Intelligent washing parameter optimization decision-making method and system based on multimodal perception fusion

Abstract

The present invention relates to the field of computer data processing technology, and specifically provides an intelligent washing parameter optimization decision-making method based on multimodal perception fusion, comprising the following steps: real-time acquisition of stain characteristic parameters of the fabric to be cleaned through a multi-source acquisition module; construction of a collaborative control model, determination of the detergent type and ratio as well as mechanical force parameters based on the stain characteristic parameters, training the collaborative control model with the goal of maximizing the stain removal rate and with energy and water consumption as constraints, and outputting an optimal washing strategy; controlling the washing device to perform corresponding operations according to the washing strategy; real-time monitoring of stain residue and energy consumption anomaly data during the washing process, dynamic correction of the washing speed and washing time, and simultaneous adjustment of the detergent replenishment amount according to the chemical composition distribution. This method breaks through the limitations of existing single-modal recognition technology in complex stain analysis, avoids resource waste and fabric damage caused by traditional extensive washing, and effectively resolves the technical contradiction between decontamination effect and resource consumption in the existing technology.

Description

Technical Field

[0001] The present invention relates to the field of computer data processing technology, and in particular to a method and system for intelligent washing parameter optimization decision-making based on multimodal perception fusion. Background Art

[0002] Smart washing technology integrates the Internet of Things (IoT), artificial intelligence (AI), and sensors to automate and intelligently upgrade the washing process. However, current technology still faces significant bottlenecks when dealing with complex stain identification and processing. For example, when dealing with mixed stains, such as oil and blood stains adhering to the fabric surface, or pigment stains penetrating deep into the fibers, existing single-modality identification technologies (such as those relying solely on RGB image color segmentation or near-infrared spectroscopy) are unable to accurately analyze the chemical composition of the stains (such as the ratio of proteins, lipids, and pigments), adhesion strength, and fiber penetration depth, leading to fundamental flaws in washing strategy formulation.

[0003] To address this issue, existing technologies often employ crude measures such as increasing detergent dosage (exceeding the standard single-time dosage), extending the number of wash cycles, and improving mechanical strength (increasing the speed). For example, repeatedly applying chlorine bleach to mixed stains and combining it with a high-temperature wash at temperatures above 60°C improves cleaning efficiency, but also increases reagent consumption, water consumption, heating energy consumption, and fabric damage.

[0004] The core contradiction lies in the lack of a dynamic control mechanism for "stains, parameters, and energy consumption" in existing technologies, resulting in a dilemma where "strong stain removal leads to greater waste, while energy conservation leads to weaker stain removal." Breaking through the collaborative control technology of precise multimodal stain identification and dynamic energy optimization, and achieving the unity of stain removal and energy conservation, is one of the core challenges of current smart laundry technology. Summary of the Invention

[0005] In order to overcome the deficiencies of the prior art, the present invention provides an intelligent washing parameter optimization decision method and system based on multimodal perception fusion to solve the problems in the prior art.

[0006] One embodiment of the present invention provides a smart washing parameter optimization decision method based on multimodal perception fusion, comprising the following steps:

[0007] S10, collecting stain characteristic parameters of the fabric to be cleaned in real time through a multi-source collection module, wherein the stain characteristic parameters include at least the chemical composition distribution of the stain, the fiber penetration depth, and the stain binding strength;

[0008] S20. Utilizing a machine learning algorithm to construct a collaborative control model, determining the type and ratio of detergent based on the chemical composition distribution of the stain, calculating mechanical parameters including wash speed, wash time, and mechanical strength based on the fiber penetration depth and stain binding strength, and training the collaborative control model with maximizing stain removal rate as the goal and energy and water consumption as constraints to output an optimal washing strategy.

[0009] S30: Generate corresponding washing control parameters according to the washing strategy to control the washing device to perform corresponding operations, wherein the operations include at least one of the following:

[0010] Detergent loading is performed in compartments based on the chemical composition of the stains;

[0011] The corresponding water temperature is adjusted based on the intersection of the fiber temperature resistance threshold and the enzyme preparation activity range;

[0012] Adjust the rotation speed based on the fiber penetration depth;

[0013] Matching the amount of water to the stain severity calculated based on the chemical composition distribution of the stain, fiber penetration depth, and stain binding strength;

[0014] S40: Real-time monitoring of stain residue and abnormal energy consumption data during the washing process, comparison of the initially collected fiber penetration depth and stain binding strength with the real-time mechanical force parameters, dynamic correction of the washing speed and washing time, and simultaneous adjustment of the detergent replenishment amount according to the chemical composition distribution.

[0015] By adopting the above solution, multi-dimensional characteristic parameters such as the chemical composition distribution of stains, fiber penetration depth and bonding strength are obtained in real time through a multi-source acquisition module. The collaborative control model constructed by the machine learning algorithm realizes the accurate calculation of detergent type ratio and mechanical force parameters, and generates the optimal washing strategy with energy and water consumption as constraints, thereby driving operations such as detergent compartment delivery, coordinated adjustment of water temperature and enzyme activity range, dynamic matching of speed and water volume, etc. At the same time, dynamic correction of washing parameters is achieved through real-time monitoring of stain residue and energy consumption data. This solution breaks through the limitations of existing single-modal recognition technology for complex stain analysis, avoids the waste of resources and fabric damage caused by traditional extensive washing, and achieves efficient removal of complex stains, optimal allocation of washing resources and targeted protection of fabric materials through the closed-loop collaborative control of "stain characteristics-washing parameters-energy consumption control", effectively solving the technical contradiction of "decontamination effect and resource consumption" in existing technologies, and providing an integrated solution for smart washing that takes into account precision, energy saving and universality.

[0016] In one embodiment, in step S20, constructing the collaborative regulation model specifically includes:

[0017] S211, extracting and converting features of the chemical composition distribution of the stain, fiber penetration depth, and stain binding strength, including:

[0018] The chemical composition distribution of stains is converted into a category feature vector through semantic analysis;

[0019] Normalize the fiber penetration depth and stain binding strength to generate a numerical feature vector;

[0020] S212: Construct a multi-branch neural network model, the model comprising:

[0021] The first processing branch receives the category feature vector and generates a predicted distribution of detergent types and ratios through a fully connected layer;

[0022] The second processing branch receives the numerical feature vector and generates optimization suggestions for mechanical force parameters including washing speed, washing time and mechanical force intensity through a convolutional layer;

[0023] S213. Establish the following mapping relationship in the multi-branch neural network model:

[0024] Detergent selection function based on stain chemical composition distribution and fiber penetration depth;

[0025] Mechanical force parameter optimization function based on fiber penetration depth and stain binding strength.

[0026] By adopting the above solution, an innovative solution based on feature extraction and transformation, a multi-branch neural network architecture, and mapping relationships is developed. Stain and fabric features are converted into vector representations that can be processed by the model. A multi-branch neural network is used to accurately predict detergent ratios and mechanical force parameters, and a mapping relationship between stain features and washing parameters is established. This solution breaks through traditional washing parameter setting methods and achieves adaptive matching of stain features to washing strategies through intelligent modeling. This effectively improves the relevance and effectiveness of washing strategies, avoids problems such as overwashing or undercleaning, and optimizes the allocation of washing resources.

[0027] In one embodiment, in step S20, training the collaborative regulation model specifically includes:

[0028] S221. Constructing a training dataset comprising input features and supervisory labels based on the chemical composition distribution of the stains, the fiber penetration depth, and the stain binding strength; wherein the input features are the categorical feature vectors and the numerical feature vectors, and the supervisory labels are the actual washing parameters corresponding to the input features and the actual constraint values ​​of energy and water consumption;

[0029] S222. Construct a multi-objective optimization framework with minimizing the deviation between the predicted washing parameters and the measured values ​​as the main objective and with the energy and water consumption not exceeding the preset thresholds as the constraint condition;

[0030] S223, using a batch stochastic gradient descent algorithm to iteratively optimize the parameters of the multi-branch neural network model, including:

[0031] Dynamically adjust network weights to minimize the objective function;

[0032] Real-time evaluation of the degree of satisfaction of energy and water consumption constraints;

[0033] When the validation set loss function converges and the energy consumption constraint is continuously satisfied, the trained collaborative control model is output.

[0034] By adopting the above solution, a complete model training system is formed through the construction of a training dataset, the design of a multi-objective optimization framework, and an iterative training strategy. A training dataset containing stain characteristics, actual washing parameters, and energy consumption constraints is constructed to provide reliable learning samples for the model; a multi-objective optimization framework is established to balance washing performance and energy consumption constraints; and an iterative optimization algorithm is used to achieve dynamic adjustment of model parameters. This solution solves problems such as poor adaptability and uncontrollable resource consumption in machine learning model training, significantly improving the model's generalization and practicality, making washing strategy predictions more accurate and achieving coordinated optimization of washing performance and resource utilization efficiency.

[0035] In one embodiment, the step S30 further includes: using graded washing to divide the washing into a pre-wash stage, a main wash stage, and a rinse stage, wherein:

[0036] During the pre-washing phase, targeted enzyme preparations are added based on the chemical composition distribution of the stains, and the mechanical force parameters are reduced for a preset operation time;

[0037] For the main wash phase, the mechanical force parameters and wash duration are dynamically adjusted based on the fiber penetration depth, stain binding strength, and stain residue monitoring results from the pre-wash phase;

[0038] During the rinsing phase, the number of rinses and the amount of water are dynamically adjusted based on the real-time monitoring of the residual detergent concentration. When the residual detergent concentration exceeds a preset threshold, an additional rinse cycle is triggered, and the amount of water used in each additional rinse is a preset proportion of the initial water volume.

[0039] In each stage of the graded washing, the data of residual stains and abnormal energy consumption are monitored in real time through step S40, and the washing parameters of each stage are corrected synchronously.

[0040] By adopting the above solution and using graded washing technology, the washing process is divided into pre-wash, main wash, and rinsing stages. In the pre-wash stage, targeted enzyme preparations are added based on the chemical composition distribution of the stains and mechanical force pretreatment is reduced. In the main wash stage, mechanical force parameters are dynamically adjusted based on the fiber penetration depth and stain binding strength. In the rinsing stage, the rinsing strategy is dynamically optimized based on the residual detergent concentration. This solution breaks through the limitations of traditional fixed-program washing and achieves a step-by-step removal of stains through the coordinated use of phased parameters. This not only avoids reagent waste in the pre-wash stage, but also resolves the contradiction between mechanical force and material matching in the main wash stage. At the same time, detergent residue is reduced through intelligent regulation in the rinsing stage. While improving the efficiency of complex stain removal, it achieves precise allocation of washing resources and targeted protection of fabric materials, effectively balancing stain removal effect, energy consumption control, and clothing protection performance.

[0041] In one embodiment, the stain characteristic parameter further includes mixed stains. When the multi-source collection module collects mixed stains, the main wash stage further includes:

[0042] Obtaining layered distribution parameters of mixed stains inside the fiber, wherein the layered distribution parameters include surface stain thickness, deep stain penetration depth, and layered interface bonding strength;

[0043] Generating a step-by-step mechanical force adjustment strategy based on the hierarchical distribution parameters includes:

[0044] When the thickness of the surface stain exceeds a first preset threshold, treating the stain with a first mechanical strength and a corresponding enzyme preparation for a first preset time;

[0045] When the penetration depth of deep stains exceeds a second preset threshold, the second mechanical force intensity is increased and the corresponding enzyme preparation is switched to treat for a second preset time;

[0046] The first mechanical force intensity is lower than the second mechanical force intensity, and the type of enzyme preparation in the two stages is dynamically selected based on the chemical composition distribution of the stain;

[0047] The ratio between the first preset time and the second preset time is dynamically adjusted based on the layered interface bonding strength. When the layered interface bonding strength exceeds a third preset threshold, the processing time of the first preset time is extended.

[0048] By adopting the above solution, within the framework of graded washing, we add layered distribution parameter collection and a step-by-step mechanical force adjustment strategy for mixed stains. By obtaining the thickness of surface stains, the depth of deep penetration, and the layered interface bonding strength, we dynamically generate a differentiated washing solution that treats the surface first and then the deep layer, combined with targeted switching of enzyme preparations and a step-by-step increase in mechanical force strength. This solution solves the problem of over-treatment of the surface layer and residue in the deep layer caused by the layered distribution of mixed stains. By dynamically adjusting the phase duration ratio of the interface bonding strength, we achieve the layer-by-layer disintegration of complex mixed stains such as "oil stains wrapped in blood stains." While ensuring the removal rate of deep stains, it reduces the risk of mechanical damage to surface fibers, significantly improving the treatment efficiency and washing uniformity of mixed stains compared to traditional single procedures.

[0049] In one embodiment, the determination condition of the mixed stain is:

[0050] The chemical composition distribution of the stains simultaneously includes characteristic components of at least two different types of stains, and the multi-source acquisition module detects that the stain distribution inside the fiber presents a stratified characteristic, and the stratified interface bonding strength is greater than a preset value.

[0051] By adopting this solution, the accuracy and reliability of mixed stain identification are ensured by limiting the dual criteria for mixed stain identification: the chemical composition distribution must contain at least two characteristic stain components and the fiber exhibits a layered distribution. This judgment logic avoids wash strategy deviations caused by misidentifying a single stain as a mixed stain, providing an accurate prerequisite for layered processing. This allows the subsequent step-by-step mechanical force adjustment strategy to be targeted in real mixed stain scenarios, ensuring the effectiveness of the wash plan from the source, avoiding reagent waste, increased energy consumption, and fabric damage caused by misidentification of stain types, and improving the robustness and adaptability of the entire smart washing system.

[0052] In one embodiment, in step S40, dynamically modifying the washing strategy specifically includes:

[0053] Real-time collection of residual stain data and energy consumption data during the washing process to determine whether there is any abnormal energy consumption;

[0054] Comparing and analyzing the fiber penetration depth and stain binding strength initially collected in step S10 with the washing speed and mechanical strength of the current washing stage;

[0055] Based on the comparison results, the washing speed and washing time are dynamically adjusted, including:

[0056] When the amount of residual dirt is higher than the preset threshold and the mechanical force parameter has not reached the upper limit, the washing speed is increased or the washing time is extended;

[0057] When the energy consumption data exceeds the preset threshold, the washing speed is reduced or the washing time is shortened;

[0058] Adjust the amount of detergent added based on the real-time monitoring of the chemical composition of the stains, including:

[0059] If an increase in the residual proportion of a certain type of stain component is detected, the corresponding type of enzyme preparation or detergent will be added in a targeted manner;

[0060] If the chemical composition of the stains tends to be stable, reduce the amount of detergent.

[0061] By adopting the above solution, a closed-loop control system of "real-time monitoring - feature comparison - parameter adjustment - detergent replenishment" is constructed. This solution realizes washing status perception through multi-dimensional data collection and energy consumption anomaly judgment. It forms a correction basis based on the comparative analysis of the initial characteristics of the fabric and the current parameters, dynamically adjusts the mechanical force parameters based on the stain residue and energy consumption threshold, and realizes precise detergent replenishment based on the chemical composition distribution. It breaks through the limitations of traditional fixed-program washing and improves the adaptability of washing strategies to complex stain scenarios through a data-driven adaptive correction mechanism, achieving the coordinated optimization of washing efficiency and resource consumption.

[0062] This application also relates to an intelligent washing parameter optimization decision system based on multimodal perception fusion, including:

[0063] A data acquisition module, configured to acquire, in real time, stain characteristic parameters of the fabric to be cleaned through a multi-source acquisition module, wherein the stain characteristic parameters include at least the chemical composition distribution of the stain, the fiber penetration depth, and the stain binding strength;

[0064] a strategy generation module for constructing a collaborative control model using a machine learning algorithm, determining the type and ratio of detergent based on the chemical composition distribution of the stain, calculating mechanical force parameters including wash speed, wash time, and mechanical force strength based on the fiber penetration depth and stain binding strength, training the collaborative control model with the goal of maximizing stain removal rate and energy and water consumption as constraints, and outputting an optimal washing strategy;

[0065] a washing execution module, configured to generate corresponding washing control parameters according to the washing strategy to control the washing device to perform corresponding operations;

[0066] The dynamic adjustment module is used to monitor the stain residue and energy consumption abnormality data during the washing process in real time, compare the initially collected fiber penetration depth and stain binding strength with the real-time mechanical force parameters, dynamically correct the washing speed and washing time, and simultaneously adjust the detergent replenishment amount according to the chemical composition distribution.

[0067] By adopting the above solution, multi-dimensional characteristic parameters such as the chemical composition distribution of stains, fiber penetration depth and bonding strength are obtained in real time through a multi-source acquisition module. The collaborative control model constructed by the machine learning algorithm realizes the accurate calculation of detergent type ratio and mechanical force parameters, and generates the optimal washing strategy with energy and water consumption as constraints, thereby driving operations such as detergent compartment delivery, coordinated adjustment of water temperature and enzyme activity range, dynamic matching of speed and water volume, etc. At the same time, dynamic correction of washing parameters is achieved through real-time monitoring of stain residue and energy consumption data. This solution breaks through the limitations of existing single-modal recognition technology for complex stain analysis, avoids the waste of resources and fabric damage caused by traditional extensive washing, and achieves efficient removal of complex stains, optimal allocation of washing resources and targeted protection of fabric materials through the closed-loop collaborative control of "stain characteristics-washing parameters-energy consumption control", effectively solving the technical contradiction of "decontamination effect and resource consumption" in existing technologies, and providing an integrated solution for smart washing that takes into account precision, energy saving and universality.

[0068] The present application also relates to a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned intelligent washing parameter optimization decision method based on multimodal perception fusion are implemented.

[0069] The present application also relates to a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-mentioned intelligent washing parameter optimization decision method based on multimodal perception fusion.

[0070] The above embodiments provide a method and system for intelligent washing parameter optimization decision-making based on multimodal perception fusion, which has the following beneficial effects:

[0071] 1. Through the multi-source acquisition module, multi-dimensional characteristic parameters such as the chemical composition distribution of stains, fiber penetration depth, and bonding strength are acquired in real time. A collaborative control model constructed using a machine learning algorithm is used to accurately calculate the detergent type ratio and mechanical force parameters. The optimal washing strategy is generated with energy and water consumption as constraints, thereby driving operations such as detergent compartment delivery, coordinated adjustment of water temperature and enzyme activity range, and dynamic matching of rotation speed and water volume. At the same time, dynamic correction of washing parameters is achieved through real-time monitoring of stain residue and energy consumption data. This solution breaks through the limitations of existing single-modal recognition technology in complex stain analysis, avoids the waste of resources and fabric damage caused by traditional extensive washing, and achieves efficient removal of complex stains, optimal allocation of washing resources, and targeted protection of fabric materials through the closed-loop collaborative control of "stain characteristics-washing parameters-energy consumption control". It effectively resolves the technical contradiction between "decontamination effect and resource consumption" in existing technologies, and provides an integrated solution for smart washing that balances precision, energy saving, and universality.

[0072] 2. In one embodiment, a graded washing technique is used to divide the washing process into pre-wash, main wash, and rinse stages. The pre-wash stage administers targeted enzyme preparations based on the chemical composition of the stains and reduces mechanical pretreatment. The main wash stage dynamically adjusts mechanical parameters based on fiber penetration depth and stain binding strength. The rinse stage dynamically optimizes the rinsing strategy based on residual detergent concentration. This solution overcomes the limitations of traditional fixed-cycle washing by achieving a step-by-step stain removal process through the coordinated use of phased parameters. This avoids reagent waste in the pre-wash stage and resolves the conflict between mechanical force and material matching in the main wash stage. Furthermore, intelligent control of the rinse stage reduces detergent residue, improving the efficiency of complex stain removal while achieving precise allocation of washing resources and targeted protection of fabric materials, effectively balancing stain removal effectiveness, energy consumption control, and garment protection.

[0073] 3. In one embodiment, within the framework of graded washing, a layered distribution parameter collection and a step-by-step mechanical force adjustment strategy are added for mixed stains. By obtaining the thickness of surface stains, the depth of deep penetration, and the layered interface bonding strength, a differentiated washing scheme is dynamically generated that treats the surface first and then the deep layers. This scheme is combined with targeted switching of enzyme preparations and a step-by-step increase in mechanical force strength. This solution solves the problem of over-treatment of the surface layer and residual residue in the deep layer caused by the layered distribution of mixed stains. By dynamically adjusting the phase duration ratio of the interface bonding strength, it achieves the layer-by-layer disintegration of complex mixed stains such as "oil stains wrapped in blood stains." While ensuring the removal rate of deep-seated stains, it reduces the risk of mechanical damage to surface fibers, significantly improving the treatment efficiency and washing uniformity of mixed stains compared to traditional single-step procedures.

[0074] 4. In one embodiment, a closed-loop control system consisting of "real-time monitoring - feature comparison - parameter adjustment - detergent replenishment" is constructed. This solution achieves washing status perception through multi-dimensional data collection and energy consumption anomaly detection. Correction is based on comparative analysis of initial fabric features and current parameters. Mechanical force parameters are dynamically adjusted based on residual stains and energy consumption thresholds, and detergent is precisely replenished based on chemical composition distribution. This overcomes the limitations of traditional fixed-program washing and, through a data-driven adaptive correction mechanism, improves the adaptability of washing strategies to complex staining scenarios, achieving the coordinated optimization of washing efficiency and resource consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0076] Figure 1 A flowchart of a smart washing parameter optimization decision-making method based on multimodal perception fusion provided by an embodiment of the present invention;

[0077] Figure 2 A block diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0078] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0079] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture. If the specific posture changes, the directional indications will also change accordingly.

[0080] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited to "first" and "second" may explicitly or implicitly include at least one of such features. In addition, if "and / or" or "and / or" appears in the full text, its meaning includes three parallel solutions. Taking "A and / or B" as an example, it includes solution A, solution B, or solutions that satisfy both A and B. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in this field to implement. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0081] Reference Figure 1 One embodiment of the present invention provides a smart washing parameter optimization decision method based on multimodal perception fusion, comprising the following steps:

[0082] S10. Collecting stain characteristic parameters of the fabric to be cleaned in real time through a multi-source collection module, wherein the stain characteristic parameters at least include the chemical composition distribution of the stain, the fiber penetration depth, and the stain binding strength.

[0083] S20. Utilize a machine learning algorithm to construct a collaborative control model, determine the detergent type and ratio based on the chemical composition distribution of the stains, calculate mechanical force parameters including washing speed, washing time, and mechanical force strength based on the fiber penetration depth and stain binding strength, and train the collaborative control model with the goal of maximizing the stain removal rate and energy and water consumption as constraints to output the optimal washing strategy.

[0084] The construction of the collaborative regulation model specifically includes:

[0085] S211, extracting and converting features of the chemical composition distribution of the stain, fiber penetration depth, and stain binding strength, including:

[0086] The chemical composition distribution of stains is converted into a category feature vector through semantic analysis;

[0087] The fiber penetration depth and stain binding strength are normalized to generate a numerical feature vector.

[0088] Specifically, natural language processing techniques are used to perform semantic analysis on the chemical composition distribution data output by the spectrometer. For example, "70% protein, 30% oil" is converted into an 8-dimensional categorical feature vector [0.7, 0.3, 0, 0, 0, 0, 0], where each dimension corresponds to one of eight stain components: protein, oil, and carbohydrates. Furthermore, a Min-Max normalization process is performed on the fiber penetration depth (range 0-2mm) and the stain binding strength (range 0-200N / m²), converting them into numerical feature vectors in the range [0, 1]. For example, a penetration depth of 0.6mm and a binding strength of 80N / m² are normalized to [0.3, 0.4].

[0089] S212: Construct a multi-branch neural network model, the model comprising:

[0090] The first processing branch receives the category feature vector and generates a predicted distribution of detergent types and ratios through a fully connected layer;

[0091] The second processing branch receives the numerical feature vector and generates mechanical force parameter optimization suggestions including washing speed, washing time and mechanical force intensity through a convolutional layer.

[0092] Specifically, the model is implemented using the TensorFlow framework, and the specific structure is as follows:

[0093] The first processing branch (detergent prediction) consists of three fully connected layers (with 64, 32, and 16 neurons, respectively). It takes as input an 8-dimensional class feature vector and outputs a 10-dimensional probability distribution of detergent types and a 3-dimensional ratio vector. For example, the output vector [0.8, 0.15, 0.05, ...] represents the probabilities of using protease, lipase, and amylase, while the ratio vector [0.7, 0.3, 0] indicates a 7:3 mixture of protease and lipase.

[0094] The second processing branch (mechanical force parameter optimization) uses two convolutional layers (3×3 kernel size, 16 and 32 channels, respectively) plus one fully connected layer (64 neurons). It takes as input a 2D numerical feature vector and outputs a 3D mechanical force parameter vector. For example, the output vector [1000, 45, 3] corresponds to a wash speed of 1000 rpm, a wash duration of 45 minutes, and a mechanical force intensity of level 3.

[0095] S213. Establish the following mapping relationship in the multi-branch neural network model:

[0096] Detergent selection function based on stain chemical composition distribution and fiber penetration depth;

[0097] Mechanical force parameter optimization function based on fiber penetration depth and stain binding strength.

[0098] Specifically:

[0099] For the detergent selection function, it is implemented through the softmax activation function, and the formula is:

[0100]

[0101] Among them, C is the category feature vector, are trainable weights and biases, outputting the probability distribution D of detergent types.

[0102] For the mechanical force parameter optimization function, the ReLU activation function is used, and the formula is:

[0103]

[0104] Among them, D is the numerical feature vector, Conv is the convolution operation, 、 is a trainable parameter, which outputs the mechanical force parameter vector M.

[0105] The training of the collaborative control model specifically includes:

[0106] S221. Based on the chemical composition distribution of the stains, the fiber penetration depth, and the stain binding strength, a training data set including input features and supervisory labels is constructed; wherein the input features are the category feature vectors and the numerical feature vectors, and the supervisory labels are the actual washing parameters corresponding to the input features and the actual constraint values ​​of energy and water consumption.

[0107] Specifically, a training dataset consisting of input features and supervisory labels is constructed based on the chemical composition distribution of stains, fiber penetration depth, and stain binding strength, as well as several sets of historical washing data collected in advance. The historical washing data is based on washing parameters under the same or similar conditions, including data on several stain types, fiber materials, washing parameters, and effect feedback.

[0108] Input features: categorical feature vectors (such as 0.7, 0.3, 0, ...), numerical feature vectors (such as 0.3, 0.4);

[0109] Supervision label: actual washing parameters (such as 1000, 45, 3, 0.7, 0.3, 0) and energy and water consumption constraints (such as 1.2kWh, 45L).

[0110] The data is divided into training set, validation set, and test set in a ratio of 8:1:1, and DataLoader is used to load data in batches.

[0111] S222. Construct a multi-objective optimization framework with the main goal of minimizing the deviation between the predicted washing parameters and the measured values, and the constraint that the energy and water consumption does not exceed the preset threshold.

[0112] Specifically, the main objective function is defined as the mean square error between the predicted parameters and the actual values:

[0113]

[0114] in, is the prediction parameter, is the actual parameter.

[0115] Introduce energy and water consumption constraints:

[0116]

[0117] Among them, E and W are the predicted energy and water consumption, 、 is a threshold value (such as 1.5kWh, 50L), 、 is the weight coefficient (initial value 0.1).

[0118] Total loss function:

[0119] S223, using a batch stochastic gradient descent algorithm to iteratively optimize the parameters of the multi-branch neural network model, including:

[0120] Dynamically adjust network weights to minimize the objective function;

[0121] Real-time evaluation of the degree of satisfaction of energy and water consumption constraints;

[0122] When the validation set loss function converges and the energy consumption constraint is continuously satisfied, the trained collaborative control model is output.

[0123] Specifically, the Adam optimizer is used to iteratively optimize the model parameters. During each batch training:

[0124] Forward propagation calculates the predicted value and loss function;

[0125] Back propagation updates the network weights, prioritizing reduction ;

[0126] If the energy and water consumption of the current batch exceeds the threshold, it will increase dynamically. 、 (e.g., for every 10% exceeding the threshold, the weight is ×1.2);

[0127] The training is stopped when the loss decreases by less than 0.01% for 10 consecutive rounds and the energy consumption constraint satisfaction rate is greater than 95%.

[0128] The trained collaborative control model is deployed on the main control chip of the washing machine, receives the characteristic parameters of step S10 in real time, and outputs the optimal washing strategy for execution in step S30.

[0129] During training, if the prediction error for a particular type of stain (such as blood) is consistently high, the system automatically increases the weight of that type of sample (e.g., repeating the sample three times) to improve the model's adaptability to that specific stain. In actual use, after each wash cycle, the actual parameters and results are fed back to the cloud. The collection of several new data sets triggers incremental model training to continuously optimize prediction accuracy.

[0130] S30: Generate corresponding washing control parameters according to the washing strategy to control the washing device to perform corresponding operations, wherein the operations include at least one of the following:

[0131] Detergent loading is performed in compartments based on the chemical composition of the stains;

[0132] The corresponding water temperature is adjusted based on the intersection of the fiber temperature resistance threshold and the enzyme preparation activity range;

[0133] Adjust the rotation speed based on the fiber penetration depth;

[0134] Matching the amount of water to the stain severity calculated based on the chemical composition distribution of the stain, fiber penetration depth, and stain binding strength;

[0135] S40: Real-time monitoring of stain residue and abnormal energy consumption data during the washing process, comparison of the initially collected fiber penetration depth and stain binding strength with the real-time mechanical force parameters, dynamic correction of the washing speed and washing time, and simultaneous adjustment of the detergent replenishment amount according to the chemical composition distribution.

[0136] In this embodiment, as described in step S10 above, real-time collection of stain characteristic parameters is implemented through a multi-source acquisition module. This module integrates multiple sensors, including a spectrum analyzer, a laser rangefinder, and an ultrasonic detector. The spectrum analyzer illuminates the stained area with a beam of light of a specific wavelength, analyzing the characteristic absorption peaks in the reflected spectrum to identify the distribution ratios of chemical components such as proteins, oils, and carbohydrates. The laser rangefinder emits pulsed laser light and measures the time difference between reflected light to calculate the penetration depth of the stain into the fiber. The ultrasonic detector emits ultrasonic waves and analyzes the attenuation characteristics of the echo signal to quantify the bond strength between the stain and the fiber. During the acquisition process, the system performs spatiotemporal registration and fusion processing on the multi-sensor data to eliminate data redundancy and improve the accuracy of the characteristic parameters. For example, for mixed stain areas, the system uses a data fusion algorithm to combine spectral and ultrasonic data to distinguish the penetration depth and bond strength of different components.

[0137] In this embodiment, as described in step S20 above, when constructing a collaborative control model using a machine learning algorithm, the chemical composition distribution of the stains collected in step S10 is first one-hot encoded and converted into a categorical feature vector that the model can process. Simultaneously, the fiber penetration depth and stain binding strength are normalized to form a numerical feature vector. The model utilizes a multi-branch neural network architecture, in which the detergent prediction branch comprises three fully connected layers, which output the probability distribution of detergent types and the optimal ratio. The mechanical force parameter prediction branch utilizes a convolutional neural network structure, which extracts spatial correlation information from the feature vectors to calculate the optimal wash speed, duration, and mechanical force intensity. During the model training phase, the primary objective function is to minimize the mean squared error between the predicted and actual stain removal rates. Energy and water consumption constraints are also introduced, and the network weights are iteratively optimized using a batch stochastic gradient descent algorithm. When the validation set loss function decreases by less than 0.1% for 10 consecutive rounds and the energy consumption constraint is satisfied at 95%, the model is considered converged and the optimal wash strategy is output.

[0138] In this embodiment, as described in step S30 above, when generating control parameters based on the washing strategy, if the chemical composition distribution of the stains indicates that protein stains account for more than 60%, the protease detergent compartment is activated for delivery and mixed with the main detergent according to the ratio output by the model. If the fiber temperature resistance threshold is 60°C and the enzyme preparation activity range is 40-55°C, the water temperature is adjusted to 50°C to balance fiber protection and enzyme activity. For deep stains with a fiber penetration depth exceeding 0.5mm, the washing speed is increased to 1200rpm to enhance mechanical scouring. Based on the stain severity calculation model, a weighted calculation is performed based on the chemical composition distribution of the stains, the fiber penetration depth, and the stain binding strength to output the corresponding stain severity. The output water volume requirement is matched to the stain severity, and the water intake is precisely controlled by a flow sensor. For example, for heavily soiled cotton and linen fabrics, the water volume is matched at a ratio of 15L of water per kilogram of fabric. During execution, the control system dynamically fine-tunes each parameter using a PID algorithm to ensure that the deviation between the actual operating state and the strategy output value does not exceed ±5%.

[0139] The stain severity calculation model uses a multi-feature weighted fusion algorithm to perform a linear weighted calculation on the stain's chemical composition distribution (weight 0.4), fiber penetration depth (weight 0.3), and stain binding strength (weight 0.3), outputting a severity value in the range [0, 1]. The specific implementation is as follows:

[0140] Input parameter processing: The chemical composition distribution of the stain (e.g., protein percentage, oil percentage), fiber penetration depth (unit: mm), and stain bonding strength (unit: N / m²) collected in step S10 are normalized and converted into a feature vector in the interval [0, 1]. For example, for a stain with a protein percentage of 70%, a penetration depth of 0.6 mm, and a bonding strength of 80 N / m², the normalized feature vector is [0.7, 0.6, 0.8].

[0141] Weighting mechanism: The weights of each feature are determined using the Analytic Hierarchy Process (AHP). The weights for the chemical composition of the stain are assigned 0.4, the fiber penetration depth 0.3, and the stain binding strength 0.3. The weight matrix is ​​iteratively optimized through correlation analysis of historical wash data. For example, the weight for penetration depth in a heavily soiled scenario can be dynamically adjusted to 0.4.

[0142] Severity is calculated using the linearly weighted formula S = 0.4 × C + 0.3 × D + 0.3 × B, where C is the normalized chemical composition distribution of the stain, D is the normalized fiber penetration depth, and B is the normalized stain binding strength. The calculated result, S, is mapped to a three-level severity scale: S < 0.4 (mild), 0.4 ≤ S < 0.7 (moderate), and S ≥ 0.7 (severe).

[0143] Water volume matching logic: The water volume requirement is output based on the severity of the contamination. Lightly contaminated fabric requires 10L of water per kilogram of fabric, moderately contaminated fabric requires 15L, and severely contaminated fabric requires 20L. For example, 0.3kg of heavily contaminated fabric requires 0.3 x 20 = 6L of water. The flow sensor uses a PID control algorithm to precisely adjust the water inlet valve opening, with a control error within ±0.5L.

[0144] In this embodiment, as described in step S40 above, the system uses a turbidity sensor to detect the concentration of suspended particles in the washing liquid during the washing process in real time, and converts it into the amount of residual stains. At the same time, the system collects current and voltage data of the washing device, and identifies energy consumption anomalies through power spectrum analysis.

[0145] For residual stains:

[0146] The current mechanical force parameters are compared with the initial fiber penetration depth and stain binding strength collected in step S10. If the stain residue is found to be above a preset threshold and the current speed has not reached the upper limit, the speed is increased in steps of 50 rpm. For example, if the initial fiber penetration depth is 0.6 mm, the upper limit of the speed is set to 1200 rpm. If the turbidity conversion residual rate is 15% (threshold 10%) after 20 minutes of washing, and the current speed of 1000 rpm has not reached the upper limit, the system increases the speed to 1050 rpm in steps of 50 rpm and extends the wash time by 5 minutes. If the stain residue is found to be above the preset threshold but the current speed has reached the upper limit (e.g., 1200 rpm), a combined "detergent replenishment + wash time extension" strategy is initiated. For example: ① 15% of the initial detergent dosage is added (e.g., if 50 ml of protease was initially added, 7.5 ml is now added); ② The wash time is extended by 10 minutes, while the mechanical force intensity is reduced to protect the fiber.

[0147] For detecting the turbidity of the washing liquid and converting the amount of stain residue, the specific method is:

[0148] Constructing a turbidity-residue mapping model: Pre-establish a polynomial regression model between turbidity (NTU) and stain residual rate (%): Where T is the turbidity value and R is the residual rate. For example, when the turbidity sensor detects that the turbidity of the washing liquid is 30 NTU, the model is substituted to calculate , which means the residual dirt rate is determined to be 14%. If the preset threshold for residual dirt is 10%, the mechanical force parameter (speed increase) adjustment is triggered.

[0149] For abnormal energy consumption:

[0150] If the instantaneous energy consumption exceeds 110% of the speed safety threshold, the speed will be automatically reduced by 20%. The speed safety threshold setting is determined by the fiber material. For example, the upper limit for cotton and linen fabrics is 1200rpm, and the upper limit for silk fabrics is 800rpm.

[0151] For chemical composition distribution monitoring, real-time spectral analysis technology is used. When it is detected that the residual proportion of a certain type of stain component increases by more than 5%, the corresponding enzyme preparation is triggered to be added, and the supplementary amount is 20% of the initial amount. If the chemical composition distribution changes by less than 3% within three consecutive monitoring cycles, the amount of detergent added is reduced by 15% to achieve dynamic optimization of the washing process.

[0152] In one possible embodiment:

[0153] Consider an intelligent drum washing machine that integrates a multi-source data acquisition module, an intelligent control module, a detergent compartment system, and a sensor array. The multi-source data acquisition module includes a spectrum analyzer, an ultrasonic sensor, and a pressure sensor; the sensor array includes a turbidity sensor, a current sensor, a temperature sensor, and a water level sensor; and the detergent compartment system has three independent detergent compartments for protease, lipase, and amylase.

[0154] Step S10: collecting stain characteristic parameters.

[0155] After a user places a cotton or linen shirt stained with blood or oil into the washing machine and activates the intelligent wash mode, the washing machine's built-in spectrometer scans the stained area using ultraviolet-visible light, identifying the blood stain as containing 70% protein and the oil stain as containing 30% fat. An ultrasonic sensor emits high-frequency sound waves into the fibers and, based on the echo attenuation, calculates the stain's penetration depth to be 0.6mm. A pressure sensor detects the compression and deformation of the garment and, based on material properties, estimates the stain's bonding strength to be 80N / m². The acquisition module integrates these three types of data and transmits them to the washing machine's main control chip.

[0156] Step S20: Coordinated regulation model construction and training.

[0157] The main control chip uses a built-in machine learning algorithm to perform one-hot encoding on the chemical composition distribution of the stains, generating a 1×8-dimensional categorical feature vector. The penetration depth and binding strength are normalized and converted into numerical feature vectors. Based on a multi-branch neural network architecture, the model's detergent prediction branch outputs a dosing plan that mixes protease and lipase in a 7:3 ratio. The mechanical force parameter branch calculates a parameter combination of a wash speed of 1000 rpm, a wash time of 45 minutes, and a mechanical force intensity level of 3. During training, the model uses historical wash data as samples and optimizes parameters using a batch stochastic gradient descent algorithm. When the error between the predicted stain removal rate and the actual value is less than 3% and the energy consumption is lower than the baseline value, the optimal wash strategy is output.

[0158] Step S30: washing control parameters are executed.

[0159] According to the washing strategy output by the model, the intelligent control module starts the corresponding operation:

[0160] Detergent delivery: Control the protease compartment and lipase compartment to release detergent in a ratio of 7:3, and at the same time open the pre-mixing tank for preliminary dissolution;

[0161] Water temperature adjustment: The temperature resistance threshold of cotton and linen fibers is 70°C. Considering the protease activity range of 40-55°C, the water temperature is set to 45°C.

[0162] Speed ​​adjustment: Based on the penetration depth of 0.6mm, set the washing speed to 1000rpm;

[0163] Water volume matching: Through the stain severity assessment model, it is calculated that a shirt weighing 0.3kg corresponds to a water requirement of 4.5L, and the water level sensor accurately controls the water intake.

[0164] Step S40: Dynamic correction of the washing process.

[0165] After 15 minutes of washing, the turbidity sensor detected that the turbidity of the detergent remained above the threshold, indicating a significant amount of blood stains. Simultaneously, the current sensor detected an abnormally high energy consumption. The main control chip compared the current speed of 1000 rpm with the initial detection of fiber penetration depth and stain binding strength, determining that mechanical force needed to be increased. The speed was then increased to 1100 rpm, and the wash time was extended by 10 minutes. The spectrometer monitored the residual protein content in real time and found that it still reached 20%, triggering the release of 10 ml of protein into the protease chamber. When energy consumption returned to normal and turbidity returned to a safe range, the current parameters were maintained until the end of the wash cycle.

[0166] Through the above process, the smart washing machine realizes the full process of intelligent washing from stain feature collection, strategy generation to dynamic optimization.

[0167] The multimodal perception fusion mentioned in the present invention is specifically the collaborative processing of multi-source data such as spectrometers, ultrasonic sensors, and pressure sensors; the parameter optimization decision is specifically implemented based on a neural network model and a reinforcement learning mechanism.

[0168] In one embodiment, step S30 further includes: using graded washing to divide the washing into a pre-wash stage, a main wash stage, and a rinse stage, wherein:

[0169] During the pre-washing phase, targeted enzyme preparations are added based on the chemical composition distribution of the stains, and the mechanical force parameters are reduced for a preset operation time;

[0170] For the main wash phase, the mechanical force parameters and wash duration are dynamically adjusted based on the fiber penetration depth, stain binding strength, and stain residue monitoring results from the pre-wash phase;

[0171] During the rinsing phase, the number of rinses and the amount of water are dynamically adjusted based on the real-time monitoring of the residual detergent concentration. When the residual detergent concentration exceeds a preset threshold, an additional rinse cycle is triggered, and the amount of water used in each additional rinse is a preset proportion of the initial water volume.

[0172] In each stage of the graded washing, the data of residual stains and abnormal energy consumption are monitored in real time through step S40, and the washing parameters of each stage are corrected synchronously.

[0173] In this embodiment, specifically:

[0174] The mechanical strength is divided into 5 levels (1-5), and differentiated control is achieved by combining the three parameters of washing speed, drum swing amplitude, and water flow impact force. The specific corresponding relationship is:

[0175] At level 1, the washing speed is 300-400rpm, the drum swing angle is 15°-20°, and the water flow impact force is 5-10kPa.

[0176] At level 2, the washing speed is 500-600rpm, the drum swing angle is 25°-30°, and the water flow impact force is 12-18kPa.

[0177] At level 3, the washing speed is 700-800rpm, the drum swing angle is 35°-40°, and the water flow impact force is 20-28kPa.

[0178] At level 4, the washing speed is 900-1000rpm, the drum swing angle is 45°-50°, and the water flow impact force is 30-40kPa.

[0179] At level 5, the washing speed is 1100-1200rpm, the drum swing angle is 55°-60°, and the water flow impact force is 45-60kPa.

[0180] As described above, in the pre-wash stage, the system automatically identifies the type of stain and administers targeted enzyme preparations based on the chemical composition distribution data of the stains collected in step S10. For example, when the spectrometer detects that the protein component in the blood stain exceeds 60%, the intelligent control module activates the protease detergent compartment and releases the protease preparation at 30% of the initial dosage; if the fat component in the oil stain exceeds 50%, the lipase preparation is administered. At the same time, the mechanical force parameters are set to level 1 intensity: the washing speed is adjusted to 400 rpm, the drum swing angle is set to 18°, the water flow impact force is maintained at 8 kPa, and the washing time is preset to 10 minutes. This stage preliminarily decomposes the stains through gentle washing, reduces the processing load of the subsequent main wash stage, and reduces fabric wear.

[0181] As described above, during the main wash phase, the system dynamically adjusts the washing parameters based on the fiber penetration depth, stain binding strength, and stain residue monitoring results from the pre-wash phase. For example, the turbidity sensor monitors the amount of stain residue after pre-wash in real time. If the residue rate exceeds a preset threshold (such as 40%), and the ultrasonic sensor detects a fiber penetration depth of more than 0.5mm, and the pressure sensor estimates a stain binding strength greater than 70N / m², the system determines that mechanical force needs to be increased. At this point, the mechanical force intensity is increased from level 1 to level 3: the washing speed is increased to 800rpm, the drum swing angle is increased to 38°, the water flow impact force is increased to 25kPa, and the washing time is extended by 15 minutes. If the residue rate is lower than the threshold, the current parameters are maintained or the mechanical force is appropriately reduced to avoid over-washing.

[0182] As described above, during the rinsing phase, the residual detergent concentration is monitored in real time through a conductivity sensor. For example, when the residual detergent concentration is detected to exceed a preset threshold (such as 50 mg / L), the system automatically triggers an additional rinse cycle. The water intake for each additional rinse is set to 30% of the initial water volume. For example, when the initial water volume is 8L, the additional rinse water intake is 2.4L. During the rinsing process, the mechanical force intensity is maintained at level 1 to reduce fabric entanglement. If the residual concentration continues to be high, it is temporarily increased to level 2 to enhance the water flow flushing effect. In addition, during the rinsing process, the system simultaneously monitors turbidity changes. If the turbidity value is stably below 10NTU within three monitoring cycles and the residual detergent concentration meets the standard, the rinse cycle is terminated early.

[0183] As described above, during each stage of the graded wash cycle, the system continuously performs real-time monitoring and parameter correction in step S40. For example, during the pre-wash phase, if the current sensor detects an abnormally high energy consumption (exceeding 120% of the baseline value), the mechanical strength is temporarily reduced from Level 1 to Level 0.5 (rotation speed at 300 rpm). During the main wash phase, if the spectrometer detects a persistently high residual percentage of a particular type of stain, in addition to supplementing enzymes, the mechanical strength can be increased from Level 3 to Level 4 or 5. During the rinse phase, if the residual detergent concentration decreases slowly and energy consumption approaches a threshold, the system automatically switches modes, temporarily increasing the mechanical strength to Level 2 and increasing the water intake. Through dynamic, coordinated control of each stage, an optimal balance between washing performance and resource consumption is achieved.

[0184] In one embodiment, the stain characteristic parameter further includes mixed stains. When the multi-source collection module collects mixed stains, the main wash stage further includes:

[0185] Obtaining layered distribution parameters of mixed stains inside the fiber, wherein the layered distribution parameters include surface stain thickness, deep stain penetration depth, and layered interface bonding strength;

[0186] Generating a step-by-step mechanical force adjustment strategy based on the hierarchical distribution parameters includes:

[0187] When the thickness of the surface stain exceeds a first preset threshold, treating the stain with a first mechanical strength and a corresponding enzyme preparation for a first preset time;

[0188] When the penetration depth of deep stains exceeds a second preset threshold, the second mechanical force intensity is increased and the corresponding enzyme preparation is switched to treat for a second preset time;

[0189] The first mechanical force intensity is lower than the second mechanical force intensity, and the type of enzyme preparation in the two stages is dynamically selected based on the chemical composition distribution of the stain;

[0190] The ratio between the first preset time and the second preset time is dynamically adjusted based on the layered interface bonding strength. When the layered interface bonding strength exceeds a third preset threshold, the processing time of the first preset time is extended.

[0191] In this embodiment, specifically:

[0192] The system uses ultrasonic tomography technology to obtain the mixed stain layer distribution parameters:

[0193] Surface stain thickness detection: emit high-frequency ultrasonic waves (frequency 10MHz) and calculate the surface stain thickness based on the time difference of the first echo signal;

[0194] Deep stain penetration depth detection: Use low-frequency penetrating waves (frequency 1MHz) to analyze the echo attenuation curve inside the fiber to locate the deepest penetration point of the stain;

[0195] Delamination interface bonding strength detection: Gradual pressure (0-50N) is applied to the fabric through a pressure sensor, and the delamination interface bonding strength is estimated based on the change in ultrasonic reflection coefficient.

[0196] After the multi-source collection module confirms the presence of mixed stains, the main wash stage performs the following operations:

[0197] Step 1: Hierarchical parameter acquisition and analysis.

[0198] The ultrasonic tomography module scans fabric stained with mixed stains (e.g., surface oil stains and deep blood stains) to obtain layer distribution parameters. For example, it detects a surface oil stain thickness of 0.3mm (exceeding the first preset threshold of 0.2mm), a deep blood stain penetration depth of 0.8mm (exceeding the second preset threshold of 0.6mm), and a layer interface bonding strength of 90N / m² (exceeding the third preset threshold of 80N / m²).

[0199] Step 2: Generation of step-by-step mechanical force strategy.

[0200] ① Surface stain treatment: Because the thickness of the surface oil stains exceeds the standard, the system starts the first mechanical force intensity (level 2, speed 600rpm, water flow impact force 15kPa), and at the same time activates the lipase detergent cabin, adding 40% of the initial amount, and the treatment time is preset to 12 minutes.

[0201] ② Deep stain treatment: When the penetration depth of deep blood stains exceeds the standard, the mechanical strength is increased to the second level (level 4, rotation speed 1000rpm, water flow impact force 35kPa), and the protease preparation is switched (dosage 30%). The treatment time is preset to 18 minutes.

[0202] ③ Dynamic adjustment of duration ratio: Due to the interface bonding strength reaching 90N / m², the system extends the duration of the first stage to 16 minutes (originally 12 minutes) to ensure that the surface stains are fully loosened before treating the deeper layers to avoid the spread of deep stains.

[0203] Step 3: Real-time monitoring and strategy correction.

[0204] The turbidity sensor detects changes in the detergent composition every 3 minutes. If the decomposition efficiency of surface oil stains is lower than expected (such as the turbidity increase within 3 minutes is less than 5NTU), the mechanical strength of the first stage will be temporarily increased to level 3; if the energy consumption exceeds the threshold when treating deep blood stains (>130% of the baseline value), the speed of the second stage will be reduced to 900rpm and the duration will be extended by 5 minutes to balance the cleaning effect and energy consumption.

[0205] The above-mentioned step-by-step strategy is deeply integrated with the aforementioned dynamic adjustment mechanism of main wash phase parameters:

[0206] If the mixed stains remaining after the pre-wash stage trigger the layered treatment of this embodiment, the stepped mechanical force strategy is preferentially executed;

[0207] During the strategy execution process, the aforementioned core parameters such as fiber penetration depth and pre-wash residue are simultaneously referenced. For example, when the fiber penetration depth is greater than 1mm, the drum swing angle of the second mechanical force intensity is automatically increased by 5° to enhance the flushing force.

[0208] The rinsing stage is still executed according to the aforementioned detergent residue monitoring logic to ensure that mixed stains and residual enzyme preparations are completely removed.

[0209] In one embodiment, the determination condition of the mixed stain is:

[0210] The chemical composition distribution of the stains simultaneously includes characteristic components of at least two different types of stains, and the multi-source acquisition module detects that the stain distribution inside the fiber presents a stratified characteristic, and the stratified interface bonding strength is greater than a preset value.

[0211] In this embodiment, specifically:

[0212] The system presets the core parameter thresholds for mixed stain determination, which can be dynamically adjusted according to the fabric type:

[0213] Characteristic component determination threshold: The proportion of a single stain characteristic component must be greater than 15% (e.g., protein > 15% and oil > 15%);

[0214] Delamination characteristics judgment criteria: ultrasonic tomography detected at least two echo intensity mutation interfaces, and the interface spacing was greater than 0.1 mm;

[0215] Bonding strength threshold: The default value of the layered interface bonding strength is 70N / m² (reduced to 50N / m² for silk fabrics and increased to 80N / m² for cotton and linen fabrics).

[0216] When executing step S10 to collect stain characteristic parameters, the system determines mixed stains according to the following process:

[0217] (1) Cross-validation of chemical composition.

[0218] The spectrometer scans the stain area across the entire wavelength range (200-1100 nm) to identify characteristic absorption peaks of components such as protein, grease, and pigments. If the relative intensities of both the protein peak (280 nm) and the grease peak (3400 cm⁻¹) exceed the judgment threshold (e.g., peak height ratio > 0.15), the stain is labeled as a "multi-component stain." For example, a fabric stain might contain 20% protein and 18% grease, meeting the chemical composition criteria.

[0219] (2) Hierarchical structure detection.

[0220] The ultrasonic sensor scans the fiber with a resolution of 0.01mm and analyzes the time-intensity curve of the echo signal. If the curve shows at least two intensity jumps (e.g., a sudden drop of more than 30% in echo intensity) with a distance greater than 0.1mm between the two points, a layered structure is determined. For example, if the surface layer 0.2mm is detected as a fat layer and the depth 0.7mm is detected as a protein layer, this meets the requirements for layered properties.

[0221] (3) Bond strength assessment.

[0222] The pressure sensor applies a gradient pressure (0-60N) to the stained area and simultaneously records the change in ultrasonic reflection coefficient.

[0223]

[0224] Where k is the calibration coefficient, is the pressure change, is the reflection coefficient change.

[0225] Calculate the interfacial bonding strength. If the calculated value is > 70 N / m² (e.g., the measured value is 85 N / m²), the bonding strength requirement is met.

[0226] (4) Comprehensive judgment and decision-making.

[0227] When all three of these conditions are met, the system marks the stain as a "mixed stain" and synchronizes the result to the stepped mechanical force adjustment module in the main wash phase. For example, a cotton and linen stain tested to contain 22% protein and 19% grease, a two-layer structure (0.5mm apart), and an interfacial strength of 82N / m² triggers the mixed stain treatment process.

[0228] Once it is determined to be mixed stains:

[0229] The system automatically enables the layered parameter collection function (surface thickness, deep penetration depth detection);

[0230] In the main wash stage, a stepped mechanical force strategy is prioritized, and the enzyme preparation type and treatment time are dynamically adjusted according to the stratification characteristics;

[0231] If changes in component distribution are detected through turbidity sensors and spectrometers during the washing process, it is reassessed whether the mixed stain judgment is maintained and the subsequent washing parameters are dynamically optimized.

[0232] In one embodiment, in step S40, dynamically modifying the washing strategy specifically includes:

[0233] Real-time collection of residual stain data and energy consumption data during the washing process to determine whether there is any abnormal energy consumption;

[0234] Comparing and analyzing the fiber penetration depth and stain binding strength initially collected in step S10 with the washing speed and mechanical strength of the current washing stage;

[0235] Based on the comparison results, the washing speed and washing time are dynamically adjusted, including:

[0236] When the amount of residual dirt is higher than the preset threshold and the mechanical force parameter has not reached the upper limit, the washing speed is increased or the washing time is extended;

[0237] When the energy consumption data exceeds the preset threshold, the washing speed is reduced or the washing time is shortened;

[0238] Adjust the amount of detergent added based on the real-time monitoring of the chemical composition of the stains, including:

[0239] If an increase in the residual proportion of a certain type of stain component is detected, the corresponding type of enzyme preparation or detergent will be added in a targeted manner;

[0240] If the chemical composition of the stains tends to be stable, reduce the amount of detergent.

[0241] In this embodiment, specifically:

[0242] For real-time data collection and abnormality judgment:

[0243] The turbidity sensor collects detergent turbidity every two minutes and converts it into the residual stain rate using a turbidity-residue mapping model. A current sensor monitors motor power in real time and calculates cumulative energy consumption based on the wash duration. If the energy consumption value at a given moment exceeds 120% of the baseline value, it is considered an energy consumption anomaly.

[0244] For parameter comparison analysis:

[0245] Compare the fiber penetration depth (e.g., 0.6 mm) and stain bond strength (e.g., 80 N / m²) collected in step S10 with the current mechanical force parameters (e.g., rotational speed 1000 rpm, mechanical strength level 3). For example, if the fiber penetration depth is greater than 0.5 mm and the bond strength is greater than 70 N / m², the corresponding mechanical strength should be ≥ level 3.

[0246] For dynamic adjustment of mechanical force parameters:

[0247] Increase strategy: If the stain residue rate is 15% (exceeds the 10% threshold) and the speed has not reached the upper limit (the upper limit for cotton and linen fabrics is 1200rpm), the speed will be increased in steps of 50rpm and the washing time will be extended by 5 minutes.

[0248] Reduction strategy: If energy consumption reaches 1.5kWh (exceeding the 1.44kWh threshold), the speed will be reduced by 100rpm and the remaining washing time will be shortened by 10%.

[0249] For detergent dosage adjustment:

[0250] The spectrometer monitors the distribution of chemical components of stains in real time. If the residual protein ratio drops from the initial 70% to 60% and then rises back to 65% (fluctuation > 5%), 5 ml of protease preparation will be added; if the fluctuation of the proportion of each component is less than 3% in three consecutive monitoring cycles, the total amount of detergent added will be reduced by 10%.

[0251] The system presets the following dynamic adjustment trigger thresholds (which can be adaptively adjusted according to fabric type):

[0252] Stain residue threshold: 10% (cotton and linen) / 5% (silk);

[0253] Energy consumption safety threshold: baseline value × 120% (e.g., if normal washing energy consumption is 1.2 kWh, the threshold is 1.44 kWh);

[0254] Residual component change threshold: 5% (adjustment is triggered when the proportion of a certain type of stain component fluctuates by >5%).

[0255] As needed, the process of dynamically modifying the washing strategy may also include:

[0256] Compare the real-time collected stain residue data and energy consumption data with the similar stain scenario data in the historical washing data to calculate the similarity score between the current washing process and the historical scenario;

[0257] Based on the similarity score, the following strategy is executed:

[0258] ① When the similarity score exceeds a preset threshold and the stain residue is higher than expected, the collaborative control model is called to recalculate the mechanical force parameters and detergent ratio;

[0259] ② When the similarity score is lower than the preset threshold, the reinforcement learning mechanism is activated, and the parameter adjustment record of the current washing process is used as a sample to update the collaborative control model;

[0260] If the energy consumption data fluctuates dramatically in a short period of time and exceeds the safety threshold, the washing process will be paused immediately and an alarm log will be generated containing the abnormality type, timestamp and current parameter status.

[0261] In this embodiment, specifically:

[0262] For similarity score calculation and policy execution:

[0263] The system obtains several sets of the most similar historical scene data (such as protein + grease mixed stains) from the cloud in real time and calculates the similarity score between the current scene and each set of historical data.

[0264] High similarity scenario (score > 0.85): If the stain residue rate is 20% higher than the historical average level, the collaborative control model is called and the current feature vector is input to recalculate the optimal mechanical force parameters and detergent ratio.

[0265] Low similarity scenario (score < 0.85): The parameter adjustment record of the current washing process (such as "speed from 1000→1100rpm, residual rate from 15%→12%") is used as a new sample, and the weight parameters of the collaborative control model are updated through the online learning algorithm.

[0266] Among them, for the similarity score calculation method, cosine similarity is used to calculate the matching degree between the current scene and historical data:

[0267]

[0268] Vector A includes five dimensions: current stain component ratio, penetration depth, bonding strength, residual rate, and energy consumption. Vector B represents the corresponding parameters for similar stain scenarios in historical data. The preset similarity threshold is 0.85.

[0269] For abnormal energy consumption processing mechanism:

[0270] If the system detects that energy consumption fluctuates by more than ±30% of the baseline value within 5 minutes (for example, a sudden increase from 1.2 kWh to 1.6 kWh), the system immediately pauses the wash cycle and generates an alarm log containing the following information:

[0271] Exception Type: "Power Overload"

[0272] Timestamp: 2025-06-25T14:30:45;

[0273] Current parameters: speed 1100rpm, water temperature 45℃, detergent dosage 40ml;

[0274] Predicted impact: Potentially shorten motor life by 5%.

[0275] For continuous model optimization:

[0276] After each wash is completed, the system compares the actual washing effect (residue rate, energy consumption, water consumption) with the predicted value. If the error exceeds 5%, the washing data will be added to the training set.

[0277] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0278] In one embodiment, a smart washing parameter optimization decision system based on multimodal perception fusion is provided. The smart washing parameter optimization decision system based on multimodal perception fusion corresponds to the smart washing parameter optimization decision method based on multimodal perception fusion in the above embodiment. The smart washing parameter optimization decision system based on multimodal perception fusion includes:

[0279] A data acquisition module, configured to acquire, in real time, stain characteristic parameters of the fabric to be cleaned through a multi-source acquisition module, wherein the stain characteristic parameters include at least the chemical composition distribution of the stain, the fiber penetration depth, and the stain binding strength;

[0280] a strategy generation module for constructing a collaborative control model using a machine learning algorithm, determining the type and ratio of detergent based on the chemical composition distribution of the stain, calculating mechanical force parameters including wash speed, wash time, and mechanical force strength based on the fiber penetration depth and stain binding strength, training the collaborative control model with the goal of maximizing stain removal rate and energy and water consumption as constraints, and outputting an optimal washing strategy;

[0281] a washing execution module, configured to generate corresponding washing control parameters according to the washing strategy to control the washing device to perform corresponding operations;

[0282] The dynamic adjustment module is used to monitor the stain residue and energy consumption abnormality data during the washing process in real time, compare the initially collected fiber penetration depth and stain binding strength with the real-time mechanical force parameters, dynamically correct the washing speed and washing time, and simultaneously adjust the detergent replenishment amount according to the chemical composition distribution.

[0283] Regarding the specific definition of an intelligent washing parameter optimization decision system based on multimodal perception fusion, please refer to the definition of an intelligent washing parameter optimization decision method based on multimodal perception fusion mentioned above, which will not be repeated here. Each module in the above-mentioned intelligent washing parameter optimization decision system based on multimodal perception fusion can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0284] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 2 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used for data storage, data processing, data analysis, etc. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it realizes an intelligent washing parameter optimization decision method based on multimodal perception fusion.

[0285] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, an intelligent washing parameter optimization decision method based on multimodal perception fusion is implemented.

[0286] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, an intelligent washing parameter optimization decision method based on multimodal perception fusion is implemented.

[0287] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0288] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0289] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A smart washing parameter optimization decision method based on multimodal perception fusion, characterized in that: The steps include: S10, collecting stain characteristic parameters of the fabric to be cleaned in real time through a multi-source collection module, wherein the stain characteristic parameters include at least the chemical composition distribution of the stain, the fiber penetration depth, and the stain binding strength; S20. Utilizing a machine learning algorithm to construct a collaborative control model, determining the type and ratio of detergent based on the chemical composition distribution of the stain, calculating mechanical parameters including wash speed, wash time, and mechanical strength based on the fiber penetration depth and stain binding strength, and training the collaborative control model with maximizing stain removal rate as the goal and energy and water consumption as constraints to output an optimal washing strategy. S30: Generate corresponding washing control parameters according to the washing strategy to control the washing device to perform corresponding operations, wherein the operations include at least one of the following: Detergent loading is performed in compartments based on the chemical composition of the stains; The corresponding water temperature is adjusted based on the intersection of the fiber temperature resistance threshold and the enzyme preparation activity range; Adjust the rotation speed based on the fiber penetration depth; Matching the amount of water to the stain severity calculated based on the chemical composition distribution of the stain, fiber penetration depth, and stain binding strength; S40: Real-time monitoring of stain residue and abnormal energy consumption data during the washing process, comparison of the initially collected fiber penetration depth and stain binding strength with the real-time mechanical force parameters, dynamic correction of the washing speed and washing time, and simultaneous adjustment of the detergent replenishment amount according to the chemical composition distribution.

2. The intelligent washing parameter optimization decision method based on multimodal perception fusion according to claim 1 is characterized in that: In step S20, constructing the collaborative control model specifically includes: S211, extracting and converting features of the chemical composition distribution of the stain, fiber penetration depth, and stain binding strength, including: The chemical composition distribution of stains is converted into a category feature vector through semantic analysis; Normalize the fiber penetration depth and stain binding strength to generate a numerical feature vector; S212: Construct a multi-branch neural network model, the model comprising: The first processing branch receives the category feature vector and generates a predicted distribution of detergent types and ratios through a fully connected layer; The second processing branch receives the numerical feature vector and generates optimization suggestions for mechanical force parameters including washing speed, washing time and mechanical force intensity through a convolutional layer; S213. Establish the following mapping relationship in the multi-branch neural network model: Detergent selection function based on stain chemical composition distribution and fiber penetration depth; Mechanical force parameter optimization function based on fiber penetration depth and stain binding strength.

3. The intelligent washing parameter optimization decision method based on multimodal perception fusion according to claim 2 is characterized in that: In step S20, training the collaborative control model specifically includes: S221. Constructing a training dataset comprising input features and supervisory labels based on the chemical composition distribution of the stains, the fiber penetration depth, and the stain binding strength; wherein the input features are the categorical feature vectors and the numerical feature vectors, and the supervisory labels are the actual washing parameters corresponding to the input features and the actual constraint values ​​of energy and water consumption; S222. Construct a multi-objective optimization framework with minimizing the deviation between the predicted washing parameters and the measured values ​​as the main objective and with the energy and water consumption not exceeding the preset thresholds as the constraint condition; S223, using a batch stochastic gradient descent algorithm to iteratively optimize the parameters of the multi-branch neural network model, including: Dynamically adjust network weights to minimize the objective function; Real-time evaluation of the degree of satisfaction of energy and water consumption constraints; When the validation set loss function converges and the energy consumption constraint is continuously satisfied, the trained collaborative control model is output.

4. The intelligent washing parameter optimization decision method based on multimodal perception fusion according to claim 1 is characterized in that: The step S30 further includes: using graded washing to divide the washing into a pre-washing stage, a main wash stage and a rinsing stage, wherein: During the pre-washing phase, targeted enzyme preparations are added based on the chemical composition distribution of the stains, and the mechanical force parameters are reduced for a preset operation time; For the main wash phase, the mechanical force parameters and wash duration are dynamically adjusted based on the fiber penetration depth, stain binding strength, and stain residue monitoring results from the pre-wash phase; During the rinsing phase, the number of rinses and the amount of water are dynamically adjusted based on the real-time monitoring of the residual detergent concentration. When the residual detergent concentration exceeds a preset threshold, an additional rinse cycle is triggered, and the amount of water used in each additional rinse is a preset proportion of the initial water volume. In each stage of the graded washing, the data of residual stains and abnormal energy consumption are monitored in real time through step S40, and the washing parameters of each stage are corrected synchronously.

5. The intelligent washing parameter optimization decision method based on multimodal perception fusion according to claim 4 is characterized in that: The stain characteristic parameters also include mixed stains. When the multi-source collection module collects mixed stains, the main wash stage further includes: Obtaining layered distribution parameters of mixed stains inside the fiber, wherein the layered distribution parameters include surface stain thickness, deep stain penetration depth, and layered interface bonding strength; Generating a step-by-step mechanical force adjustment strategy based on the hierarchical distribution parameters includes: When the thickness of the surface stain exceeds a first preset threshold, treating the stain with a first mechanical strength and a corresponding enzyme preparation for a first preset time; When the penetration depth of deep stains exceeds a second preset threshold, the second mechanical force intensity is increased and the corresponding enzyme preparation is switched to treat for a second preset time; The first mechanical force intensity is lower than the second mechanical force intensity, and the type of enzyme preparation in the two stages is dynamically selected based on the chemical composition distribution of the stain; The ratio between the first preset time and the second preset time is dynamically adjusted based on the layered interface bonding strength. When the layered interface bonding strength exceeds a third preset threshold, the processing time of the first preset time is extended.

6. The intelligent washing parameter optimization decision method based on multimodal perception fusion according to claim 5 is characterized in that: The determination conditions for the mixed stains are: The chemical composition distribution of the stains simultaneously includes characteristic components of at least two different types of stains, and the multi-source acquisition module detects that the stain distribution inside the fiber presents a stratified characteristic, and the stratified interface bonding strength is greater than a preset value.

7. The intelligent washing parameter optimization decision method based on multimodal perception fusion according to claim 1 is characterized in that: In step S40, the dynamic modification of the washing strategy specifically includes: Real-time collection of residual stain data and energy consumption data during the washing process to determine whether there is any abnormal energy consumption; Comparing and analyzing the fiber penetration depth and stain binding strength initially collected in step S10 with the washing speed and mechanical strength of the current washing stage; Based on the comparison results, the washing speed and washing time are dynamically adjusted, including: When the amount of residual dirt is higher than the preset threshold and the mechanical force parameter has not reached the upper limit, the washing speed is increased or the washing time is extended; When the energy consumption data exceeds the preset threshold, the washing speed is reduced or the washing time is shortened; Adjust the amount of detergent added based on the real-time monitoring of the chemical composition of the stains, including: If an increase in the residual proportion of a certain type of stain component is detected, the corresponding type of enzyme preparation or detergent will be added in a targeted manner; If the chemical composition of the stains tends to be stable, reduce the amount of detergent.

8. A smart washing parameter optimization decision system based on multimodal perception fusion, used to implement the steps of a smart washing parameter optimization decision method based on multimodal perception fusion as described in any one of claims 1 to 7, characterized in that: include: A data acquisition module, configured to acquire, in real time, stain characteristic parameters of the fabric to be cleaned through a multi-source acquisition module, wherein the stain characteristic parameters include at least the chemical composition distribution of the stain, the fiber penetration depth, and the stain binding strength; a strategy generation module for constructing a collaborative control model using a machine learning algorithm, determining the type and ratio of detergent based on the chemical composition distribution of the stain, calculating mechanical force parameters including wash speed, wash time, and mechanical force strength based on the fiber penetration depth and stain binding strength, training the collaborative control model with the goal of maximizing stain removal rate and energy and water consumption as constraints, and outputting an optimal washing strategy; a washing execution module, configured to generate corresponding washing control parameters according to the washing strategy to control the washing device to perform corresponding operations; The dynamic adjustment module is used to monitor the stain residue and energy consumption abnormality data during the washing process in real time, compare the initially collected fiber penetration depth and stain binding strength with the real-time mechanical force parameters, dynamically correct the washing speed and washing time, and simultaneously adjust the detergent replenishment amount according to the chemical composition distribution.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, it implements the steps of an intelligent washing parameter optimization decision method based on multimodal perception fusion as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, the steps of the intelligent washing parameter optimization decision method based on multimodal perception fusion as described in any one of claims 1 to 7 are implemented.

Images