74 results about "Dynamic channel" patented technology

The invention provides an intelligent short message scheduling method and device based on multi-dimensional dynamic optimization, and the method comprises the steps: obtaining the performance data of a plurality of short message channels, and calculating a channel health score based on a weight dynamic adjustment model; determining a scheduling strategy according to the priority identifier of the to-be-sent message, and performing channel screening and optimal matching; executing message sending and monitoring a sending state; terminal state detection is carried out on the failure message through operator base station signaling, and a decision tree model is applied to determine a retry strategy; and performing Huffman coding compression processing on the P2-level marketing messages which fail in retry, and performing batch sending in an idle window. According to the method, a comprehensive performance evaluation index and reward function model is also constructed, and parameter optimization is performed by applying a reinforcement learning algorithm. According to the invention, multi-dimensional dynamic channel scoring, intelligent retry decision making based on terminal state perception, batch processing with balanced cost-time efficiency and a closed-loop self-optimization system are realized, the short message delivery rate is obviously improved, and the invalid retry rate and the sending cost are reduced.

The invention provides a leather defect identification method based on double-branch frequency domain feature decoupling and a dynamic channel enhancement network. For the problem that the characteristics of a leather image are difficult to correctly extract due to the characteristics of complexity and non-uniformity of the leather image, a high and low frequency domain decoupling thought is introduced in the invention to extract main body and detail information respectively; a DCT (Discrete Cosine Transformation) guided high / low-frequency double-branch network is designed to respectively capture a local structure (high-frequency branch) of a microdefect and a continuity feature (low-frequency branch) of a macroscopic texture, a structure in which an SE (Square-and-Excitation) module is embedded in a feature fusion stage is designed, and channel weight and spatial position response are jointly optimized, so that the feature fusion is realized. And the significance expression of the key defect area is enhanced. In some public data sets, the method has better classification performance and higher diagnosis speed.

The invention discloses a key value cache grouping quantification method and device, a storage medium and electronic equipment. When the key value cache grouping quantification method provided by the specification is adopted to quantify key value cache data, key vector and value vector data are divided based on channel dimensions to obtain a plurality of grouped data; determining quantization parameters of the grouped data, and performing asymmetric quantization on elements in the grouped data based on the quantization parameters; finally, quantization results and corresponding quantization parameter partitions may be stored in physical blocks. According to the method, on the premise of ensuring the model generation precision, the video memory occupation of key value cache is greatly compressed, and meanwhile, the reasoning throughput is improved. Through fusion of dynamic channel grouping quantization and implicit inverse quantization, a key value cache quantization solution capable of guaranteeing the generation precision is provided for edge device end-side deployment, and the technical defect that the occupation scale of a video memory is too large when the occupation amount of the key value cache video memory is linearly increased along with the sequence length in the autoregression decoding process of a large language model is relieved.

The invention relates to the technical field of data fusion, in particular to a multi-source electrocardiosignal correction method and system based on adaptive fusion, and the method comprises the following steps: constructing a multi-channel input tensor, extracting local features through a weight calculation network, carrying out the adaptive weight fusion and dimension reduction of multiple paths of signals, and carrying out the correction of the multi-source electrocardiosignal. A nonlinear mapping relation is established through a deep reconstruction network, a standard waveform is reconstructed, and network parameters are optimized based on reconstruction error reverse iteration. According to the method, local neighborhood features of multichannel signals are extracted by constructing a weight calculation network, a dynamic channel weight sequence reflecting the real-time contribution degree of a signal source is constructed, the amplitude intensity is adaptively adjusted according to the signal quality, unstable channel noise interference is effectively inhibited, and high-quality signal components are enhanced; a deep reconstruction network is used for carrying out nonlinear feature transformation on a fusion sequence, accurate mapping from non-standard input to standard lead waveforms is established, and weight distribution and optimization of signal reconstruction parameters are achieved in combination with an error back propagation mechanism.

The invention discloses an electrocardiosignal artifact elimination method and system based on dynamic channel weighting, and the method comprises the steps: a data collection and preprocessing stage: collecting original ECG physiological signal data, and building a data set suitable for subsequent model training; a network model construction stage: designing a network architecture with a dynamic channel weighting mechanism for electrocardiosignal motion artifact confrontation elimination; a model training stage: training the model by using the constructed data set; in the practical application stage, ECG signals to be processed are input into the trained model, artifact filtering is achieved through multi-stage feature processing, and purified ECG signals are output. The system comprises a data set construction unit, a model construction unit, a model training unit and an artifact elimination unit. By using the method and the device, artifacts in the electrocardiosignals can be effectively eliminated. The method can be widely applied to the field of signal processing.

The invention provides an eddy current sensor signal noise reduction method based on frequency division detection low-pass filtering and a monitoring system, belongs to the technical field of sensor signal processing, and can at least partially solve the problems of unstable signals and inaccurate measurement in the prior art. The signal noise reduction method for the eddy current sensor comprises the following steps: processing an original output signal of the eddy current sensor by adopting a synchronous detection module, demodulating a dynamic vibration alternating current component of the original output signal into a direct current voltage signal by utilizing a reference signal, and separating a static displacement direct current component signal; an independent dynamic signal channel and an independent static signal channel are formed; filtering high-frequency noise in the dynamic signal channel through a dynamic channel low-pass filter with a first preset cut-off frequency, and filtering power frequency interference in the static signal channel through a static channel low-pass filter with a second cut-off frequency; and reversely injecting compensation voltage at the output end of the static channel low-pass filter so as to eliminate the direct-current offset error of the static displacement component signal.

The invention discloses a double-flow fault diagnosis method based on state space modeling and dynamic edge context awareness, the method constructs a time domain state flow and a time frequency edge flow, and the specific implementation process is as follows: the time domain state flow is subjected to state space modeling under a Transform framework to form a MambaT encoder, long and short term dependence is efficiently captured with linear complexity, and the time domain state flow is subjected to dynamic edge context awareness; time domain sensitive features are accurately represented; the time-frequency edge stream converts vibration data into a two-dimensional time-frequency image through wavelet transformation, a differentiable edge sensitive mask is designed to adaptively separate time-frequency image impact characteristics and background noise, boundary content perception adaptive filling is designed to maintain image boundary topology integrity, and the two parts cooperate to form dynamic edge context perception convolution, so that the image boundary topology integrity is improved. Time-frequency edge features are effectively extracted; and a channel enhancement and channel gating module is designed to establish a dynamic channel interaction gating mechanism, so that time domain-time frequency domain double-current cascade characterization is realized, and the fault discrimination capability is remarkably improved. Experiments prove that the method has higher fault diagnosis performance.

The invention discloses a security risk dynamic filtering and feedback control method for large model output content. The method comprises the step of executing a multi-dimensional risk feature library construction operation. According to the method, semantic features are extracted by adopting a BERT-BiLSTM hybrid model, risk modes are automatically found in combination with an improved DBSCAN algorithm, the updating efficiency of a feature library is improved by 40%, feature complementation is achieved through a text / image / voice three-mode heterogeneous network design and a cross-mode attention mechanism, the composite violation content detection accuracy rate reaches 92.3%, and the method has the advantages that the method is simple and convenient to operate and high in practicability. A static channel is adopted to ensure basic rule adherence, a dynamic channel is adopted to capture context risks, a three-level filtering strategy is adopted, illegal content is clear to be completely shielded within 0.3 second, a visual warning label is added to suspected content, edge content is manually rechecked, and the time is shorter than or equal to 5 minutes, so that when the model is output, content safety is ensured, and the probability of illegal content detection is lowered. And the application value of the large model is furthest maintained.

The invention discloses a model generalization ability optimization method and system based on hybrid experts, and aims to solve the problems of weak generalization ability, difficulty in knowledge migration, low semantic alignment efficiency and the like when a multi-modal large model performs fine adjustment on downstream tasks in a new field. The method comprises the following steps: embedding a hybrid expert module in a low-rank adaptive fine tuning framework of a self-attention module in the visual encoder, introducing a heterogeneous hybrid convolution structure, innovatively realizing transverse and longitudinal independent scaling through bilinear interpolation in the convolution structure, supporting a multi-scale convolution kernel, and realizing dynamic combination through an MoE framework; constructing a cross-modal expert adapter of channel perception, and adopting dynamic channel recombination, cross-modal conditional convolution and weighted spatial feature aggregation methods; according to the method, the problem of local semantic confusion can be well solved, meanwhile, semantic fusion of image features and text features is enhanced, more effective spatial information and priori knowledge are injected into the model, and therefore the generalization of the model is improved.

The invention belongs to the technical field of computer vision and neuromorphic computing. The invention provides a multi-scale dynamic fusion target detection method and system based on a spiking neural network. According to the embodiment of the invention, the network adopts a trainable integer LIF neuron, the SABC module adopts a dynamic channel expansion mechanism, the problem of deep feature degradation caused by pulse signal time sequence sparsity is effectively solved, the SMGFDH realizes multi-scale feature fusion of pulse distribution rate self-adaption through a differential attention mechanism, and the accuracy of the feature fusion is improved. And redundant convolution operation before traditional splicing is innovatively canceled, and an efficient detection framework completely adaptive to pulse sparsity is constructed. The problems of insufficient deep feature expression and low multi-scale fusion efficiency of the existing SNN in a complex visual task are solved.

The invention discloses a multi-channel data transmission control method and system. The method comprises the following steps: acquiring current to-be-transmitted data and an idle channel set; judging the relationship between the number of channels and the data volume, and when the channels are sufficient and the data volume is large, splitting the data into reference data and corresponding split data by adopting an intelligent data division strategy; establishing a matching relationship between the data set and the channel; inputting the reference data and the split data into corresponding channels for transmission in a cross arrangement mode; and performing data fusion based on the same data field at a receiving end. Through dynamic channel matching, intelligent data splitting and optimization recombination mechanisms, the channel utilization rate and the transmission efficiency are effectively improved, the data recombination complexity is reduced, the method is particularly suitable for high-concurrency and low-delay data transmission scenes, and compared with a traditional method, the transmission performance can be remarkably improved.

The invention provides a multi-branch sequence recommendation method based on dynamic channel fusion. The method aims at solving the problems that in sequence recommendation, data have a large amount of noise and are sparse, and existing model prediction is smooth. The method comprises the following steps: on one hand, enabling a user sequence to pass through an embedding layer, introducing position information by utilizing RoPE rotation position coding, and extracting long-term dependency of a user in the sequence through a multi-feature channel feature network to obtain a new embedded Eglobal, a short-term interest and long-term dependency interactive embedded Ecross, an embedded Eema after exponential smoothing filtering and an embedded Efreq after frequency domain modulation; and on the other hand, in order to fully fuse the characteristics of different channels, a traditional hyper-parameter fusion mode is abandoned, and a pooling network layered adaptive fusion mode is adopted. Firstly, four channel features are divided into two groups including a self-attention group and a trend group, features of different channels are learned through Squeeze average pooling, learned parameters are activated through Excitation by adopting sigmoid to obtain channel weights, and feature representations of different groups are sent into a gating network for weighted summation loss calculation.

The invention discloses a back door defense method and system based on adaptive feature blocking, and the method comprises the steps: 1, constructing a blocking module, and embedding the blocking module into an image classification model; the blocking module comprises a self-adaptive instance statistical calibration layer and a dynamic channel suppression layer, the self-adaptive instance statistical calibration layer is used for eliminating cross-sample statistical offset caused by backdoor attack, and the dynamic channel suppression layer is used for suppressing abnormal activation of a polluted channel; and 2, training the embedded classification network to obtain a final classification network, and completing backdoor attack defense according to the final classification network. Through lightweight modular design and a fine adjustment mechanism, the bottleneck problem of a traditional model defense method in the aspects of calculation cost and flexibility is effectively broken through.

The invention discloses an SV message dynamic channel real-time mapping device based on a modification instruction set, comprising a configuration computer which generates a data packet corresponding to the SV modification instruction set and a data packet corresponding to SV message static data, and the message modification module modifies SV message dynamic data in a to-be-matched SV message received by the message receiving module from the to-be-matched source device based on the SV modification instruction set, then generates a modified SV message in combination with the SV message static data, and sends the modified SV message to the target device. The invention further discloses an SV message dynamic channel real-time mapping method based on the modification instruction set, and the to-be-matched SV message from the to-be-matched source device is modified and converted into the message which can be received by the target device in real time. The problems that cross-interval equipment stops for multiple times and SV messages of new and old devices are not matched in the transformation process of the secondary equipment of the intelligent substation are solved, and new equipment debugging, synchronous operation of the old equipment and seamless transformation switching of the new and old equipment are achieved.

The invention provides an access point communication method with a dynamic channel selection mechanism, which is applied to an access point device and comprises the following steps: performing wireless communication with a station device through an adjacent relay access point device; periodically collecting a plurality of communication parameters related to the wireless communication; obtaining a required data flow of the station device according to the communication parameters, and calculating a plurality of available data flows corresponding to a plurality of wireless channels between the access point device and the relay access point device; calculating a plurality of ratios of the required data traffic relative to the available data traffic as a plurality of channel congestion parameters corresponding to the wireless channel; and selecting one of the wireless channels corresponding to one of the channel congestion parameters with the minimum value as the selected wireless channel to perform packet transmission to the site device through the relay access point device.

The invention discloses a cigarette carton cache channel tide type flexible control scheduling method and system, and belongs to the technical field of cigarette carton channel control scheduling. A part of cache channels are used as flexible dynamic channels, and a system takes over brand allocation decisions of the part of channels; carrying out wave treatment on the replenishment demand; in the wave time, according to the demand quantity of replenishment of the channel type cache system for the sorting line, the number of cache channels matched with the demand is dynamically distributed for each brand; the warehouse-in and warehouse-out processes of the cache channel are managed, and warehouse-in locking is carried out on the dynamic channel in good time so as to ensure that the inventory of the dynamic channel is emptied when the replenishment wave is finished, automatic recovery of the channel is realized, the brand of the channel is redistributed for the subsequent replenishment wave, and the configuration of the cache channel of the brand and the tide-type following change of the replenishment demand are realized on the whole. After the method is applied to control scheduling of the channel type cache, the resource utilization rate of the cache channel can be improved, human intervention is reduced, and the workload and management difficulty of manual control are reduced.

The invention discloses a dynamic channel normalization method based on shift operation, and belongs to the technical field of artificial intelligence and integrated circuit design. According to the method, a division-free and floating-point-exponent-free normalization operator is constructed through piecewise linear fitting and a binarization power interpolation mechanism, so that hardware-friendly dynamic channel normalization based on shift operation is realized, traditional high-complexity nonlinear mapping is converted into most efficient shift and additive operation of bottom hardware, and the method has the advantages that the complexity of the hardware is reduced, and the reliability of the hardware is improved. The area and the power consumption of the circuit are greatly reduced, and the feature extraction capability of the model is maintained.

The application discloses a traffic sign detection method based on an adaptive double-path local-global fusion network and belongs to the technical field of computer vision and intelligent traffic. Three innovative modules work together: the nonlinear adaptive feature extraction module adopts a wide-narrow parallel structure, combines an adaptive deformation activation function and a partial convolution mechanism, and enhances the capture ability of weak features; the dynamic channel calibration attention module realizes dynamic calibration of key channels through a local-global double-path interaction strategy and double-view attention fusion; and the local-global feature fusion module efficiently fuses multi-scale features based on multi-granularity parallel attention and reparameterization convolution technology. The application constructs an end-to-end detection network on the basis of a YOLO framework, significantly improves the precision and environmental adaptability of traffic sign detection, balances the calculation efficiency and resource consumption through light-weight design, and is suitable for high real-time requirement scenes such as automatic driving and intelligent traffic systems.

The invention provides a DBLE-based bearing surface feature fusion method and system and a storage medium, and the method comprises the steps: extracting bottom layer features based on a preset convolution kernel according to a first feature image of a middle layer in a coding stage of a bearing surface, sequentially carrying out the global averaging and activation processing of each channel of the bottom layer features, and carrying out the dynamic distribution of channel weights, fusing to obtain a first fused feature map; and then, performing splicing or element-by-element addition on the high-resolution second feature image of the bearing surface in the encoding stage and the first fusion feature map, and outputting a multi-fusion feature map based on a preset ECA mechanism and convolution operation. According to the collaborative method, the adaptability to diversified defects is enhanced through multi-scale feature extraction; dynamic channel weight adjustment is carried out; multi-scale information is fused through splicing or element-by-element addition, and detail and semantic expressions are balanced; and high-precision and high-efficiency feature fusion operation is realized in a bearing surface defect detection task.

The present application discloses a method and device for allocating channel weights for monitoring the status of port machinery equipment. The method includes: multiple sensors transmit vibration signal data via Modbus and CAN protocols, and perform FFT transformation on the data to generate an energy spectrum distribution; based on the distribution, wavelet packet decomposition is used to determine the main characteristic frequency band, and its association with the neural network channel is established; based on this association, a multi-layer perceptron is used to perform nonlinear mapping of the frequency band energy, and a dynamic channel weight is generated via a Sigmoid function; the weight is combined with the time series features extracted by Bi‑LSTM for channel weighting, and equipment fault diagnosis is achieved based on the weighted fusion features. The present application extracts key characteristic frequency bands through FFT transformation and wavelet packet decomposition, establishes its association with the neural network channel, and uses a multi-layer perceptron to generate dynamic weights to optimize feature fusion, making equipment fault diagnosis more accurate and improving monitoring efficiency and accuracy.

The application discloses a kind of dynamic channel-based numerical control system M code control method and related equipment, it is related to numerical control machine tool and industrial automation control technical field.Numerical control equipment first extracts M code set from current program segment, then according to preset channel distribution strategy, multiple M codes in M code set are distributed to different preset logical channels.M code is delivered according to channel distribution table, avoid the complex decoding logic problem caused by writing order uncertainty in traditional multiway control, PLC program only needs to process specific range of M code for fixed channel, reduce the complexity and memory occupation of ladder diagram program.Effectively solve the technical problems that the PLC program complexity is high in the prior art due to the coupling of M code channel allocation and writing order depth, while improving parallel processing efficiency, ensure the reliability of program execution, and then realize the scalability and maintainability of numerical control system by improving the parallel processing capacity of numerical control system.

Instead of having more lanes than the bus can support, the number of lanes can match the number the bus can support. In a system with six lanes and two ports, this means each port has two permanent lanes and two lanes shared with the other port. When changing the configuration from four lanes on the first port to four lanes on the second port, the shared lanes are put into a low-power state from the first port's perspective and into a fully operational state from the second port's. Thus, each port assumes there are four lanes available for it, resulting in a total of eight lanes for the device. In reality, however, there are only six lanes available for the device, thereby saving costs and meeting bandwidth and power limitations.

The present application belongs to the technical field of speech conversion, and in particular relates to a singing speech conversion method based on CBAM and dynamic convolution decomposition, which comprises a training stage and a conversion stage, and a model network comprises a generator, a discriminator and a style encoder. First, dynamic convolution decomposition is introduced in the generator, dynamic channel fusion is used to replace the dynamic attention of the channel group, the problem that dynamic convolution in the generator causes the number of convolution weights to increase by K times is solved, the difficulty of joint optimization is reduced, and the model requires fewer parameters without sacrificing accuracy, thereby improving the running performance of the algorithm. Further, CBAM attention modules are introduced in the encoding network and the decoding network of the generator, attention is applied to the channels and the space, the attention and capture of the detailed information in the spectrum are improved, and the quality of the converted singing speech is significantly improved.

This invention discloses a secure authentication method and system based on a general server interface, belonging to the field of interface security control technology. The method involves acquiring the original request message, extracting the interface identifier, timestamp, request parameters, and a first signature value; generating an interface context feature code based on the interface identifier and an interface mapping table, containing call hierarchy and permission domain information; and comparing this with a second signature value generated using a pre-shared key to determine message legitimacy. After confirming message legitimacy, a dynamic channel identifier bound to the interface context feature code is generated and written into the call context stack. The consistency between the dynamic channel identifier and the interface path vector is verified using a Merkle path binding verification algorithm, controlling the execution of the target business module. This invention achieves strong binding between interface message authentication and the server's internal call path, effectively preventing cross-module replay and path mismatch risks, and improving the security and reliability of general interfaces in complex operating environments.

The application relates to the technical field of communication signal processing, and provides a parallel receiving and processing device and method for multi-system signals based on dynamic configuration, wherein through the cooperative design of a data receiving module, a channel scheduling module, a tracking despreading module and a decoding module, a limited number of general signal processing channels are designed into a dynamic channel resource pool, dynamic allocation and recovery according to decoding results are supported, the decoupling of channels and signal systems is realized, and the resource utilization rate is greatly improved. A hardware scheduling module is realized in an FPGA, fast scheduling and parameter configuration are realized. The scheduling strategy is flexibly configured, the priority allocation of channel resources of key signals is guaranteed, each general channel is designed in a parameterized mode, different system signals can be seamlessly switched, FPGA logic does not need to be reconfigured, parallel tracking and despreading are realized, signal receiving and processing with multiple systems, multiple channels, high concurrency and low resource occupation can be realized in a single machine, and the system reliability is improved.

The application discloses a dynamic channel attention-based lightweight agricultural pest detection method and system, and particularly relates to the field of agricultural video target detection, and is used for solving the problem that channel weights are difficult to dynamically adjust with interframe changes in pest identification. The scale jump judgment of texture and contour features is used, continuous coefficients generated by a Gaussian process are taken as weight split gates, channel mapping and asymmetric transition are synchronously responded to interframe texture burst and form breaking, weight real-time rearrangement is realized, noise channels are rapidly inhibited through exponential contraction, high-confidence channels are stabilized through linear expansion, and it is ensured that the lightweight backbone maintains focusing on the smallest pests in power-limited occasions; the prior weight of a front frame is injected into subsequent mapping, the detection head always receives the latest effective features, and stable inference is maintained under light, shielding and parallax variation; meanwhile, delay caused by redundant splitting is avoided, deep convolution storage is saved, and original frame rate output is maintained.

The present application relates to the technical field of data fusion, in particular to a multi-source electrocardio signal correction method and system based on adaptive fusion, comprising the following steps: constructing a multi-channel input tensor, extracting local features by using a weight calculation network, adaptively weighting and fusing and reducing dimensions of the multi-channel signals, establishing a nonlinear mapping relationship by using a deep reconstruction network, reconstructing a standard waveform, and iteratively optimizing network parameters based on reconstruction error. In the present application, local neighborhood features of multi-channel signals are extracted by constructing a weight calculation network, a dynamic channel weight sequence reflecting real-time contribution of signal sources is constructed, amplitude intensity is adaptively adjusted according to signal quality, unstable channel noise interference is effectively suppressed and high-quality signal components are enhanced, a nonlinear feature transformation of the fused sequence is performed by using a deep reconstruction network, an accurate mapping from non-standard input to standard lead waveform is established, and weight distribution and signal reconstruction parameter optimization are realized in combination with an error back propagation mechanism.

The invention discloses a multi-element echelon soft and hard combined unmanned aerial vehicle countering system and method, and relates to the technical field of unmanned aerial vehicle defense. The method comprises the following steps: identifying a behavior intention and an anti-interference level of an unmanned aerial vehicle through information fusion of a multi-source sensor; according to an identification result, generating a plurality of candidate countering strategy chains containing different soft and hard means combinations and time sequences in parallel; executing a first step means of the selected chain, and synchronously calculating an actual interference efficiency index; comparing the actual efficiency with an expected target to generate an efficiency deviation signal, and dynamically deciding and adjusting subsequent parameters or switching a strategy chain through a finite-state machine based on the signal; through a dynamic channel guiding mechanism, the unmanned aerial vehicle is restrained in a dynamically generated expected interception channel by a soft interference means, and hard interception is triggered according to a predicted trajectory in the channel. According to the method, the problems of rigid response mode and insufficient means collaboration in the prior art are solved, and integrated countering from sensing, decision making to self-adaptive processing is realized.

The invention discloses a shared variable dependency chain breakage detection method and device, and relates to the technical field of multiprocessor programming, the method comprises the steps that two detection channels including a static disassembling channel and a dynamic PMU channel are coupled, the static disassembling channel uses the principle that related disassembling codes can change when a shared variable dependency chain breaks, and the dynamic PMU channel is used for detecting whether the shared variable dependency chain breaks or not; the dynamic PMU channel constructs a shared variable dependency chain fracture detection architecture by utilizing the principle that the number of related PMU events in the program running process changes when the shared variable dependency chain fractures, and the problems that in the related technology, the performance overhead of a dynamic analysis method is large, and the automation degree of a static analysis method is limited are solved; besides, the technical problems that the coverage scene of a formalized verification method is limited, and complex competition conditions of the production environment are difficult to reproduce are solved, and the technical effects that the detection accuracy and efficiency are effectively improved, the dynamic detection overhead is reduced, and the competition conditions in the real production environment can be covered are achieved.

This invention relates to a multi-channel abnormal flow detection method for power systems based on an adversarial autoencoder. By employing protocol-application dual-dimensional fine-grained dynamic channel partitioning, three-stage progressive collaborative training within dedicated channels, and a memory-enhanced reconstruction mechanism with Top-k sparse constraints, it effectively addresses core shortcomings in existing technologies, such as feature interference caused by coarse-grained channel partitioning, model instability due to missing three-stage training, and abnormal smooth reconstruction caused by over-generalization of ordinary autoencoders. The solution achieves accurate modeling under business semantic isolation, explicit anchoring of normal patterns, and high-sensitivity amplification of subtle deviations. While maintaining an extremely low false alarm rate, it significantly improves the detection capability for high-risk, low-frequency disguised anomalies. Furthermore, through an online channel adaptive evolution mechanism, it endows the system with the ability to autonomously adapt to dynamic changes in power grid business over the long term, meeting the stringent requirements of power systems for high reliability and real-time performance.