introducing a population diversity index, the convergence coefficient, adaptive adjustment factor, and spiral shape parameter are dynamically adjusted. Specifically...

[0176] The convergence coefficient is calculated based on a non-linearly decaying convergence factor, which decays exponentially from its initial value to its minimum value. This factor controls the algorithm's transition from global exploration to local exploitation, and is expressed as... ;

[0177] The adaptive adjustment factor is obtained by multiplying a random number by a diversity adaptive factor based on the population standard deviation. When the population location distribution is divergent, the adaptive factor increases, expanding the search step size to enhance exploration; when the population is clustered, the adaptive factor decreases, shrinking the step size to enhance local exploitation. ;

[0178] The spiral shape parameter adopts an oscillating decay model, with its initial value decaying exponentially with the iteration process, while simultaneously being superimposed with sinusoidal periodic fluctuations. This ensures that the spiral trajectory gradually converges in the later stages of the search while retaining a certain degree of ergodicity, represented as follows: ;

[0179] in, The convergence factor, which decreases non-linearly from its initial value to its minimum, controls the transition of the algorithm from global exploration to local exploitation. Its calculation method is expressed as follows: ;

[0180] This represents the initial value of the convergence factor, which determines the initial step size of the exploration phase. An example value is 2.

[0181] Indicates the index of the current iteration number;

[0182] This represents the maximum number of iterations, which determines the total number of iterations the algorithm will run; an example value is 500.

[0183] Represents the decay exponent, controlling the convergence factor. The shape of the descent curve affects the conversion speed of exploration and development; an example value is 1.5.

[0184] This represents the minimum convergence factor, ensuring sufficient local exploitation capability in the later stages of iteration; an example value is 0.

[0185] Indicates in The first random number, uniformly distributed within the interval, is used to introduce randomness and adjust the direction and step size of the encirclement search;

[0186] Indicates in The second random number, uniformly distributed within the interval, is used to adjust the adaptive adjustment factor. The random component;

[0187] This represents the adaptive adjustment coefficient for population diversity, used to dynamically adjust the adaptive adjustment factor. In response to changes in population distribution, the calculation method is expressed as follows: ;

[0188] The standard deviation represents the current population position, characterizing the degree of population diversity, and is calculated as follows: ;

[0189] This represents the mean vector of the positions of all individuals in the current population. It is obtained by averaging the positions of all individuals in the current population and reflects the central position of the population. It is used to calculate diversity.

[0190] This represents a very small constant used to prevent the denominator from being zero. Examples of its values ​​are shown below. ;

[0191] This represents the hyperbolic tangent function, used to map the input to... interval;

[0192] Represents the diversity sensitivity coefficient, controlling the effect of population diversity on... The degree of influence, with an example value of 0.5;

[0193] This represents the initial value of the helical shape parameter, which determines the initial pitch of the helical search. An example value is 1.

[0194] This represents the oscillation amplitude coefficient, controlling the spiral shape parameter. The oscillation amplitude, with an example value of 0.2;

[0195] This represents a sine function, used to introduce periodic oscillations and enhance the ergodicity of the search;

[0196] This represents the decay time constant, used to control the helix shape parameters. The decay rate, with examples of values ​​as follows: .

[0197] It should be noted that, Based on real-time adjustment of population diversity, when population diversity is high... Expand the search scope; when diversity is low, Shrink the step size and focus on local development. For example, in the initial iteration, Larger ,but Later Approaching 0 , .

[0198] It should also be noted that the convergence factor The iterations decrease linearly from 2 to 0 to ensure that the later stages of the iteration fully enter local development. This application uses nonlinear decay to 0, which can generate fine-tuning perturbations around the optimal solution without causing the search to stagnate.

[0199] It should also be noted that the standard whale optimization algorithm uses fixed parameters ( The standard whale optimization algorithm, with its fixed parameters, cannot adapt to the dynamic changes in exploration and development needs during iteration. Given the unique characteristics of dynamic control in the warehousing environment—environmental parameters (temperature, humidity) exhibiting strong time-varying, multi-peak, and uncertainties—it struggles to balance global exploration and local development, and is prone to getting trapped in local optima. The convergence coefficient of this application… Adaptive Factor Spiral shape parameters Based on the dynamic changes in population diversity and iteration process, exploration is enhanced when the population disperses and development is strengthened when the population gathers, so as to achieve an adaptive balance between exploration and mining and adapt to the dynamic changes in the storage environment.

[0200] Step 140: Optimize the convergence coefficient, adaptive adjustment factor and spiral shape parameters involved in the whale optimization algorithm by introducing a population diversity index; adopt a fitness function that integrates the total energy consumption target and the constraint violation penalty, and adopt a dynamic penalty coefficient adjustment strategy based on the total historical violation amount.

[0201] In some embodiments, the convergence coefficient, adaptive adjustment factor, and spiral shape parameters involved in the whale optimization algorithm are optimized by introducing a population diversity index, specifically including:

[0202] Calculate the mean vector of the current positions of all individual whales, and then calculate the standard deviation of the positions of all individuals in the population based on the mean vector;

[0203] The convergence coefficient is calculated using the nonlinearly decaying convergence factor and the first random number; wherein the convergence factor gradually decays from the initial value to the preset minimum value according to a nonlinear exponential law.

[0204] The adaptive adjustment factor is obtained by multiplying the second random number and the population diversity adaptive adjustment coefficient; wherein, the population diversity adaptive adjustment coefficient is calculated in real time based on the population diversity index.

[0205] An oscillating decay model is adopted, in which the initial spiral shape parameter value decays exponentially with the number of iterations, while sinusoidal periodic fluctuations are superimposed to calculate the spiral shape parameter.

[0206] In some embodiments, a fitness function that integrates the total energy consumption target and constraint violation penalties is adopted, and a dynamic penalty coefficient adjustment strategy based on the total historical violations is employed, specifically including:

[0207] Thermodynamic and mass transfer models were used to simulate the temperature and humidity distribution at different locations and times in the warehouse under the current control strategy.

[0208] Based on the equipment operating parameters and operating time, the power consumption of the equipment in the environment is added to obtain the total energy consumption value of the control strategy;

[0209] Determine whether the temperature exceeds the preset upper and lower limits at each location and at each time point, calculate the square value of the temperature deviation exceeding the limit and sum them to obtain the total amount of temperature constraint violation;

[0210] The sum of squares of deviations from the preset humidity range is calculated synchronously to obtain the total amount of humidity constraint violation;

[0211] Based on the basic penalty coefficients for temperature and humidity, and combined with the total number of constraint violations in the current iteration and the historical maximum total number of violations, the basic penalty magnitude is calculated.

[0212] The sigmoid function is used to make the penalty coefficient increase smoothly with the number of iterations;

[0213] Generate dynamic penalty coefficients for temperature and humidity respectively;

[0214] Calculate the absolute deviation between the simulated temperature and the optimal temperature setpoint at each location and time, and generate a reward value through an exponential decay function;

[0215] The total reward of the control strategy is obtained by summing the reward values ​​at all spatiotemporal points.

[0216] The core optimization objective is to use the negative value of total energy consumption, plus the dynamic penalty negative value for violations of temperature and humidity constraints, and the reward term for the optimal temperature.

[0217] Each iteration recalculates the total number of constraint violations, dynamic penalty coefficient, and reward term for the current population.

[0218] More specifically, this step can be defined as follows:

[0219] A fitness function that integrates the total energy consumption target and constraint violation penalties is adopted, and a dynamic penalty coefficient adjustment strategy based on the total historical violations is used. Fuzzy logic is combined to make the penalty change smoothly with the iteration process, thereby effectively handling multi-constraint optimization problems. The specific steps are as follows:

[0220] 1) Fitness calculation:

[0221] The fitness function achieves multi-objective optimization by comprehensively balancing total energy consumption targets and environmental constraints. Specifically,

[0222] Based on the thermodynamic and mass transfer model, the hourly temperature and humidity distribution under each control strategy is simulated, and then the total energy consumption is obtained by accumulating the product of the power of all devices and the time step.

[0223] Then, the square-shaped excesses of temperature and humidity at each location and time relative to the preset allowable range are calculated, and the total temperature and humidity violations are summed over all spatiotemporal points.

[0224] Then, based on the overall violation rate and iteration progress of the population in the current iteration, the penalty coefficients of temperature and humidity are dynamically adjusted using the Sigmoid smoothing mechanism. This allows the algorithm to tolerate certain violations in the early stage to strengthen exploration, and to strictly punish out-of-bounds violations in the later stage to ensure feasibility. At the same time, a reward is added to encourage the temperature to approach the optimal set value. This reward measures the degree of closeness between the temperature and the optimal value at each time point through an exponential decay function and is accumulated.

[0225] Finally, the normalized total energy consumption negative value, the two dynamic penalty negative values ​​and the reward term are weighted and summed to obtain the individual's fitness value. The larger the value, the better the control strategy performs in terms of both energy saving and compliance.

[0226] The specific implementation formula is expressed as follows:

[0227] In the formula, Indicates the first The fitness value of the first individual whale is used to evaluate the fitness of the second whale. The quality of a control strategy is determined by its value; a higher value indicates a better strategy.

[0228] Indicates the use of the first The total energy consumption simulated under the first control strategy represents the energy consumption obtained when using the second control strategy. The total energy consumption simulated under each control strategy is one of the optimization objectives. It is calculated by accumulating the product of the power consumption of each environmental control device and the operating time interval, and is expressed as: ;

[0229] The dynamic penalty coefficient representing the temperature constraint is a comprehensive coefficient that includes dimensional transformation. It is used to adjust the strength of the temperature violation penalty, adaptively changing based on the total historical violations. The calculation method is expressed as follows: ;

[0230] This represents the base temperature penalty coefficient, which is the baseline value of the dynamic penalty coefficient. It determines the intensity of the penalty for exceeding the temperature limit during the initial iteration. The example value is 100.

[0231] This represents the rate of temperature penalty adjustment, indicating the sensitivity of the dynamic penalty coefficient controlling temperature constraints to changes in the total number of temperature violations in the current population. The larger the value, the faster the penalty coefficient increases as the total number of violations increases; an example value is 0.5.

[0232] This represents the sum of temperature constraint violations for all individuals in the current iteration. It is obtained by accumulating the temperature violation of each individual. Furthermore, the temperature violation of each individual is obtained by performing limit checks on the simulated temperature values ​​at all locations and all times, and then accumulating the results. The temperature violation of the i-th individual is equal to... ;

[0233] The dynamic penalty coefficient representing the humidity constraint is a comprehensive coefficient that includes dimension transformation. It is used to dynamically weight the penalty for humidity constraint violations. Its value is adaptively adjusted based on the basic humidity penalty coefficient, the total number of humidity violations, and the iteration progress. The calculation method is expressed as follows: ;

[0234] This represents the basic humidity penalty coefficient, which serves as the baseline value for humidity penalty. An example value is 50.

[0235] This indicates the humidity penalty adjustment rate, controlling the adjustment speed of the humidity penalty coefficient based on the total amount of violation. An example value is 0.5.

[0236] This represents the sum of humidity constraint violations for all individuals in the current iteration. It is the sum of humidity constraint violations for all individuals in the current iteration, obtained by accumulating the humidity violation of each individual. The humidity violation of the i-th individual is equal to... ;

[0237] This represents the maximum total number of constraint violations recorded in all iterations to date, and is used to normalize the penalty coefficient.

[0238] This represents the Sigmoid function, which causes the penalty coefficient to increase smoothly with the number of iterations;

[0239] Indicates the index of the current iteration number;

[0240] This indicates the maximum number of iterations, preferably set to 500.

[0241] This represents the function that takes the maximum value.

[0242] Indicates the first Under the first control strategy, the warehouse contents were calculated using a thermodynamic model. The location is in The dry-bulb temperature at any given time, for example, can be obtained by a thermodynamic model based on the control setpoint (temperature, humidity), outdoor meteorological parameters, thermal characteristics of the building envelope, internal heat sources, etc. A lumped parameter RC model (Resistance-Capacitance model) can be used to simplify the warehouse into a single heat capacity node, and the indoor temperature change can be solved by the thermal balance differential equation.

[0243] This indicates the lower limit of the permissible temperature range; for example, setting... ;

[0244] Indicates the upper limit of the allowable temperature range, for example, setting ;

[0245] Indicates the first Under the first strategy, the... The power of the refrigeration equipment at any given time is calculated by the equipment model based on the control setpoint and environmental data. For example, assuming the refrigeration equipment is a fixed-frequency / variable-frequency air conditioner, its power is directly proportional to its cooling capacity. The power of the refrigeration equipment can be approximated by the difference between the room load (heat transfer from the building envelope, fresh air load, and internal heat sources) and the set temperature. Where EER is the energy efficiency ratio, which is a constant;

[0246] Indicates the first Under the first strategy, the... The power of the heating equipment is calculated by the equipment model based on the control setpoint and environmental data. For example, for an electric heater, the power of the heating equipment is... (Assuming an electrothermal conversion efficiency of 100%), where, The heat load is calculated based on the indoor-outdoor temperature difference and the thermal resistance of the building envelope; or, for a heat pump, the power of the heating equipment. Where COP is the performance coefficient, which is a constant;

[0247] Indicates the first Under the first strategy, the... The power of the dehumidifier is calculated by the equipment model based on the control setpoint and environmental data. For example, for a refrigeration dehumidifier, the power of the dehumidifier is mainly related to the operating status of the compressor. A multinomial regression model is used, with the input being the set humidity, current humidity, and inlet air temperature, and the output being the power of the dehumidifier.

[0248] Indicates the first Under the first strategy, the... The power of the humidifier is calculated by the device model based on the control setpoint and environmental data. For example, for electrode humidification, the power of the humidifier is directly proportional to the humidification amount. ,in The steam generation rate is expressed as kg / s. Latent heat of vaporization (kWh / kg);

[0249] This indicates a time interval, with an example value of 1 hour.

[0250] Indicates the first Under the first control strategy, the mass transfer model simulation obtained the first... Position at Humidity at any given time, for example, using a lumped-parameter moisture balance model, treating the warehouse as a node, and considering factors such as supply air moisture content, indoor moisture generation, moisture transfer from the building envelope, and fresh air infiltration, the differential equation is: ,in, Moisture content (kg / kg dry air) air density (kg / m³). The effective volume of the warehouse (m³). The air supply mass flow rate is (kg / s). Moisture content of the supply air (kg / kg dry air). Indoor air humidity (kg / kg dry air) The infiltration air mass flow rate is (kg / s). Outdoor air humidity (kg / kg dry air). If the internal moisture load is (kg / s), the humidity can be calculated from the moisture content and temperature using the properties of moist air.

[0251] This indicates the lower limit of the allowable humidity range; for example, a possible value is... ;

[0252] This indicates the upper limit of the allowable humidity range; for example, a possible value is... ;

[0253] This indicates the weight of the reward item, expressed in units of 1 / 2. A consistent energy unit (default is kWh) controls the importance of reward items in fitness; an example value is 10 kWh.

[0254] This represents a reward item used to encourage temperatures to approach the set optimal value; the calculation method is expressed as follows: ;

[0255] This represents the reward decay coefficient, which controls the rate at which the reward decays when the temperature deviates from the optimal value. An example value is 0.5.

[0256] This represents the optimal temperature setting, which is set according to the storage requirements or comfort standards of the stored goods. An example value is shown below. .

[0257] Step 150: After each iteration, construct the next generation of whale population through elite retention, hybridization operation and diversity maintenance; when the preset termination condition is met, output the current global best individual whale position; obtain the optimal storage environment data based on the current global best individual whale position.

[0258] In some embodiments, after each iteration, the next generation of whale population is constructed through elite retention, crossbreeding operations, and diversity maintenance, specifically including:

[0259] Calculate the fitness value of all current whale individuals and sort them from highest to lowest fitness.

[0260] According to the set elite ratio, select the top few individuals with the highest fitness and directly retain them to enter the next generation of the population;

[0261] From the non-elite individuals that were never retained, two different individuals were randomly selected as parents.

[0262] The control strategy vectors of the two parents are convexly combined and fused according to random weights to generate a basic new individual;

[0263] Injecting adaptively normally distributed noise into the base new individuals generates new hybrid individuals;

[0264] Repeat this process to generate a sufficient number of new hybrid individuals to supplement the next generation of the population;

[0265] Calculate the overall distribution standard deviation of the current population to obtain population diversity;

[0266] If the diversity is higher than the preset diversity threshold, no adjustment is made and the population construction is completed directly; if the diversity is lower than the preset diversity threshold, the population reset mechanism is triggered.

[0267] Select several individuals with the worst fitness in the current population, and add uniform random perturbation to the global optimal control strategy to generate new replacement individuals.

[0268] Replace the worst individual with the newly generated replacement individual;

[0269] By integrating elite individuals, hybridizing to generate new individuals, and maintaining diversity among the reset individuals, a complete next-generation whale population is formed.

[0270] More specifically, this step can be defined as follows:

[0271] The next generation population is constructed using elite preservation, crossover operations, and diversity maintenance mechanisms. The algorithm's convergence speed and global exploration capability are balanced, and the termination condition comprehensively considers the number of iterations and fitness convergence. The specific steps are as follows:

[0272] 1) Population renewal mechanism:

[0273] After each iteration, the following operations are performed to update the population (generating the next generation population). Through elite preservation, crossover operations, and diversity maintenance, the convergence speed and global exploration capability are balanced to avoid premature convergence:

[0274] a) Elite Preservation: Selecting the highest-fitting individuals from the current population. Each individual's genes are directly passed on to the next generation, ensuring that superior genes are not lost.

[0275] in, The number of elite individuals is represented by the following calculation method: ;

[0276] This represents the proportion of elite individuals, determining how many of the best individuals are retained for the next generation; an example value is 0.1.

[0277] This represents the function for rounding up.

[0278] b) Crossbreeding operation: Randomly select two individuals from non-elite individuals as parent individuals (e.g., and By generating new individuals through convex combination and adaptive noise injection, the randomness of the local search is enhanced, as expressed as:

[0279] ;

[0280] In the formula, Indicates the location of the new individual generated through hybridization;

[0281] Indicates the first non-elite individual. This indicates the second non-elite individual;

[0282] The crossover weight is represented in Uniformly distributed random numbers within the interval control the mixing ratio of two parent individuals when generating a new individual;

[0283] This represents the hybridization noise factor, which controls the intensity of the injected noise; an example value is 0.01.

[0284] This indicates that the mean is 0 and the covariance matrix is... Multivariate normal distribution noise is used to introduce random perturbations in hybridization to increase diversity;

[0285] The covariance matrix represents the noise, controlling its distribution and amplitude; it is preferably set as a diagonal matrix. ;

[0286] This represents a diagonal matrix, that is, a matrix in which only the diagonal elements are non-zero;

[0287] This represents the standard deviation of the noise, controlling the intensity of hybridization noise; an example value is 0.05.

[0288] c) Diversity Maintenance: Real-time monitoring of population distribution breadth. When diversity falls below a threshold, some of the worst-performing individuals are reset within the neighborhood of the optimal solution to avoid premature convergence. Specifically, the current population diversity index is calculated. ,like ( If the diversity threshold is set to 0.1 (for example), then the population with the worst fitness will be reinitialized. Individual, This indicates the number of individuals to be reset, and the calculation method is as follows: ;

[0289] in This indicates the reset ratio, which determines how many worst-case individuals are reset; an example value is 0.2.

[0290] Furthermore, the new individual, based on the current optimal solution, is generated by adding perturbations as follows:

[0291] ;

[0292] In the formula, This indicates the reset perturbation magnitude, controlling the range of perturbation around the optimal solution for each reset individual. An example value is 0.1.

[0293] Indicates in A uniformly distributed random vector within the interval is used to generate the reset perturbation.

[0294] 2) Termination of judgment:

[0295] The iteration terminates and the current globally optimal control strategy is output when any of the following conditions are met:

[0296] Condition 1: Current iteration number Reaching the preset maximum number of iterations ;

[0297] Condition 2: Global optimal fitness value In continuous In the next iteration, the change is less than a preset threshold (preferably set to...). ),in, This represents the globally optimal fitness value. This represents the number of iterations for determining convergence. An example value of 50 indicates that if the global optimal fitness changes by less than 50% in 50 consecutive iterations... If the algorithm fails to converge, it is considered to have converged and terminated prematurely.

[0298] Based on the globally optimal individual whale position output after the iterative convergence of the improved whale optimization algorithm, the optimal control strategy vector that satisfies the conditions of minimizing energy consumption and temperature and humidity constraints is obtained. This vector fully describes the temperature setpoint and humidity setpoint of each independent control zone per hour within the future optimization time window.

[0299] In the actual control of the storage environment, the first hour set value of the optimal strategy is first issued as an instruction to the programmable logic controllers of the air conditioning units, dehumidifiers, humidifiers and ventilation systems in the corresponding areas through the central monitoring system. The equipment then adjusts its operating conditions according to the new set value, so as to operate with the lowest energy consumption while ensuring the safety of the stored items.

[0300] Because the warehousing environment is constantly changing due to factors such as outdoor weather disturbances, goods entering and leaving, and equipment start-up and shutdown, a single optimization cannot cover the entire time period. Therefore, a rolling optimization mechanism is adopted to achieve dynamic closed-loop control: at the end of each control cycle (e.g., every hour), the system re-collects all sensor-measured temperature and humidity data, the latest outdoor weather forecast data, and real-time energy consumption feedback of the equipment at the current moment, and inputs them as new initial states into the improved whale optimization algorithm; based on the updated multi-scale normalized features and cluster center information, the algorithm re-initializes the population and executes an adaptive search and constraint processing flow to quickly solve for the optimal control strategy for the next optimization window (e.g., the next six hours); this rolling process is repeated to ensure that the control decision is always adjusted in real time to follow the environmental evolution, which not only keeps the long-term energy-saving goal close but also flexibly responds to short-term disturbances.

[0301] The entire dynamic optimization process is embedded in the warehouse environment management platform. The algorithm module, data acquisition module, energy consumption simulation module, and equipment control module form a closed-loop link: the data layer continuously provides fresh samples for normalization and feature fusion, and the cluster centers and statistical parameters are recalculated at fixed intervals (e.g., weekly) to maintain an accurate characterization of the current environmental pattern; each iteration of the optimization layer generates an initial population based on the latest fused features, and automatically balances constraint satisfaction and energy consumption targets through dynamic penalty coefficients and reward items in fitness evaluation; the execution layer is responsible for strategy parsing and instruction issuance, and feeds back the actual running data to the database for subsequent model calibration and parameter adaptive adjustment.

[0302] Through the above mechanism, this application can achieve a leap from static setting to dynamic optimization in warehouse environment control, while ensuring that the temperature and humidity are always within the allowable range, and significantly reducing the long-term operating energy consumption of refrigeration, heating and dehumidification equipment.

[0303] In one embodiment, such as Figure 2As shown, a biaxial bar chart analysis comparing temperature and humidity constraint violations was conducted to evaluate the degree to which the preset allowable ranges of temperature and humidity in the storage environment were met under different control strategies, thus verifying the application's ability to handle hard constraints. The allowable temperature range was set to 18 to 25 degrees Celsius, and the allowable humidity range was 40% to 60%. The horizontal axis of the figure also represents six control strategies, the left vertical axis represents the temperature violation amount in degrees Celsius squared, and the right vertical axis represents the humidity violation amount in percentage squared. The violation amount is defined as the sum of squares of the deviations of the simulated temperature or humidity values ​​at each location and time from the allowable range, and is a key indicator for measuring the compliance of the strategy. Data shows that this technology almost completely satisfies the constraints, while the violation amounts of the standard whale optimization algorithm, genetic algorithm, proportional-integral-derivative controller, and rule control method are all higher than that of this technology. Experimental results show that this application introduces a dynamic penalty coefficient based on the total historical violation amount and a sigmoid smoothing mechanism in the fitness evaluation. In the early stage of iteration, slight overshoots are appropriately tolerated to broaden the search space, while in the later stage of iteration, any violation is strictly punished, and a reward item is set to guide the temperature to approach the optimal set value. Therefore, this technology can optimize energy consumption while keeping temperature and humidity within the specified range, proving that the constraint processing strategy proposed in this application has extremely high feasibility and robustness.

[0304] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for dynamic control and optimization of a warehouse environment based on artificial intelligence, characterized in that, The method includes: Perform wavelet decomposition on the original warehousing environment data and calculate the normalized wavelet coefficients; Based on normalized wavelet coefficients, historical control strategy, and random weights, the initial position of each individual whale involved in the whale optimization algorithm is calculated. Based on the randomly generated switching probability, the search method is selected, and chaotic perturbation and local enhancement factor are introduced to iteratively update the position of individual whales; The convergence coefficient, adaptive adjustment factor, and spiral shape parameters involved in the whale optimization algorithm are adjusted by introducing a population diversity index; a fitness function that integrates the total energy consumption target and the constraint violation penalty is adopted, and a dynamic penalty coefficient adjustment strategy based on the total historical violation is used, combined with fuzzy logic to make the penalty change smoothly with the iteration process. After each iteration, the next generation of whale population is constructed through elite preservation, crossbreeding operations, and diversity maintenance; when the preset termination condition is met, the current global best individual whale position is output. Based on the current location of the best individual whale in the global system, obtain the optimal storage environment data; The process of selecting the search method based on randomly generated switching probabilities, and introducing chaotic perturbations and local enhancement factors to iteratively update the position of individual whales specifically includes: Before updating the position in each iteration, a random number uniformly distributed between 0 and 1 is generated. If the generated switching probability is less than the preset probability threshold, an encirclement search is performed; otherwise, a spiral search is performed. When performing a bounding search, a whale individual's position is randomly selected from the current population as a reference position; the distance between the current individual and the reference position is calculated, and the basic search step size is determined by combining the adaptively adjusted convergence coefficient and the adaptive adjustment factor; a logistic chaotic perturbation factor is introduced to perturb and adjust the basic search results; after completing the calculation, the updated whale individual positions are obtained. When performing a spiral search, the distance between the current whale individual and the current global best individual is calculated; a basic spiral movement trajectory is generated according to the spiral update rule; a normal distribution local enhancement factor is superimposed to perturb the optimal solution; after the calculation is completed, the position of the whale individual after this spiral search update is obtained. Complete a single iteration of position update, and determine the final position of the current whale individual after this iteration based on the selected search method and corresponding perturbation enhancement processing, thus completing the position iteration update of a single individual; The entire population is traversed and updated, repeating this process sequentially for all individual whales in the population to complete the current round of position updates for all individuals; In specific implementation, the first The whale in the The position update method for the next iteration is represented as follows: In the formula, Indicates the first The whale in the The position at the nth iteration represents the... The position of each whale in the next iteration, i.e., the updated control strategy; This indicates a randomly selected whale location within the current population, used to guide exploration of unknown areas; This represents the convergence coefficient, used to control the encirclement step size, and its value decreases adaptively with iteration; This represents an adaptive adjustment factor, which is used to adjust the search step size based on the dynamic changes in population diversity. Indicates the first The whale in the The position at the nth iteration represents the... The position of each whale in the current iteration, i.e., the current control strategy; The chaotic perturbation factor is obtained through iterative logistic mapping, i.e. ; This indicates an assignment operation; Indicates the probability of behavior switching; This represents the distance vector between the current individual and the current global best individual; This indicates the position of the whale with the best global fitness in the current iteration; Indicates the parameters of the spiral shape; Indicates a A random number that is uniformly distributed within an interval; Represents the natural constant. Characterizes the magnitude of spiral renewal; Represents the cosine function; This represents a local enhancement factor.

2. The method for dynamic control and optimization of warehouse environment based on artificial intelligence according to claim 1, characterized in that, Before performing wavelet decomposition on the original warehousing environment data and calculating the normalized wavelet coefficients, the method further includes: In the current warehousing environment, deploy a network of temperature and humidity sensors covering all storage areas; Furthermore, the temperature and humidity sensor network involves sensors that automatically collect temperature and humidity data at fixed time intervals; Simultaneously, environmental data and environmental control equipment data corresponding to the warehouse's external envelope structure are collected as boundary conditions.

3. The method for dynamic control and optimization of warehouse environment based on artificial intelligence according to claim 1, characterized in that, Wavelet decomposition is performed on the original warehousing environment data to calculate the normalized wavelet coefficients, specifically including: Wavelet decomposition is performed on the original warehousing environment data to obtain multi-scale wavelet coefficients; For each scale and spatial location, based on historical wavelet coefficient data, the mean and standard deviation of the wavelet coefficients at that scale and location are calculated respectively. Using a dynamic time warping algorithm, wavelet coefficient data sequences at different spatial locations are time-phase aligned, and the time offset parameters corresponding to each scale and each location are calculated and determined. K-means clustering analysis was performed on historical wavelet coefficient data. The number of clusters was determined by the elbow method, and the cluster centers of wavelet coefficients at each scale and location were obtained. Based on the mean and standard deviation, the wavelet coefficients are scaled. Then, a Gaussian window function is used to weight the time offset, improving the smoothness and consistency of the time features. Finally, the deviation of the current wavelet coefficients from the typical environmental pattern is calculated by using the natural exponential function, combined with the normalized decay coefficient and the cluster center, and the normalized wavelet coefficients are obtained.

4. The method for dynamic control and optimization of warehouse environment based on artificial intelligence according to claim 1, characterized in that, Based on normalized wavelet coefficients, historical control strategies, and random weights, the initial position of each individual whale involved in the whale optimization algorithm is calculated, specifically including: Extract the normalized wavelet coefficients that have been calculated and the historical control strategy data of the warehousing environment, and generate the random weight parameters required by the algorithm. Based on the number of target parameters for warehouse environment control optimization, the total number of individuals in the whale optimization algorithm and the dimension size corresponding to each whale individual are set. For each individual whale, the normalized wavelet coefficients, the matched historical control strategy parameters, and the generated random weight values ​​for each dimension parameter of the individual are extracted. According to the allocated basic weight ratio, the three types of data are weighted and fused. Through numerical superposition and constraint adjustment, the initial position parameter value of the current dimension of the individual whale is calculated. By comparing the actual range of values ​​for the warehouse environment control parameters, boundary constraints are applied to the calculated initial position parameters, and values ​​that exceed the reasonable range are eliminated. This continues until all dimensional parameters of all individual whales have been calculated, forming a complete initial position set for the whale optimization algorithm.

5. The method for dynamic control and optimization of warehouse environment based on artificial intelligence according to claim 1, characterized in that, The convergence coefficient, adaptive adjustment factor, and spiral shape parameters involved in the whale optimization algorithm are adjusted by introducing a population diversity index, specifically including: Calculate the mean vector of the current positions of all individual whales, and then calculate the standard deviation of the positions of all individuals in the population based on the mean vector; The convergence coefficient is calculated using the nonlinearly decaying convergence factor and the first random number; wherein the convergence factor gradually decays from the initial value to the preset minimum value according to a nonlinear exponential law. The adaptive adjustment factor is obtained by multiplying the second random number and the population diversity adaptive adjustment coefficient; wherein, the population diversity adaptive adjustment coefficient is calculated in real time based on the population diversity index. An oscillating decay model is adopted, in which the initial spiral shape parameter value decays exponentially with the number of iterations, while sinusoidal periodic fluctuations are superimposed to calculate the spiral shape parameter.

6. The method for dynamic control and optimization of warehouse environment based on artificial intelligence according to claim 1, characterized in that, A fitness function that integrates the total energy consumption target and constraint violation penalties is adopted, and a dynamic penalty coefficient adjustment strategy based on the total historical violations is employed, specifically including: Thermodynamic and mass transfer models were used to simulate the temperature and humidity distribution at different locations and times in the warehouse under the current control strategy. Based on the equipment operating parameters and operating time, the power consumption of the equipment in the environment is added to obtain the total energy consumption value of the control strategy; Determine whether the temperature exceeds the preset upper and lower limits at each location and at each time point, calculate the square value of the temperature deviation exceeding the limit and sum them to obtain the total amount of temperature constraint violation; The sum of squares of deviations from the preset humidity range is calculated synchronously to obtain the total amount of humidity constraint violation; Based on the basic penalty coefficients for temperature and humidity, and combined with the total number of constraint violations in the current iteration and the historical maximum total number of violations, the basic penalty magnitude is calculated. The sigmoid function is used to make the penalty coefficient increase smoothly with the number of iterations; Generate dynamic penalty coefficients for temperature and humidity respectively; Calculate the absolute deviation between the simulated temperature and the optimal temperature setpoint at each location and time, and generate a reward value through an exponential decay function; The total reward of the control strategy is obtained by summing the reward values ​​at all spatiotemporal points. The core optimization objective is to use the negative value of total energy consumption, plus the dynamic penalty negative value for violations of temperature and humidity constraints, and the reward term for the optimal temperature. Each iteration recalculates the total number of constraint violations, dynamic penalty coefficient, and reward term for the current population.

7. The method for dynamic control and optimization of warehouse environment based on artificial intelligence according to claim 1, characterized in that, After each iteration, the next generation of whale populations is constructed through elite preservation, crossbreeding, and diversity maintenance, specifically including: Calculate the fitness value of all current whale individuals and sort them from highest to lowest fitness. According to the set elite ratio, select the top few individuals with the highest fitness and directly retain them to enter the next generation of the population; From the non-elite individuals that were never retained, two different individuals were randomly selected as parents. The control strategy vectors of the two parents are convexly combined and fused according to random weights to generate a basic new individual; Injecting adaptively normally distributed noise into the base new individuals generates new hybrid individuals; Repeat this process to generate a sufficient number of new hybrid individuals to supplement the next generation of the population; Calculate the overall distribution standard deviation of the current population to obtain population diversity; If the diversity is higher than the preset diversity threshold, no adjustment is made and the population construction is completed directly; if the diversity is lower than the preset diversity threshold, the population reset mechanism is triggered. Select several individuals with the worst fitness in the current population, and add uniform random perturbation to the global optimal control strategy to generate new replacement individuals; Replace the worst individual with the newly generated replacement individual; By integrating elite individuals, hybridizing to generate new individuals, and maintaining diversity among the reset individuals, a complete next-generation whale population is formed.

Images