Intelligent washing control method and system based on multi-modal data

Abstract

The invention relates to the field of intelligent control, in particular to an intelligent washing control method and system based on multi-modal data, and the method comprises the steps: carrying out the time sequence alignment and feature fusion of a running state parameter set and a communication demand parameter set, and obtaining a multi-modal communication feature matrix; analyzing the coupling relationship between the dehydration stage identifier and the real-time energy consumption value of the washing equipment through a preset conflict prediction model, dynamically calculating the channel preemption priority of each washing equipment according to an instruction execution time window, and generating a clustering type communication resource scheduling sequence in combination with the priority of a control instruction; obtaining the number of instruction conflicts identified by each washing device in the dehydration stage, and constructing a load balancing optimization function with the instantaneous circuit load; and when the output value of the load balancing optimization function is minimized, obtaining a dehydration peak shifting control instruction. The problems of communication quality reduction and control delay during cooperative control of a plurality of washing devices are solved, and the timeliness and safety of control of the washing devices are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control, and in particular to a washing intelligent control method and system based on multimodal data. Background Art

[0002] In the modern laundry industry, with the increasing intelligence and automation of laundry equipment, commercial laundries and industrial laundry centers often require multiple machines to operate simultaneously to meet large-scale laundry needs. However, when multiple machines operate in concert, electromagnetic interference (EMI) issues become increasingly prominent, leading to degraded communication quality and potentially causing delays, errors, and even out-of-order transmission of control commands. Specifically, when multiple machines simultaneously send data to or receive control commands from a control terminal, wireless channels become congested, increasing the probability of signal collisions and causing delays in command transmission. If control commands fail to reach the machines promptly and accurately, the machines' operating status cannot be adjusted as expected. This not only impacts the normal execution of the washing process and reduces washing efficiency, but can also cause equipment failure due to erroneous control commands. The dehydration process in particular consumes high power. If control commands cannot be transmitted promptly, the machines will simultaneously dehydrate, resulting in a sudden increase in circuit load and serious safety concerns.

[0003] Therefore, there is an urgent need for an intelligent washing control method based on multimodal data to solve the problems of communication quality degradation and control delay caused by electromagnetic interference when multiple washing equipment are collaboratively controlled, to achieve high efficiency, timeliness and accuracy in washing equipment control, and thus improve the operating efficiency and stability of the entire washing system. Summary of the Invention

[0004] (1) Technical problems to be solved

[0005] The purpose of the present invention is to provide an intelligent washing control method and system based on multimodal data to solve the problem of communication quality degradation and control delay caused by electromagnetic interference when multiple washing equipment are collaboratively controlled, which in turn leads to high circuit load in the dehydration link of the washing equipment.

[0006] (2) Technical solution

[0007] To achieve the above objectives, the present invention provides, on the one hand, a washing intelligent control method based on multimodal data, the method comprising:

[0008] S1. The operating status parameter set and communication requirement parameter set of each washing device are collected in real time through the wireless sensor network of the washing device communication nodes; the operating status parameter set includes water level parameters, real-time energy consumption values ​​and dehydration stage identifiers, and the communication requirement parameter set includes control instruction priority, maximum allowable transmission delay threshold and instruction execution time window; the operating status parameter set and communication requirement parameter set are time-aligned and feature-fused to obtain a multimodal communication feature matrix.

[0009] S2. Analyze the coupling relationship between the dehydration stage identifier and the real-time energy consumption value of the washing equipment using the multimodal communication feature matrix through a preset conflict prediction model, and dynamically calculate the channel preemption priority of each washing equipment according to the instruction execution time window based on the coupling relationship; generate a clustered communication resource scheduling sequence based on the channel preemption priority and the control instruction priority.

[0010] S3. After performing hybrid channel allocation according to the clustered communication resource scheduling sequence, obtain the number of instruction conflicts identified by each washing device during the dehydration phase, and construct a load balancing optimization function with the instantaneous circuit load.

[0011] S4. When the output value of the load balancing optimization function is minimized by dynamically adjusting the channel capacity of the orthogonal frequency division time slot and the competition parameters of the carrier sensing window, a dehydration peak-shifting control instruction is obtained and sent to the control end; the transmission delay of the dehydration peak-shifting control instruction is lower than the maximum allowable transmission delay threshold.

[0012] Furthermore, the method of performing time alignment and feature fusion on the operating status parameter set and the communication requirement parameter set to obtain a multimodal communication feature matrix includes:

[0013] Generate a global timestamp sequence T={t1…,t k …,t n}, t k is the reference time point of the kth synchronization cycle; for the water level parameter L in the operating state parameter set i (t) and real-time energy consumption value E i (t) Perform piecewise linear interpolation to generate a discrete sequence L aligned on the global timestamp sequence T i (t k ) and E i (t k ), where i is the washing equipment number.

[0014] The instruction execution time window [τ start ,τ end ] is mapped to a binary indicator function I corresponding to the global timestamp sequence T i (tk ), if and only if t k ∈[τ start ,τ end ] Time i (t k )=1, otherwise I i (t k )=0;

[0015] The sliding window algorithm is used to calculate the dehydration stage marker S i (t) The difference sequence ΔS on the global timestamp sequence T i (t k )=S i (t k )―S i (t k―1 ), when ΔS i (t k )≥θ s When t is the starting time of the dehydration phase, s , and construct the dehydration stage feature vector D i =[Δt s ,ΔE s ]; where θ s is the preset dehydration stage state mutation threshold; Δt s The starting time of the dehydration phase is t s The time deviation from the reference time point of the most recent synchronization cycle; ΔE s It is the sliding average difference of energy consumption before and after dehydration starts.

[0016] After the timing is aligned, L i (t k ), E i (t k ), I i (t k ) and D i According to the washing equipment number i and the reference time point t k They are combined into a multi-dimensional tensor and a multimodal communication feature matrix is ​​generated through the feature fusion function.

[0017] Furthermore, the method of analyzing the coupling relationship between the dehydration stage identifier and the real-time energy consumption value of the washing equipment using the multimodal communication feature matrix through a preset conflict prediction model, and dynamically calculating the channel preemption priority of each washing equipment according to the instruction execution time window based on the coupling relationship; and generating a clustered communication resource scheduling sequence based on the channel preemption priority and the control instruction priority includes:

[0018] The preset conflict prediction model adopts a double-layer attention network, and the dehydration stage conflict probability P of washing equipment i and washing equipment j is obtained through the double-layer attention network.ij According to the conflict probability P ij and the instruction execution time window [τ start ,τ end ] dynamically generates the channel preemption priority score ρ of washing equipment i i ; The channel preemption priority score ρ i The calculation formula is:

[0019]

[0020] Where ξ is the instruction execution time window overlap penalty factor; is the real-time energy consumption change rate of washing equipment i after standardization; Ω ij is the overlap of the instruction execution time windows of washing equipment i and washing equipment j, and its value range is [0,1].

[0021] According to the channel preemption priority score ρ i and control instruction priority π i Get the weighted sum of channel preemption priority η i =ρ i +νπ i , where ν is a preset weight coefficient; according to the weighted sum of the channel preemption priority η i Divide the washing equipment into high-preemption cluster C H ={i|η i ≥η th} with low preemption cluster C L and for high preemption cluster C H and low preemption cluster C L Allocate an orthogonal frequency division time slot resource block set, and generate a clustered communication resource scheduling sequence according to the orthogonal frequency division time slot resource block set; wherein, n th The classification threshold is the weighted sum of the preset channel preemption priorities.

[0022] Furthermore, the dehydration phase conflict probability P of washing equipment i and washing equipment j is obtained through the two-layer attention network. ij The methods include:

[0023] The first layer of the two-layer attention network uses the temporal attention mechanism to calculate the dehydration stage identifier S i (t k ) and real-time energy consumption value E i (t k )’s temporal association weight Among them, Q and K are learnable parameter matrices, and d is the feature dimension.

[0024] The second layer of the two-layer attention network uses the spatial attention mechanism to calculate the conflict probability of the dehydration stage of washing equipment i and washing equipment j Among them, σ is the Sigmoid activation function; b c is the conflict detection bias term; W c is the conflict detection weight matrix; Represents vector concatenation.

[0025] Furthermore, the high preemption cluster C H and low preemption cluster C L The method of allocating an orthogonal frequency division time slot resource block set and generating a clustered communication resource scheduling sequence according to the orthogonal frequency division time slot resource block set includes:

[0026] According to the high preemption cluster C H and low preemption cluster C L The ratio of the number of washing equipment to dynamically divide the orthogonal frequency division time slot resource block set R H With R L , generate clustered communication resource scheduling sequence S = {(i,r i ,t i )|r i ∈R H ∪R L}.

[0027] in, is the scheduling time slot offset of washing equipment i, δ is the time slot interval unit, rank(η i ) represents the weighted sum of channel preemption priority and η i Sequence number in descending order; resource block r i =(f i ,t i ) frequency component f i With the time slot component t i Satisfy the orthogonality constraint |f i ―f j |≥Δf min or |t i ―t j |≥δ min ; Among them, f i The frequency component of the resource block allocated to the washing device i; f j The frequency component of the resource block allocated to scrubbing device j; t i The resource block time slot component allocated to scrubber device i; t j The resource block time slot component allocated to scrubber j; Δf min is the minimum frequency interval, δ min To prevent collision time slot interval.

[0028] Furthermore, the high preemption cluster CH and low preemption cluster C L The ratio of the number of washing equipment to dynamically divide the orthogonal frequency division time slot resource block set R H With R L The methods include:

[0029] When allocating resource blocks to high-preemption cluster scrubbing devices, establish a frequency-time two-dimensional mapping function Among them, N f is the number of preset frequency band divisions, f0 is the starting frequency, mod is the modulo operation, To round down; The start time of the instruction execution time window of washing equipment i.

[0030] A competitive resource allocation strategy is adopted for low-preemptive cluster washing equipment, and its available resource pool is defined as R L =R total \R H , and in each scheduling cycle according to the channel preemption priority η i Select the ones that meet the requirements in descending order Resource block r i ; Where λ is the time-frequency weight coefficient, ||·|| is the normalized frequency interval distance; R total For all orthogonal frequency division time slot resource blocks.

[0031] When the dehydration stage conflict probability P is detected ij When the collision probability is greater than a preset threshold, resource blocks are forcibly isolated for the washing device i and the washing device j.

[0032] Furthermore, when the dehydration stage conflict probability P is detected ij When the probability of conflict is greater than a preset threshold, the method for forcibly isolating the resource blocks of the scrubbing device i and the scrubbing device j includes:

[0033] If the resource block r i ∈R H And resource block r j ∈R H , then the frequency interval |f′ is reassigned to washing equipment j j ―f i |≥2Δf min And the time slot interval |t′ j ―t i |≥2δ min Resource block r′ j ; If resource block r j ∈R L , then upgrade it to R H And trigger inter-cluster resource redistribution.

[0034] The inter-cluster resources are redistributed into L The resource block that conflicts with washing equipment j is moved to R H , and recalculate R H With R L The division ratio.

[0035] Furthermore, the method of obtaining the number of instruction conflicts identified by each washing device during the dehydration stage and constructing a load balancing optimization function with the instantaneous circuit load includes:

[0036] The number of instruction conflicts is the number of times the cumulative conflict probability between washing devices exceeds a preset conflict probability threshold; the number of instruction conflicts for each washing device is normalized according to a preset circuit load safety threshold, and a conflict penalty term is applied, and a load overlimit penalty term is introduced when the instantaneous circuit load exceeds the preset circuit load safety threshold; the hard constraint of time window overlap in the dehydration stage is converted into an exponential decay function of the time difference between the dehydration start times of the washing devices to obtain a time window overlap risk term; the conflict penalty term, the load overlimit penalty term and the time window overlap risk term are weighted and fused, and the regularization constraints of the channel capacity adjustment amount and the competition parameter offset are superimposed to obtain a load balancing optimization function.

[0037] Furthermore, the method of obtaining the dehydration peak shifting control instruction by dynamically adjusting the channel capacity of the orthogonal frequency division time slot and the contention parameter of the carrier sensing window to minimize the output value of the load balancing optimization function includes:

[0038] The channel capacity of the orthogonal frequency division time slot and the contention parameters of the carrier sensing window are iteratively adjusted through the gradient descent algorithm, so that the output value of the load balancing optimization function gradually converges to the minimum value. At the same time, the transmission delay constraint of the dehydration peak-shifting control instruction is verified in real time. If there is a washing equipment whose transmission delay exceeds the maximum allowable transmission delay threshold, the optimization is re-performed until a dehydration peak-shifting control instruction that satisfies both circuit load safety and transmission delay constraints is generated.

[0039] On the other hand, based on the same inventive concept, the present invention also provides a washing intelligent control system based on multimodal data, the system comprising: a multimodal communication feature matrix acquisition module, a clustered communication resource scheduling sequence generation module, a load balancing analysis module, and a dehydration peak shifting management module, wherein the modules are sequentially connected to each other in communication;

[0040] A multimodal communication feature matrix acquisition module is used to collect, in real time, a set of operating status parameters and a set of communication demand parameters of each washing device through a wireless sensor network of the washing device communication nodes. The operating status parameter set includes a water level parameter, a real-time energy consumption value, and a dehydration stage identifier, and the communication demand parameter set includes a control instruction priority, a maximum allowable transmission delay threshold, and an instruction execution time window. The operating status parameter set and the communication demand parameter set are time-series aligned and feature-fused to obtain a multimodal communication feature matrix.

[0041] A clustered communication resource scheduling sequence generation module is used to analyze the coupling relationship between the dehydration stage identification and the real-time energy consumption value of the washing equipment using the multimodal communication feature matrix through a preset conflict prediction model, dynamically calculate the channel preemption priority of each washing equipment according to the instruction execution time window based on the coupling relationship; and generate a clustered communication resource scheduling sequence based on the channel preemption priority and the control instruction priority.

[0042] A load balancing analysis module is configured to obtain the number of instruction conflicts identified by each washing device during the dehydration phase after performing hybrid channel allocation according to the clustered communication resource scheduling sequence, and to construct a load balancing optimization function based on the instantaneous circuit load;

[0043] The dehydration peak-shifting management module is used to obtain a dehydration peak-shifting control instruction and send it to the control end by dynamically adjusting the channel capacity of the orthogonal frequency division time slot and the competition parameters of the carrier sensing window to minimize the output value of the load balancing optimization function; the transmission delay of the dehydration peak-shifting control instruction is lower than the maximum allowable transmission delay threshold.

[0044] (3) Beneficial effects

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. Through multimodal data fusion and conflict prediction models, a comprehensive analysis of the operating status and communication needs of washing equipment is conducted. The channel preemption priority of each device is dynamically calculated, and a clustered communication resource scheduling sequence is generated. A hybrid channel allocation strategy, combining the priorities of washing equipment and actual communication needs, rationally allocates channel resources, effectively reducing the number of command conflicts between washing equipment and improving communication efficiency. Especially during the dehydration phase, the risk of conflicts is further reduced by forcibly isolating resource blocks with high conflict probabilities, ensuring a smooth washing process.

[0047] 2. By constructing a load balancing optimization function that comprehensively considers the washing machine's channel utilization, command conflict counts, and instantaneous circuit load, this system dynamically adjusts the washing machine's communication resources. Furthermore, by considering circuit load safety thresholds and introducing conflict penalties, load-exceeding penalties, and time window overlap risk factors, this ensures that overload during the dehydration phase of the washing machine will not cause circuit safety issues. Furthermore, by dynamically adjusting the channel capacity of the orthogonal frequency division time slots and the contention parameters of the carrier sense window, the system further optimizes load balancing and improves the overall stability of the washing system.

[0048] 3. By generating and sending dehydration peak-shifting control instructions, staggered operation of the washing equipment during the dehydration phase is achieved. By optimizing the dehydration phase start time and channel resource allocation, while ensuring that communication latency is below the maximum allowable transmission latency threshold, the generation process of dehydration peak-shifting control instructions is further optimized, ensuring the accuracy and timeliness of the control instructions. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flowchart of the intelligent washing control method based on multimodal data according to Example 1 of the present invention.

[0050] Figure 2 This is a schematic diagram of the module composition of the intelligent washing control system based on multimodal data according to Example 2 of the present invention. DETAILED DESCRIPTION

[0051] 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] Before giving examples, it is necessary to explain the application scenarios of the present invention. The present invention is an intelligent washing control method and system based on multimodal data, which is applied to the coordinated control of multiple washing equipment when the communication quality deteriorates and the control delay is caused by electromagnetic interference, etc., and because the power of the washing equipment in the dehydration link is very high, if the control instructions cannot be transmitted in a timely manner, the washing equipment will be dehydrated at the same time, and the circuit load will suddenly increase, which will seriously affect the safety of the circuit.

[0053] Example 1: Figure 1 As shown, this embodiment provides a washing intelligent control method based on multimodal data, the method comprising:

[0054] S1. The operating status parameter set and communication requirement parameter set of each washing device are collected in real time through the wireless sensor network of the washing device communication nodes; the operating status parameter set includes water level parameters, real-time energy consumption values ​​and dehydration stage identifiers, and the communication requirement parameter set includes control instruction priority, maximum allowable transmission delay threshold and instruction execution time window; the operating status parameter set and communication requirement parameter set are time-aligned and feature-fused to obtain a multimodal communication feature matrix.

[0055] S2. Analyze the coupling relationship between the dehydration stage identifier and the real-time energy consumption value of the washing equipment using the multimodal communication feature matrix through a preset conflict prediction model, and dynamically calculate the channel preemption priority of each washing equipment according to the instruction execution time window based on the coupling relationship; generate a clustered communication resource scheduling sequence based on the channel preemption priority and the control instruction priority.

[0056] S3. After performing hybrid channel allocation according to the clustered communication resource scheduling sequence, the number of command conflicts identified by each washing machine during the dehydration phase is obtained and combined with the instantaneous circuit load to construct a load balancing optimization function. Clustered scheduling can group washing machines according to communication requirements (such as control command priority and dehydration phase identification), allocating independent resource blocks to each group. The hybrid channel allocation strategy uses orthogonal channels for high-priority commands and contention channels for low-priority commands.

[0057] S4. When the output value of the load balancing optimization function is minimized by dynamically adjusting the channel capacity of the orthogonal frequency division time slot and the competition parameters of the carrier sensing window, a dehydration peak-shifting control instruction is obtained and sent to the control end; the transmission delay of the dehydration peak-shifting control instruction is lower than the maximum allowable transmission delay threshold.

[0058] The method of performing time alignment and feature fusion on the operating status parameter set and the communication requirement parameter set to obtain a multimodal communication feature matrix includes:

[0059] Generate a global timestamp sequence T={t1…,t k …,t n}, t k is the reference time point of the kth synchronization cycle; for the water level parameter L in the operating state parameter set i (t) and real-time energy consumption value E i (t) Perform piecewise linear interpolation to generate a discrete sequence L aligned on the global timestamp sequence T i (t k ) and E i (t k ), where i is the washing equipment number.

[0060] The instruction execution time window [τ start ,τ end ] is mapped to a binary indicator function I corresponding to the global timestamp sequence T i (t k ), if and only if t k ∈[τ start ,τ end ] Time i (t k )=1, otherwise I i (t k )=0.

[0061] The sliding window algorithm is used to calculate the dehydration stage marker S i (t) The difference sequence ΔS on the global timestamp sequence T i (t k )=S i (t k )―S i (t k―1 ), when ΔS i (t k )≥θ s When t is the starting time of the dehydration phase, s , and construct the dehydration stage feature vector D i =[Δt s ,ΔE s ]; where θ s is the preset dehydration stage state mutation threshold; Δt s The starting time of the dehydration phase is t s The time deviation from the reference time point of the most recent synchronization cycle; ΔE s It is the sliding average difference of energy consumption before and after dehydration starts.

[0062] After the timing is aligned, L i (t k ), E i (t k ), I i (t k ) and D i According to the washing equipment number i and the reference time point t k They are combined into a multi-dimensional tensor and a multimodal communication feature matrix is ​​generated through the feature fusion function.

[0063] For example, in a commercial laundry environment, data collection times for individual washing machines often differ due to varying startup times and operating cycles. By using the built-in clock synchronization signals of each washing machine to generate a unified global timestamp sequence, the data collection times of all washing machines are aligned to the same benchmark time point. For example, a synchronization period of 5 seconds is set, and all washing machines sample and report data at these standard time points. For two continuously changing physical quantities, water level parameters and real-time energy consumption, piecewise linear interpolation is used to address the time alignment issue. When data for a washing machine is updated between standard sampling time points, linear interpolation is used to estimate the corresponding value for that machine at the standard time point. For example, if the water level of washing machine A is 85% at the 10th second and 60% at the 20th second, it can be estimated that its water level at the standard time point of 15 seconds is approximately 72.5%. This approach ensures that data from all washing machines can be compared and analyzed on a unified time scale.

[0064] The instruction execution time window is processed using a binary indicator function mapping method. Each control instruction for a washing machine has a clear execution time range. For example, if a dehydration instruction needs to start within three minutes after the wash cycle is completed, this time window is converted into an indicator sequence consisting of 0s and 1s. Time points within the time window are marked as 1, and time points outside the window are marked as 0. This mapping method facilitates subsequent mathematical calculations and logical judgments.

[0065] Processing the dehydration phase identifier is crucial, as it represents peak power consumption for washing machines and is also the most susceptible to electromagnetic interference. A sliding window algorithm is used to monitor changes in dehydration status, detecting sudden state changes by calculating the difference between dehydration identifiers at adjacent time points. When a washing machine switches from a non-dehydration state to a dehydration state, this critical moment is recorded. The energy consumption change before and after the dehydration start is calculated to form a dehydration phase feature vector, providing important input for subsequent conflict prediction.

[0066] The time-aligned water level data, energy consumption data, time window indicator data, and dehydration feature vectors are organized into a multidimensional tensor based on the two-dimensional structure of equipment number and time point. This multidimensional tensor is converted into a standardized multimodal communication feature matrix using a feature fusion function. This multimodal communication feature matrix contains comprehensive status information for all washing equipment at all critical time points, providing a comprehensive data foundation for intelligent control decisions.

[0067] The method of analyzing the coupling relationship between the dehydration stage identifier and the real-time energy consumption value of the washing equipment using the multimodal communication feature matrix through a preset conflict prediction model, and dynamically calculating the channel preemption priority of each washing equipment according to the instruction execution time window based on the coupling relationship; and generating a clustered communication resource scheduling sequence based on the channel preemption priority and the control instruction priority includes:

[0068] The preset conflict prediction model adopts a double-layer attention network, and the dehydration stage conflict probability P of washing equipment i and washing equipment j is obtained through the double-layer attention network. ij According to the conflict probability P ij and the instruction execution time window [τ start ,τ end ] dynamically generates the channel preemption priority score ρ of washing equipment i i ; The channel preemption priority score ρ i The calculation formula is:

[0069]

[0070] Where ξ is the instruction execution time window overlap penalty factor; is the real-time energy consumption change rate of washing equipment i after standardization; Ω ij is the overlap of the instruction execution time windows of washing equipment i and washing equipment j, and its value range is [0,1].

[0071] According to the channel preemption priority score ρ i and control instruction priority π i Get the weighted sum of channel preemption priority η i =ρ i +νπ i , where ν is a preset weight coefficient; according to the weighted sum of the channel preemption priority η i Divide the washing equipment into high-preemption cluster C H ={i|η i ≥η th} with low preemption cluster C L and for high preemption cluster C H and low preemption cluster C L Allocate an orthogonal frequency division time slot resource block set, and generate a clustered communication resource scheduling sequence according to the orthogonal frequency division time slot resource block set; wherein, n th The classification threshold is the weighted sum of the preset channel preemption priorities.

[0072] The pre-defined conflict prediction model uses a two-layer attention network architecture to analyze potential conflicts between devices. Its core task is to identify which washing machines are likely to enter the dehydration phase at the same time, thereby causing communication interference and circuit load conflicts. This two-layer attention mechanism simultaneously captures the temporal dimension and inter-device correlation patterns, improving the accuracy of conflict prediction. After calculating the probability of dehydration phase conflict between any two washing machines using the two-layer attention network, it dynamically generates a channel preemption priority score for each device based on the overlap of their instruction execution time windows. The channel preemption priority scoring formula comprehensively considers factors such as the device's energy consumption change rate, the probability of conflict with other devices, and the degree of time window overlap. Washing machines with high energy consumption change rates are typically about to enter or are currently dehydrating and therefore require higher communication priority. Washing machines with low conflict probability and minimal time window overlap with other machines receive relatively high priority scores.

[0073] The dynamic calculation of channel preemption priority takes into account the operational characteristics of real-world laundry facilities. During peak hours, multiple washing machines may simultaneously complete their wash cycle and enter the spin stage. Communication resources must be allocated appropriately based on the importance, urgency, and conflict risk of each washing machine. The exponential decay term in the channel preemption priority scoring formula ensures that combinations of washing machines with higher time window overlap receive a greater priority penalty, thereby preventing multiple washing machines from competing for limited communication resources. The calculated channel preemption priority score is then combined with the priority of the control command itself to generate a weighted sum of the channel preemption priorities. The control command priority is a pre-set static priority that reflects the importance of the control command for each washing machine and is a dimensionless continuous score. For example, certain control commands, such as emergency shutdown and safety alarms, have inherently high priority and should receive priority communication resources even if the washing machine's current operating state is not at high risk of conflict. The weighting coefficients are set to balance the importance of operating state priority and command priority.

[0074] The high-preemption cluster contains scrubbing machines whose weighted channel preemption priority sum exceeds a preset classification threshold. These scrubbing machines are typically in critical operating stages or need to execute important control instructions. The low-preemption cluster contains scrubbing machines in normal operation. Orthogonal frequency-division time slot resource blocks are allocated to each cluster to ensure reliable communication needs for high-priority scrubbing machines while reserving necessary communication resources for low-priority scrubbing machines, preventing complete monopoly of communication resources.

[0075] The generation of clustered communication resource scheduling sequences adheres to the principle of orthogonality, ensuring that frequencies and time slots used by different washing machines at the same time do not interfere with each other. This scheduling method is particularly suitable for scenarios where multiple washing machines are running concurrently, maximizing communication resource utilization while ensuring communication quality.

[0076] The dehydration stage conflict probability P of washing equipment i and washing equipment j is obtained by the double-layer attention network ij The methods include:

[0077] The first layer of the two-layer attention network uses the temporal attention mechanism to calculate the dehydration stage identifier S i (t k ) and real-time energy consumption value E i (t k )’s temporal association weight Among them, Q and K are learnable parameter matrices, and d is the feature dimension.

[0078] The second layer of the two-layer attention network uses the spatial attention mechanism to calculate the conflict probability of the dehydration stage of washing equipment i and washing equipment j Among them, σ is the Sigmoid activation function; b c is the conflict detection bias term; W c is the conflict detection weight matrix; Represents vector concatenation.

[0079] The design of a two-layer attention network is specifically optimized for the temporal characteristics of laundry equipment operation data and the correlations between laundry equipment. The first layer employs a temporal attention mechanism, focusing on analyzing the temporal correlation patterns between the dehydration stage indicator and real-time energy consumption within a single laundry equipment unit. In actual operation, a gradual increase in energy consumption is often seen before a laundry equipment enters the dehydration stage. Energy consumption jumps significantly when dehydration starts, and energy consumption remains high throughout the dehydration process. By learning these typical temporal patterns, the temporal attention mechanism accurately identifies the transitions between dehydration stages. The temporal correlation weight is calculated using the classic attention mechanism formula, capturing the strength of the correlation between the dehydration indicator sequence and the energy consumption sequence through the inner product operation of the query matrix Q and the key matrix K. When the time of the dehydration indicator jump coincides closely with the time of the sudden increase in energy consumption, the corresponding temporal correlation weight increases significantly, indicating that the laundry equipment has indeed entered the dehydration stage during that time period. This attention-based temporal analysis is more accurate and reliable than traditional threshold judgment methods.

[0080] The second layer uses a spatial attention mechanism to calculate the probability of collisions between different washing machines during the dehydration phase. The second layer's input is the temporal correlation weights of each washing machine output from the first layer. The temporal features of the two washing machines are combined through vector concatenation. The collision probability is then calculated using a fully connected layer and a sigmoid activation function. The output of the sigmoid function ranges from 0 to 1, directly corresponding to the numerical meaning of the collision probability. The spatial attention mechanism can learn the interaction patterns between different machines. For example, in a real-world laundry environment, adjacent washing machines, washing machines that share power lines, or washing machines of the same model often pose a greater risk of electromagnetic interference. Through training, the spatial attention network can automatically identify these associations between washing machines, providing a more accurate basis for collision probability prediction.

[0081] The conflict detection weight matrix and bias are optimized through supervised learning using historical operating data. For example, historical laundry operation logs are collected to identify time periods and the laundry equipment involved when communication interference or control delays actually occurred. These are used as positive samples to train the network parameters. Negative samples are taken from time periods when operations were normal and there were no obvious conflicts. By training with a large number of positive and negative samples, the two-layer attention network can accurately identify potential conflict risks between laundry equipment.

[0082] The above is the high preemption cluster C H and low preemption cluster C L The method of allocating an orthogonal frequency division time slot resource block set and generating a clustered communication resource scheduling sequence according to the orthogonal frequency division time slot resource block set includes:

[0083] According to the high preemption cluster C H and low preemption cluster C L The ratio of the number of washing equipment to dynamically divide the orthogonal frequency division time slot resource block set R H With R L , generate clustered communication resource scheduling sequence S = {(i,r i ,t i )|r i ∈R H ∪R L}.

[0084] in, is the scheduling time slot offset of washing equipment i, δ is the time slot interval unit, rank(η i ) represents the weighted sum of channel preemption priority and η i Sequence number in descending order; resource block r i =(f i ,t i ) frequency component f i With the time slot component t i Satisfy the orthogonality constraint |fi ―f j |≥Δf min or |t i ―t j |≥δ min ; Among them, f i The frequency component of the resource block allocated to the washing device i; f j The frequency component of the resource block allocated to scrubbing device j; t i The resource block time slot component allocated to scrubber device i; t j The resource block time slot component allocated to scrubber j; Δf min is the minimum frequency interval, δ min To prevent collision time slot interval.

[0085] The allocation of orthogonal frequency division time slot resource block sets adopts a dynamic partitioning strategy, flexibly adjusting the resource allocation ratio based on the ratio of the number of washing machines in high-preemption clusters and low-preemption clusters. This dynamic allocation mechanism ensures that communication resources can be optimally configured according to actual needs, avoiding resource waste or insufficient supply. During the generation of the clustered communication resource scheduling sequence, each washing machine is allocated a unique resource block, which consists of a frequency component and a time slot component. The calculation of the scheduling time slot offset is based on the weighted and sorted channel preemption priority of the washing machine. The washing machine with a higher priority will obtain the communication time slot sooner, ensuring that key control instructions can be transmitted first. The setting of the time slot interval unit needs to comprehensively consider the transmission time of the control instruction and the response speed requirements of the system.

[0086] Orthogonality constraints are a key technical requirement for ensuring communication quality. The resource blocks of two washing machines must be sufficiently separated in frequency or staggered in time slots to avoid signal interference caused by simultaneous communication on the same frequency. The minimum frequency separation is determined based on the spectrum characteristics and anti-interference capabilities of the washing machine's communication module, while the anti-collision time slot spacing takes into account signal propagation delay and device response time. In practical applications, implementing orthogonality constraints requires precise frequency and timing control. The system maintains a global resource allocation table and records the occupancy status of each frequency and time slot in real time. When allocating resources to a new washing machine, candidate resource blocks are automatically checked to see if they meet the orthogonality constraints. Only those that meet the constraints are allocated to the washing machine. This constraint mechanism ensures reliable concurrent communication among multiple washing machines. The specific values ​​of the frequency and time slot components of the resource blocks must be determined based on the washing machine's communication protocol specifications and the actual wireless environment. Frequency components are typically selected within the industrial, scientific, and medical frequency bands to avoid interference with other wireless devices. Time slot components must have millisecond-level accuracy to ensure that control instructions for critical operations such as dehydration can be transmitted within a strict time window.

[0087] The high preemption cluster CH and low preemption cluster C L The ratio of the number of washing equipment to dynamically divide the orthogonal frequency division time slot resource block set R H With R L The methods include:

[0088] When allocating resource blocks to high-preemption cluster scrubbing devices, establish a frequency-time two-dimensional mapping function Among them, N f is the number of preset frequency band divisions, f0 is the starting frequency, mod is the modulo operation, To round down; The start time of the instruction execution time window of washing equipment i.

[0089] A competitive resource allocation strategy is adopted for low-preemptive cluster washing equipment, and its available resource pool is defined as R L =R total \R H , and in each scheduling cycle according to the channel preemption priority η i Select the ones that meet the requirements in descending order Resource block r i ; Where λ is the time-frequency weight coefficient, ||·|| is the normalized frequency interval distance; R total For all orthogonal frequency division time slot resource blocks.

[0090] When the dehydration stage conflict probability P is detected ij When the collision probability is greater than a preset threshold, resource blocks are forcibly isolated for the washing device i and the washing device j.

[0091] Different resource allocation strategies are used for high-preemption clusters and low-preemption clusters to suit their respective characteristics and needs. Due to the high priority and strict time requirements of the washing equipment in the high-preemption cluster, a deterministic frequency-time two-dimensional mapping function is used for resource allocation. The frequency-time two-dimensional mapping function automatically calculates the corresponding frequency and time slot positions based on the priority ranking of the washing equipment, ensuring the determinism and predictability of resource allocation. The design of the frequency-time two-dimensional mapping function takes into account the uniformity of resource utilization and the need to avoid conflicts. Through modulo operations and floor operations, the linear priority ranking can be mapped to the two-dimensional frequency-time resource space to achieve a uniform distribution of resources. The selection of the number of preset frequency band divisions needs to balance the full utilization of frequency resources and the interference control requirements of channel spacing.

[0092] Low-preemption cluster devices use a competitive resource allocation strategy, selecting the optimal resource blocks from the remaining resource pool in order of priority. Optimality is determined by maximizing the frequency and time slot spacing from already allocated resource blocks, thereby minimizing potential communication interference. The time-frequency weighting coefficients reflect the relative importance of frequency isolation and time slot isolation, with frequency isolation typically given a higher weight because frequency interference is more difficult to handle than time slot conflicts. The competitive allocation strategy's advantage lies in its ability to fully utilize remaining resources and avoid idle resources. During each scheduling cycle, low-preemption cluster washing devices re-compete for resources based on their current priority ranking, ensuring fair and dynamic resource allocation. The dynamic competition mechanism is particularly well-suited for handling situations where the operating status of multiple washing machines frequently changes, such as multiple washing machines in a laundry room.

[0093] The mandatory resource block isolation mechanism is a special protection measure for devices with high conflict risk. The conflict probability threshold should be set based on historical data analysis and actual operational experience. A too low conflict probability threshold can lead to frequent resource reallocation, impacting stability; a too high threshold can prevent timely detection of potential conflicts, compromising control effectiveness. Generally, a conflict probability threshold between 0.7 and 0.8 is appropriate, effectively identifying high-risk situations without being overly sensitive.

[0094] When the dehydration stage conflict probability P is detected ij When the probability of conflict is greater than a preset threshold, the method for forcibly isolating the resource blocks of the scrubbing device i and the scrubbing device j includes:

[0095] If the resource block r i ∈R H And resource block r j ∈R H , then the frequency interval |f′ is reassigned to washing equipment j j ―f i |≥2Δf min And the time slot interval |t′ j ―t i |≥2δ min Resource block r′ j ; If resource block r j ∈R L , then upgrade it to R H And trigger inter-cluster resource redistribution.

[0096] The inter-cluster resources are redistributed into L The resource block that conflicts with washing equipment j is moved to R H , and recalculate R H With R L The division ratio.

[0097] The implementation of mandatory resource block isolation requires distinguishing different conflict situations and taking appropriate action. When a high conflict risk is detected between two scrubbing machines in a high-preemption cluster, resource blocks are reassigned to the lower-priority scrubbing machine first. The new resource blocks must meet stricter frequency and time slot spacing requirements to ensure that communications between the two scrubbing machines do not interfere with each other. The doubling of the frequency and time slot spacing requirements reflects the stringency of mandatory isolation. While the minimum anti-collision time slot requirement under normal circumstances can prevent most communication interference, the doubled anti-collision time slot requirement provides an additional safety margin for high-risk devices. While this conservative isolation strategy consumes more communication resources, it ensures that critical control command transmission is not affected. When a conflict occurs between scrubbing machines in a low-preemption cluster and a high-preemption cluster, an escalation strategy is implemented, upgrading the conflicting device from the low-preemption cluster to the high-preemption cluster, triggering a reallocation of inter-cluster resources. This escalation mechanism ensures that all high-risk scrubbing machines have reliable communication resources, avoiding control errors caused by improper priority assignments.

[0098] Repartitioning resources between clusters is a complex dynamic adjustment process. The number of washing machines and resource requirements in the high- and low-preemption clusters must be recalculated, and the resource allocation ratio between the two clusters adjusted. Some resources originally allocated to the low-preemption cluster must be migrated to the high-preemption cluster to provide sufficient communication resources for the newly upgraded washing machines.

[0099] The method of obtaining the number of instruction conflicts identified by each washing device during the dehydration stage and constructing a load balancing optimization function with the instantaneous circuit load includes:

[0100] The number of instruction conflicts is the number of times the cumulative conflict probability between washing devices exceeds a preset conflict probability threshold; the number of instruction conflicts for each washing device is normalized according to a preset circuit load safety threshold, and a conflict penalty term is applied, and a load overlimit penalty term is introduced when the instantaneous circuit load exceeds the preset circuit load safety threshold; the hard constraint of time window overlap in the dehydration stage is converted into an exponential decay function of the time difference between the dehydration start times of the washing devices to obtain a time window overlap risk term; the conflict penalty term, the load overlimit penalty term and the time window overlap risk term are weighted and fused, and the regularization constraints of the channel capacity adjustment amount and the competition parameter offset are superimposed to obtain a load balancing optimization function.

[0101] The construction of the load balancing optimization function comprehensively considers the two core goals of communication efficiency and circuit load safety. The number of instruction conflicts is counted cumulatively, recording the total number of times the probability of conflict between washing machines exceeds the preset circuit load safety threshold within a specific time window, which directly reflects the conflict control effect of the current resource allocation scheme. The greater the number of conflicts, the more serious the communication interference between washing machines, which needs to be improved by adjusting resource allocation. The design of the conflict penalty term adopts normalization processing, standardizing the number of instruction conflicts for each washing machine according to the circuit load safety threshold. This is used to ensure that the conflict penalties for different washing machines are comparable and avoid unfair penalties caused by differences in washing machine power. The intensity of the penalty is proportional to the frequency of conflicts. The more frequent the conflicts, the heavier the penalty for the device.

[0102] The load overload penalty is specifically designed to address circuit load safety risks. When the instantaneous circuit load of a washing machine exceeds the preset circuit load safety threshold, an additional penalty is imposed, forcing the optimization algorithm to reduce the resource usage of the washing machine or delay the execution of its high-power operations. This penalty mechanism is an important guarantee for ensuring the safe operation of the circuit system. The time window overlap risk term converts hard constraints into soft penalties, quantifying the impact of the time difference between the dehydration start times of different washing machines on the system risk through an exponential decay function. The smaller the time difference, the higher the risk and the heavier the corresponding penalty. The weighting of the conflict penalty, load overload penalty, and time window overlap risk term needs to reflect their respective importance in system safety and efficiency. Generally, the load overload penalty has the highest weight because circuit safety is the bottom line requirement for system operation.

[0103] The introduction of regularization constraints prevents over-adjustment during the optimization process. Regularization constraints on channel capacity adjustment and competition parameter offset limit the magnitude of a single adjustment, ensuring system stability and controllability. Excessive parameter adjustment can disrupt the existing communication balance, leading to oscillation or instability.

[0104] The method for obtaining a dehydration peak shifting control instruction by dynamically adjusting the channel capacity of the orthogonal frequency division time slot and the contention parameter of the carrier sensing window to minimize the output value of the load balancing optimization function includes:

[0105] The channel capacity of the orthogonal frequency division time slot and the contention parameters of the carrier sensing window are iteratively adjusted through the gradient descent algorithm, so that the output value of the load balancing optimization function gradually converges to the minimum value. At the same time, the transmission delay constraint of the dehydration peak-shifting control instruction is verified in real time. If there is a washing equipment whose transmission delay exceeds the maximum allowable transmission delay threshold, the optimization is re-performed until a dehydration peak-shifting control instruction that satisfies both circuit load safety and transmission delay constraints is generated.

[0106] Dehydration peak-shifting control instructions are generated using an iterative optimization gradient descent algorithm. This algorithm determines the direction and magnitude of parameter adjustments by calculating the partial derivatives of the load balancing optimization function with respect to each adjustment parameter. Adjusting the channel capacity of orthogonal frequency-division time slots affects the data transmission capacity within each time slot, while adjusting the contention parameters of the carrier sense window influences the waiting strategy and backoff mechanism of devices competing for channels.

[0107] Convergence during iterative optimization is determined based on the trend of the load balancing optimization function value. When the improvement in the objective function value over several consecutive iterations is less than a preset convergence threshold, the optimal solution for the current conditions is considered to have been found. Setting the convergence threshold requires a balance between optimization accuracy and computational efficiency. A too small convergence threshold results in excessive computation time, while a too large one may lead to incomplete optimization.

[0108] Real-time verification of transmission delay constraints is crucial for ensuring the timeliness of control commands. After each parameter adjustment, the expected transmission delay for each device's control commands is recalculated and compared to the maximum allowable transmission delay threshold. If the transmission delay of any washing machine is found to exceed the limit, the resource allocation or priority setting for that washing machine is adjusted, and the optimization calculation is re-performed. Circuit load safety constraints ensure that too many washing machines do not enter the dehydration stage simultaneously, causing circuit overload. Transmission delay constraints ensure that all control commands are accurately transmitted within the specified time. Meeting these two constraints simultaneously is a fundamental requirement for the safe and reliable operation of intelligent washing control systems.

[0109] Example 2: Based on the same inventive concept, Figure 2 As shown, this embodiment also provides a washing intelligent control system based on multimodal data, the system comprising: a multimodal communication feature matrix acquisition module, a clustered communication resource scheduling sequence generation module, a load balancing analysis module, and a dehydration peak shifting management module, wherein each module is sequentially connected to each other;

[0110] A multimodal communication feature matrix acquisition module is used to collect, in real time, a set of operating status parameters and a set of communication demand parameters of each washing device through a wireless sensor network of the washing device communication nodes. The operating status parameter set includes a water level parameter, a real-time energy consumption value, and a dehydration stage identifier, and the communication demand parameter set includes a control instruction priority, a maximum allowable transmission delay threshold, and an instruction execution time window. The operating status parameter set and the communication demand parameter set are time-series aligned and feature-fused to obtain a multimodal communication feature matrix.

[0111] A clustered communication resource scheduling sequence generation module is used to analyze the coupling relationship between the dehydration stage identification and the real-time energy consumption value of the washing equipment using the multimodal communication feature matrix through a preset conflict prediction model, dynamically calculate the channel preemption priority of each washing equipment according to the instruction execution time window based on the coupling relationship; and generate a clustered communication resource scheduling sequence based on the channel preemption priority and the control instruction priority.

[0112] A load balancing analysis module is configured to obtain the number of instruction conflicts identified by each washing device during the dehydration phase after performing hybrid channel allocation according to the clustered communication resource scheduling sequence, and to construct a load balancing optimization function based on the instantaneous circuit load;

[0113] The dehydration peak-shifting management module is used to obtain a dehydration peak-shifting control instruction and send it to the control end by dynamically adjusting the channel capacity of the orthogonal frequency division time slot and the competition parameters of the carrier sensing window to minimize the output value of the load balancing optimization function; the transmission delay of the dehydration peak-shifting control instruction is lower than the maximum allowable transmission delay threshold.

[0114] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated on here.

[0115] Finally, it should be noted that although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The intelligent washing control method based on multimodal data is characterized by: The method comprises: The wireless sensor network of the washing equipment communication nodes collects in real time a set of operating status parameters and a set of communication demand parameters of each washing equipment; the set of operating status parameters includes a water level parameter, a real-time energy consumption value, and a dehydration stage identifier; the set of communication demand parameters includes a control instruction priority, a maximum allowable transmission delay threshold, and an instruction execution time window; the set of operating status parameters and the set of communication demand parameters are time-series aligned and feature-fused to obtain a multimodal communication feature matrix; The multimodal communication feature matrix is ​​used to analyze the coupling relationship between the dehydration stage identifier and the real-time energy consumption value of the washing equipment through a preset conflict prediction model, and the coupling relationship is dynamically calculated according to the instruction execution time window to calculate the channel preemption priority of each washing equipment; a clustered communication resource scheduling sequence is generated based on the channel preemption priority and the control instruction priority; After performing hybrid channel allocation according to the clustered communication resource scheduling sequence, the number of instruction conflicts identified by each washing device during the dehydration phase is obtained, and a load balancing optimization function is constructed with the instantaneous circuit load; By dynamically adjusting the channel capacity of the orthogonal frequency division time slot and the competition parameters of the carrier sensing window to minimize the output value of the load balancing optimization function, a dehydration peak-shifting control instruction is obtained and sent to the control end; the transmission delay of the dehydration peak-shifting control instruction is lower than the maximum allowable transmission delay threshold.

2. The intelligent washing control method based on multimodal data according to claim 1, characterized in that: The method of performing time alignment and feature fusion on the operating status parameter set and the communication requirement parameter set to obtain a multimodal communication feature matrix includes: Generate a global timestamp sequence T={t1…,t k …,t n }, t k is the reference time point of the kth synchronization cycle; for the water level parameter L in the operating state parameter set i (t) and real-time energy consumption value E i (t) Perform piecewise linear interpolation to generate a discrete sequence L aligned on the global timestamp sequence T i (t k ) and E i (t k ), where i is the washing equipment number; The instruction execution time window [τ start ,τ end ] is mapped to a binary indicator function I corresponding to the global timestamp sequence T i (t k ), if and only if t k ∈[τ start ,τ end ] Time i (t k )=1, otherwise I i (t k )=0; The sliding window algorithm is used to calculate the dehydration stage marker S i (t) The difference sequence ΔS on the global timestamp sequence T i (t k )=S i (t k )―S i (t k―1 ), when ΔS i (t k )≥θ s When t is the starting time of the dehydration phase, s , and construct the dehydration stage feature vector D i =[Δt s ,ΔE s ]; where θ s is the preset dehydration stage state mutation threshold; Δt s The starting time of the dehydration phase is t s The time deviation from the reference time point of the most recent synchronization cycle; ΔE s It is the sliding average difference of energy consumption before and after dehydration starts; After the timing is aligned, L i (t k ), E i (t k ), I i (t k ) and D i According to the washing equipment number i and the reference time point t k They are combined into a multi-dimensional tensor and a multimodal communication feature matrix is ​​generated through the feature fusion function.

3. The intelligent washing control method based on multimodal data according to claim 2, characterized in that: The multimodal communication feature matrix is ​​analyzed using a preset conflict prediction model to analyze the coupling relationship between the dehydration stage identifier and the real-time energy consumption value of the washing equipment, and the coupling relationship is dynamically calculated according to the instruction execution time window for the channel preemption priority of each washing equipment; The method for generating a clustered communication resource scheduling sequence according to the channel preemption priority and the control instruction priority includes: The preset conflict prediction model adopts a double-layer attention network, and the dehydration stage conflict probability P of washing equipment i and washing equipment j is obtained through the double-layer attention network. ij According to the conflict probability P ij and the instruction execution time window [τ start ,τ end ] dynamically generates the channel preemption priority score ρ of washing equipment i i ; The channel preemption priority score ρ i The calculation formula is: Where ξ is the instruction execution time window overlap penalty factor; is the real-time energy consumption change rate of washing equipment i after standardization; Ω ij is the overlap of the instruction execution time windows of washing equipment i and washing equipment j, and its value range is [0,1]; According to the channel preemption priority score ρ i and control instruction priority π i Get the weighted sum of channel preemption priority η i =ρ i +νπ i , where ν is a preset weight coefficient; according to the weighted sum of the channel preemption priority η i Divide the washing equipment into high-preemption cluster C H ={i|η i ≥η th } with low preemption cluster C L and for high preemption cluster C H and low preemption cluster C L Allocate an orthogonal frequency division time slot resource block set, and generate a clustered communication resource scheduling sequence according to the orthogonal frequency division time slot resource block set; wherein, n th The classification threshold is the weighted sum of the preset channel preemption priorities.

4. The intelligent washing control method based on multimodal data according to claim 3, characterized in that: The dehydration stage conflict probability P of washing equipment i and washing equipment j is obtained by the double-layer attention network ij The methods include: The first layer of the two-layer attention network uses the temporal attention mechanism to calculate the dehydration stage identifier S i (t k ) and real-time energy consumption value E i (t k )’s temporal association weight Among them, Q and K are learnable parameter matrices, and d is the feature dimension; The second layer of the two-layer attention network uses the spatial attention mechanism to calculate the conflict probability of the dehydration stage of washing equipment i and washing equipment j Among them, σ is the Sigmoid activation function; b c is the conflict detection bias term; W c is the conflict detection weight matrix; Represents vector concatenation.

5. The intelligent washing control method based on multimodal data according to claim 3, characterized in that: The above is the high preemption cluster C H and low preemption cluster C L The method of allocating an orthogonal frequency division time slot resource block set and generating a clustered communication resource scheduling sequence according to the orthogonal frequency division time slot resource block set includes: According to the high preemption cluster C H and low preemption cluster C L The ratio of the number of washing equipment to dynamically divide the orthogonal frequency division time slot resource block set R H With R L , generate clustered communication resource scheduling sequence S = {(i,r i ,t i )|r i ∈R H ∪R L }; in, is the scheduling time slot offset of washing equipment i, δ is the time slot interval unit, rank(η i ) represents the weighted sum of channel preemption priority and η i Sequence number in descending order; resource block r i =(f i ,t i ) frequency component f i With the time slot component t i Satisfy the orthogonality constraint |f i ―f j |≥Δf min or |t i ―t j |≥δ min ; Among them, f i The frequency component of the resource block allocated to the washing device i; f j The frequency component of the resource block allocated to scrubbing device j; t i The resource block time slot component allocated to scrubber device i; t j The resource block time slot component allocated to scrubber j; Δf min is the minimum frequency interval, δ min To prevent collision time slot interval.

6. The intelligent washing control method based on multimodal data according to claim 5, characterized in that: The high preemption cluster C H and low preemption cluster C L The ratio of the number of washing equipment to dynamically divide the orthogonal frequency division time slot resource block set R H With R L The methods include: When allocating resource blocks to high-preemption cluster scrubbing devices, establish a frequency-time two-dimensional mapping function Among them, N f is the number of preset frequency band divisions, f0 is the starting frequency, mod is the modulo operation, To round down; The start time of the instruction execution window of washing equipment i; A competitive resource allocation strategy is adopted for low-preemptive cluster washing equipment, and its available resource pool is defined as R L =R total \R H , and in each scheduling cycle according to the channel preemption priority η i Select the ones that meet the requirements in descending order Resource block r i ; Where λ is the time-frequency weight coefficient, ||·|| is the normalized frequency interval distance; R total For all orthogonal frequency division time slot resource blocks; When the dehydration stage conflict probability P is detected ij When the collision probability is greater than a preset threshold, resource blocks are forcibly isolated for the washing device i and the washing device j.

7. The intelligent washing control method based on multimodal data according to claim 6, characterized in that: When the dehydration stage conflict probability P is detected ij When the probability of conflict is greater than a preset threshold, the method for forcibly isolating the resource blocks of the scrubbing device i and the scrubbing device j includes: If the resource block r i ∈R H And resource block r j ∈R H , then the frequency interval |f′ is reassigned to washing equipment j j ―f i |≥2Δf min And the time slot interval |t′ j ―t i |≥2δ min Resource block r′ j ; If resource block r j ∈R L , then upgrade it to R H And trigger the redistribution of resources between clusters; The inter-cluster resources are redistributed into L The resource block that conflicts with washing equipment j is moved to R H , and recalculate R H With R L The division ratio.

8. The intelligent washing control method based on multimodal data according to claim 1, characterized in that: The method of obtaining the number of instruction conflicts identified by each washing device during the dehydration stage and constructing a load balancing optimization function with the instantaneous circuit load includes: The number of instruction conflicts is the number of times the cumulative conflict probability between washing devices exceeds a preset conflict probability threshold; the number of instruction conflicts for each washing device is normalized according to a preset circuit load safety threshold, and a conflict penalty term is applied, and a load overlimit penalty term is introduced when the instantaneous circuit load exceeds the preset circuit load safety threshold; the hard constraint of time window overlap in the dehydration stage is converted into an exponential decay function of the time difference between the dehydration start times of the washing devices to obtain a time window overlap risk term; the conflict penalty term, the load overlimit penalty term and the time window overlap risk term are weighted and fused, and the regularization constraints of the channel capacity adjustment amount and the competition parameter offset are superimposed to obtain a load balancing optimization function.

9. The intelligent washing control method based on multimodal data according to claim 8, characterized in that: The method for obtaining a dehydration peak shifting control instruction by dynamically adjusting the channel capacity of the orthogonal frequency division time slot and the contention parameter of the carrier sensing window to minimize the output value of the load balancing optimization function includes: The channel capacity of the orthogonal frequency division time slot and the contention parameters of the carrier sensing window are iteratively adjusted through the gradient descent algorithm, so that the output value of the load balancing optimization function gradually converges to the minimum value. At the same time, the transmission delay constraint of the dehydration peak-shifting control instruction is verified in real time. If there is a washing equipment whose transmission delay exceeds the maximum allowable transmission delay threshold, the optimization is re-performed until a dehydration peak-shifting control instruction that satisfies both circuit load safety and transmission delay constraints is generated.

10. The intelligent washing control system based on multimodal data is characterized by: The system includes: a multimodal communication feature matrix acquisition module, a clustered communication resource scheduling sequence generation module, a load balancing analysis module, and a dehydration peak shifting management module, and each module is sequentially connected to each other; A multimodal communication feature matrix acquisition module is configured to collect, in real time, a set of operating status parameters and a set of communication requirement parameters of each washing device through a wireless sensor network of washing device communication nodes; the set of operating status parameters includes a water level parameter, real-time energy consumption value, and dehydration stage identifier; and the set of communication requirement parameters includes a control instruction priority, a maximum allowable transmission delay threshold, and an instruction execution time window; and perform time alignment and feature fusion on the set of operating status parameters and the set of communication requirement parameters to obtain a multimodal communication feature matrix. A clustered communication resource scheduling sequence generation module is configured to analyze the coupling relationship between the dehydration stage identifier and the real-time energy consumption value of the washing equipment using the multimodal communication feature matrix through a preset conflict prediction model, dynamically calculate the channel preemption priority of each washing equipment based on the instruction execution time window according to the coupling relationship, and generate a clustered communication resource scheduling sequence based on the channel preemption priority and the control instruction priority; A load balancing analysis module is configured to obtain the number of instruction conflicts identified by each washing device during the dehydration phase after performing hybrid channel allocation according to the clustered communication resource scheduling sequence, and to construct a load balancing optimization function based on the instantaneous circuit load; The dehydration peak-shifting management module is used to obtain a dehydration peak-shifting control instruction and send it to the control end by dynamically adjusting the channel capacity of the orthogonal frequency division time slot and the competition parameters of the carrier sensing window to minimize the output value of the load balancing optimization function; the transmission delay of the dehydration peak-shifting control instruction is lower than the maximum allowable transmission delay threshold.

Images